diff options
Diffstat (limited to 'src/gallium/auxiliary')
110 files changed, 3915 insertions, 2897 deletions
diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile index 53a0847f03..ff355c4783 100644 --- a/src/gallium/auxiliary/Makefile +++ b/src/gallium/auxiliary/Makefile @@ -128,12 +128,12 @@ C_SOURCES = \ util/u_linkage.c \ util/u_network.c \ util/u_math.c \ - util/u_mempool.c \ util/u_mm.c \ util/u_rect.c \ util/u_ringbuffer.c \ util/u_sampler.c \ util/u_simple_shaders.c \ + util/u_slab.c \ util/u_snprintf.c \ util/u_staging.c \ util/u_surface.c \ @@ -185,7 +185,7 @@ GALLIVM_SOURCES = \ draw/draw_pt_fetch_shade_pipeline_llvm.c GALLIVM_CPP_SOURCES = \ - gallivm/lp_bld_misc.cpp + gallivm/lp_bld_misc.cpp GENERATED_SOURCES = \ indices/u_indices_gen.c \ @@ -203,9 +203,6 @@ CPP_SOURCES += \ endif -LIBRARY_DEFINES += -D__STDC_CONSTANT_MACROS - - include ../Makefile.template diff --git a/src/gallium/auxiliary/SConscript b/src/gallium/auxiliary/SConscript index 75c27dd242..fca7e5fd11 100644 --- a/src/gallium/auxiliary/SConscript +++ b/src/gallium/auxiliary/SConscript @@ -175,13 +175,13 @@ source = [ 'util/u_linkage.c', 'util/u_network.c', 'util/u_math.c', - 'util/u_mempool.c', 'util/u_mm.c', 'util/u_rect.c', 'util/u_resource.c', 'util/u_ringbuffer.c', 'util/u_sampler.c', 'util/u_simple_shaders.c', + 'util/u_slab.c', 'util/u_snprintf.c', 'util/u_staging.c', 'util/u_surface.c', diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c index 39d82f3289..e045313b94 100644 --- a/src/gallium/auxiliary/draw/draw_context.c +++ b/src/gallium/auxiliary/draw/draw_context.c @@ -35,6 +35,7 @@ #include "util/u_memory.h" #include "util/u_math.h" #include "util/u_cpu_detect.h" +#include "util/u_inlines.h" #include "draw_context.h" #include "draw_vs.h" #include "draw_gs.h" @@ -63,19 +64,32 @@ draw_get_option_use_llvm(void) } #endif -struct draw_context *draw_create( struct pipe_context *pipe ) + + +/** + * Create new draw module context. + */ +struct draw_context * +draw_create(struct pipe_context *pipe) +{ + return draw_create_gallivm(pipe, NULL); +} + + + +/** + * Create new draw module context with gallivm state for LLVM JIT. + */ +struct draw_context * +draw_create_gallivm(struct pipe_context *pipe, struct gallivm_state *gallivm) { struct draw_context *draw = CALLOC_STRUCT( draw_context ); if (draw == NULL) goto fail; #if HAVE_LLVM - if(draw_get_option_use_llvm()) - { - lp_build_init(); - assert(lp_build_engine); - draw->engine = lp_build_engine; - draw->llvm = draw_llvm_create(draw); + if (draw_get_option_use_llvm() && gallivm) { + draw->llvm = draw_llvm_create(draw, gallivm); } #endif @@ -91,6 +105,8 @@ fail: return NULL; } + + boolean draw_init(struct draw_context *draw) { /* @@ -149,6 +165,10 @@ void draw_destroy( struct draw_context *draw ) } } + for (i = 0; i < draw->pt.nr_vertex_buffers; i++) { + pipe_resource_reference(&draw->pt.vertex_buffer[i].buffer, NULL); + } + /* Not so fast -- we're just borrowing this at the moment. * if (draw->render) @@ -292,8 +312,9 @@ draw_set_vertex_buffers(struct draw_context *draw, { assert(count <= PIPE_MAX_ATTRIBS); - memcpy(draw->pt.vertex_buffer, buffers, count * sizeof(buffers[0])); - draw->pt.nr_vertex_buffers = count; + util_copy_vertex_buffers(draw->pt.vertex_buffer, + &draw->pt.nr_vertex_buffers, + buffers, count); } diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h index ff4f753604..a0b217e4d3 100644 --- a/src/gallium/auxiliary/draw/draw_context.h +++ b/src/gallium/auxiliary/draw/draw_context.h @@ -48,10 +48,15 @@ struct draw_vertex_shader; struct draw_geometry_shader; struct draw_fragment_shader; struct tgsi_sampler; +struct gallivm_state; + struct draw_context *draw_create( struct pipe_context *pipe ); +struct draw_context * +draw_create_gallivm(struct pipe_context *pipe, struct gallivm_state *gallivm); + void draw_destroy( struct draw_context *draw ); void draw_flush(struct draw_context *draw); diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c index eb162fb0f6..41269ee869 100644 --- a/src/gallium/auxiliary/draw/draw_llvm.c +++ b/src/gallium/auxiliary/draw/draw_llvm.c @@ -42,6 +42,7 @@ #include "gallivm/lp_bld_printf.h" #include "gallivm/lp_bld_intr.h" #include "gallivm/lp_bld_init.h" +#include "gallivm/lp_bld_type.h" #include "tgsi/tgsi_exec.h" #include "tgsi/tgsi_dump.h" @@ -49,259 +50,352 @@ #include "util/u_math.h" #include "util/u_pointer.h" #include "util/u_string.h" +#include "util/u_simple_list.h" -#include <llvm-c/Transforms/Scalar.h> #define DEBUG_STORE 0 -/* generates the draw jit function */ + +/** + * This function is called by the gallivm "garbage collector" when + * the LLVM global data structures are freed. We must free all LLVM-related + * data. Specifically, all JIT'd shader variants. + */ +static void +draw_llvm_garbage_collect_callback(void *cb_data) +{ + struct draw_llvm *llvm = (struct draw_llvm *) cb_data; + struct draw_llvm_variant_list_item *li; + + /* free all shader variants */ + li = first_elem(&llvm->vs_variants_list); + while (!at_end(&llvm->vs_variants_list, li)) { + struct draw_llvm_variant_list_item *next = next_elem(li); + draw_llvm_destroy_variant(li->base); + li = next; + } + + /* Null-out these pointers so they get remade next time they're needed. + * See the accessor functions below. + */ + llvm->context_ptr_type = NULL; + llvm->buffer_ptr_type = NULL; + llvm->vb_ptr_type = NULL; + llvm->vertex_header_ptr_type = NULL; +} + + static void draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *var); + static void draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *var); -static void -init_globals(struct draw_llvm *llvm) + +/** + * Create LLVM type for struct draw_jit_texture + */ +static LLVMTypeRef +create_jit_texture_type(struct gallivm_state *gallivm) { + LLVMTargetDataRef target = gallivm->target; LLVMTypeRef texture_type; + LLVMTypeRef elem_types[DRAW_JIT_TEXTURE_NUM_FIELDS]; + LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context); + + elem_types[DRAW_JIT_TEXTURE_WIDTH] = + elem_types[DRAW_JIT_TEXTURE_HEIGHT] = + elem_types[DRAW_JIT_TEXTURE_DEPTH] = + elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = int32_type; + elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] = + elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] = + LLVMArrayType(int32_type, PIPE_MAX_TEXTURE_LEVELS); + elem_types[DRAW_JIT_TEXTURE_DATA] = + LLVMArrayType(LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0), + PIPE_MAX_TEXTURE_LEVELS); + elem_types[DRAW_JIT_TEXTURE_MIN_LOD] = + elem_types[DRAW_JIT_TEXTURE_MAX_LOD] = + elem_types[DRAW_JIT_TEXTURE_LOD_BIAS] = LLVMFloatTypeInContext(gallivm->context); + elem_types[DRAW_JIT_TEXTURE_BORDER_COLOR] = + LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4); + + texture_type = LLVMStructTypeInContext(gallivm->context, elem_types, + Elements(elem_types), 0); + + /* Make sure the target's struct layout cache doesn't return + * stale/invalid data. + */ + LLVMInvalidateStructLayout(gallivm->target, texture_type); + + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, width, + target, texture_type, + DRAW_JIT_TEXTURE_WIDTH); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, height, + target, texture_type, + DRAW_JIT_TEXTURE_HEIGHT); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, depth, + target, texture_type, + DRAW_JIT_TEXTURE_DEPTH); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, last_level, + target, texture_type, + DRAW_JIT_TEXTURE_LAST_LEVEL); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, row_stride, + target, texture_type, + DRAW_JIT_TEXTURE_ROW_STRIDE); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, img_stride, + target, texture_type, + DRAW_JIT_TEXTURE_IMG_STRIDE); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, data, + target, texture_type, + DRAW_JIT_TEXTURE_DATA); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, min_lod, + target, texture_type, + DRAW_JIT_TEXTURE_MIN_LOD); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, max_lod, + target, texture_type, + DRAW_JIT_TEXTURE_MAX_LOD); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, lod_bias, + target, texture_type, + DRAW_JIT_TEXTURE_LOD_BIAS); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, border_color, + target, texture_type, + DRAW_JIT_TEXTURE_BORDER_COLOR); + + LP_CHECK_STRUCT_SIZE(struct draw_jit_texture, target, texture_type); + + return texture_type; +} - /* struct draw_jit_texture */ - { - LLVMTypeRef elem_types[DRAW_JIT_TEXTURE_NUM_FIELDS]; - - elem_types[DRAW_JIT_TEXTURE_WIDTH] = LLVMInt32Type(); - elem_types[DRAW_JIT_TEXTURE_HEIGHT] = LLVMInt32Type(); - elem_types[DRAW_JIT_TEXTURE_DEPTH] = LLVMInt32Type(); - elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type(); - elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] = - LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS); - elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] = - LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS); - elem_types[DRAW_JIT_TEXTURE_DATA] = - LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0), - PIPE_MAX_TEXTURE_LEVELS); - elem_types[DRAW_JIT_TEXTURE_MIN_LOD] = LLVMFloatType(); - elem_types[DRAW_JIT_TEXTURE_MAX_LOD] = LLVMFloatType(); - elem_types[DRAW_JIT_TEXTURE_LOD_BIAS] = LLVMFloatType(); - elem_types[DRAW_JIT_TEXTURE_BORDER_COLOR] = - LLVMArrayType(LLVMFloatType(), 4); - - texture_type = LLVMStructType(elem_types, Elements(elem_types), 0); - - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, width, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_WIDTH); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, height, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_HEIGHT); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, depth, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_DEPTH); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, last_level, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_LAST_LEVEL); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, row_stride, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_ROW_STRIDE); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, img_stride, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_IMG_STRIDE); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, data, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_DATA); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, min_lod, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_MIN_LOD); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, max_lod, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_MAX_LOD); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, lod_bias, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_LOD_BIAS); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, border_color, - llvm->target, texture_type, - DRAW_JIT_TEXTURE_BORDER_COLOR); - LP_CHECK_STRUCT_SIZE(struct draw_jit_texture, - llvm->target, texture_type); - - LLVMAddTypeName(llvm->module, "texture", texture_type); - } +/** + * Create LLVM type for struct draw_jit_texture + */ +static LLVMTypeRef +create_jit_context_type(struct gallivm_state *gallivm, + LLVMTypeRef texture_type) +{ + LLVMTargetDataRef target = gallivm->target; + LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context); + LLVMTypeRef elem_types[5]; + LLVMTypeRef context_type; + + elem_types[0] = LLVMPointerType(float_type, 0); /* vs_constants */ + elem_types[1] = LLVMPointerType(float_type, 0); /* gs_constants */ + elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4), 12), 0); /* planes */ + elem_types[3] = LLVMPointerType(float_type, 0); /* viewport */ + elem_types[4] = LLVMArrayType(texture_type, + PIPE_MAX_VERTEX_SAMPLERS); /* textures */ + + context_type = LLVMStructTypeInContext(gallivm->context, elem_types, + Elements(elem_types), 0); + + LLVMInvalidateStructLayout(gallivm->target, context_type); + + LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, vs_constants, + target, context_type, 0); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, gs_constants, + target, context_type, 1); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes, + target, context_type, 2); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures, + target, context_type, + DRAW_JIT_CTX_TEXTURES); + LP_CHECK_STRUCT_SIZE(struct draw_jit_context, + target, context_type); + + return context_type; +} - /* struct draw_jit_context */ - { - LLVMTypeRef elem_types[5]; - LLVMTypeRef context_type; - - elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */ - elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* gs_constants */ - elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(LLVMFloatType(), 4), 12), 0); /* planes */ - elem_types[3] = LLVMPointerType(LLVMFloatType(), 0); /* viewport */ - elem_types[4] = LLVMArrayType(texture_type, - PIPE_MAX_VERTEX_SAMPLERS); /* textures */ - - context_type = LLVMStructType(elem_types, Elements(elem_types), 0); - - LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, vs_constants, - llvm->target, context_type, 0); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, gs_constants, - llvm->target, context_type, 1); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes, - llvm->target, context_type, 2); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures, - llvm->target, context_type, - DRAW_JIT_CTX_TEXTURES); - LP_CHECK_STRUCT_SIZE(struct draw_jit_context, - llvm->target, context_type); - - LLVMAddTypeName(llvm->module, "draw_jit_context", context_type); - - llvm->context_ptr_type = LLVMPointerType(context_type, 0); - } - { - LLVMTypeRef buffer_ptr = LLVMPointerType(LLVMIntType(8), 0); - llvm->buffer_ptr_type = LLVMPointerType(buffer_ptr, 0); - } - /* struct pipe_vertex_buffer */ - { - LLVMTypeRef elem_types[4]; - LLVMTypeRef vb_type; - elem_types[0] = LLVMInt32Type(); - elem_types[1] = LLVMInt32Type(); - elem_types[2] = LLVMInt32Type(); - elem_types[3] = LLVMPointerType(LLVMOpaqueType(), 0); /* vs_constants */ +/** + * Create LLVM type for struct pipe_vertex_buffer + */ +static LLVMTypeRef +create_jit_vertex_buffer_type(struct gallivm_state *gallivm) +{ + LLVMTargetDataRef target = gallivm->target; + LLVMTypeRef elem_types[4]; + LLVMTypeRef vb_type; - vb_type = LLVMStructType(elem_types, Elements(elem_types), 0); + elem_types[0] = + elem_types[1] = + elem_types[2] = LLVMInt32TypeInContext(gallivm->context); + elem_types[3] = LLVMPointerType(LLVMOpaqueTypeInContext(gallivm->context), 0); /* vs_constants */ - LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, stride, - llvm->target, vb_type, 0); - LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, buffer_offset, - llvm->target, vb_type, 2); - LP_CHECK_STRUCT_SIZE(struct pipe_vertex_buffer, - llvm->target, vb_type); + vb_type = LLVMStructTypeInContext(gallivm->context, elem_types, + Elements(elem_types), 0); - LLVMAddTypeName(llvm->module, "pipe_vertex_buffer", vb_type); + LLVMInvalidateStructLayout(gallivm->target, vb_type); - llvm->vb_ptr_type = LLVMPointerType(vb_type, 0); - } + LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, stride, + target, vb_type, 0); + LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, buffer_offset, + target, vb_type, 2); + + LP_CHECK_STRUCT_SIZE(struct pipe_vertex_buffer, target, vb_type); + + return vb_type; } + +/** + * Create LLVM type for struct vertex_header; + */ static LLVMTypeRef -create_vertex_header(struct draw_llvm *llvm, int data_elems) +create_jit_vertex_header(struct gallivm_state *gallivm, int data_elems) { - /* struct vertex_header */ + LLVMTargetDataRef target = gallivm->target; LLVMTypeRef elem_types[3]; LLVMTypeRef vertex_header; char struct_name[24]; util_snprintf(struct_name, 23, "vertex_header%d", data_elems); - elem_types[0] = LLVMIntType(32); - elem_types[1] = LLVMArrayType(LLVMFloatType(), 4); + elem_types[0] = LLVMIntTypeInContext(gallivm->context, 32); + elem_types[1] = LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4); elem_types[2] = LLVMArrayType(elem_types[1], data_elems); - vertex_header = LLVMStructType(elem_types, Elements(elem_types), 0); + vertex_header = LLVMStructTypeInContext(gallivm->context, elem_types, + Elements(elem_types), 0); + + LLVMInvalidateStructLayout(gallivm->target, vertex_header); /* these are bit-fields and we can't take address of them LP_CHECK_MEMBER_OFFSET(struct vertex_header, clipmask, - llvm->target, vertex_header, + target, vertex_header, DRAW_JIT_VERTEX_CLIPMASK); LP_CHECK_MEMBER_OFFSET(struct vertex_header, edgeflag, - llvm->target, vertex_header, + target, vertex_header, DRAW_JIT_VERTEX_EDGEFLAG); LP_CHECK_MEMBER_OFFSET(struct vertex_header, pad, - llvm->target, vertex_header, + target, vertex_header, DRAW_JIT_VERTEX_PAD); LP_CHECK_MEMBER_OFFSET(struct vertex_header, vertex_id, - llvm->target, vertex_header, + target, vertex_header, DRAW_JIT_VERTEX_VERTEX_ID); */ LP_CHECK_MEMBER_OFFSET(struct vertex_header, clip, - llvm->target, vertex_header, + target, vertex_header, DRAW_JIT_VERTEX_CLIP); LP_CHECK_MEMBER_OFFSET(struct vertex_header, data, - llvm->target, vertex_header, + target, vertex_header, DRAW_JIT_VERTEX_DATA); - LLVMAddTypeName(llvm->module, struct_name, vertex_header); + LLVMAddTypeName(gallivm->module, struct_name, vertex_header); - return LLVMPointerType(vertex_header, 0); + return vertex_header; } -struct draw_llvm * -draw_llvm_create(struct draw_context *draw) + +/** + * Create LLVM types for various structures. + */ +static void +create_jit_types(struct draw_llvm *llvm) { - struct draw_llvm *llvm; + struct gallivm_state *gallivm = llvm->gallivm; + LLVMTypeRef texture_type, context_type, buffer_type, vb_type; - llvm = CALLOC_STRUCT( draw_llvm ); - if (!llvm) - return NULL; + texture_type = create_jit_texture_type(gallivm); + LLVMAddTypeName(gallivm->module, "texture", texture_type); - llvm->draw = draw; - llvm->engine = draw->engine; + context_type = create_jit_context_type(gallivm, texture_type); + LLVMAddTypeName(gallivm->module, "draw_jit_context", context_type); + llvm->context_ptr_type = LLVMPointerType(context_type, 0); - debug_assert(llvm->engine); + buffer_type = LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 8), 0); + LLVMAddTypeName(gallivm->module, "buffer", buffer_type); + llvm->buffer_ptr_type = LLVMPointerType(buffer_type, 0); - llvm->module = LLVMModuleCreateWithName("draw_llvm"); - llvm->provider = LLVMCreateModuleProviderForExistingModule(llvm->module); + vb_type = create_jit_vertex_buffer_type(gallivm); + LLVMAddTypeName(gallivm->module, "pipe_vertex_buffer", vb_type); + llvm->vb_ptr_type = LLVMPointerType(vb_type, 0); +} - LLVMAddModuleProvider(llvm->engine, llvm->provider); - llvm->target = LLVMGetExecutionEngineTargetData(llvm->engine); +static LLVMTypeRef +get_context_ptr_type(struct draw_llvm *llvm) +{ + if (!llvm->context_ptr_type) + create_jit_types(llvm); + return llvm->context_ptr_type; +} - llvm->pass = LLVMCreateFunctionPassManager(llvm->provider); - LLVMAddTargetData(llvm->target, llvm->pass); - if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) { - /* These are the passes currently listed in llvm-c/Transforms/Scalar.h, - * but there are more on SVN. */ - /* TODO: Add more passes */ +static LLVMTypeRef +get_buffer_ptr_type(struct draw_llvm *llvm) +{ + if (!llvm->buffer_ptr_type) + create_jit_types(llvm); + return llvm->buffer_ptr_type; +} - LLVMAddCFGSimplificationPass(llvm->pass); - if (HAVE_LLVM >= 0x207 && sizeof(void*) == 4) { - /* For LLVM >= 2.7 and 32-bit build, use this order of passes to - * avoid generating bad code. - * Test with piglit glsl-vs-sqrt-zero test. - */ - LLVMAddConstantPropagationPass(llvm->pass); - LLVMAddPromoteMemoryToRegisterPass(llvm->pass); - } - else { - LLVMAddPromoteMemoryToRegisterPass(llvm->pass); - LLVMAddConstantPropagationPass(llvm->pass); - } +static LLVMTypeRef +get_vb_ptr_type(struct draw_llvm *llvm) +{ + if (!llvm->vb_ptr_type) + create_jit_types(llvm); + return llvm->vb_ptr_type; +} - LLVMAddInstructionCombiningPass(llvm->pass); - LLVMAddGVNPass(llvm->pass); - } else { - /* We need at least this pass to prevent the backends to fail in - * unexpected ways. - */ - LLVMAddPromoteMemoryToRegisterPass(llvm->pass); - } +static LLVMTypeRef +get_vertex_header_ptr_type(struct draw_llvm *llvm) +{ + if (!llvm->vertex_header_ptr_type) + create_jit_types(llvm); + return llvm->vertex_header_ptr_type; +} - init_globals(llvm); + +/** + * Create per-context LLVM info. + */ +struct draw_llvm * +draw_llvm_create(struct draw_context *draw, struct gallivm_state *gallivm) +{ + struct draw_llvm *llvm; + + llvm = CALLOC_STRUCT( draw_llvm ); + if (!llvm) + return NULL; + + lp_build_init(); + + llvm->draw = draw; + llvm->gallivm = gallivm; if (gallivm_debug & GALLIVM_DEBUG_IR) { - LLVMDumpModule(llvm->module); + LLVMDumpModule(llvm->gallivm->module); } llvm->nr_variants = 0; make_empty_list(&llvm->vs_variants_list); + gallivm_register_garbage_collector_callback( + draw_llvm_garbage_collect_callback, llvm); + return llvm; } + +/** + * Free per-context LLVM info. + */ void draw_llvm_destroy(struct draw_llvm *llvm) { - LLVMDisposePassManager(llvm->pass); + gallivm_remove_garbage_collector_callback( + draw_llvm_garbage_collect_callback, llvm); + /* XXX free other draw_llvm data? */ FREE(llvm); } + +/** + * Create LLVM-generated code for a vertex shader. + */ struct draw_llvm_variant * draw_llvm_create_variant(struct draw_llvm *llvm, unsigned num_inputs, @@ -310,6 +404,7 @@ draw_llvm_create_variant(struct draw_llvm *llvm, struct draw_llvm_variant *variant; struct llvm_vertex_shader *shader = llvm_vertex_shader(llvm->draw->vs.vertex_shader); + LLVMTypeRef vertex_header; variant = MALLOC(sizeof *variant + shader->variant_key_size - @@ -321,7 +416,9 @@ draw_llvm_create_variant(struct draw_llvm *llvm, memcpy(&variant->key, key, shader->variant_key_size); - llvm->vertex_header_ptr_type = create_vertex_header(llvm, num_inputs); + vertex_header = create_jit_vertex_header(llvm->gallivm, num_inputs); + + llvm->vertex_header_ptr_type = LLVMPointerType(vertex_header, 0); draw_llvm_generate(llvm, variant); draw_llvm_generate_elts(llvm, variant); @@ -346,7 +443,7 @@ generate_vs(struct draw_llvm *llvm, { const struct tgsi_token *tokens = llvm->draw->vs.vertex_shader->state.tokens; struct lp_type vs_type; - LLVMValueRef consts_ptr = draw_jit_context_vs_constants(builder, context_ptr); + LLVMValueRef consts_ptr = draw_jit_context_vs_constants(llvm->gallivm, context_ptr); struct lp_build_sampler_soa *sampler = 0; memset(&vs_type, 0, sizeof vs_type); @@ -367,7 +464,7 @@ generate_vs(struct draw_llvm *llvm, llvm->draw->num_samplers) sampler = draw_sampler; - lp_build_tgsi_soa(builder, + lp_build_tgsi_soa(llvm->gallivm, tokens, vs_type, NULL /*struct lp_build_mask_context *mask*/, @@ -386,20 +483,20 @@ static void print_vectorf(LLVMBuilderRef builder, { LLVMValueRef val[4]; val[0] = LLVMBuildExtractElement(builder, vec, - LLVMConstInt(LLVMInt32Type(), 0, 0), ""); + lp_build_const_int32(gallivm, 0), ""); val[1] = LLVMBuildExtractElement(builder, vec, - LLVMConstInt(LLVMInt32Type(), 1, 0), ""); + lp_build_const_int32(gallivm, 1), ""); val[2] = LLVMBuildExtractElement(builder, vec, - LLVMConstInt(LLVMInt32Type(), 2, 0), ""); + lp_build_const_int32(gallivm, 2), ""); val[3] = LLVMBuildExtractElement(builder, vec, - LLVMConstInt(LLVMInt32Type(), 3, 0), ""); + lp_build_const_int32(gallivm, 3), ""); lp_build_printf(builder, "vector = [%f, %f, %f, %f]\n", val[0], val[1], val[2], val[3]); } #endif static void -generate_fetch(LLVMBuilderRef builder, +generate_fetch(struct gallivm_state *gallivm, LLVMValueRef vbuffers_ptr, LLVMValueRef *res, struct pipe_vertex_element *velem, @@ -407,19 +504,22 @@ generate_fetch(LLVMBuilderRef builder, LLVMValueRef index, LLVMValueRef instance_id) { - LLVMValueRef indices = LLVMConstInt(LLVMInt64Type(), velem->vertex_buffer_index, 0); + LLVMBuilderRef builder = gallivm->builder; + LLVMValueRef indices = + LLVMConstInt(LLVMInt64TypeInContext(gallivm->context), + velem->vertex_buffer_index, 0); LLVMValueRef vbuffer_ptr = LLVMBuildGEP(builder, vbuffers_ptr, &indices, 1, ""); - LLVMValueRef vb_stride = draw_jit_vbuffer_stride(builder, vbuf); - LLVMValueRef vb_max_index = draw_jit_vbuffer_max_index(builder, vbuf); - LLVMValueRef vb_buffer_offset = draw_jit_vbuffer_offset(builder, vbuf); + LLVMValueRef vb_stride = draw_jit_vbuffer_stride(gallivm, vbuf); + LLVMValueRef vb_max_index = draw_jit_vbuffer_max_index(gallivm, vbuf); + LLVMValueRef vb_buffer_offset = draw_jit_vbuffer_offset(gallivm, vbuf); LLVMValueRef cond; LLVMValueRef stride; if (velem->instance_divisor) { /* array index = instance_id / instance_divisor */ index = LLVMBuildUDiv(builder, instance_id, - LLVMConstInt(LLVMInt32Type(), velem->instance_divisor, 0), + lp_build_const_int32(gallivm, velem->instance_divisor), "instance_divisor"); } @@ -435,23 +535,24 @@ generate_fetch(LLVMBuilderRef builder, vb_buffer_offset, ""); stride = LLVMBuildAdd(builder, stride, - LLVMConstInt(LLVMInt32Type(), velem->src_offset, 0), + lp_build_const_int32(gallivm, velem->src_offset), ""); /*lp_build_printf(builder, "vbuf index = %d, stride is %d\n", indices, stride);*/ vbuffer_ptr = LLVMBuildGEP(builder, vbuffer_ptr, &stride, 1, ""); - *res = draw_llvm_translate_from(builder, vbuffer_ptr, velem->src_format); + *res = draw_llvm_translate_from(gallivm, vbuffer_ptr, velem->src_format); } static LLVMValueRef -aos_to_soa(LLVMBuilderRef builder, +aos_to_soa(struct gallivm_state *gallivm, LLVMValueRef val0, LLVMValueRef val1, LLVMValueRef val2, LLVMValueRef val3, LLVMValueRef channel) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef ex, res; ex = LLVMBuildExtractElement(builder, val0, @@ -459,38 +560,39 @@ aos_to_soa(LLVMBuilderRef builder, res = LLVMBuildInsertElement(builder, LLVMConstNull(LLVMTypeOf(val0)), ex, - LLVMConstInt(LLVMInt32Type(), 0, 0), + lp_build_const_int32(gallivm, 0), ""); ex = LLVMBuildExtractElement(builder, val1, channel, ""); res = LLVMBuildInsertElement(builder, res, ex, - LLVMConstInt(LLVMInt32Type(), 1, 0), + lp_build_const_int32(gallivm, 1), ""); ex = LLVMBuildExtractElement(builder, val2, channel, ""); res = LLVMBuildInsertElement(builder, res, ex, - LLVMConstInt(LLVMInt32Type(), 2, 0), + lp_build_const_int32(gallivm, 2), ""); ex = LLVMBuildExtractElement(builder, val3, channel, ""); res = LLVMBuildInsertElement(builder, res, ex, - LLVMConstInt(LLVMInt32Type(), 3, 0), + lp_build_const_int32(gallivm, 3), ""); return res; } static void -soa_to_aos(LLVMBuilderRef builder, +soa_to_aos(struct gallivm_state *gallivm, LLVMValueRef soa[NUM_CHANNELS], LLVMValueRef aos[NUM_CHANNELS]) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef comp; int i = 0; @@ -500,29 +602,29 @@ soa_to_aos(LLVMBuilderRef builder, aos[1] = aos[2] = aos[3] = aos[0]; for (i = 0; i < NUM_CHANNELS; ++i) { - LLVMValueRef channel = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef channel = lp_build_const_int32(gallivm, i); comp = LLVMBuildExtractElement(builder, soa[i], - LLVMConstInt(LLVMInt32Type(), 0, 0), ""); + lp_build_const_int32(gallivm, 0), ""); aos[0] = LLVMBuildInsertElement(builder, aos[0], comp, channel, ""); comp = LLVMBuildExtractElement(builder, soa[i], - LLVMConstInt(LLVMInt32Type(), 1, 0), ""); + lp_build_const_int32(gallivm, 1), ""); aos[1] = LLVMBuildInsertElement(builder, aos[1], comp, channel, ""); comp = LLVMBuildExtractElement(builder, soa[i], - LLVMConstInt(LLVMInt32Type(), 2, 0), ""); + lp_build_const_int32(gallivm, 2), ""); aos[2] = LLVMBuildInsertElement(builder, aos[2], comp, channel, ""); comp = LLVMBuildExtractElement(builder, soa[i], - LLVMConstInt(LLVMInt32Type(), 3, 0), ""); + lp_build_const_int32(gallivm, 3), ""); aos[3] = LLVMBuildInsertElement(builder, aos[3], comp, channel, ""); } } static void -convert_to_soa(LLVMBuilderRef builder, +convert_to_soa(struct gallivm_state *gallivm, LLVMValueRef (*aos)[NUM_CHANNELS], LLVMValueRef (*soa)[NUM_CHANNELS], int num_attribs) @@ -537,36 +639,37 @@ convert_to_soa(LLVMBuilderRef builder, LLVMValueRef val2 = aos[i][2]; LLVMValueRef val3 = aos[i][3]; - soa[i][0] = aos_to_soa(builder, val0, val1, val2, val3, - LLVMConstInt(LLVMInt32Type(), 0, 0)); - soa[i][1] = aos_to_soa(builder, val0, val1, val2, val3, - LLVMConstInt(LLVMInt32Type(), 1, 0)); - soa[i][2] = aos_to_soa(builder, val0, val1, val2, val3, - LLVMConstInt(LLVMInt32Type(), 2, 0)); - soa[i][3] = aos_to_soa(builder, val0, val1, val2, val3, - LLVMConstInt(LLVMInt32Type(), 3, 0)); + soa[i][0] = aos_to_soa(gallivm, val0, val1, val2, val3, + lp_build_const_int32(gallivm, 0)); + soa[i][1] = aos_to_soa(gallivm, val0, val1, val2, val3, + lp_build_const_int32(gallivm, 1)); + soa[i][2] = aos_to_soa(gallivm, val0, val1, val2, val3, + lp_build_const_int32(gallivm, 2)); + soa[i][3] = aos_to_soa(gallivm, val0, val1, val2, val3, + lp_build_const_int32(gallivm, 3)); } } static void -store_aos(LLVMBuilderRef builder, +store_aos(struct gallivm_state *gallivm, LLVMValueRef io_ptr, LLVMValueRef index, LLVMValueRef value, LLVMValueRef clipmask) { - LLVMValueRef id_ptr = draw_jit_header_id(builder, io_ptr); - LLVMValueRef data_ptr = draw_jit_header_data(builder, io_ptr); + LLVMBuilderRef builder = gallivm->builder; + LLVMValueRef id_ptr = draw_jit_header_id(gallivm, io_ptr); + LLVMValueRef data_ptr = draw_jit_header_data(gallivm, io_ptr); LLVMValueRef indices[3]; LLVMValueRef val, shift; - indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); + indices[0] = lp_build_const_int32(gallivm, 0); indices[1] = index; - indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0); + indices[2] = lp_build_const_int32(gallivm, 0); /* initialize vertex id:16 = 0xffff, pad:3 = 0, edgeflag:1 = 1 */ - val = LLVMConstInt(LLVMInt32Type(), 0xffff1, 0); - shift = LLVMConstInt(LLVMInt32Type(), 12, 0); + val = lp_build_const_int32(gallivm, 0xffff1); + shift = lp_build_const_int32(gallivm, 12); val = LLVMBuildShl(builder, val, shift, ""); /* add clipmask:12 */ val = LLVMBuildOr(builder, val, clipmask, ""); @@ -582,7 +685,7 @@ store_aos(LLVMBuilderRef builder, /*lp_build_printf(builder, " ---- %p storing at %d (%p) ", io_ptr, index, data_ptr); print_vectorf(builder, value);*/ data_ptr = LLVMBuildBitCast(builder, data_ptr, - LLVMPointerType(LLVMArrayType(LLVMVectorType(LLVMFloatType(), 4), 0), 0), + LLVMPointerType(LLVMArrayType(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4), 0), 0), "datavec"); data_ptr = LLVMBuildGEP(builder, data_ptr, indices, 2, ""); @@ -594,10 +697,10 @@ store_aos(LLVMBuilderRef builder, LLVMValueRef gep0, gep1, gep2, gep3; data_ptr = LLVMBuildGEP(builder, data_ptr, indices, 3, ""); - idx0 = LLVMConstInt(LLVMInt32Type(), 0, 0); - idx1 = LLVMConstInt(LLVMInt32Type(), 1, 0); - idx2 = LLVMConstInt(LLVMInt32Type(), 2, 0); - idx3 = LLVMConstInt(LLVMInt32Type(), 3, 0); + idx0 = lp_build_const_int32(gallivm, 0); + idx1 = lp_build_const_int32(gallivm, 1); + idx2 = lp_build_const_int32(gallivm, 2); + idx3 = lp_build_const_int32(gallivm, 3); x = LLVMBuildExtractElement(builder, value, idx0, ""); @@ -624,18 +727,19 @@ store_aos(LLVMBuilderRef builder, } static void -store_aos_array(LLVMBuilderRef builder, +store_aos_array(struct gallivm_state *gallivm, LLVMValueRef io_ptr, LLVMValueRef aos[NUM_CHANNELS], int attrib, int num_outputs, LLVMValueRef clipmask) { - LLVMValueRef attr_index = LLVMConstInt(LLVMInt32Type(), attrib, 0); - LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0); - LLVMValueRef ind1 = LLVMConstInt(LLVMInt32Type(), 1, 0); - LLVMValueRef ind2 = LLVMConstInt(LLVMInt32Type(), 2, 0); - LLVMValueRef ind3 = LLVMConstInt(LLVMInt32Type(), 3, 0); + LLVMBuilderRef builder = gallivm->builder; + LLVMValueRef attr_index = lp_build_const_int32(gallivm, attrib); + LLVMValueRef ind0 = lp_build_const_int32(gallivm, 0); + LLVMValueRef ind1 = lp_build_const_int32(gallivm, 1); + LLVMValueRef ind2 = lp_build_const_int32(gallivm, 2); + LLVMValueRef ind3 = lp_build_const_int32(gallivm, 3); LLVMValueRef io0_ptr, io1_ptr, io2_ptr, io3_ptr; LLVMValueRef clipmask0, clipmask1, clipmask2, clipmask3; @@ -664,20 +768,21 @@ store_aos_array(LLVMBuilderRef builder, io_ptr, ind0, ind1, ind2, ind3, clipmask0, clipmask1, clipmask2, clipmask3); #endif /* store for each of the 4 vertices */ - store_aos(builder, io0_ptr, attr_index, aos[0], clipmask0); - store_aos(builder, io1_ptr, attr_index, aos[1], clipmask1); - store_aos(builder, io2_ptr, attr_index, aos[2], clipmask2); - store_aos(builder, io3_ptr, attr_index, aos[3], clipmask3); + store_aos(gallivm, io0_ptr, attr_index, aos[0], clipmask0); + store_aos(gallivm, io1_ptr, attr_index, aos[1], clipmask1); + store_aos(gallivm, io2_ptr, attr_index, aos[2], clipmask2); + store_aos(gallivm, io3_ptr, attr_index, aos[3], clipmask3); } static void -convert_to_aos(LLVMBuilderRef builder, +convert_to_aos(struct gallivm_state *gallivm, LLVMValueRef io, LLVMValueRef (*outputs)[NUM_CHANNELS], LLVMValueRef clipmask, int num_outputs, int max_vertices) { + LLVMBuilderRef builder = gallivm->builder; unsigned chan, attrib; #if DEBUG_STORE @@ -698,8 +803,8 @@ convert_to_aos(LLVMBuilderRef builder, } else soa[chan] = 0; } - soa_to_aos(builder, soa, aos); - store_aos_array(builder, + soa_to_aos(gallivm, soa, aos); + store_aos_array(gallivm, io, aos, attrib, @@ -717,10 +822,11 @@ convert_to_aos(LLVMBuilderRef builder, * rather than extracting each element one by one. */ static void -store_clip(LLVMBuilderRef builder, +store_clip(struct gallivm_state *gallivm, LLVMValueRef io_ptr, LLVMValueRef (*outputs)[NUM_CHANNELS]) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef out[4]; LLVMValueRef indices[2]; LLVMValueRef io0_ptr, io1_ptr, io2_ptr, io3_ptr; @@ -729,13 +835,13 @@ store_clip(LLVMBuilderRef builder, LLVMValueRef out0elem, out1elem, out2elem, out3elem; int i; - LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0); - LLVMValueRef ind1 = LLVMConstInt(LLVMInt32Type(), 1, 0); - LLVMValueRef ind2 = LLVMConstInt(LLVMInt32Type(), 2, 0); - LLVMValueRef ind3 = LLVMConstInt(LLVMInt32Type(), 3, 0); + LLVMValueRef ind0 = lp_build_const_int32(gallivm, 0); + LLVMValueRef ind1 = lp_build_const_int32(gallivm, 1); + LLVMValueRef ind2 = lp_build_const_int32(gallivm, 2); + LLVMValueRef ind3 = lp_build_const_int32(gallivm, 3); - indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); - indices[1] = LLVMConstInt(LLVMInt32Type(), 0, 0); + indices[0] = + indices[1] = lp_build_const_int32(gallivm, 0); out[0] = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/ out[1] = LLVMBuildLoad(builder, outputs[0][1], ""); /*y0 y1 y2 y3*/ @@ -747,29 +853,21 @@ store_clip(LLVMBuilderRef builder, io2_ptr = LLVMBuildGEP(builder, io_ptr, &ind2, 1, ""); io3_ptr = LLVMBuildGEP(builder, io_ptr, &ind3, 1, ""); - clip_ptr0 = draw_jit_header_clip(builder, io0_ptr); - clip_ptr1 = draw_jit_header_clip(builder, io1_ptr); - clip_ptr2 = draw_jit_header_clip(builder, io2_ptr); - clip_ptr3 = draw_jit_header_clip(builder, io3_ptr); + clip_ptr0 = draw_jit_header_clip(gallivm, io0_ptr); + clip_ptr1 = draw_jit_header_clip(gallivm, io1_ptr); + clip_ptr2 = draw_jit_header_clip(gallivm, io2_ptr); + clip_ptr3 = draw_jit_header_clip(gallivm, io3_ptr); for (i = 0; i<4; i++){ - clip0_ptr = LLVMBuildGEP(builder, clip_ptr0, - indices, 2, ""); //x0 - clip1_ptr = LLVMBuildGEP(builder, clip_ptr1, - indices, 2, ""); //x1 - clip2_ptr = LLVMBuildGEP(builder, clip_ptr2, - indices, 2, ""); //x2 - clip3_ptr = LLVMBuildGEP(builder, clip_ptr3, - indices, 2, ""); //x3 - - out0elem = LLVMBuildExtractElement(builder, out[i], - ind0, ""); //x0 - out1elem = LLVMBuildExtractElement(builder, out[i], - ind1, ""); //x1 - out2elem = LLVMBuildExtractElement(builder, out[i], - ind2, ""); //x2 - out3elem = LLVMBuildExtractElement(builder, out[i], - ind3, ""); //x3 + clip0_ptr = LLVMBuildGEP(builder, clip_ptr0, indices, 2, ""); /* x0 */ + clip1_ptr = LLVMBuildGEP(builder, clip_ptr1, indices, 2, ""); /* x1 */ + clip2_ptr = LLVMBuildGEP(builder, clip_ptr2, indices, 2, ""); /* x2 */ + clip3_ptr = LLVMBuildGEP(builder, clip_ptr3, indices, 2, ""); /* x3 */ + + out0elem = LLVMBuildExtractElement(builder, out[i], ind0, ""); /* x0 */ + out1elem = LLVMBuildExtractElement(builder, out[i], ind1, ""); /* x1 */ + out2elem = LLVMBuildExtractElement(builder, out[i], ind2, ""); /* x2 */ + out3elem = LLVMBuildExtractElement(builder, out[i], ind3, ""); /* x3 */ LLVMBuildStore(builder, out0elem, clip0_ptr); LLVMBuildStore(builder, out1elem, clip1_ptr); @@ -783,16 +881,19 @@ store_clip(LLVMBuilderRef builder, /* Equivalent of _mm_set1_ps(a) */ -static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld, - LLVMValueRef a, - const char *name) +static LLVMValueRef +vec4f_from_scalar(struct gallivm_state *gallivm, + LLVMValueRef a, + const char *name) { - LLVMValueRef res = LLVMGetUndef(LLVMVectorType(LLVMFloatType(), 4)); + LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context); + LLVMValueRef res = LLVMGetUndef(LLVMVectorType(float_type, 4)); int i; for(i = 0; i < 4; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); - res = LLVMBuildInsertElement(bld, res, a, index, i == 3 ? name : ""); + LLVMValueRef index = lp_build_const_int32(gallivm, i); + res = LLVMBuildInsertElement(gallivm->builder, res, a, + index, i == 3 ? name : ""); } return res; @@ -808,10 +909,11 @@ generate_viewport(struct draw_llvm *llvm, LLVMValueRef context_ptr) { int i; + struct gallivm_state *gallivm = llvm->gallivm; struct lp_type f32_type = lp_type_float_vec(32); LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/ - LLVMValueRef const1 = lp_build_const_vec(f32_type, 1.0); /*1.0 1.0 1.0 1.0*/ - LLVMValueRef vp_ptr = draw_jit_context_viewport(builder, context_ptr); + LLVMValueRef const1 = lp_build_const_vec(gallivm, f32_type, 1.0); /*1.0 1.0 1.0 1.0*/ + LLVMValueRef vp_ptr = draw_jit_context_viewport(gallivm, context_ptr); /* for 1/w convention*/ out3 = LLVMBuildFDiv(builder, const1, out3, ""); @@ -826,14 +928,14 @@ generate_viewport(struct draw_llvm *llvm, LLVMValueRef trans_i; LLVMValueRef index; - index = LLVMConstInt(LLVMInt32Type(), i, 0); + index = lp_build_const_int32(gallivm, i); scale_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, ""); - index = LLVMConstInt(LLVMInt32Type(), i+4, 0); + index = lp_build_const_int32(gallivm, i+4); trans_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, ""); - scale = vec4f_from_scalar(builder, LLVMBuildLoad(builder, scale_i, ""), "scale"); - trans = vec4f_from_scalar(builder, LLVMBuildLoad(builder, trans_i, ""), "trans"); + scale = vec4f_from_scalar(gallivm, LLVMBuildLoad(builder, scale_i, ""), "scale"); + trans = vec4f_from_scalar(gallivm, LLVMBuildLoad(builder, trans_i, ""), "trans"); /* divide by w */ out = LLVMBuildFMul(builder, out, out3, ""); @@ -853,7 +955,7 @@ generate_viewport(struct draw_llvm *llvm, * Returns clipmask as 4xi32 bitmask for the 4 vertices */ static LLVMValueRef -generate_clipmask(LLVMBuilderRef builder, +generate_clipmask(struct gallivm_state *gallivm, LLVMValueRef (*outputs)[NUM_CHANNELS], boolean clip_xy, boolean clip_z, @@ -862,6 +964,7 @@ generate_clipmask(LLVMBuilderRef builder, unsigned nr, LLVMValueRef context_ptr) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef mask; /* stores the <4xi32> clipmasks */ LLVMValueRef test, temp; LLVMValueRef zero, shift; @@ -872,10 +975,10 @@ generate_clipmask(LLVMBuilderRef builder, struct lp_type f32_type = lp_type_float_vec(32); - mask = lp_build_const_int_vec(lp_type_int_vec(32), 0); - temp = lp_build_const_int_vec(lp_type_int_vec(32), 0); - zero = lp_build_const_vec(f32_type, 0); /* 0.0f 0.0f 0.0f 0.0f */ - shift = lp_build_const_int_vec(lp_type_int_vec(32), 1); /* 1 1 1 1 */ + mask = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 0); + temp = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 0); + zero = lp_build_const_vec(gallivm, f32_type, 0); /* 0.0f 0.0f 0.0f 0.0f */ + shift = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 1); /* 1 1 1 1 */ /* Assuming position stored at output[0] */ pos_x = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/ @@ -886,92 +989,92 @@ generate_clipmask(LLVMBuilderRef builder, /* Cliptest, for hardwired planes */ if (clip_xy){ /* plane 1 */ - test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w); + test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w); temp = shift; test = LLVMBuildAnd(builder, test, temp, ""); mask = test; /* plane 2 */ test = LLVMBuildFAdd(builder, pos_x, pos_w, ""); - test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test); + test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test); temp = LLVMBuildShl(builder, temp, shift, ""); test = LLVMBuildAnd(builder, test, temp, ""); mask = LLVMBuildOr(builder, mask, test, ""); /* plane 3 */ - test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w); + test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w); temp = LLVMBuildShl(builder, temp, shift, ""); test = LLVMBuildAnd(builder, test, temp, ""); mask = LLVMBuildOr(builder, mask, test, ""); /* plane 4 */ test = LLVMBuildFAdd(builder, pos_y, pos_w, ""); - test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test); + test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test); temp = LLVMBuildShl(builder, temp, shift, ""); test = LLVMBuildAnd(builder, test, temp, ""); mask = LLVMBuildOr(builder, mask, test, ""); } if (clip_z){ - temp = lp_build_const_int_vec(lp_type_int_vec(32), 16); + temp = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 16); if (clip_halfz){ /* plane 5 */ - test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, pos_z); + test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, pos_z); test = LLVMBuildAnd(builder, test, temp, ""); mask = LLVMBuildOr(builder, mask, test, ""); } else{ /* plane 5 */ test = LLVMBuildFAdd(builder, pos_z, pos_w, ""); - test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test); + test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test); test = LLVMBuildAnd(builder, test, temp, ""); mask = LLVMBuildOr(builder, mask, test, ""); } /* plane 6 */ - test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_z, pos_w); + test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_z, pos_w); temp = LLVMBuildShl(builder, temp, shift, ""); test = LLVMBuildAnd(builder, test, temp, ""); mask = LLVMBuildOr(builder, mask, test, ""); } if (clip_user){ - LLVMValueRef planes_ptr = draw_jit_context_planes(builder, context_ptr); + LLVMValueRef planes_ptr = draw_jit_context_planes(gallivm, context_ptr); LLVMValueRef indices[3]; - temp = lp_build_const_int_vec(lp_type_int_vec(32), 32); + temp = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 32); /* userclip planes */ for (i = 6; i < nr; i++) { - indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); - indices[1] = LLVMConstInt(LLVMInt32Type(), i, 0); + indices[0] = lp_build_const_int32(gallivm, 0); + indices[1] = lp_build_const_int32(gallivm, i); - indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0); + indices[2] = lp_build_const_int32(gallivm, 0); plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, ""); plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_x"); - planes = vec4f_from_scalar(builder, plane1, "plane4_x"); + planes = vec4f_from_scalar(gallivm, plane1, "plane4_x"); sum = LLVMBuildFMul(builder, planes, pos_x, ""); - indices[2] = LLVMConstInt(LLVMInt32Type(), 1, 0); + indices[2] = lp_build_const_int32(gallivm, 1); plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, ""); plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_y"); - planes = vec4f_from_scalar(builder, plane1, "plane4_y"); + planes = vec4f_from_scalar(gallivm, plane1, "plane4_y"); test = LLVMBuildFMul(builder, planes, pos_y, ""); sum = LLVMBuildFAdd(builder, sum, test, ""); - indices[2] = LLVMConstInt(LLVMInt32Type(), 2, 0); + indices[2] = lp_build_const_int32(gallivm, 2); plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, ""); plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_z"); - planes = vec4f_from_scalar(builder, plane1, "plane4_z"); + planes = vec4f_from_scalar(gallivm, plane1, "plane4_z"); test = LLVMBuildFMul(builder, planes, pos_z, ""); sum = LLVMBuildFAdd(builder, sum, test, ""); - indices[2] = LLVMConstInt(LLVMInt32Type(), 3, 0); + indices[2] = lp_build_const_int32(gallivm, 3); plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, ""); plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_w"); - planes = vec4f_from_scalar(builder, plane1, "plane4_w"); + planes = vec4f_from_scalar(gallivm, plane1, "plane4_w"); test = LLVMBuildFMul(builder, planes, pos_w, ""); sum = LLVMBuildFAdd(builder, sum, test, ""); - test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, sum); + test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, sum); temp = LLVMBuildShl(builder, temp, shift, ""); test = LLVMBuildAnd(builder, test, temp, ""); mask = LLVMBuildOr(builder, mask, test, ""); @@ -985,17 +1088,18 @@ generate_clipmask(LLVMBuilderRef builder, * Used zero/non-zero i32 value to represent boolean */ static void -clipmask_bool(LLVMBuilderRef builder, +clipmask_bool(struct gallivm_state *gallivm, LLVMValueRef clipmask, LLVMValueRef ret_ptr) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef ret = LLVMBuildLoad(builder, ret_ptr, ""); LLVMValueRef temp; int i; for (i=0; i<4; i++){ temp = LLVMBuildExtractElement(builder, clipmask, - LLVMConstInt(LLVMInt32Type(), i, 0) , ""); + lp_build_const_int32(gallivm, i) , ""); ret = LLVMBuildOr(builder, ret, temp, ""); } @@ -1005,6 +1109,9 @@ clipmask_bool(LLVMBuilderRef builder, static void draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) { + struct gallivm_state *gallivm = llvm->gallivm; + LLVMContextRef context = gallivm->context; + LLVMTypeRef int32_type = LLVMInt32TypeInContext(context); LLVMTypeRef arg_types[8]; LLVMTypeRef func_type; LLVMValueRef context_ptr; @@ -1029,18 +1136,19 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) variant->key.clip_z || variant->key.clip_user; - arg_types[0] = llvm->context_ptr_type; /* context */ - arg_types[1] = llvm->vertex_header_ptr_type; /* vertex_header */ - arg_types[2] = llvm->buffer_ptr_type; /* vbuffers */ - arg_types[3] = LLVMInt32Type(); /* start */ - arg_types[4] = LLVMInt32Type(); /* count */ - arg_types[5] = LLVMInt32Type(); /* stride */ - arg_types[6] = llvm->vb_ptr_type; /* pipe_vertex_buffer's */ - arg_types[7] = LLVMInt32Type(); /* instance_id */ - - func_type = LLVMFunctionType(LLVMInt32Type(), arg_types, Elements(arg_types), 0); - - variant->function = LLVMAddFunction(llvm->module, "draw_llvm_shader", func_type); + arg_types[0] = get_context_ptr_type(llvm); /* context */ + arg_types[1] = get_vertex_header_ptr_type(llvm); /* vertex_header */ + arg_types[2] = get_buffer_ptr_type(llvm); /* vbuffers */ + arg_types[3] = int32_type; /* start */ + arg_types[4] = int32_type; /* count */ + arg_types[5] = int32_type; /* stride */ + arg_types[6] = get_vb_ptr_type(llvm); /* pipe_vertex_buffer's */ + arg_types[7] = int32_type; /* instance_id */ + + func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0); + + variant->function = LLVMAddFunction(gallivm->module, "draw_llvm_shader", + func_type); LLVMSetFunctionCallConv(variant->function, LLVMCCallConv); for(i = 0; i < Elements(arg_types); ++i) if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind) @@ -1068,22 +1176,23 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) * Function body */ - block = LLVMAppendBasicBlock(variant->function, "entry"); - builder = LLVMCreateBuilder(); + block = LLVMAppendBasicBlockInContext(gallivm->context, variant->function, "entry"); + builder = gallivm->builder; + assert(builder); LLVMPositionBuilderAtEnd(builder, block); - lp_build_context_init(&bld, builder, lp_type_int(32)); + lp_build_context_init(&bld, llvm->gallivm, lp_type_int(32)); - system_values_array = lp_build_system_values_array(builder, vs_info, + system_values_array = lp_build_system_values_array(gallivm, vs_info, instance_id, NULL); end = lp_build_add(&bld, start, count); - step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0); + step = lp_build_const_int32(gallivm, max_vertices); /* function will return non-zero i32 value if any clipped vertices */ - ret_ptr = lp_build_alloca(builder, LLVMInt32Type(), ""); - LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), ret_ptr); + ret_ptr = lp_build_alloca(gallivm, int32_type, ""); + LLVMBuildStore(builder, lp_build_const_int32(gallivm, 0), ret_ptr); /* code generated texture sampling */ sampler = draw_llvm_sampler_soa_create( @@ -1094,7 +1203,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) lp_build_printf(builder, "start = %d, end = %d, step = %d\n", start, end, step); #endif - lp_build_loop_begin(builder, start, &lp_loop); + lp_build_loop_begin(&lp_loop, llvm->gallivm, start); { LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } }; @@ -1112,20 +1221,18 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) LLVMValueRef true_index = LLVMBuildAdd( builder, lp_loop.counter, - LLVMConstInt(LLVMInt32Type(), i, 0), ""); + lp_build_const_int32(gallivm, i), ""); for (j = 0; j < draw->pt.nr_vertex_elements; ++j) { struct pipe_vertex_element *velem = &draw->pt.vertex_element[j]; - LLVMValueRef vb_index = LLVMConstInt(LLVMInt32Type(), - velem->vertex_buffer_index, - 0); + LLVMValueRef vb_index = lp_build_const_int32(gallivm, velem->vertex_buffer_index); LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr, &vb_index, 1, ""); - generate_fetch(builder, vbuffers_ptr, + generate_fetch(llvm->gallivm, vbuffers_ptr, &aos_attribs[j][i], velem, vb, true_index, instance_id); } } - convert_to_soa(builder, aos_attribs, inputs, + convert_to_soa(gallivm, aos_attribs, inputs, draw->pt.nr_vertex_elements); ptr_aos = (const LLVMValueRef (*)[NUM_CHANNELS]) inputs; @@ -1138,12 +1245,12 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) sampler); /* store original positions in clip before further manipulation */ - store_clip(builder, io, outputs); + store_clip(gallivm, io, outputs); /* do cliptest */ if (enable_cliptest){ /* allocate clipmask, assign it integer type */ - clipmask = generate_clipmask(builder, outputs, + clipmask = generate_clipmask(gallivm, outputs, variant->key.clip_xy, variant->key.clip_z, variant->key.clip_user, @@ -1151,10 +1258,10 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) variant->key.nr_planes, context_ptr); /* return clipping boolean value for function */ - clipmask_bool(builder, clipmask, ret_ptr); + clipmask_bool(gallivm, clipmask, ret_ptr); } else{ - clipmask = lp_build_const_int_vec(lp_type_int_vec(32), 0); + clipmask = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 0); } /* do viewport mapping */ @@ -1163,19 +1270,17 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) } /* store clipmask in vertex header and positions in data */ - convert_to_aos(builder, io, outputs, clipmask, + convert_to_aos(gallivm, io, outputs, clipmask, vs_info->num_outputs, max_vertices); } - lp_build_loop_end_cond(builder, end, step, LLVMIntUGE, &lp_loop); + lp_build_loop_end_cond(&lp_loop, end, step, LLVMIntUGE); sampler->destroy(sampler); ret = LLVMBuildLoad(builder, ret_ptr,""); LLVMBuildRet(builder, ret); - LLVMDisposeBuilder(builder); - /* * Translate the LLVM IR into machine code. */ @@ -1186,14 +1291,14 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) } #endif - LLVMRunFunctionPassManager(llvm->pass, variant->function); + LLVMRunFunctionPassManager(gallivm->passmgr, variant->function); if (gallivm_debug & GALLIVM_DEBUG_IR) { lp_debug_dump_value(variant->function); debug_printf("\n"); } - code = LLVMGetPointerToGlobal(llvm->draw->engine, variant->function); + code = LLVMGetPointerToGlobal(gallivm->engine, variant->function); variant->jit_func = (draw_jit_vert_func)pointer_to_func(code); if (gallivm_debug & GALLIVM_DEBUG_ASM) { @@ -1206,6 +1311,9 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) static void draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *variant) { + struct gallivm_state *gallivm = llvm->gallivm; + LLVMContextRef context = gallivm->context; + LLVMTypeRef int32_type = LLVMInt32TypeInContext(context); LLVMTypeRef arg_types[8]; LLVMTypeRef func_type; LLVMValueRef context_ptr; @@ -1231,18 +1339,18 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian variant->key.clip_z || variant->key.clip_user; - arg_types[0] = llvm->context_ptr_type; /* context */ - arg_types[1] = llvm->vertex_header_ptr_type; /* vertex_header */ - arg_types[2] = llvm->buffer_ptr_type; /* vbuffers */ - arg_types[3] = LLVMPointerType(LLVMInt32Type(), 0); /* fetch_elts * */ - arg_types[4] = LLVMInt32Type(); /* fetch_count */ - arg_types[5] = LLVMInt32Type(); /* stride */ - arg_types[6] = llvm->vb_ptr_type; /* pipe_vertex_buffer's */ - arg_types[7] = LLVMInt32Type(); /* instance_id */ - - func_type = LLVMFunctionType(LLVMInt32Type(), arg_types, Elements(arg_types), 0); - - variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts", func_type); + arg_types[0] = get_context_ptr_type(llvm); /* context */ + arg_types[1] = get_vertex_header_ptr_type(llvm); /* vertex_header */ + arg_types[2] = get_buffer_ptr_type(llvm); /* vbuffers */ + arg_types[3] = LLVMPointerType(int32_type, 0); /* fetch_elts * */ + arg_types[4] = int32_type; /* fetch_count */ + arg_types[5] = int32_type; /* stride */ + arg_types[6] = get_vb_ptr_type(llvm); /* pipe_vertex_buffer's */ + arg_types[7] = int32_type; /* instance_id */ + + func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0); + + variant->function_elts = LLVMAddFunction(gallivm->module, "draw_llvm_shader_elts", func_type); LLVMSetFunctionCallConv(variant->function_elts, LLVMCCallConv); for(i = 0; i < Elements(arg_types); ++i) if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind) @@ -1271,17 +1379,17 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian * Function body */ - block = LLVMAppendBasicBlock(variant->function_elts, "entry"); - builder = LLVMCreateBuilder(); + block = LLVMAppendBasicBlockInContext(gallivm->context, variant->function_elts, "entry"); + builder = gallivm->builder; LLVMPositionBuilderAtEnd(builder, block); - lp_build_context_init(&bld, builder, lp_type_int(32)); + lp_build_context_init(&bld, gallivm, lp_type_int(32)); - system_values_array = lp_build_system_values_array(builder, vs_info, + system_values_array = lp_build_system_values_array(gallivm, vs_info, instance_id, NULL); - step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0); + step = lp_build_const_int32(gallivm, max_vertices); /* code generated texture sampling */ sampler = draw_llvm_sampler_soa_create( @@ -1289,14 +1397,14 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian context_ptr); fetch_max = LLVMBuildSub(builder, fetch_count, - LLVMConstInt(LLVMInt32Type(), 1, 0), + lp_build_const_int32(gallivm, 1), "fetch_max"); /* function returns non-zero i32 value if any clipped vertices */ - ret_ptr = lp_build_alloca(builder, LLVMInt32Type(), ""); - LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), ret_ptr); + ret_ptr = lp_build_alloca(gallivm, int32_type, ""); + LLVMBuildStore(builder, lp_build_const_int32(gallivm, 0), ret_ptr); - lp_build_loop_begin(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), &lp_loop); + lp_build_loop_begin(&lp_loop, gallivm, lp_build_const_int32(gallivm, 0)); { LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]; LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } }; @@ -1314,7 +1422,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian LLVMValueRef true_index = LLVMBuildAdd( builder, lp_loop.counter, - LLVMConstInt(LLVMInt32Type(), i, 0), ""); + lp_build_const_int32(gallivm, i), ""); LLVMValueRef fetch_ptr; /* make sure we're not out of bounds which can happen @@ -1327,17 +1435,15 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian true_index = LLVMBuildLoad(builder, fetch_ptr, "fetch_elt"); for (j = 0; j < draw->pt.nr_vertex_elements; ++j) { struct pipe_vertex_element *velem = &draw->pt.vertex_element[j]; - LLVMValueRef vb_index = LLVMConstInt(LLVMInt32Type(), - velem->vertex_buffer_index, - 0); + LLVMValueRef vb_index = lp_build_const_int32(gallivm, velem->vertex_buffer_index); LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr, &vb_index, 1, ""); - generate_fetch(builder, vbuffers_ptr, + generate_fetch(gallivm, vbuffers_ptr, &aos_attribs[j][i], velem, vb, true_index, instance_id); } } - convert_to_soa(builder, aos_attribs, inputs, + convert_to_soa(gallivm, aos_attribs, inputs, draw->pt.nr_vertex_elements); ptr_aos = (const LLVMValueRef (*)[NUM_CHANNELS]) inputs; @@ -1350,12 +1456,12 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian sampler); /* store original positions in clip before further manipulation */ - store_clip(builder, io, outputs); + store_clip(gallivm, io, outputs); /* do cliptest */ if (enable_cliptest){ /* allocate clipmask, assign it integer type */ - clipmask = generate_clipmask(builder, outputs, + clipmask = generate_clipmask(gallivm, outputs, variant->key.clip_xy, variant->key.clip_z, variant->key.clip_user, @@ -1363,10 +1469,10 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian variant->key.nr_planes, context_ptr); /* return clipping boolean value for function */ - clipmask_bool(builder, clipmask, ret_ptr); + clipmask_bool(gallivm, clipmask, ret_ptr); } else{ - clipmask = lp_build_const_int_vec(lp_type_int_vec(32), 0); + clipmask = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 0); } /* do viewport mapping */ @@ -1378,19 +1484,17 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian * original positions in clip * and transformed positions in data */ - convert_to_aos(builder, io, outputs, clipmask, + convert_to_aos(gallivm, io, outputs, clipmask, vs_info->num_outputs, max_vertices); } - lp_build_loop_end_cond(builder, fetch_count, step, LLVMIntUGE, &lp_loop); + lp_build_loop_end_cond(&lp_loop, fetch_count, step, LLVMIntUGE); sampler->destroy(sampler); ret = LLVMBuildLoad(builder, ret_ptr,""); LLVMBuildRet(builder, ret); - LLVMDisposeBuilder(builder); - /* * Translate the LLVM IR into machine code. */ @@ -1401,14 +1505,14 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian } #endif - LLVMRunFunctionPassManager(llvm->pass, variant->function_elts); + LLVMRunFunctionPassManager(gallivm->passmgr, variant->function_elts); if (gallivm_debug & GALLIVM_DEBUG_IR) { lp_debug_dump_value(variant->function_elts); debug_printf("\n"); } - code = LLVMGetPointerToGlobal(llvm->draw->engine, variant->function_elts); + code = LLVMGetPointerToGlobal(gallivm->engine, variant->function_elts); variant->jit_func_elts = (draw_jit_vert_func_elts)pointer_to_func(code); if (gallivm_debug & GALLIVM_DEBUG_ASM) { @@ -1517,19 +1621,16 @@ void draw_llvm_destroy_variant(struct draw_llvm_variant *variant) { struct draw_llvm *llvm = variant->llvm; - struct draw_context *draw = llvm->draw; if (variant->function_elts) { - if (variant->function_elts) - LLVMFreeMachineCodeForFunction(draw->engine, - variant->function_elts); + LLVMFreeMachineCodeForFunction(llvm->gallivm->engine, + variant->function_elts); LLVMDeleteFunction(variant->function_elts); } if (variant->function) { - if (variant->function) - LLVMFreeMachineCodeForFunction(draw->engine, - variant->function); + LLVMFreeMachineCodeForFunction(llvm->gallivm->engine, + variant->function); LLVMDeleteFunction(variant->function); } diff --git a/src/gallium/auxiliary/draw/draw_llvm.h b/src/gallium/auxiliary/draw/draw_llvm.h index c3c30c07c6..9f038f1f04 100644 --- a/src/gallium/auxiliary/draw/draw_llvm.h +++ b/src/gallium/auxiliary/draw/draw_llvm.h @@ -103,41 +103,41 @@ struct draw_jit_context }; -#define draw_jit_context_vs_constants(_builder, _ptr) \ - lp_build_struct_get(_builder, _ptr, 0, "vs_constants") +#define draw_jit_context_vs_constants(_gallivm, _ptr) \ + lp_build_struct_get(_gallivm, _ptr, 0, "vs_constants") -#define draw_jit_context_gs_constants(_builder, _ptr) \ - lp_build_struct_get(_builder, _ptr, 1, "gs_constants") +#define draw_jit_context_gs_constants(_gallivm, _ptr) \ + lp_build_struct_get(_gallivm, _ptr, 1, "gs_constants") -#define draw_jit_context_planes(_builder, _ptr) \ - lp_build_struct_get(_builder, _ptr, 2, "planes") +#define draw_jit_context_planes(_gallivm, _ptr) \ + lp_build_struct_get(_gallivm, _ptr, 2, "planes") -#define draw_jit_context_viewport(_builder, _ptr) \ - lp_build_struct_get(_builder, _ptr, 3, "viewport") +#define draw_jit_context_viewport(_gallivm, _ptr) \ + lp_build_struct_get(_gallivm, _ptr, 3, "viewport") #define DRAW_JIT_CTX_TEXTURES 4 -#define draw_jit_context_textures(_builder, _ptr) \ - lp_build_struct_get_ptr(_builder, _ptr, DRAW_JIT_CTX_TEXTURES, "textures") +#define draw_jit_context_textures(_gallivm, _ptr) \ + lp_build_struct_get_ptr(_gallivm, _ptr, DRAW_JIT_CTX_TEXTURES, "textures") -#define draw_jit_header_id(_builder, _ptr) \ - lp_build_struct_get_ptr(_builder, _ptr, 0, "id") +#define draw_jit_header_id(_gallivm, _ptr) \ + lp_build_struct_get_ptr(_gallivm, _ptr, 0, "id") -#define draw_jit_header_clip(_builder, _ptr) \ - lp_build_struct_get_ptr(_builder, _ptr, 1, "clip") +#define draw_jit_header_clip(_gallivm, _ptr) \ + lp_build_struct_get_ptr(_gallivm, _ptr, 1, "clip") -#define draw_jit_header_data(_builder, _ptr) \ - lp_build_struct_get_ptr(_builder, _ptr, 2, "data") +#define draw_jit_header_data(_gallivm, _ptr) \ + lp_build_struct_get_ptr(_gallivm, _ptr, 2, "data") -#define draw_jit_vbuffer_stride(_builder, _ptr) \ - lp_build_struct_get(_builder, _ptr, 0, "stride") +#define draw_jit_vbuffer_stride(_gallivm, _ptr) \ + lp_build_struct_get(_gallivm, _ptr, 0, "stride") -#define draw_jit_vbuffer_max_index(_builder, _ptr) \ - lp_build_struct_get(_builder, _ptr, 1, "max_index") +#define draw_jit_vbuffer_max_index(_gallivm, _ptr) \ + lp_build_struct_get(_gallivm, _ptr, 1, "max_index") -#define draw_jit_vbuffer_offset(_builder, _ptr) \ - lp_build_struct_get(_builder, _ptr, 2, "buffer_offset") +#define draw_jit_vbuffer_offset(_gallivm, _ptr) \ + lp_build_struct_get(_gallivm, _ptr, 2, "buffer_offset") typedef int @@ -229,7 +229,6 @@ struct draw_llvm_variant /* key is variable-sized, must be last */ struct draw_llvm_variant_key key; - /* key is variable-sized, must be last */ }; struct llvm_vertex_shader { @@ -246,21 +245,19 @@ struct draw_llvm { struct draw_jit_context jit_context; + struct gallivm_state *gallivm; + struct draw_llvm_variant_list_item vs_variants_list; int nr_variants; - LLVMModuleRef module; - LLVMExecutionEngineRef engine; - LLVMModuleProviderRef provider; - LLVMTargetDataRef target; - LLVMPassManagerRef pass; - + /* LLVM JIT builder types */ LLVMTypeRef context_ptr_type; - LLVMTypeRef vertex_header_ptr_type; LLVMTypeRef buffer_ptr_type; LLVMTypeRef vb_ptr_type; + LLVMTypeRef vertex_header_ptr_type; }; + static INLINE struct llvm_vertex_shader * llvm_vertex_shader(struct draw_vertex_shader *vs) { @@ -269,7 +266,7 @@ llvm_vertex_shader(struct draw_vertex_shader *vs) struct draw_llvm * -draw_llvm_create(struct draw_context *draw); +draw_llvm_create(struct draw_context *draw, struct gallivm_state *gallivm); void draw_llvm_destroy(struct draw_llvm *llvm); @@ -286,7 +283,7 @@ struct draw_llvm_variant_key * draw_llvm_make_variant_key(struct draw_llvm *llvm, char *store); LLVMValueRef -draw_llvm_translate_from(LLVMBuilderRef builder, +draw_llvm_translate_from(struct gallivm_state *gallivm, LLVMValueRef vbuffer, enum pipe_format from_format); diff --git a/src/gallium/auxiliary/draw/draw_llvm_sample.c b/src/gallium/auxiliary/draw/draw_llvm_sample.c index ac1fbb179c..574c7cc452 100644 --- a/src/gallium/auxiliary/draw/draw_llvm_sample.c +++ b/src/gallium/auxiliary/draw/draw_llvm_sample.c @@ -32,6 +32,7 @@ #include "pipe/p_defines.h" #include "pipe/p_shader_tokens.h" +#include "gallivm/lp_bld_const.h" #include "gallivm/lp_bld_debug.h" #include "gallivm/lp_bld_type.h" #include "gallivm/lp_bld_sample.h" @@ -84,12 +85,13 @@ struct draw_llvm_sampler_soa */ static LLVMValueRef draw_llvm_texture_member(const struct lp_sampler_dynamic_state *base, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, unsigned unit, unsigned member_index, const char *member_name, boolean emit_load) { + LLVMBuilderRef builder = gallivm->builder; struct draw_llvm_sampler_dynamic_state *state = (struct draw_llvm_sampler_dynamic_state *)base; LLVMValueRef indices[4]; @@ -99,13 +101,13 @@ draw_llvm_texture_member(const struct lp_sampler_dynamic_state *base, debug_assert(unit < PIPE_MAX_VERTEX_SAMPLERS); /* context[0] */ - indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); + indices[0] = lp_build_const_int32(gallivm, 0); /* context[0].textures */ - indices[1] = LLVMConstInt(LLVMInt32Type(), DRAW_JIT_CTX_TEXTURES, 0); + indices[1] = lp_build_const_int32(gallivm, DRAW_JIT_CTX_TEXTURES); /* context[0].textures[unit] */ - indices[2] = LLVMConstInt(LLVMInt32Type(), unit, 0); + indices[2] = lp_build_const_int32(gallivm, unit); /* context[0].textures[unit].member */ - indices[3] = LLVMConstInt(LLVMInt32Type(), member_index, 0); + indices[3] = lp_build_const_int32(gallivm, member_index); ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), ""); @@ -132,10 +134,10 @@ draw_llvm_texture_member(const struct lp_sampler_dynamic_state *base, #define DRAW_LLVM_TEXTURE_MEMBER(_name, _index, _emit_load) \ static LLVMValueRef \ draw_llvm_texture_##_name( const struct lp_sampler_dynamic_state *base, \ - LLVMBuilderRef builder, \ + struct gallivm_state *gallivm, \ unsigned unit) \ { \ - return draw_llvm_texture_member(base, builder, unit, _index, #_name, _emit_load ); \ + return draw_llvm_texture_member(base, gallivm, unit, _index, #_name, _emit_load ); \ } @@ -165,7 +167,7 @@ draw_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler) */ static void draw_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, struct lp_type type, unsigned unit, unsigned num_coords, @@ -180,7 +182,7 @@ draw_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base, assert(unit < PIPE_MAX_VERTEX_SAMPLERS); - lp_build_sample_soa(builder, + lp_build_sample_soa(gallivm, &sampler->dynamic_state.static_state[unit], &sampler->dynamic_state.base, type, diff --git a/src/gallium/auxiliary/draw/draw_llvm_translate.c b/src/gallium/auxiliary/draw/draw_llvm_translate.c index 5171327ce2..77d0af7473 100644 --- a/src/gallium/auxiliary/draw/draw_llvm_translate.c +++ b/src/gallium/auxiliary/draw/draw_llvm_translate.c @@ -3,6 +3,7 @@ #include "draw_llvm.h" +#include "gallivm/lp_bld_const.h" #include "gallivm/lp_bld_struct.h" #include "gallivm/lp_bld_format.h" #include "gallivm/lp_bld_debug.h" @@ -16,272 +17,279 @@ #define DRAW_DBG 0 static LLVMValueRef -from_64_float(LLVMBuilderRef builder, LLVMValueRef val) +from_64_float(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMDoubleType(), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - return LLVMBuildFPTrunc(builder, l, LLVMFloatType(), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMDoubleTypeInContext(gallivm->context), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + return LLVMBuildFPTrunc(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); } static LLVMValueRef -from_32_float(LLVMBuilderRef builder, LLVMValueRef val) +from_32_float(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMFloatType(), 0) , ""); - return LLVMBuildLoad(builder, bc, ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMFloatTypeInContext(gallivm->context), 0) , ""); + return LLVMBuildLoad(gallivm->builder, bc, ""); } static INLINE LLVMValueRef -from_8_uscaled(LLVMBuilderRef builder, LLVMValueRef val) +from_8_uscaled(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef l = LLVMBuildLoad(builder, val, ""); - return LLVMBuildUIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, ""); + return LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); } static INLINE LLVMValueRef -from_16_uscaled(LLVMBuilderRef builder, LLVMValueRef val) +from_16_uscaled(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(16), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - return LLVMBuildUIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 16), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + return LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); } static INLINE LLVMValueRef -from_32_uscaled(LLVMBuilderRef builder, LLVMValueRef val) +from_32_uscaled(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(32), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - return LLVMBuildUIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + return LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); } static INLINE LLVMValueRef -from_8_sscaled(LLVMBuilderRef builder, LLVMValueRef val) +from_8_sscaled(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef l = LLVMBuildLoad(builder, val, ""); - return LLVMBuildSIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, ""); + return LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); } static INLINE LLVMValueRef -from_16_sscaled(LLVMBuilderRef builder, LLVMValueRef val) +from_16_sscaled(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(16), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - return LLVMBuildSIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 16), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + return LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); } static INLINE LLVMValueRef -from_32_sscaled(LLVMBuilderRef builder, LLVMValueRef val) +from_32_sscaled(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(32), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - return LLVMBuildSIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + return LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); } static INLINE LLVMValueRef -from_8_unorm(LLVMBuilderRef builder, LLVMValueRef val) +from_8_unorm(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef l = LLVMBuildLoad(builder, val, ""); - LLVMValueRef uscaled = LLVMBuildUIToFP(builder, l, LLVMFloatType(), ""); - return LLVMBuildFDiv(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 255.), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, ""); + LLVMValueRef uscaled = LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); + return LLVMBuildFDiv(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 255.), ""); } static INLINE LLVMValueRef -from_16_unorm(LLVMBuilderRef builder, LLVMValueRef val) +from_16_unorm(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(16), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - LLVMValueRef uscaled = LLVMBuildUIToFP(builder, l, LLVMFloatType(), ""); - return LLVMBuildFDiv(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 65535.), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 16), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + LLVMValueRef uscaled = LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); + return LLVMBuildFDiv(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 65535.), ""); } static INLINE LLVMValueRef -from_32_unorm(LLVMBuilderRef builder, LLVMValueRef val) +from_32_unorm(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(32), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - LLVMValueRef uscaled = LLVMBuildUIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + LLVMValueRef uscaled = LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); - return LLVMBuildFDiv(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 4294967295.), ""); + return LLVMBuildFDiv(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 4294967295.), ""); } static INLINE LLVMValueRef -from_8_snorm(LLVMBuilderRef builder, LLVMValueRef val) +from_8_snorm(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef l = LLVMBuildLoad(builder, val, ""); - LLVMValueRef uscaled = LLVMBuildSIToFP(builder, l, LLVMFloatType(), ""); - return LLVMBuildFDiv(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 127.0), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, ""); + LLVMValueRef uscaled = LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); + return LLVMBuildFDiv(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 127.0), ""); } static INLINE LLVMValueRef -from_16_snorm(LLVMBuilderRef builder, LLVMValueRef val) +from_16_snorm(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(16), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - LLVMValueRef uscaled = LLVMBuildSIToFP(builder, l, LLVMFloatType(), ""); - return LLVMBuildFDiv(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 32767.0f), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 16), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + LLVMValueRef uscaled = LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); + return LLVMBuildFDiv(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 32767.0f), ""); } static INLINE LLVMValueRef -from_32_snorm(LLVMBuilderRef builder, LLVMValueRef val) +from_32_snorm(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(32), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - LLVMValueRef uscaled = LLVMBuildSIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + LLVMValueRef uscaled = LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); - return LLVMBuildFDiv(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 2147483647.0), ""); + return LLVMBuildFDiv(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 2147483647.0), ""); } static INLINE LLVMValueRef -from_32_fixed(LLVMBuilderRef builder, LLVMValueRef val) +from_32_fixed(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef bc = LLVMBuildBitCast(builder, val, - LLVMPointerType(LLVMIntType(32), 0) , ""); - LLVMValueRef l = LLVMBuildLoad(builder, bc, ""); - LLVMValueRef uscaled = LLVMBuildSIToFP(builder, l, LLVMFloatType(), ""); + LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val, + LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, ""); + LLVMValueRef uscaled = LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), ""); - return LLVMBuildFDiv(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 65536.0), ""); + return LLVMBuildFDiv(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 65536.0), ""); } static LLVMValueRef -to_64_float(LLVMBuilderRef builder, LLVMValueRef fp) +to_64_float(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - return LLVMBuildFPExt(builder, l, LLVMDoubleType(), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + return LLVMBuildFPExt(gallivm->builder, l, LLVMDoubleTypeInContext(gallivm->context), ""); } static LLVMValueRef -to_32_float(LLVMBuilderRef builder, LLVMValueRef fp) +to_32_float(struct gallivm_state *gallivm, LLVMValueRef fp) { - return LLVMBuildLoad(builder, fp, ""); + return LLVMBuildLoad(gallivm->builder, fp, ""); } static INLINE LLVMValueRef -to_8_uscaled(LLVMBuilderRef builder, LLVMValueRef fp) +to_8_uscaled(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - return LLVMBuildFPToUI(builder, l, LLVMIntType(8), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + return LLVMBuildFPToUI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 8), ""); } static INLINE LLVMValueRef -to_16_uscaled(LLVMBuilderRef builder, LLVMValueRef fp) +to_16_uscaled(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - return LLVMBuildFPToUI(builder, l, LLVMIntType(16), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + return LLVMBuildFPToUI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 16), ""); } static INLINE LLVMValueRef -to_32_uscaled(LLVMBuilderRef builder, LLVMValueRef fp) +to_32_uscaled(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - return LLVMBuildFPToUI(builder, l, LLVMIntType(32), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + return LLVMBuildFPToUI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 32), ""); } static INLINE LLVMValueRef -to_8_sscaled(LLVMBuilderRef builder, LLVMValueRef fp) +to_8_sscaled(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - return LLVMBuildFPToSI(builder, l, LLVMIntType(8), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + return LLVMBuildFPToSI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 8), ""); } static INLINE LLVMValueRef -to_16_sscaled(LLVMBuilderRef builder, LLVMValueRef fp) +to_16_sscaled(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - return LLVMBuildFPToSI(builder, l, LLVMIntType(16), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + return LLVMBuildFPToSI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 16), ""); } static INLINE LLVMValueRef -to_32_sscaled(LLVMBuilderRef builder, LLVMValueRef fp) +to_32_sscaled(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - return LLVMBuildFPToSI(builder, l, LLVMIntType(32), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + return LLVMBuildFPToSI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 32), ""); } static INLINE LLVMValueRef -to_8_unorm(LLVMBuilderRef builder, LLVMValueRef fp) +to_8_unorm(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - LLVMValueRef uscaled = LLVMBuildFPToUI(builder, l, LLVMIntType(8), ""); - return LLVMBuildFMul(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 255.), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + LLVMValueRef uscaled = LLVMBuildFPToUI(gallivm->builder, l, + LLVMIntTypeInContext(gallivm->context, 8), ""); + return LLVMBuildFMul(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 255.), ""); } static INLINE LLVMValueRef -to_16_unorm(LLVMBuilderRef builder, LLVMValueRef fp) +to_16_unorm(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - LLVMValueRef uscaled = LLVMBuildFPToUI(builder, l, LLVMIntType(32), ""); - return LLVMBuildFMul(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 65535.), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + LLVMValueRef uscaled = LLVMBuildFPToUI(gallivm->builder, l, + LLVMIntTypeInContext(gallivm->context, 32), ""); + return LLVMBuildFMul(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 65535.), ""); } static INLINE LLVMValueRef -to_32_unorm(LLVMBuilderRef builder, LLVMValueRef fp) +to_32_unorm(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - LLVMValueRef uscaled = LLVMBuildFPToUI(builder, l, LLVMIntType(32), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + LLVMValueRef uscaled = LLVMBuildFPToUI(gallivm->builder, l, + LLVMIntTypeInContext(gallivm->context, 32), ""); - return LLVMBuildFMul(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 4294967295.), ""); + return LLVMBuildFMul(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 4294967295.), ""); } static INLINE LLVMValueRef -to_8_snorm(LLVMBuilderRef builder, LLVMValueRef val) +to_8_snorm(struct gallivm_state *gallivm, LLVMValueRef val) { - LLVMValueRef l = LLVMBuildLoad(builder, val, ""); - LLVMValueRef uscaled = LLVMBuildFPToSI(builder, l, LLVMIntType(8), ""); - return LLVMBuildFMul(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 127.0), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, ""); + LLVMValueRef uscaled = LLVMBuildFPToSI(gallivm->builder, l, + LLVMIntTypeInContext(gallivm->context, 8), ""); + return LLVMBuildFMul(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 127.0), ""); } static INLINE LLVMValueRef -to_16_snorm(LLVMBuilderRef builder, LLVMValueRef fp) +to_16_snorm(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - LLVMValueRef uscaled = LLVMBuildFPToSI(builder, l, LLVMIntType(16), ""); - return LLVMBuildFMul(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 32767.0f), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + LLVMValueRef uscaled = LLVMBuildFPToSI(gallivm->builder, l, + LLVMIntTypeInContext(gallivm->context, 16), ""); + return LLVMBuildFMul(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 32767.0f), ""); } static INLINE LLVMValueRef -to_32_snorm(LLVMBuilderRef builder, LLVMValueRef fp) +to_32_snorm(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - LLVMValueRef uscaled = LLVMBuildFPToSI(builder, l, LLVMIntType(32), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + LLVMValueRef uscaled = LLVMBuildFPToSI(gallivm->builder, l, + LLVMIntTypeInContext(gallivm->context, 32), ""); - return LLVMBuildFMul(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 2147483647.0), ""); + return LLVMBuildFMul(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 2147483647.0), ""); } static INLINE LLVMValueRef -to_32_fixed(LLVMBuilderRef builder, LLVMValueRef fp) +to_32_fixed(struct gallivm_state *gallivm, LLVMValueRef fp) { - LLVMValueRef l = LLVMBuildLoad(builder, fp, ""); - LLVMValueRef uscaled = LLVMBuildFPToSI(builder, l, LLVMIntType(32), ""); + LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, ""); + LLVMValueRef uscaled = LLVMBuildFPToSI(gallivm->builder, l, + LLVMIntTypeInContext(gallivm->context, 32), ""); - return LLVMBuildFMul(builder, uscaled, - LLVMConstReal(LLVMFloatType(), 65536.0), ""); + return LLVMBuildFMul(gallivm->builder, uscaled, + lp_build_const_float(gallivm, 65536.0), ""); } -typedef LLVMValueRef (*from_func)(LLVMBuilderRef, LLVMValueRef); -typedef LLVMValueRef (*to_func)(LLVMBuilderRef, LLVMValueRef); +typedef LLVMValueRef (*from_func)(struct gallivm_state *, LLVMValueRef); +typedef LLVMValueRef (*to_func)(struct gallivm_state *, LLVMValueRef); /* so that underneath can avoid function calls which are prohibited * for static initialization we need this conversion */ @@ -294,21 +302,21 @@ enum ll_type { }; static INLINE LLVMTypeRef -ll_type_to_llvm(enum ll_type type) +ll_type_to_llvm(struct gallivm_state *gallivm, enum ll_type type) { switch (type) { case LL_Double: - return LLVMDoubleType(); + return LLVMDoubleTypeInContext(gallivm->context); case LL_Float: - return LLVMFloatType(); + return LLVMFloatTypeInContext(gallivm->context); case LL_Int32: - return LLVMInt32Type(); + return LLVMInt32TypeInContext(gallivm->context); case LL_Int16: - return LLVMIntType(16); + return LLVMIntTypeInContext(gallivm->context, 16); case LL_Int8: - return LLVMIntType(8); + return LLVMIntTypeInContext(gallivm->context, 8); } - return LLVMIntType(8); + return LLVMIntTypeInContext(gallivm->context, 8); } static INLINE int @@ -414,42 +422,42 @@ struct draw_llvm_translate { static LLVMValueRef -fetch(LLVMBuilderRef builder, +fetch(struct gallivm_state *gallivm, LLVMValueRef ptr, int val_size, int nr_components, from_func func) { int i; int offset = 0; - LLVMValueRef res = LLVMConstNull( - LLVMVectorType(LLVMFloatType(), 4)); + LLVMValueRef res = + LLVMConstNull(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4)); LLVMValueRef defaults[4]; - defaults[0] = LLVMConstReal(LLVMFloatType(), 0); - defaults[1] = LLVMConstReal(LLVMFloatType(), 0); - defaults[2] = LLVMConstReal(LLVMFloatType(), 0); - defaults[3] = LLVMConstReal(LLVMFloatType(), 1); + defaults[0] = + defaults[1] = + defaults[2] = lp_build_const_float(gallivm, 0.0); + defaults[3] = lp_build_const_float(gallivm, 1.0); for (i = 0; i < nr_components; ++i) { - LLVMValueRef src_index = LLVMConstInt(LLVMInt32Type(), offset, 0); - LLVMValueRef dst_index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef src_index = lp_build_const_int32(gallivm, offset); + LLVMValueRef dst_index = lp_build_const_int32(gallivm, i); LLVMValueRef src_tmp; LLVMValueRef component; - src_tmp = LLVMBuildGEP(builder, ptr, &src_index, 1, "src_tmp"); + src_tmp = LLVMBuildGEP(gallivm->builder, ptr, &src_index, 1, "src_tmp"); /* convert src_tmp to float */ - component = func(builder, src_tmp); + component = func(gallivm, src_tmp); /* vec.comp = component */ - res = LLVMBuildInsertElement(builder, + res = LLVMBuildInsertElement(gallivm->builder, res, component, dst_index, ""); offset += val_size; } for (; i < 4; ++i) { - LLVMValueRef dst_index = LLVMConstInt(LLVMInt32Type(), i, 0); - res = LLVMBuildInsertElement(builder, + LLVMValueRef dst_index = lp_build_const_int32(gallivm, i); + res = LLVMBuildInsertElement(gallivm->builder, res, defaults[i], dst_index, ""); @@ -459,7 +467,7 @@ fetch(LLVMBuilderRef builder, LLVMValueRef -draw_llvm_translate_from(LLVMBuilderRef builder, +draw_llvm_translate_from(struct gallivm_state *gallivm, LLVMValueRef vbuffer, enum pipe_format from_format) { @@ -476,7 +484,7 @@ draw_llvm_translate_from(LLVMBuilderRef builder, for (i = 0; i < Elements(translates); ++i) { if (translates[i].format == from_format) { /*LLVMTypeRef type = ll_type_to_llvm(translates[i].type);*/ - return fetch(builder, + return fetch(gallivm, vbuffer, ll_type_size(translates[i].type), translates[i].num_components, @@ -493,6 +501,6 @@ draw_llvm_translate_from(LLVMBuilderRef builder, */ format_desc = util_format_description(from_format); - zero = LLVMConstNull(LLVMInt32Type()); - return lp_build_fetch_rgba_aos(builder, format_desc, type, vbuffer, zero, zero, zero); + zero = LLVMConstNull(LLVMInt32TypeInContext(gallivm->context)); + return lp_build_fetch_rgba_aos(gallivm, format_desc, type, vbuffer, zero, zero, zero); } diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c index d1aba76309..0851b9acc0 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c +++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c @@ -406,6 +406,7 @@ aaline_create_texture(struct aaline_stage *aaline) texTemp.width0 = 1 << MAX_TEXTURE_LEVEL; texTemp.height0 = 1 << MAX_TEXTURE_LEVEL; texTemp.depth0 = 1; + texTemp.array_size = 1; texTemp.bind = PIPE_BIND_SAMPLER_VIEW; aaline->texture = screen->resource_create(screen, &texTemp); @@ -441,10 +442,10 @@ aaline_create_texture(struct aaline_stage *aaline) /* This texture is new, no need to flush. */ transfer = pipe->get_transfer(pipe, - aaline->texture, - u_subresource(0, level), - PIPE_TRANSFER_WRITE, - &box); + aaline->texture, + level, + PIPE_TRANSFER_WRITE, + &box); data = pipe->transfer_map(pipe, transfer); if (data == NULL) diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c index ed9a53e154..f5515c1df7 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c +++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c @@ -393,8 +393,8 @@ pstip_update_texture(struct pstip_stage *pstip) */ pipe->flush( pipe, PIPE_FLUSH_TEXTURE_CACHE, NULL ); - transfer = pipe_get_transfer(pipe, pstip->texture, 0, 0, 0, - PIPE_TRANSFER_WRITE, 0, 0, 32, 32); + transfer = pipe_get_transfer(pipe, pstip->texture, 0, 0, + PIPE_TRANSFER_WRITE, 0, 0, 32, 32); data = pipe->transfer_map(pipe, transfer); /* @@ -440,6 +440,7 @@ pstip_create_texture(struct pstip_stage *pstip) texTemp.width0 = 32; texTemp.height0 = 32; texTemp.depth0 = 1; + texTemp.array_size = 1; texTemp.bind = PIPE_BIND_SAMPLER_VIEW; pstip->texture = screen->resource_create(screen, &texTemp); diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index 54163d7f9e..06ed4d60ef 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -286,7 +286,6 @@ struct draw_context #ifdef HAVE_LLVM struct draw_llvm *llvm; - LLVMExecutionEngineRef engine; #endif struct pipe_sampler_view *sampler_views[PIPE_MAX_VERTEX_SAMPLERS]; diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 7c198c6026..c98fb3d520 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -58,8 +58,8 @@ struct fetch_shade_emit { const ubyte *src[PIPE_MAX_ATTRIBS]; unsigned prim; - struct draw_vs_varient_key key; - struct draw_vs_varient *active; + struct draw_vs_variant_key key; + struct draw_vs_variant *active; const struct vertex_info *vinfo; @@ -150,7 +150,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle, } - fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader, + fse->active = draw_vs_lookup_variant( draw->vs.vertex_shader, &fse->key ); if (!fse->active) { diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c index a53a768d02..2e3afb22c4 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c @@ -34,6 +34,7 @@ #include "draw/draw_pt.h" #include "draw/draw_vs.h" #include "draw/draw_llvm.h" +#include "gallivm/lp_bld_init.h" struct llvm_middle_end { @@ -72,19 +73,18 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle, struct draw_llvm_variant_list_item *li; unsigned i; unsigned instance_id_index = ~0; - - - unsigned out_prim = (draw->gs.geometry_shader ? - draw->gs.geometry_shader->output_primitive : - in_prim); + const unsigned out_prim = (draw->gs.geometry_shader ? + draw->gs.geometry_shader->output_primitive : + in_prim); /* Add one to num_outputs because the pipeline occasionally tags on * an additional texcoord, eg for AA lines. */ - unsigned nr = MAX2( shader->base.info.num_inputs, - shader->base.info.num_outputs + 1 ); + const unsigned nr = MAX2( shader->base.info.num_inputs, + shader->base.info.num_outputs + 1 ); /* Scan for instanceID system value. + * XXX but we never use instance_id_index?! */ for (i = 0; i < shader->base.info.num_inputs; i++) { if (shader->base.info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) { @@ -133,9 +133,10 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle, key = draw_llvm_make_variant_key(fpme->llvm, store); + /* Search shader's list of variants for the key */ li = first_elem(&shader->variants); - while(!at_end(&shader->variants, li)) { - if(memcmp(&li->base->key, key, shader->variant_key_size) == 0) { + while (!at_end(&shader->variants, li)) { + if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) { variant = li->base; break; } @@ -143,10 +144,16 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle, } if (variant) { + /* found the variant, move to head of global list (for LRU) */ move_to_head(&fpme->llvm->vs_variants_list, &variant->list_item_global); } else { + /* Need to create new variant */ unsigned i; + + /* First check if we've created too many variants. If so, free + * 25% of the LRU to avoid using too much memory. + */ if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) { /* * XXX: should we flush here ? @@ -422,7 +429,7 @@ draw_pt_fetch_pipeline_or_emit_llvm(struct draw_context *draw) { struct llvm_middle_end *fpme = 0; - if (!draw->engine) + if (!draw->llvm->gallivm->engine) return NULL; fpme = CALLOC_STRUCT( llvm_middle_end ); diff --git a/src/gallium/auxiliary/draw/draw_pt_vsplit_tmp.h b/src/gallium/auxiliary/draw/draw_pt_vsplit_tmp.h index 3f66f962e1..75dba8c39a 100644 --- a/src/gallium/auxiliary/draw/draw_pt_vsplit_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_vsplit_tmp.h @@ -258,9 +258,10 @@ vsplit_segment_fan_linear(struct vsplit_frontend *vsplit, unsigned flags, boolean use_spoken = ((flags & DRAW_SPLIT_BEFORE) != 0); unsigned nr = 0, i; - assert(icount + !!use_spoken <= vsplit->segment_size); + assert(icount <= vsplit->segment_size); if (use_spoken) { + /* replace istart by i0 */ vsplit->fetch_elts[nr++] = i0; for (i = 1 ; i < icount; i++) vsplit->fetch_elts[nr++] = istart + i; diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c index fb665b08ff..7caad6f005 100644 --- a/src/gallium/auxiliary/draw/draw_vs.c +++ b/src/gallium/auxiliary/draw/draw_vs.c @@ -165,10 +165,10 @@ draw_delete_vertex_shader(struct draw_context *draw, { unsigned i; - for (i = 0; i < dvs->nr_varients; i++) - dvs->varient[i]->destroy( dvs->varient[i] ); + for (i = 0; i < dvs->nr_variants; i++) + dvs->variant[i]->destroy( dvs->variant[i] ); - dvs->nr_varients = 0; + dvs->nr_variants = 0; dvs->delete( dvs ); } @@ -225,40 +225,40 @@ draw_vs_destroy( struct draw_context *draw ) } -struct draw_vs_varient * -draw_vs_lookup_varient( struct draw_vertex_shader *vs, - const struct draw_vs_varient_key *key ) +struct draw_vs_variant * +draw_vs_lookup_variant( struct draw_vertex_shader *vs, + const struct draw_vs_variant_key *key ) { - struct draw_vs_varient *varient; + struct draw_vs_variant *variant; unsigned i; - /* Lookup existing varient: + /* Lookup existing variant: */ - for (i = 0; i < vs->nr_varients; i++) - if (draw_vs_varient_key_compare(key, &vs->varient[i]->key) == 0) - return vs->varient[i]; + for (i = 0; i < vs->nr_variants; i++) + if (draw_vs_variant_key_compare(key, &vs->variant[i]->key) == 0) + return vs->variant[i]; /* Else have to create a new one: */ - varient = vs->create_varient( vs, key ); - if (varient == NULL) + variant = vs->create_variant( vs, key ); + if (variant == NULL) return NULL; /* Add it to our list, could be smarter: */ - if (vs->nr_varients < Elements(vs->varient)) { - vs->varient[vs->nr_varients++] = varient; + if (vs->nr_variants < Elements(vs->variant)) { + vs->variant[vs->nr_variants++] = variant; } else { - vs->last_varient++; - vs->last_varient %= Elements(vs->varient); - vs->varient[vs->last_varient]->destroy(vs->varient[vs->last_varient]); - vs->varient[vs->last_varient] = varient; + vs->last_variant++; + vs->last_variant %= Elements(vs->variant); + vs->variant[vs->last_variant]->destroy(vs->variant[vs->last_variant]); + vs->variant[vs->last_variant] = variant; } /* Done */ - return varient; + return variant; } diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index f9a038788f..bfb72d50ef 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -38,7 +38,7 @@ struct draw_context; struct pipe_shader_state; -struct draw_varient_input +struct draw_variant_input { enum pipe_format format; unsigned buffer; @@ -46,19 +46,19 @@ struct draw_varient_input unsigned instance_divisor; }; -struct draw_varient_output +struct draw_variant_output { enum pipe_format format; /* output format */ unsigned vs_output:8; /* which vertex shader output is this? */ unsigned offset:24; /* offset into output vertex */ }; -struct draw_varient_element { - struct draw_varient_input in; - struct draw_varient_output out; +struct draw_variant_element { + struct draw_variant_input in; + struct draw_variant_output out; }; -struct draw_vs_varient_key { +struct draw_vs_variant_key { unsigned output_stride; unsigned nr_elements:8; /* max2(nr_inputs, nr_outputs) */ unsigned nr_inputs:8; @@ -66,34 +66,34 @@ struct draw_vs_varient_key { unsigned viewport:1; unsigned clip:1; unsigned const_vbuffers:5; - struct draw_varient_element element[PIPE_MAX_ATTRIBS]; + struct draw_variant_element element[PIPE_MAX_ATTRIBS]; }; -struct draw_vs_varient; +struct draw_vs_variant; -struct draw_vs_varient { - struct draw_vs_varient_key key; +struct draw_vs_variant { + struct draw_vs_variant_key key; struct draw_vertex_shader *vs; - void (*set_buffer)( struct draw_vs_varient *, + void (*set_buffer)( struct draw_vs_variant *, unsigned i, const void *ptr, unsigned stride, unsigned max_stride ); - void (PIPE_CDECL *run_linear)( struct draw_vs_varient *shader, + void (PIPE_CDECL *run_linear)( struct draw_vs_variant *shader, unsigned start, unsigned count, void *output_buffer ); - void (PIPE_CDECL *run_elts)( struct draw_vs_varient *shader, + void (PIPE_CDECL *run_elts)( struct draw_vs_variant *shader, const unsigned *elts, unsigned count, void *output_buffer ); - void (*destroy)( struct draw_vs_varient * ); + void (*destroy)( struct draw_vs_variant * ); }; @@ -117,11 +117,11 @@ struct draw_vertex_shader { /* */ - struct draw_vs_varient *varient[16]; - unsigned nr_varients; - unsigned last_varient; - struct draw_vs_varient *(*create_varient)( struct draw_vertex_shader *shader, - const struct draw_vs_varient_key *key ); + struct draw_vs_variant *variant[16]; + unsigned nr_variants; + unsigned last_variant; + struct draw_vs_variant *(*create_variant)( struct draw_vertex_shader *shader, + const struct draw_vs_variant_key *key ); void (*prepare)( struct draw_vertex_shader *shader, @@ -144,9 +144,9 @@ struct draw_vertex_shader { }; -struct draw_vs_varient * -draw_vs_lookup_varient( struct draw_vertex_shader *base, - const struct draw_vs_varient_key *key ); +struct draw_vs_variant * +draw_vs_lookup_variant( struct draw_vertex_shader *base, + const struct draw_vs_variant_key *key ); /******************************************************************************** @@ -166,12 +166,12 @@ draw_create_vs_ppc(struct draw_context *draw, const struct pipe_shader_state *templ); -struct draw_vs_varient_key; +struct draw_vs_variant_key; struct draw_vertex_shader; -struct draw_vs_varient * -draw_vs_create_varient_aos_sse( struct draw_vertex_shader *vs, - const struct draw_vs_varient_key *key ); +struct draw_vs_variant * +draw_vs_create_variant_aos_sse( struct draw_vertex_shader *vs, + const struct draw_vs_variant_key *key ); #if HAVE_LLVM struct draw_vertex_shader * @@ -181,7 +181,7 @@ draw_create_vs_llvm(struct draw_context *draw, /******************************************************************************** - * Helpers for vs implementations that don't do their own fetch/emit varients. + * Helpers for vs implementations that don't do their own fetch/emit variants. * Means these can be shared between shaders. */ struct translate; @@ -194,21 +194,21 @@ struct translate *draw_vs_get_fetch( struct draw_context *draw, struct translate *draw_vs_get_emit( struct draw_context *draw, struct translate_key *key ); -struct draw_vs_varient * -draw_vs_create_varient_generic( struct draw_vertex_shader *vs, - const struct draw_vs_varient_key *key ); +struct draw_vs_variant * +draw_vs_create_variant_generic( struct draw_vertex_shader *vs, + const struct draw_vs_variant_key *key ); -static INLINE int draw_vs_varient_keysize( const struct draw_vs_varient_key *key ) +static INLINE int draw_vs_variant_keysize( const struct draw_vs_variant_key *key ) { - return 2 * sizeof(int) + key->nr_elements * sizeof(struct draw_varient_element); + return 2 * sizeof(int) + key->nr_elements * sizeof(struct draw_variant_element); } -static INLINE int draw_vs_varient_key_compare( const struct draw_vs_varient_key *a, - const struct draw_vs_varient_key *b ) +static INLINE int draw_vs_variant_key_compare( const struct draw_vs_variant_key *a, + const struct draw_vs_variant_key *b ) { - int keysize = draw_vs_varient_keysize(a); + int keysize = draw_vs_variant_keysize(a); return memcmp(a, b, keysize); } diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 19f49e34c8..7b90dba0cd 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -1918,7 +1918,7 @@ static void find_last_write_outputs( struct aos_compilation *cp ) #define ARG_OUTBUF 4 -static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, +static boolean build_vertex_program( struct draw_vs_variant_aos_sse *variant, boolean linear ) { struct tgsi_parse_context parse; @@ -1927,14 +1927,14 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, util_init_math(); - tgsi_parse_init( &parse, varient->base.vs->state.tokens ); + tgsi_parse_init( &parse, variant->base.vs->state.tokens ); memset(&cp, 0, sizeof(cp)); cp.insn_counter = 1; - cp.vaos = varient; + cp.vaos = variant; cp.have_sse2 = 1; - cp.func = &varient->func[ linear ? 0 : 1 ]; + cp.func = &variant->func[ linear ? 0 : 1 ]; cp.tmp_EAX = x86_make_reg(file_REG32, reg_AX); cp.idx_EBX = x86_make_reg(file_REG32, reg_BX); @@ -2090,20 +2090,20 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, /** cast wrapper */ -static INLINE struct draw_vs_varient_aos_sse * -draw_vs_varient_aos_sse(struct draw_vs_varient *varient) +static INLINE struct draw_vs_variant_aos_sse * +draw_vs_variant_aos_sse(struct draw_vs_variant *variant) { - return (struct draw_vs_varient_aos_sse *) varient; + return (struct draw_vs_variant_aos_sse *) variant; } -static void vaos_set_buffer( struct draw_vs_varient *varient, +static void vaos_set_buffer( struct draw_vs_variant *variant, unsigned buf, const void *ptr, unsigned stride, unsigned max_stride) { - struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient); + struct draw_vs_variant_aos_sse *vaos = draw_vs_variant_aos_sse(variant); if (buf < vaos->nr_vb) { vaos->buffer[buf].base_ptr = (char *)ptr; @@ -2115,12 +2115,12 @@ static void vaos_set_buffer( struct draw_vs_varient *varient, -static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient, +static void PIPE_CDECL vaos_run_elts( struct draw_vs_variant *variant, const unsigned *elts, unsigned count, void *output_buffer ) { - struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient); + struct draw_vs_variant_aos_sse *vaos = draw_vs_variant_aos_sse(variant); struct aos_machine *machine = vaos->draw->vs.aos_machine; unsigned i; @@ -2139,12 +2139,12 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient, output_buffer ); } -static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient, +static void PIPE_CDECL vaos_run_linear( struct draw_vs_variant *variant, unsigned start, unsigned count, void *output_buffer ) { - struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient); + struct draw_vs_variant_aos_sse *vaos = draw_vs_variant_aos_sse(variant); struct aos_machine *machine = vaos->draw->vs.aos_machine; unsigned i; @@ -2171,9 +2171,9 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient, -static void vaos_destroy( struct draw_vs_varient *varient ) +static void vaos_destroy( struct draw_vs_variant *variant ) { - struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient); + struct draw_vs_variant_aos_sse *vaos = draw_vs_variant_aos_sse(variant); FREE( vaos->buffer ); @@ -2185,11 +2185,11 @@ static void vaos_destroy( struct draw_vs_varient *varient ) -static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, - const struct draw_vs_varient_key *key ) +static struct draw_vs_variant *variant_aos_sse( struct draw_vertex_shader *vs, + const struct draw_vs_variant_key *key ) { unsigned i; - struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse); + struct draw_vs_variant_aos_sse *vaos = CALLOC_STRUCT(draw_vs_variant_aos_sse); if (!vaos) goto fail; @@ -2249,17 +2249,17 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, } -struct draw_vs_varient * -draw_vs_create_varient_aos_sse( struct draw_vertex_shader *vs, - const struct draw_vs_varient_key *key ) +struct draw_vs_variant * +draw_vs_create_variant_aos_sse( struct draw_vertex_shader *vs, + const struct draw_vs_variant_key *key ) { - struct draw_vs_varient *varient = varient_aos_sse( vs, key ); + struct draw_vs_variant *variant = variant_aos_sse( vs, key ); - if (varient == NULL) { - varient = draw_vs_create_varient_generic( vs, key ); + if (variant == NULL) { + variant = draw_vs_create_variant_generic( vs, key ); } - return varient; + return variant; } diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index 68e8295b5e..55e63d8b9f 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -98,9 +98,9 @@ struct aos_buffer { -/* This is the temporary storage used by all the aos_sse vs varients. +/* This is the temporary storage used by all the aos_sse vs variants. * Create one per context and reuse by passing a pointer in at - * vs_varient creation?? + * vs_variant creation?? */ struct aos_machine { float input [MAX_INPUTS ][4]; @@ -134,7 +134,7 @@ struct aos_machine { struct aos_compilation { struct x86_function *func; - struct draw_vs_varient_aos_sse *vaos; + struct draw_vs_variant_aos_sse *vaos; unsigned insn_counter; unsigned num_immediates; @@ -234,8 +234,8 @@ typedef void (PIPE_CDECL *vaos_run_linear_func)( struct aos_machine *, void *output_buffer); -struct draw_vs_varient_aos_sse { - struct draw_vs_varient base; +struct draw_vs_variant_aos_sse { + struct draw_vs_variant base; struct draw_context *draw; struct aos_buffer *buffer; diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c index 08608b480e..c41d7c42a0 100644 --- a/src/gallium/auxiliary/draw/draw_vs_exec.c +++ b/src/gallium/auxiliary/draw/draw_vs_exec.c @@ -209,7 +209,7 @@ draw_create_vs_exec(struct draw_context *draw, vs->base.prepare = vs_exec_prepare; vs->base.run_linear = vs_exec_run_linear; vs->base.delete = vs_exec_delete; - vs->base.create_varient = draw_vs_create_varient_generic; + vs->base.create_variant = draw_vs_create_variant_generic; vs->machine = draw->vs.machine; return &vs->base; diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c index fa9992db78..54a5574f73 100644 --- a/src/gallium/auxiliary/draw/draw_vs_llvm.c +++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c @@ -65,19 +65,7 @@ static void vs_llvm_delete( struct draw_vertex_shader *dvs ) { struct llvm_vertex_shader *shader = llvm_vertex_shader(dvs); - struct pipe_fence_handle *fence = NULL; struct draw_llvm_variant_list_item *li; - struct pipe_context *pipe = dvs->draw->pipe; - - /* - * XXX: This might be not neccessary at all. - */ - pipe->flush(pipe, 0, &fence); - if (fence) { - pipe->screen->fence_finish(pipe->screen, fence, 0); - pipe->screen->fence_reference(pipe->screen, &fence, NULL); - } - li = first_elem(&shader->variants); while(!at_end(&shader->variants, li)) { @@ -119,7 +107,7 @@ draw_create_vs_llvm(struct draw_context *draw, vs->base.prepare = vs_llvm_prepare; vs->base.run_linear = vs_llvm_run_linear; vs->base.delete = vs_llvm_delete; - vs->base.create_varient = draw_vs_create_varient_generic; + vs->base.create_variant = draw_vs_create_variant_generic; make_empty_list(&vs->variants); diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c index 5df84916c5..cf894bbe8a 100644 --- a/src/gallium/auxiliary/draw/draw_vs_ppc.c +++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c @@ -187,10 +187,10 @@ draw_create_vs_ppc(struct draw_context *draw, vs->base.draw = draw; #if 0 if (1) - vs->base.create_varient = draw_vs_varient_aos_ppc; + vs->base.create_variant = draw_vs_variant_aos_ppc; else #endif - vs->base.create_varient = draw_vs_create_varient_generic; + vs->base.create_variant = draw_vs_create_variant_generic; vs->base.prepare = vs_ppc_prepare; vs->base.run_linear = vs_ppc_run_linear; vs->base.delete = vs_ppc_delete; diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index 672efe25cc..d55b9b0807 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -172,9 +172,9 @@ draw_create_vs_sse(struct draw_context *draw, vs->base.draw = draw; if (1) - vs->base.create_varient = draw_vs_create_varient_aos_sse; + vs->base.create_variant = draw_vs_create_variant_aos_sse; else - vs->base.create_varient = draw_vs_create_varient_generic; + vs->base.create_variant = draw_vs_create_variant_generic; vs->base.prepare = vs_sse_prepare; vs->base.run_linear = vs_sse_run_linear; vs->base.delete = vs_sse_delete; diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c index eacd160187..d8f030f61e 100644 --- a/src/gallium/auxiliary/draw/draw_vs_varient.c +++ b/src/gallium/auxiliary/draw/draw_vs_varient.c @@ -41,8 +41,8 @@ /* A first pass at incorporating vertex fetch/emit functionality into */ -struct draw_vs_varient_generic { - struct draw_vs_varient base; +struct draw_vs_variant_generic { + struct draw_vs_variant base; struct draw_vertex_shader *shader; struct draw_context *draw; @@ -63,13 +63,13 @@ struct draw_vs_varient_generic { -static void vsvg_set_buffer( struct draw_vs_varient *varient, +static void vsvg_set_buffer( struct draw_vs_variant *variant, unsigned buffer, const void *ptr, unsigned stride, unsigned max_index ) { - struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; + struct draw_vs_variant_generic *vsvg = (struct draw_vs_variant_generic *)variant; vsvg->fetch->set_buffer(vsvg->fetch, buffer, @@ -81,7 +81,7 @@ static void vsvg_set_buffer( struct draw_vs_varient *varient, /* Mainly for debug at this stage: */ -static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg, +static void do_rhw_viewport( struct draw_vs_variant_generic *vsvg, unsigned count, void *output_buffer ) { @@ -104,7 +104,7 @@ static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg, } } -static void do_viewport( struct draw_vs_varient_generic *vsvg, +static void do_viewport( struct draw_vs_variant_generic *vsvg, unsigned count, void *output_buffer ) { @@ -126,12 +126,12 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg, } -static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient, +static void PIPE_CDECL vsvg_run_elts( struct draw_vs_variant *variant, const unsigned *elts, unsigned count, void *output_buffer) { - struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; + struct draw_vs_variant_generic *vsvg = (struct draw_vs_variant_generic *)variant; unsigned temp_vertex_stride = vsvg->temp_vertex_stride; void *temp_buffer = MALLOC( align(count,4) * temp_vertex_stride ); @@ -193,12 +193,12 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient, } -static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient, +static void PIPE_CDECL vsvg_run_linear( struct draw_vs_variant *variant, unsigned start, unsigned count, void *output_buffer ) { - struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; + struct draw_vs_variant_generic *vsvg = (struct draw_vs_variant_generic *)variant; unsigned temp_vertex_stride = vsvg->temp_vertex_stride; void *temp_buffer = MALLOC( align(count,4) * temp_vertex_stride ); @@ -259,20 +259,20 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient, -static void vsvg_destroy( struct draw_vs_varient *varient ) +static void vsvg_destroy( struct draw_vs_variant *variant ) { - FREE(varient); + FREE(variant); } -struct draw_vs_varient * -draw_vs_create_varient_generic( struct draw_vertex_shader *vs, - const struct draw_vs_varient_key *key ) +struct draw_vs_variant * +draw_vs_create_variant_generic( struct draw_vertex_shader *vs, + const struct draw_vs_variant_key *key ) { unsigned i; struct translate_key fetch, emit; - struct draw_vs_varient_generic *vsvg = CALLOC_STRUCT( draw_vs_varient_generic ); + struct draw_vs_variant_generic *vsvg = CALLOC_STRUCT( draw_vs_variant_generic ); if (vsvg == NULL) return NULL; diff --git a/src/gallium/auxiliary/gallivm/lp_bld.h b/src/gallium/auxiliary/gallivm/lp_bld.h index 8103bc917f..ee05c6ba01 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld.h +++ b/src/gallium/auxiliary/gallivm/lp_bld.h @@ -55,4 +55,33 @@ #endif +/** + * Redefine these LLVM entrypoints as invalid macros to make sure we + * don't accidentally use them. We need to use the functions which + * take an explicit LLVMContextRef parameter. + */ +#define LLVMInt1Type ILLEGAL_LLVM_FUNCTION +#define LLVMInt8Type ILLEGAL_LLVM_FUNCTION +#define LLVMInt16Type ILLEGAL_LLVM_FUNCTION +#define LLVMInt32Type ILLEGAL_LLVM_FUNCTION +#define LLVMInt64Type ILLEGAL_LLVM_FUNCTION +#define LLVMIntType ILLEGAL_LLVM_FUNCTION +#define LLVMFloatType ILLEGAL_LLVM_FUNCTION +#define LLVMDoubleType ILLEGAL_LLVM_FUNCTION +#define LLVMX86FP80Type ILLEGAL_LLVM_FUNCTION +#define LLVMFP128Type ILLEGAL_LLVM_FUNCTION +#define LLVMPPCFP128Type ILLEGAL_LLVM_FUNCTION +#define LLVMStructType ILLEGAL_LLVM_FUNCTION +#define LLVMVoidType ILLEGAL_LLVM_FUNCTION +#define LLVMLabelType ILLEGAL_LLVM_FUNCTION +#define LLVMOpaqueType ILLEGAL_LLVM_FUNCTION +#define LLVMUnionType ILLEGAL_LLVM_FUNCTION +#define LLVMMDString ILLEGAL_LLVM_FUNCTION +#define LLVMMDNode ILLEGAL_LLVM_FUNCTION +#define LLVMConstString ILLEGAL_LLVM_FUNCTION +#define LLVMConstStruct ILLEGAL_LLVM_FUNCTION +#define LLVMAppendBasicBlock ILLEGAL_LLVM_FUNCTION +#define LLVMInsertBasicBlock ILLEGAL_LLVM_FUNCTION +#define LLVMCreateBuilder ILLEGAL_LLVM_FUNCTION + #endif /* LP_BLD_H */ diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c index f9a12a41a1..02b3bde789 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c @@ -53,6 +53,7 @@ #include "lp_bld_type.h" #include "lp_bld_const.h" +#include "lp_bld_init.h" #include "lp_bld_intr.h" #include "lp_bld_logic.h" #include "lp_bld_pack.h" @@ -74,6 +75,7 @@ lp_build_min_simple(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; const char *intrinsic = NULL; LLVMValueRef cond; @@ -107,7 +109,7 @@ lp_build_min_simple(struct lp_build_context *bld, } if(intrinsic) - return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b); + return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b); cond = lp_build_cmp(bld, PIPE_FUNC_LESS, a, b); return lp_build_select(bld, cond, a, b); @@ -123,6 +125,7 @@ lp_build_max_simple(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; const char *intrinsic = NULL; LLVMValueRef cond; @@ -156,7 +159,7 @@ lp_build_max_simple(struct lp_build_context *bld, } if(intrinsic) - return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b); + return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b); cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, b); return lp_build_select(bld, cond, a, b); @@ -170,6 +173,7 @@ LLVMValueRef lp_build_comp(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(lp_check_value(type, a)); @@ -183,7 +187,7 @@ lp_build_comp(struct lp_build_context *bld, if(LLVMIsConstant(a)) return LLVMConstNot(a); else - return LLVMBuildNot(bld->builder, a, ""); + return LLVMBuildNot(builder, a, ""); } if(LLVMIsConstant(a)) @@ -193,9 +197,9 @@ lp_build_comp(struct lp_build_context *bld, return LLVMConstSub(bld->one, a); else if (type.floating) - return LLVMBuildFSub(bld->builder, bld->one, a, ""); + return LLVMBuildFSub(builder, bld->one, a, ""); else - return LLVMBuildSub(bld->builder, bld->one, a, ""); + return LLVMBuildSub(builder, bld->one, a, ""); } @@ -207,6 +211,7 @@ lp_build_add(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef res; @@ -236,7 +241,7 @@ lp_build_add(struct lp_build_context *bld, } if(intrinsic) - return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b); + return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b); } if(LLVMIsConstant(a) && LLVMIsConstant(b)) @@ -246,9 +251,9 @@ lp_build_add(struct lp_build_context *bld, res = LLVMConstAdd(a, b); else if (type.floating) - res = LLVMBuildFAdd(bld->builder, a, b, ""); + res = LLVMBuildFAdd(builder, a, b, ""); else - res = LLVMBuildAdd(bld->builder, a, b, ""); + res = LLVMBuildAdd(builder, a, b, ""); /* clamp to ceiling of 1.0 */ if(bld->type.norm && (bld->type.floating || bld->type.fixed)) @@ -265,6 +270,7 @@ LLVMValueRef lp_build_sum_vector(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef index, res; unsigned i; @@ -277,19 +283,19 @@ lp_build_sum_vector(struct lp_build_context *bld, assert(!bld->type.norm); - index = LLVMConstInt(LLVMInt32Type(), 0, 0); - res = LLVMBuildExtractElement(bld->builder, a, index, ""); + index = lp_build_const_int32(bld->gallivm, 0); + res = LLVMBuildExtractElement(builder, a, index, ""); for (i = 1; i < type.length; i++) { - index = LLVMConstInt(LLVMInt32Type(), i, 0); + index = lp_build_const_int32(bld->gallivm, i); if (type.floating) - res = LLVMBuildFAdd(bld->builder, res, - LLVMBuildExtractElement(bld->builder, + res = LLVMBuildFAdd(builder, res, + LLVMBuildExtractElement(builder, a, index, ""), ""); else - res = LLVMBuildAdd(bld->builder, res, - LLVMBuildExtractElement(bld->builder, + res = LLVMBuildAdd(builder, res, + LLVMBuildExtractElement(builder, a, index, ""), ""); } @@ -306,6 +312,7 @@ lp_build_sub(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef res; @@ -335,7 +342,7 @@ lp_build_sub(struct lp_build_context *bld, } if(intrinsic) - return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b); + return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b); } if(LLVMIsConstant(a) && LLVMIsConstant(b)) @@ -345,9 +352,9 @@ lp_build_sub(struct lp_build_context *bld, res = LLVMConstSub(a, b); else if (type.floating) - res = LLVMBuildFSub(bld->builder, a, b, ""); + res = LLVMBuildFSub(builder, a, b, ""); else - res = LLVMBuildSub(bld->builder, a, b, ""); + res = LLVMBuildSub(builder, a, b, ""); if(bld->type.norm && (bld->type.floating || bld->type.fixed)) res = lp_build_max_simple(bld, res, bld->zero); @@ -398,10 +405,11 @@ lp_build_sub(struct lp_build_context *bld, * http://www.stereopsis.com/doubleblend.html */ static LLVMValueRef -lp_build_mul_u8n(LLVMBuilderRef builder, +lp_build_mul_u8n(struct gallivm_state *gallivm, struct lp_type i16_type, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef c8; LLVMValueRef ab; @@ -409,12 +417,12 @@ lp_build_mul_u8n(LLVMBuilderRef builder, assert(lp_check_value(i16_type, a)); assert(lp_check_value(i16_type, b)); - c8 = lp_build_const_int_vec(i16_type, 8); + c8 = lp_build_const_int_vec(gallivm, i16_type, 8); #if 0 /* a*b/255 ~= (a*(b + 1)) >> 256 */ - b = LLVMBuildAdd(builder, b, lp_build_const_int_vec(i16_type, 1), ""); + b = LLVMBuildAdd(builder, b, lp_build_const_int_vec(gallium, i16_type, 1), ""); ab = LLVMBuildMul(builder, a, b, ""); #else @@ -422,7 +430,7 @@ lp_build_mul_u8n(LLVMBuilderRef builder, /* ab/255 ~= (ab + (ab >> 8) + 0x80) >> 8 */ ab = LLVMBuildMul(builder, a, b, ""); ab = LLVMBuildAdd(builder, ab, LLVMBuildLShr(builder, ab, c8, ""), ""); - ab = LLVMBuildAdd(builder, ab, lp_build_const_int_vec(i16_type, 0x80), ""); + ab = LLVMBuildAdd(builder, ab, lp_build_const_int_vec(gallivm, i16_type, 0x80), ""); #endif @@ -440,6 +448,7 @@ lp_build_mul(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef shift; LLVMValueRef res; @@ -463,14 +472,14 @@ lp_build_mul(struct lp_build_context *bld, struct lp_type i16_type = lp_wider_type(type); LLVMValueRef al, ah, bl, bh, abl, abh, ab; - lp_build_unpack2(bld->builder, type, i16_type, a, &al, &ah); - lp_build_unpack2(bld->builder, type, i16_type, b, &bl, &bh); + lp_build_unpack2(bld->gallivm, type, i16_type, a, &al, &ah); + lp_build_unpack2(bld->gallivm, type, i16_type, b, &bl, &bh); /* PMULLW, PSRLW, PADDW */ - abl = lp_build_mul_u8n(bld->builder, i16_type, al, bl); - abh = lp_build_mul_u8n(bld->builder, i16_type, ah, bh); + abl = lp_build_mul_u8n(bld->gallivm, i16_type, al, bl); + abh = lp_build_mul_u8n(bld->gallivm, i16_type, ah, bh); - ab = lp_build_pack2(bld->builder, i16_type, type, abl, abh); + ab = lp_build_pack2(bld->gallivm, i16_type, type, abl, abh); return ab; } @@ -480,7 +489,7 @@ lp_build_mul(struct lp_build_context *bld, } if(type.fixed) - shift = lp_build_const_int_vec(type, type.width/2); + shift = lp_build_const_int_vec(bld->gallivm, type, type.width/2); else shift = NULL; @@ -498,14 +507,14 @@ lp_build_mul(struct lp_build_context *bld, } else { if (type.floating) - res = LLVMBuildFMul(bld->builder, a, b, ""); + res = LLVMBuildFMul(builder, a, b, ""); else - res = LLVMBuildMul(bld->builder, a, b, ""); + res = LLVMBuildMul(builder, a, b, ""); if(shift) { if(type.sign) - res = LLVMBuildAShr(bld->builder, res, shift, ""); + res = LLVMBuildAShr(builder, res, shift, ""); else - res = LLVMBuildLShr(bld->builder, res, shift, ""); + res = LLVMBuildLShr(builder, res, shift, ""); } } @@ -521,6 +530,7 @@ lp_build_mul_imm(struct lp_build_context *bld, LLVMValueRef a, int b) { + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef factor; assert(lp_check_value(bld->type, a)); @@ -550,20 +560,20 @@ lp_build_mul_imm(struct lp_build_context *bld, * for Inf and NaN. */ unsigned mantissa = lp_mantissa(bld->type); - factor = lp_build_const_int_vec(bld->type, (unsigned long long)shift << mantissa); - a = LLVMBuildBitCast(bld->builder, a, lp_build_int_vec_type(bld->type), ""); - a = LLVMBuildAdd(bld->builder, a, factor, ""); - a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(bld->type), ""); + factor = lp_build_const_int_vec(bld->gallivm, bld->type, (unsigned long long)shift << mantissa); + a = LLVMBuildBitCast(builder, a, lp_build_int_vec_type(bld->type), ""); + a = LLVMBuildAdd(builder, a, factor, ""); + a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, bld->type), ""); return a; #endif } else { - factor = lp_build_const_vec(bld->type, shift); - return LLVMBuildShl(bld->builder, a, factor, ""); + factor = lp_build_const_vec(bld->gallivm, bld->type, shift); + return LLVMBuildShl(builder, a, factor, ""); } } - factor = lp_build_const_vec(bld->type, (double)b); + factor = lp_build_const_vec(bld->gallivm, bld->type, (double)b); return lp_build_mul(bld, a, factor); } @@ -576,6 +586,7 @@ lp_build_div(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(lp_check_value(type, a)); @@ -605,11 +616,11 @@ lp_build_div(struct lp_build_context *bld, return lp_build_mul(bld, a, lp_build_rcp(bld, b)); if (type.floating) - return LLVMBuildFDiv(bld->builder, a, b, ""); + return LLVMBuildFDiv(builder, a, b, ""); else if (type.sign) - return LLVMBuildSDiv(bld->builder, a, b, ""); + return LLVMBuildSDiv(builder, a, b, ""); else - return LLVMBuildUDiv(bld->builder, a, b, ""); + return LLVMBuildUDiv(builder, a, b, ""); } @@ -624,6 +635,7 @@ lp_build_lerp_simple(struct lp_build_context *bld, LLVMValueRef v0, LLVMValueRef v1) { + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef delta; LLVMValueRef res; @@ -642,7 +654,7 @@ lp_build_lerp_simple(struct lp_build_context *bld, * but it will be wrong for other uses. Basically we need a more * powerful lp_type, capable of further distinguishing the values * interpretation from the value storage. */ - res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(bld->type, (1 << bld->type.width/2) - 1), ""); + res = LLVMBuildAnd(builder, res, lp_build_const_int_vec(bld->gallivm, bld->type, (1 << bld->type.width/2) - 1), ""); } return res; @@ -658,6 +670,7 @@ lp_build_lerp(struct lp_build_context *bld, LLVMValueRef v0, LLVMValueRef v1) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef res; @@ -683,22 +696,22 @@ lp_build_lerp(struct lp_build_context *bld, wide_type.width = type.width*2; wide_type.length = type.length/2; - lp_build_context_init(&wide_bld, bld->builder, wide_type); + lp_build_context_init(&wide_bld, bld->gallivm, wide_type); - lp_build_unpack2(bld->builder, type, wide_type, x, &xl, &xh); - lp_build_unpack2(bld->builder, type, wide_type, v0, &v0l, &v0h); - lp_build_unpack2(bld->builder, type, wide_type, v1, &v1l, &v1h); + lp_build_unpack2(bld->gallivm, type, wide_type, x, &xl, &xh); + lp_build_unpack2(bld->gallivm, type, wide_type, v0, &v0l, &v0h); + lp_build_unpack2(bld->gallivm, type, wide_type, v1, &v1l, &v1h); /* * Scale x from [0, 255] to [0, 256] */ - shift = lp_build_const_int_vec(wide_type, type.width - 1); + shift = lp_build_const_int_vec(bld->gallivm, wide_type, type.width - 1); xl = lp_build_add(&wide_bld, xl, - LLVMBuildAShr(bld->builder, xl, shift, "")); + LLVMBuildAShr(builder, xl, shift, "")); xh = lp_build_add(&wide_bld, xh, - LLVMBuildAShr(bld->builder, xh, shift, "")); + LLVMBuildAShr(builder, xh, shift, "")); /* * Lerp both halves. @@ -707,7 +720,7 @@ lp_build_lerp(struct lp_build_context *bld, resl = lp_build_lerp_simple(&wide_bld, xl, v0l, v1l); resh = lp_build_lerp_simple(&wide_bld, xh, v0h, v1h); - res = lp_build_pack2(bld->builder, wide_type, type, resl, resh); + res = lp_build_pack2(bld->gallivm, wide_type, type, resl, resh); } else { res = lp_build_lerp_simple(bld, x, v0, v1); } @@ -820,8 +833,9 @@ LLVMValueRef lp_build_abs(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); assert(lp_check_value(type, a)); @@ -830,27 +844,27 @@ lp_build_abs(struct lp_build_context *bld, if(type.floating) { /* Mask out the sign bit */ - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type); unsigned long long absMask = ~(1ULL << (type.width - 1)); - LLVMValueRef mask = lp_build_const_int_vec(type, ((unsigned long long) absMask)); - a = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); - a = LLVMBuildAnd(bld->builder, a, mask, ""); - a = LLVMBuildBitCast(bld->builder, a, vec_type, ""); + LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type, ((unsigned long long) absMask)); + a = LLVMBuildBitCast(builder, a, int_vec_type, ""); + a = LLVMBuildAnd(builder, a, mask, ""); + a = LLVMBuildBitCast(builder, a, vec_type, ""); return a; } if(type.width*type.length == 128 && util_cpu_caps.has_ssse3) { switch(type.width) { case 8: - return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.b.128", vec_type, a); + return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.b.128", vec_type, a); case 16: - return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.w.128", vec_type, a); + return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.w.128", vec_type, a); case 32: - return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.d.128", vec_type, a); + return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.d.128", vec_type, a); } } - return lp_build_max(bld, a, LLVMBuildNeg(bld->builder, a, "")); + return lp_build_max(bld, a, LLVMBuildNeg(builder, a, "")); } @@ -858,14 +872,16 @@ LLVMValueRef lp_build_negate(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; + assert(lp_check_value(bld->type, a)); #if HAVE_LLVM >= 0x0207 if (bld->type.floating) - a = LLVMBuildFNeg(bld->builder, a, ""); + a = LLVMBuildFNeg(builder, a, ""); else #endif - a = LLVMBuildNeg(bld->builder, a, ""); + a = LLVMBuildNeg(builder, a, ""); return a; } @@ -876,6 +892,7 @@ LLVMValueRef lp_build_sgn(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef cond; LLVMValueRef res; @@ -895,20 +912,20 @@ lp_build_sgn(struct lp_build_context *bld, LLVMValueRef one; unsigned long long maskBit = (unsigned long long)1 << (type.width - 1); - int_type = lp_build_int_vec_type(type); - vec_type = lp_build_vec_type(type); - mask = lp_build_const_int_vec(type, maskBit); + int_type = lp_build_int_vec_type(bld->gallivm, type); + vec_type = lp_build_vec_type(bld->gallivm, type); + mask = lp_build_const_int_vec(bld->gallivm, type, maskBit); /* Take the sign bit and add it to 1 constant */ - sign = LLVMBuildBitCast(bld->builder, a, int_type, ""); - sign = LLVMBuildAnd(bld->builder, sign, mask, ""); + sign = LLVMBuildBitCast(builder, a, int_type, ""); + sign = LLVMBuildAnd(builder, sign, mask, ""); one = LLVMConstBitCast(bld->one, int_type); - res = LLVMBuildOr(bld->builder, sign, one, ""); - res = LLVMBuildBitCast(bld->builder, res, vec_type, ""); + res = LLVMBuildOr(builder, sign, one, ""); + res = LLVMBuildBitCast(builder, res, vec_type, ""); } else { - LLVMValueRef minus_one = lp_build_const_vec(type, -1.0); + LLVMValueRef minus_one = lp_build_const_vec(bld->gallivm, type, -1.0); cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, bld->zero); res = lp_build_select(bld, cond, bld->one, minus_one); } @@ -931,11 +948,12 @@ LLVMValueRef lp_build_set_sign(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef sign) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); - LLVMTypeRef vec_type = lp_build_vec_type(type); - LLVMValueRef shift = lp_build_const_int_vec(type, type.width - 1); - LLVMValueRef mask = lp_build_const_int_vec(type, + LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); + LLVMValueRef shift = lp_build_const_int_vec(bld->gallivm, type, type.width - 1); + LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type, ~((unsigned long long) 1 << (type.width - 1))); LLVMValueRef val, res; @@ -943,15 +961,15 @@ lp_build_set_sign(struct lp_build_context *bld, assert(lp_check_value(type, a)); /* val = reinterpret_cast<int>(a) */ - val = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); + val = LLVMBuildBitCast(builder, a, int_vec_type, ""); /* val = val & mask */ - val = LLVMBuildAnd(bld->builder, val, mask, ""); + val = LLVMBuildAnd(builder, val, mask, ""); /* sign = sign << shift */ - sign = LLVMBuildShl(bld->builder, sign, shift, ""); + sign = LLVMBuildShl(builder, sign, shift, ""); /* res = val | sign */ - res = LLVMBuildOr(bld->builder, val, sign, ""); + res = LLVMBuildOr(builder, val, sign, ""); /* res = reinterpret_cast<float>(res) */ - res = LLVMBuildBitCast(bld->builder, res, vec_type, ""); + res = LLVMBuildBitCast(builder, res, vec_type, ""); return res; } @@ -964,12 +982,13 @@ LLVMValueRef lp_build_int_to_float(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); assert(type.floating); - return LLVMBuildSIToFP(bld->builder, a, vec_type, ""); + return LLVMBuildSIToFP(builder, a, vec_type, ""); } @@ -994,8 +1013,9 @@ lp_build_round_sse41(struct lp_build_context *bld, LLVMValueRef a, enum lp_build_round_sse41_mode mode) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef i32t = LLVMInt32Type(); + LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context); const char *intrinsic; LLVMValueRef res; @@ -1027,13 +1047,13 @@ lp_build_round_sse41(struct lp_build_context *bld, undef = LLVMGetUndef(vec_type); args[0] = undef; - args[1] = LLVMBuildInsertElement(bld->builder, undef, a, index0, ""); + args[1] = LLVMBuildInsertElement(builder, undef, a, index0, ""); args[2] = LLVMConstInt(i32t, mode, 0); - res = lp_build_intrinsic(bld->builder, intrinsic, + res = lp_build_intrinsic(builder, intrinsic, vec_type, args, Elements(args)); - res = LLVMBuildExtractElement(bld->builder, res, index0, ""); + res = LLVMBuildExtractElement(builder, res, index0, ""); } else { assert(type.width*type.length == 128); @@ -1050,7 +1070,7 @@ lp_build_round_sse41(struct lp_build_context *bld, return bld->undef; } - res = lp_build_intrinsic_binary(bld->builder, intrinsic, + res = lp_build_intrinsic_binary(builder, intrinsic, bld->vec_type, a, LLVMConstInt(i32t, mode, 0)); } @@ -1063,9 +1083,10 @@ static INLINE LLVMValueRef lp_build_iround_nearest_sse2(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef i32t = LLVMInt32Type(); - LLVMTypeRef ret_type = lp_build_int_vec_type(type); + LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context); + LLVMTypeRef ret_type = lp_build_int_vec_type(bld->gallivm, type); const char *intrinsic; LLVMValueRef res; @@ -1089,9 +1110,9 @@ lp_build_iround_nearest_sse2(struct lp_build_context *bld, undef = LLVMGetUndef(vec_type); - arg = LLVMBuildInsertElement(bld->builder, undef, a, index0, ""); + arg = LLVMBuildInsertElement(builder, undef, a, index0, ""); - res = lp_build_intrinsic_unary(bld->builder, intrinsic, + res = lp_build_intrinsic_unary(builder, intrinsic, ret_type, arg); } else { @@ -1099,7 +1120,7 @@ lp_build_iround_nearest_sse2(struct lp_build_context *bld, intrinsic = "llvm.x86.sse2.cvtps2dq"; - res = lp_build_intrinsic_unary(bld->builder, intrinsic, + res = lp_build_intrinsic_unary(builder, intrinsic, ret_type, a); } @@ -1116,6 +1137,7 @@ LLVMValueRef lp_build_trunc(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(type.floating); @@ -1126,11 +1148,11 @@ lp_build_trunc(struct lp_build_context *bld, return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_TRUNCATE); } else { - LLVMTypeRef vec_type = lp_build_vec_type(type); - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type); LLVMValueRef res; - res = LLVMBuildFPToSI(bld->builder, a, int_vec_type, ""); - res = LLVMBuildSIToFP(bld->builder, res, vec_type, ""); + res = LLVMBuildFPToSI(builder, a, int_vec_type, ""); + res = LLVMBuildSIToFP(builder, res, vec_type, ""); return res; } } @@ -1146,6 +1168,7 @@ LLVMValueRef lp_build_round(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(type.floating); @@ -1156,10 +1179,10 @@ lp_build_round(struct lp_build_context *bld, return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST); } else { - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); LLVMValueRef res; res = lp_build_iround(bld, a); - res = LLVMBuildSIToFP(bld->builder, res, vec_type, ""); + res = LLVMBuildSIToFP(builder, res, vec_type, ""); return res; } } @@ -1174,6 +1197,7 @@ LLVMValueRef lp_build_floor(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(type.floating); @@ -1184,10 +1208,10 @@ lp_build_floor(struct lp_build_context *bld, return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_FLOOR); } else { - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); LLVMValueRef res; res = lp_build_ifloor(bld, a); - res = LLVMBuildSIToFP(bld->builder, res, vec_type, ""); + res = LLVMBuildSIToFP(builder, res, vec_type, ""); return res; } } @@ -1202,6 +1226,7 @@ LLVMValueRef lp_build_ceil(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(type.floating); @@ -1212,10 +1237,10 @@ lp_build_ceil(struct lp_build_context *bld, return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL); } else { - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); LLVMValueRef res; res = lp_build_iceil(bld, a); - res = LLVMBuildSIToFP(bld->builder, res, vec_type, ""); + res = LLVMBuildSIToFP(builder, res, vec_type, ""); return res; } } @@ -1243,13 +1268,14 @@ LLVMValueRef lp_build_itrunc(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type); assert(type.floating); assert(lp_check_value(type, a)); - return LLVMBuildFPToSI(bld->builder, a, int_vec_type, ""); + return LLVMBuildFPToSI(builder, a, int_vec_type, ""); } @@ -1263,6 +1289,7 @@ LLVMValueRef lp_build_iround(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMTypeRef int_vec_type = bld->int_vec_type; LLVMValueRef res; @@ -1282,27 +1309,28 @@ lp_build_iround(struct lp_build_context *bld, else { LLVMValueRef half; - half = lp_build_const_vec(type, 0.5); + half = lp_build_const_vec(bld->gallivm, type, 0.5); if (type.sign) { LLVMTypeRef vec_type = bld->vec_type; - LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1)); + LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type, + (unsigned long long)1 << (type.width - 1)); LLVMValueRef sign; /* get sign bit */ - sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); - sign = LLVMBuildAnd(bld->builder, sign, mask, ""); + sign = LLVMBuildBitCast(builder, a, int_vec_type, ""); + sign = LLVMBuildAnd(builder, sign, mask, ""); /* sign * 0.5 */ - half = LLVMBuildBitCast(bld->builder, half, int_vec_type, ""); - half = LLVMBuildOr(bld->builder, sign, half, ""); - half = LLVMBuildBitCast(bld->builder, half, vec_type, ""); + half = LLVMBuildBitCast(builder, half, int_vec_type, ""); + half = LLVMBuildOr(builder, sign, half, ""); + half = LLVMBuildBitCast(builder, half, vec_type, ""); } - res = LLVMBuildFAdd(bld->builder, a, half, ""); + res = LLVMBuildFAdd(builder, a, half, ""); } - res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, ""); + res = LLVMBuildFPToSI(builder, res, int_vec_type, ""); return res; } @@ -1317,6 +1345,7 @@ LLVMValueRef lp_build_ifloor(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMTypeRef int_vec_type = bld->int_vec_type; LLVMValueRef res; @@ -1335,29 +1364,34 @@ lp_build_ifloor(struct lp_build_context *bld, /* Take the sign bit and add it to 1 constant */ LLVMTypeRef vec_type = bld->vec_type; unsigned mantissa = lp_mantissa(type); - LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1)); + LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type, + (unsigned long long)1 << (type.width - 1)); LLVMValueRef sign; LLVMValueRef offset; /* sign = a < 0 ? ~0 : 0 */ - sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); - sign = LLVMBuildAnd(bld->builder, sign, mask, ""); - sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "ifloor.sign"); + sign = LLVMBuildBitCast(builder, a, int_vec_type, ""); + sign = LLVMBuildAnd(builder, sign, mask, ""); + sign = LLVMBuildAShr(builder, sign, + lp_build_const_int_vec(bld->gallivm, type, + type.width - 1), + "ifloor.sign"); /* offset = -0.99999(9)f */ - offset = lp_build_const_vec(type, -(double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa)); + offset = lp_build_const_vec(bld->gallivm, type, + -(double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa)); offset = LLVMConstBitCast(offset, int_vec_type); /* offset = a < 0 ? offset : 0.0f */ - offset = LLVMBuildAnd(bld->builder, offset, sign, ""); - offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "ifloor.offset"); + offset = LLVMBuildAnd(builder, offset, sign, ""); + offset = LLVMBuildBitCast(builder, offset, vec_type, "ifloor.offset"); - res = LLVMBuildFAdd(bld->builder, res, offset, "ifloor.res"); + res = LLVMBuildFAdd(builder, res, offset, "ifloor.res"); } } /* round to nearest (toward zero) */ - res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, "ifloor.res"); + res = LLVMBuildFPToSI(builder, res, int_vec_type, "ifloor.res"); return res; } @@ -1372,6 +1406,7 @@ LLVMValueRef lp_build_iceil(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMTypeRef int_vec_type = bld->int_vec_type; LLVMValueRef res; @@ -1389,29 +1424,34 @@ lp_build_iceil(struct lp_build_context *bld, LLVMValueRef offset; /* offset = 0.99999(9)f */ - offset = lp_build_const_vec(type, (double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa)); + offset = lp_build_const_vec(bld->gallivm, type, + (double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa)); if (type.sign) { - LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1)); + LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type, + (unsigned long long)1 << (type.width - 1)); LLVMValueRef sign; /* sign = a < 0 ? 0 : ~0 */ - sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); - sign = LLVMBuildAnd(bld->builder, sign, mask, ""); - sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "iceil.sign"); - sign = LLVMBuildNot(bld->builder, sign, "iceil.not"); + sign = LLVMBuildBitCast(builder, a, int_vec_type, ""); + sign = LLVMBuildAnd(builder, sign, mask, ""); + sign = LLVMBuildAShr(builder, sign, + lp_build_const_int_vec(bld->gallivm, type, + type.width - 1), + "iceil.sign"); + sign = LLVMBuildNot(builder, sign, "iceil.not"); /* offset = a < 0 ? 0.0 : offset */ offset = LLVMConstBitCast(offset, int_vec_type); - offset = LLVMBuildAnd(bld->builder, offset, sign, ""); - offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "iceil.offset"); + offset = LLVMBuildAnd(builder, offset, sign, ""); + offset = LLVMBuildBitCast(builder, offset, vec_type, "iceil.offset"); } - res = LLVMBuildFAdd(bld->builder, a, offset, "iceil.res"); + res = LLVMBuildFAdd(builder, a, offset, "iceil.res"); } /* round to nearest (toward zero) */ - res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, "iceil.res"); + res = LLVMBuildFPToSI(builder, res, int_vec_type, "iceil.res"); return res; } @@ -1429,6 +1469,7 @@ lp_build_ifloor_fract(struct lp_build_context *bld, LLVMValueRef *out_ipart, LLVMValueRef *out_fpart) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef ipart; @@ -1442,8 +1483,8 @@ lp_build_ifloor_fract(struct lp_build_context *bld, */ ipart = lp_build_floor(bld, a); - *out_fpart = LLVMBuildFSub(bld->builder, a, ipart, "fpart"); - *out_ipart = LLVMBuildFPToSI(bld->builder, ipart, bld->int_vec_type, "ipart"); + *out_fpart = LLVMBuildFSub(builder, a, ipart, "fpart"); + *out_ipart = LLVMBuildFPToSI(builder, ipart, bld->int_vec_type, "ipart"); } else { /* @@ -1451,8 +1492,8 @@ lp_build_ifloor_fract(struct lp_build_context *bld, */ *out_ipart = lp_build_ifloor(bld, a); - ipart = LLVMBuildSIToFP(bld->builder, *out_ipart, bld->vec_type, "ipart"); - *out_fpart = LLVMBuildFSub(bld->builder, a, ipart, "fpart"); + ipart = LLVMBuildSIToFP(builder, *out_ipart, bld->vec_type, "ipart"); + *out_fpart = LLVMBuildFSub(builder, a, ipart, "fpart"); } } @@ -1461,8 +1502,9 @@ LLVMValueRef lp_build_sqrt(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); char intrinsic[32]; assert(lp_check_value(type, a)); @@ -1473,7 +1515,7 @@ lp_build_sqrt(struct lp_build_context *bld, assert(type.floating); util_snprintf(intrinsic, sizeof intrinsic, "llvm.sqrt.v%uf%u", type.length, type.width); - return lp_build_intrinsic_unary(bld->builder, intrinsic, vec_type, a); + return lp_build_intrinsic_unary(builder, intrinsic, vec_type, a); } @@ -1496,12 +1538,13 @@ lp_build_rcp_refine(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef rcp_a) { - LLVMValueRef two = lp_build_const_vec(bld->type, 2.0); + LLVMBuilderRef builder = bld->gallivm->builder; + LLVMValueRef two = lp_build_const_vec(bld->gallivm, bld->type, 2.0); LLVMValueRef res; - res = LLVMBuildFMul(bld->builder, a, rcp_a, ""); - res = LLVMBuildFSub(bld->builder, two, res, ""); - res = LLVMBuildFMul(bld->builder, rcp_a, res, ""); + res = LLVMBuildFMul(builder, a, rcp_a, ""); + res = LLVMBuildFSub(builder, two, res, ""); + res = LLVMBuildFMul(builder, rcp_a, res, ""); return res; } @@ -1511,6 +1554,7 @@ LLVMValueRef lp_build_rcp(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(lp_check_value(type, a)); @@ -1545,7 +1589,7 @@ lp_build_rcp(struct lp_build_context *bld, LLVMValueRef res; unsigned i; - res = lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rcp.ps", bld->vec_type, a); + res = lp_build_intrinsic_unary(builder, "llvm.x86.sse.rcp.ps", bld->vec_type, a); for (i = 0; i < num_iterations; ++i) { res = lp_build_rcp_refine(bld, a, res); @@ -1554,7 +1598,7 @@ lp_build_rcp(struct lp_build_context *bld, return res; } - return LLVMBuildFDiv(bld->builder, bld->one, a, ""); + return LLVMBuildFDiv(builder, bld->one, a, ""); } @@ -1571,15 +1615,16 @@ lp_build_rsqrt_refine(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef rsqrt_a) { - LLVMValueRef half = lp_build_const_vec(bld->type, 0.5); - LLVMValueRef three = lp_build_const_vec(bld->type, 3.0); + LLVMBuilderRef builder = bld->gallivm->builder; + LLVMValueRef half = lp_build_const_vec(bld->gallivm, bld->type, 0.5); + LLVMValueRef three = lp_build_const_vec(bld->gallivm, bld->type, 3.0); LLVMValueRef res; - res = LLVMBuildFMul(bld->builder, rsqrt_a, rsqrt_a, ""); - res = LLVMBuildFMul(bld->builder, a, res, ""); - res = LLVMBuildFSub(bld->builder, three, res, ""); - res = LLVMBuildFMul(bld->builder, rsqrt_a, res, ""); - res = LLVMBuildFMul(bld->builder, half, res, ""); + res = LLVMBuildFMul(builder, rsqrt_a, rsqrt_a, ""); + res = LLVMBuildFMul(builder, a, res, ""); + res = LLVMBuildFSub(builder, three, res, ""); + res = LLVMBuildFMul(builder, rsqrt_a, res, ""); + res = LLVMBuildFMul(builder, half, res, ""); return res; } @@ -1592,6 +1637,7 @@ LLVMValueRef lp_build_rsqrt(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(lp_check_value(type, a)); @@ -1603,7 +1649,7 @@ lp_build_rsqrt(struct lp_build_context *bld, LLVMValueRef res; unsigned i; - res = lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rsqrt.ps", bld->vec_type, a); + res = lp_build_intrinsic_unary(builder, "llvm.x86.sse.rsqrt.ps", bld->vec_type, a); for (i = 0; i < num_iterations; ++i) { res = lp_build_rsqrt_refine(bld, a, res); @@ -1617,17 +1663,17 @@ lp_build_rsqrt(struct lp_build_context *bld, static inline LLVMValueRef -lp_build_const_v4si(unsigned long value) +lp_build_const_v4si(struct gallivm_state *gallivm, unsigned long value) { - LLVMValueRef element = LLVMConstInt(LLVMInt32Type(), value, 0); + LLVMValueRef element = lp_build_const_int32(gallivm, value); LLVMValueRef elements[4] = { element, element, element, element }; return LLVMConstVector(elements, 4); } static inline LLVMValueRef -lp_build_const_v4sf(float value) +lp_build_const_v4sf(struct gallivm_state *gallivm, float value) { - LLVMValueRef element = LLVMConstReal(LLVMFloatType(), value); + LLVMValueRef element = lp_build_const_float(gallivm, value); LLVMValueRef elements[4] = { element, element, element, element }; return LLVMConstVector(elements, 4); } @@ -1640,17 +1686,19 @@ LLVMValueRef lp_build_sin(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; + struct gallivm_state *gallivm = bld->gallivm; struct lp_type int_type = lp_int_type(bld->type); - LLVMBuilderRef b = bld->builder; - LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatType(), 4); - LLVMTypeRef v4si = LLVMVectorType(LLVMInt32Type(), 4); + LLVMBuilderRef b = builder; + LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(bld->gallivm->context), 4); + LLVMTypeRef v4si = LLVMVectorType(LLVMInt32TypeInContext(bld->gallivm->context), 4); /* * take the absolute value, * x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask); */ - LLVMValueRef inv_sig_mask = lp_build_const_v4si(~0x80000000); + LLVMValueRef inv_sig_mask = lp_build_const_v4si(bld->gallivm, ~0x80000000); LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, v4si, "a_v4si"); LLVMValueRef absi = LLVMBuildAnd(b, a_v4si, inv_sig_mask, "absi"); @@ -1660,7 +1708,7 @@ lp_build_sin(struct lp_build_context *bld, * extract the sign bit (upper one) * sign_bit = _mm_and_ps(sign_bit, *(v4sf*)_ps_sign_mask); */ - LLVMValueRef sig_mask = lp_build_const_v4si(0x80000000); + LLVMValueRef sig_mask = lp_build_const_v4si(bld->gallivm, 0x80000000); LLVMValueRef sign_bit_i = LLVMBuildAnd(b, a_v4si, sig_mask, "sign_bit_i"); /* @@ -1668,7 +1716,7 @@ lp_build_sin(struct lp_build_context *bld, * y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI); */ - LLVMValueRef FOPi = lp_build_const_v4sf(1.27323954473516); + LLVMValueRef FOPi = lp_build_const_v4sf(gallivm, 1.27323954473516); LLVMValueRef scale_y = LLVMBuildFMul(b, x_abs, FOPi, "scale_y"); /* @@ -1683,12 +1731,12 @@ lp_build_sin(struct lp_build_context *bld, * emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1); */ - LLVMValueRef all_one = lp_build_const_v4si(1); + LLVMValueRef all_one = lp_build_const_v4si(bld->gallivm, 1); LLVMValueRef emm2_add = LLVMBuildAdd(b, emm2_i, all_one, "emm2_add"); /* * emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1); */ - LLVMValueRef inv_one = lp_build_const_v4si(~1); + LLVMValueRef inv_one = lp_build_const_v4si(bld->gallivm, ~1); LLVMValueRef emm2_and = LLVMBuildAnd(b, emm2_add, inv_one, "emm2_and"); /* @@ -1699,13 +1747,13 @@ lp_build_sin(struct lp_build_context *bld, /* get the swap sign flag * emm0 = _mm_and_si128(emm2, *(v4si*)_pi32_4); */ - LLVMValueRef pi32_4 = lp_build_const_v4si(4); + LLVMValueRef pi32_4 = lp_build_const_v4si(bld->gallivm, 4); LLVMValueRef emm0_and = LLVMBuildAnd(b, emm2_add, pi32_4, "emm0_and"); /* * emm2 = _mm_slli_epi32(emm0, 29); */ - LLVMValueRef const_29 = lp_build_const_v4si(29); + LLVMValueRef const_29 = lp_build_const_v4si(bld->gallivm, 29); LLVMValueRef swap_sign_bit = LLVMBuildShl(b, emm0_and, const_29, "swap_sign_bit"); /* @@ -1718,10 +1766,11 @@ lp_build_sin(struct lp_build_context *bld, * emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128()); */ - LLVMValueRef pi32_2 = lp_build_const_v4si(2); + LLVMValueRef pi32_2 = lp_build_const_v4si(bld->gallivm, 2); LLVMValueRef emm2_3 = LLVMBuildAnd(b, emm2_and, pi32_2, "emm2_3"); - LLVMValueRef poly_mask = lp_build_compare(b, int_type, PIPE_FUNC_EQUAL, - emm2_3, lp_build_const_v4si(0)); + LLVMValueRef poly_mask = lp_build_compare(bld->gallivm, + int_type, PIPE_FUNC_EQUAL, + emm2_3, lp_build_const_v4si(bld->gallivm, 0)); /* * sign_bit = _mm_xor_ps(sign_bit, swap_sign_bit); */ @@ -1732,9 +1781,9 @@ lp_build_sin(struct lp_build_context *bld, * _PS_CONST(minus_cephes_DP2, -2.4187564849853515625e-4); * _PS_CONST(minus_cephes_DP3, -3.77489497744594108e-8); */ - LLVMValueRef DP1 = lp_build_const_v4sf(-0.78515625); - LLVMValueRef DP2 = lp_build_const_v4sf(-2.4187564849853515625e-4); - LLVMValueRef DP3 = lp_build_const_v4sf(-3.77489497744594108e-8); + LLVMValueRef DP1 = lp_build_const_v4sf(gallivm, -0.78515625); + LLVMValueRef DP2 = lp_build_const_v4sf(gallivm, -2.4187564849853515625e-4); + LLVMValueRef DP3 = lp_build_const_v4sf(gallivm, -3.77489497744594108e-8); /* * The magic pass: "Extended precision modular arithmetic" @@ -1769,9 +1818,9 @@ lp_build_sin(struct lp_build_context *bld, * _PS_CONST(coscof_p1, -1.388731625493765E-003); * _PS_CONST(coscof_p2, 4.166664568298827E-002); */ - LLVMValueRef coscof_p0 = lp_build_const_v4sf(2.443315711809948E-005); - LLVMValueRef coscof_p1 = lp_build_const_v4sf(-1.388731625493765E-003); - LLVMValueRef coscof_p2 = lp_build_const_v4sf(4.166664568298827E-002); + LLVMValueRef coscof_p0 = lp_build_const_v4sf(gallivm, 2.443315711809948E-005); + LLVMValueRef coscof_p1 = lp_build_const_v4sf(gallivm, -1.388731625493765E-003); + LLVMValueRef coscof_p2 = lp_build_const_v4sf(gallivm, 4.166664568298827E-002); /* * y = *(v4sf*)_ps_coscof_p0; @@ -1790,10 +1839,10 @@ lp_build_sin(struct lp_build_context *bld, * y = _mm_sub_ps(y, tmp); * y = _mm_add_ps(y, *(v4sf*)_ps_1); */ - LLVMValueRef half = lp_build_const_v4sf(0.5); + LLVMValueRef half = lp_build_const_v4sf(gallivm, 0.5); LLVMValueRef tmp = LLVMBuildFMul(b, z, half, "tmp"); LLVMValueRef y_9 = LLVMBuildFSub(b, y_8, tmp, "y_8"); - LLVMValueRef one = lp_build_const_v4sf(1.0); + LLVMValueRef one = lp_build_const_v4sf(gallivm, 1.0); LLVMValueRef y_10 = LLVMBuildFAdd(b, y_9, one, "y_9"); /* @@ -1801,9 +1850,9 @@ lp_build_sin(struct lp_build_context *bld, * _PS_CONST(sincof_p1, 8.3321608736E-3); * _PS_CONST(sincof_p2, -1.6666654611E-1); */ - LLVMValueRef sincof_p0 = lp_build_const_v4sf(-1.9515295891E-4); - LLVMValueRef sincof_p1 = lp_build_const_v4sf(8.3321608736E-3); - LLVMValueRef sincof_p2 = lp_build_const_v4sf(-1.6666654611E-1); + LLVMValueRef sincof_p0 = lp_build_const_v4sf(gallivm, -1.9515295891E-4); + LLVMValueRef sincof_p1 = lp_build_const_v4sf(gallivm, 8.3321608736E-3); + LLVMValueRef sincof_p2 = lp_build_const_v4sf(gallivm, -1.6666654611E-1); /* * Evaluate the second polynom (Pi/4 <= x <= 0) @@ -1836,7 +1885,7 @@ lp_build_sin(struct lp_build_context *bld, LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, v4si, "y2_i"); LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, v4si, "y_i"); LLVMValueRef y2_and = LLVMBuildAnd(b, y2_i, poly_mask, "y2_and"); - LLVMValueRef inv = lp_build_const_v4si(~0); + LLVMValueRef inv = lp_build_const_v4si(bld->gallivm, ~0); LLVMValueRef poly_mask_inv = LLVMBuildXor(b, poly_mask, inv, "poly_mask_inv"); LLVMValueRef y_and = LLVMBuildAnd(b, y_i, poly_mask_inv, "y_and"); LLVMValueRef y_combine = LLVMBuildAdd(b, y_and, y2_and, "y_combine"); @@ -1858,17 +1907,19 @@ LLVMValueRef lp_build_cos(struct lp_build_context *bld, LLVMValueRef a) { + LLVMBuilderRef builder = bld->gallivm->builder; + struct gallivm_state *gallivm = bld->gallivm; struct lp_type int_type = lp_int_type(bld->type); - LLVMBuilderRef b = bld->builder; - LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatType(), 4); - LLVMTypeRef v4si = LLVMVectorType(LLVMInt32Type(), 4); + LLVMBuilderRef b = builder; + LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(bld->gallivm->context), 4); + LLVMTypeRef v4si = LLVMVectorType(LLVMInt32TypeInContext(bld->gallivm->context), 4); /* * take the absolute value, * x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask); */ - LLVMValueRef inv_sig_mask = lp_build_const_v4si(~0x80000000); + LLVMValueRef inv_sig_mask = lp_build_const_v4si(bld->gallivm, ~0x80000000); LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, v4si, "a_v4si"); LLVMValueRef absi = LLVMBuildAnd(b, a_v4si, inv_sig_mask, "absi"); @@ -1879,7 +1930,7 @@ lp_build_cos(struct lp_build_context *bld, * y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI); */ - LLVMValueRef FOPi = lp_build_const_v4sf(1.27323954473516); + LLVMValueRef FOPi = lp_build_const_v4sf(gallivm, 1.27323954473516); LLVMValueRef scale_y = LLVMBuildFMul(b, x_abs, FOPi, "scale_y"); /* @@ -1894,12 +1945,12 @@ lp_build_cos(struct lp_build_context *bld, * emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1); */ - LLVMValueRef all_one = lp_build_const_v4si(1); + LLVMValueRef all_one = lp_build_const_v4si(bld->gallivm, 1); LLVMValueRef emm2_add = LLVMBuildAdd(b, emm2_i, all_one, "emm2_add"); /* * emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1); */ - LLVMValueRef inv_one = lp_build_const_v4si(~1); + LLVMValueRef inv_one = lp_build_const_v4si(bld->gallivm, ~1); LLVMValueRef emm2_and = LLVMBuildAnd(b, emm2_add, inv_one, "emm2_and"); /* @@ -1911,22 +1962,22 @@ lp_build_cos(struct lp_build_context *bld, /* * emm2 = _mm_sub_epi32(emm2, *(v4si*)_pi32_2); */ - LLVMValueRef const_2 = lp_build_const_v4si(2); + LLVMValueRef const_2 = lp_build_const_v4si(bld->gallivm, 2); LLVMValueRef emm2_2 = LLVMBuildSub(b, emm2_and, const_2, "emm2_2"); /* get the swap sign flag * emm0 = _mm_andnot_si128(emm2, *(v4si*)_pi32_4); */ - LLVMValueRef inv = lp_build_const_v4si(~0); + LLVMValueRef inv = lp_build_const_v4si(bld->gallivm, ~0); LLVMValueRef emm0_not = LLVMBuildXor(b, emm2_2, inv, "emm0_not"); - LLVMValueRef pi32_4 = lp_build_const_v4si(4); + LLVMValueRef pi32_4 = lp_build_const_v4si(bld->gallivm, 4); LLVMValueRef emm0_and = LLVMBuildAnd(b, emm0_not, pi32_4, "emm0_and"); /* * emm2 = _mm_slli_epi32(emm0, 29); */ - LLVMValueRef const_29 = lp_build_const_v4si(29); + LLVMValueRef const_29 = lp_build_const_v4si(bld->gallivm, 29); LLVMValueRef sign_bit = LLVMBuildShl(b, emm0_and, const_29, "sign_bit"); /* @@ -1939,19 +1990,20 @@ lp_build_cos(struct lp_build_context *bld, * emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128()); */ - LLVMValueRef pi32_2 = lp_build_const_v4si(2); + LLVMValueRef pi32_2 = lp_build_const_v4si(bld->gallivm, 2); LLVMValueRef emm2_3 = LLVMBuildAnd(b, emm2_2, pi32_2, "emm2_3"); - LLVMValueRef poly_mask = lp_build_compare(b, int_type, PIPE_FUNC_EQUAL, - emm2_3, lp_build_const_v4si(0)); + LLVMValueRef poly_mask = lp_build_compare(bld->gallivm, + int_type, PIPE_FUNC_EQUAL, + emm2_3, lp_build_const_v4si(bld->gallivm, 0)); /* * _PS_CONST(minus_cephes_DP1, -0.78515625); * _PS_CONST(minus_cephes_DP2, -2.4187564849853515625e-4); * _PS_CONST(minus_cephes_DP3, -3.77489497744594108e-8); */ - LLVMValueRef DP1 = lp_build_const_v4sf(-0.78515625); - LLVMValueRef DP2 = lp_build_const_v4sf(-2.4187564849853515625e-4); - LLVMValueRef DP3 = lp_build_const_v4sf(-3.77489497744594108e-8); + LLVMValueRef DP1 = lp_build_const_v4sf(gallivm, -0.78515625); + LLVMValueRef DP2 = lp_build_const_v4sf(gallivm, -2.4187564849853515625e-4); + LLVMValueRef DP3 = lp_build_const_v4sf(gallivm, -3.77489497744594108e-8); /* * The magic pass: "Extended precision modular arithmetic" @@ -1986,9 +2038,9 @@ lp_build_cos(struct lp_build_context *bld, * _PS_CONST(coscof_p1, -1.388731625493765E-003); * _PS_CONST(coscof_p2, 4.166664568298827E-002); */ - LLVMValueRef coscof_p0 = lp_build_const_v4sf(2.443315711809948E-005); - LLVMValueRef coscof_p1 = lp_build_const_v4sf(-1.388731625493765E-003); - LLVMValueRef coscof_p2 = lp_build_const_v4sf(4.166664568298827E-002); + LLVMValueRef coscof_p0 = lp_build_const_v4sf(gallivm, 2.443315711809948E-005); + LLVMValueRef coscof_p1 = lp_build_const_v4sf(gallivm, -1.388731625493765E-003); + LLVMValueRef coscof_p2 = lp_build_const_v4sf(gallivm, 4.166664568298827E-002); /* * y = *(v4sf*)_ps_coscof_p0; @@ -2007,10 +2059,10 @@ lp_build_cos(struct lp_build_context *bld, * y = _mm_sub_ps(y, tmp); * y = _mm_add_ps(y, *(v4sf*)_ps_1); */ - LLVMValueRef half = lp_build_const_v4sf(0.5); + LLVMValueRef half = lp_build_const_v4sf(gallivm, 0.5); LLVMValueRef tmp = LLVMBuildFMul(b, z, half, "tmp"); LLVMValueRef y_9 = LLVMBuildFSub(b, y_8, tmp, "y_8"); - LLVMValueRef one = lp_build_const_v4sf(1.0); + LLVMValueRef one = lp_build_const_v4sf(gallivm, 1.0); LLVMValueRef y_10 = LLVMBuildFAdd(b, y_9, one, "y_9"); /* @@ -2018,9 +2070,9 @@ lp_build_cos(struct lp_build_context *bld, * _PS_CONST(sincof_p1, 8.3321608736E-3); * _PS_CONST(sincof_p2, -1.6666654611E-1); */ - LLVMValueRef sincof_p0 = lp_build_const_v4sf(-1.9515295891E-4); - LLVMValueRef sincof_p1 = lp_build_const_v4sf(8.3321608736E-3); - LLVMValueRef sincof_p2 = lp_build_const_v4sf(-1.6666654611E-1); + LLVMValueRef sincof_p0 = lp_build_const_v4sf(gallivm, -1.9515295891E-4); + LLVMValueRef sincof_p1 = lp_build_const_v4sf(gallivm, 8.3321608736E-3); + LLVMValueRef sincof_p2 = lp_build_const_v4sf(gallivm, -1.6666654611E-1); /* * Evaluate the second polynom (Pi/4 <= x <= 0) @@ -2094,7 +2146,8 @@ lp_build_exp(struct lp_build_context *bld, LLVMValueRef x) { /* log2(e) = 1/log(2) */ - LLVMValueRef log2e = lp_build_const_vec(bld->type, 1.4426950408889634); + LLVMValueRef log2e = lp_build_const_vec(bld->gallivm, bld->type, + 1.4426950408889634); assert(lp_check_value(bld->type, x)); @@ -2110,7 +2163,8 @@ lp_build_log(struct lp_build_context *bld, LLVMValueRef x) { /* log(2) */ - LLVMValueRef log2 = lp_build_const_vec(bld->type, 0.69314718055994529); + LLVMValueRef log2 = lp_build_const_vec(bld->gallivm, bld->type, + 0.69314718055994529); assert(lp_check_value(bld->type, x)); @@ -2144,7 +2198,7 @@ lp_build_polynomial(struct lp_build_context *bld, for (i = num_coeffs; i--; ) { LLVMValueRef coeff; - coeff = lp_build_const_vec(type, coeffs[i]); + coeff = lp_build_const_vec(bld->gallivm, type, coeffs[i]); if(res) res = lp_build_add(bld, coeff, lp_build_mul(bld, x, res)); @@ -2198,9 +2252,10 @@ lp_build_exp2_approx(struct lp_build_context *bld, LLVMValueRef *p_frac_part, LLVMValueRef *p_exp2) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef vec_type = lp_build_vec_type(type); - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type); LLVMValueRef ipart = NULL; LLVMValueRef fpart = NULL; LLVMValueRef expipart = NULL; @@ -2219,29 +2274,31 @@ lp_build_exp2_approx(struct lp_build_context *bld, assert(type.floating && type.width == 32); - x = lp_build_min(bld, x, lp_build_const_vec(type, 129.0)); - x = lp_build_max(bld, x, lp_build_const_vec(type, -126.99999)); + x = lp_build_min(bld, x, lp_build_const_vec(bld->gallivm, type, 129.0)); + x = lp_build_max(bld, x, lp_build_const_vec(bld->gallivm, type, -126.99999)); /* ipart = floor(x) */ ipart = lp_build_floor(bld, x); /* fpart = x - ipart */ - fpart = LLVMBuildFSub(bld->builder, x, ipart, ""); + fpart = LLVMBuildFSub(builder, x, ipart, ""); } if(p_exp2_int_part || p_exp2) { /* expipart = (float) (1 << ipart) */ - ipart = LLVMBuildFPToSI(bld->builder, ipart, int_vec_type, ""); - expipart = LLVMBuildAdd(bld->builder, ipart, lp_build_const_int_vec(type, 127), ""); - expipart = LLVMBuildShl(bld->builder, expipart, lp_build_const_int_vec(type, 23), ""); - expipart = LLVMBuildBitCast(bld->builder, expipart, vec_type, ""); + ipart = LLVMBuildFPToSI(builder, ipart, int_vec_type, ""); + expipart = LLVMBuildAdd(builder, ipart, + lp_build_const_int_vec(bld->gallivm, type, 127), ""); + expipart = LLVMBuildShl(builder, expipart, + lp_build_const_int_vec(bld->gallivm, type, 23), ""); + expipart = LLVMBuildBitCast(builder, expipart, vec_type, ""); } if(p_exp2) { expfpart = lp_build_polynomial(bld, fpart, lp_build_exp2_polynomial, Elements(lp_build_exp2_polynomial)); - res = LLVMBuildFMul(bld->builder, expipart, expfpart, ""); + res = LLVMBuildFMul(builder, expipart, expfpart, ""); } if(p_exp2_int_part) @@ -2279,6 +2336,7 @@ lp_build_extract_exponent(struct lp_build_context *bld, LLVMValueRef x, int bias) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; unsigned mantissa = lp_mantissa(type); LLVMValueRef res; @@ -2287,11 +2345,14 @@ lp_build_extract_exponent(struct lp_build_context *bld, assert(lp_check_value(bld->type, x)); - x = LLVMBuildBitCast(bld->builder, x, bld->int_vec_type, ""); + x = LLVMBuildBitCast(builder, x, bld->int_vec_type, ""); - res = LLVMBuildLShr(bld->builder, x, lp_build_const_int_vec(type, mantissa), ""); - res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(type, 255), ""); - res = LLVMBuildSub(bld->builder, res, lp_build_const_int_vec(type, 127 - bias), ""); + res = LLVMBuildLShr(builder, x, + lp_build_const_int_vec(bld->gallivm, type, mantissa), ""); + res = LLVMBuildAnd(builder, res, + lp_build_const_int_vec(bld->gallivm, type, 255), ""); + res = LLVMBuildSub(builder, res, + lp_build_const_int_vec(bld->gallivm, type, 127 - bias), ""); return res; } @@ -2308,9 +2369,11 @@ LLVMValueRef lp_build_extract_mantissa(struct lp_build_context *bld, LLVMValueRef x) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; unsigned mantissa = lp_mantissa(type); - LLVMValueRef mantmask = lp_build_const_int_vec(type, (1ULL << mantissa) - 1); + LLVMValueRef mantmask = lp_build_const_int_vec(bld->gallivm, type, + (1ULL << mantissa) - 1); LLVMValueRef one = LLVMConstBitCast(bld->one, bld->int_vec_type); LLVMValueRef res; @@ -2318,12 +2381,12 @@ lp_build_extract_mantissa(struct lp_build_context *bld, assert(type.floating); - x = LLVMBuildBitCast(bld->builder, x, bld->int_vec_type, ""); + x = LLVMBuildBitCast(builder, x, bld->int_vec_type, ""); /* res = x / 2**ipart */ - res = LLVMBuildAnd(bld->builder, x, mantmask, ""); - res = LLVMBuildOr(bld->builder, res, one, ""); - res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, ""); + res = LLVMBuildAnd(builder, x, mantmask, ""); + res = LLVMBuildOr(builder, res, one, ""); + res = LLVMBuildBitCast(builder, res, bld->vec_type, ""); return res; } @@ -2374,12 +2437,13 @@ lp_build_log2_approx(struct lp_build_context *bld, LLVMValueRef *p_floor_log2, LLVMValueRef *p_log2) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; - LLVMTypeRef vec_type = lp_build_vec_type(type); - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type); - LLVMValueRef expmask = lp_build_const_int_vec(type, 0x7f800000); - LLVMValueRef mantmask = lp_build_const_int_vec(type, 0x007fffff); + LLVMValueRef expmask = lp_build_const_int_vec(bld->gallivm, type, 0x7f800000); + LLVMValueRef mantmask = lp_build_const_int_vec(bld->gallivm, type, 0x007fffff); LLVMValueRef one = LLVMConstBitCast(bld->one, int_vec_type); LLVMValueRef i = NULL; @@ -2401,35 +2465,35 @@ lp_build_log2_approx(struct lp_build_context *bld, assert(type.floating && type.width == 32); - i = LLVMBuildBitCast(bld->builder, x, int_vec_type, ""); + i = LLVMBuildBitCast(builder, x, int_vec_type, ""); /* exp = (float) exponent(x) */ - exp = LLVMBuildAnd(bld->builder, i, expmask, ""); + exp = LLVMBuildAnd(builder, i, expmask, ""); } if(p_floor_log2 || p_log2) { - logexp = LLVMBuildLShr(bld->builder, exp, lp_build_const_int_vec(type, 23), ""); - logexp = LLVMBuildSub(bld->builder, logexp, lp_build_const_int_vec(type, 127), ""); - logexp = LLVMBuildSIToFP(bld->builder, logexp, vec_type, ""); + logexp = LLVMBuildLShr(builder, exp, lp_build_const_int_vec(bld->gallivm, type, 23), ""); + logexp = LLVMBuildSub(builder, logexp, lp_build_const_int_vec(bld->gallivm, type, 127), ""); + logexp = LLVMBuildSIToFP(builder, logexp, vec_type, ""); } if(p_log2) { /* mant = (float) mantissa(x) */ - mant = LLVMBuildAnd(bld->builder, i, mantmask, ""); - mant = LLVMBuildOr(bld->builder, mant, one, ""); - mant = LLVMBuildBitCast(bld->builder, mant, vec_type, ""); + mant = LLVMBuildAnd(builder, i, mantmask, ""); + mant = LLVMBuildOr(builder, mant, one, ""); + mant = LLVMBuildBitCast(builder, mant, vec_type, ""); logmant = lp_build_polynomial(bld, mant, lp_build_log2_polynomial, Elements(lp_build_log2_polynomial)); /* This effectively increases the polynomial degree by one, but ensures that log2(1) == 0*/ - logmant = LLVMBuildFMul(bld->builder, logmant, LLVMBuildFSub(bld->builder, mant, bld->one, ""), ""); + logmant = LLVMBuildFMul(builder, logmant, LLVMBuildFSub(builder, mant, bld->one, ""), ""); - res = LLVMBuildFAdd(bld->builder, logmant, logexp, ""); + res = LLVMBuildFAdd(builder, logmant, logexp, ""); } if(p_exp) { - exp = LLVMBuildBitCast(bld->builder, exp, vec_type, ""); + exp = LLVMBuildBitCast(builder, exp, vec_type, ""); *p_exp = exp; } @@ -2465,6 +2529,7 @@ LLVMValueRef lp_build_fast_log2(struct lp_build_context *bld, LLVMValueRef x) { + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef ipart; LLVMValueRef fpart; @@ -2474,13 +2539,13 @@ lp_build_fast_log2(struct lp_build_context *bld, /* ipart = floor(log2(x)) - 1 */ ipart = lp_build_extract_exponent(bld, x, -1); - ipart = LLVMBuildSIToFP(bld->builder, ipart, bld->vec_type, ""); + ipart = LLVMBuildSIToFP(builder, ipart, bld->vec_type, ""); /* fpart = x / 2**ipart */ fpart = lp_build_extract_mantissa(bld, x); /* ipart + fpart */ - return LLVMBuildFAdd(bld->builder, ipart, fpart, ""); + return LLVMBuildFAdd(builder, ipart, fpart, ""); } @@ -2493,7 +2558,8 @@ LLVMValueRef lp_build_ilog2(struct lp_build_context *bld, LLVMValueRef x) { - LLVMValueRef sqrt2 = lp_build_const_vec(bld->type, M_SQRT2); + LLVMBuilderRef builder = bld->gallivm->builder; + LLVMValueRef sqrt2 = lp_build_const_vec(bld->gallivm, bld->type, M_SQRT2); LLVMValueRef ipart; assert(bld->type.floating); @@ -2501,7 +2567,7 @@ lp_build_ilog2(struct lp_build_context *bld, assert(lp_check_value(bld->type, x)); /* x * 2^(0.5) i.e., add 0.5 to the log2(x) */ - x = LLVMBuildFMul(bld->builder, x, sqrt2, ""); + x = LLVMBuildFMul(builder, x, sqrt2, ""); /* ipart = floor(log2(x) + 0.5) */ ipart = lp_build_extract_exponent(bld, x, 0); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_assert.c b/src/gallium/auxiliary/gallivm/lp_bld_assert.c index f2ebd868a8..9de5e8e7b5 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_assert.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_assert.c @@ -56,20 +56,21 @@ lp_assert(int condition, const char *msg) * \param msg a string to print if the assertion fails. */ LLVMValueRef -lp_build_assert(LLVMBuilderRef builder, LLVMValueRef condition, +lp_build_assert(struct gallivm_state *gallivm, + LLVMValueRef condition, const char *msg) { - LLVMModuleRef module; + LLVMBuilderRef builder = gallivm->builder; + LLVMContextRef context = gallivm->context; + LLVMModuleRef module = gallivm->module; LLVMTypeRef arg_types[2]; LLVMValueRef msg_string, assert_func, params[2], r; - module = LLVMGetGlobalParent(LLVMGetBasicBlockParent( - LLVMGetInsertBlock(builder))); + msg_string = lp_build_const_string_variable(module, context, + msg, strlen(msg) + 1); - msg_string = lp_build_const_string_variable(module, msg, strlen(msg) + 1); - - arg_types[0] = LLVMInt32Type(); - arg_types[1] = LLVMPointerType(LLVMInt8Type(), 0); + arg_types[0] = LLVMInt32TypeInContext(context); + arg_types[1] = LLVMPointerType(LLVMInt8TypeInContext(context), 0); /* lookup the lp_assert function */ assert_func = LLVMGetNamedFunction(module, "lp_assert"); @@ -77,12 +78,12 @@ lp_build_assert(LLVMBuilderRef builder, LLVMValueRef condition, /* Create the assertion function if not found */ if (!assert_func) { LLVMTypeRef func_type = - LLVMFunctionType(LLVMVoidType(), arg_types, 2, 0); + LLVMFunctionType(LLVMVoidTypeInContext(context), arg_types, 2, 0); assert_func = LLVMAddFunction(module, "lp_assert", func_type); LLVMSetFunctionCallConv(assert_func, LLVMCCallConv); LLVMSetLinkage(assert_func, LLVMExternalLinkage); - LLVMAddGlobalMapping(lp_build_engine, assert_func, + LLVMAddGlobalMapping(gallivm->engine, assert_func, func_to_pointer((func_pointer)lp_assert)); } assert(assert_func); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_assert.h b/src/gallium/auxiliary/gallivm/lp_bld_assert.h index ddd879dc2c..1d2baab30a 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_assert.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_assert.h @@ -30,10 +30,12 @@ #include "lp_bld.h" +#include "lp_bld_init.h" LLVMValueRef -lp_build_assert(LLVMBuilderRef builder, LLVMValueRef condition, +lp_build_assert(struct gallivm_state *gallivm, + LLVMValueRef condition, const char *msg); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_bitarit.c b/src/gallium/auxiliary/gallivm/lp_bld_bitarit.c index 706479b4d5..a9c57d682f 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_bitarit.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_bitarit.c @@ -40,6 +40,7 @@ LLVMValueRef lp_build_or(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef res; @@ -48,14 +49,14 @@ lp_build_or(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) /* can't do bitwise ops on floating-point values */ if (type.floating) { - a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, ""); - b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, ""); + a = LLVMBuildBitCast(builder, a, bld->int_vec_type, ""); + b = LLVMBuildBitCast(builder, b, bld->int_vec_type, ""); } - res = LLVMBuildOr(bld->builder, a, b, ""); + res = LLVMBuildOr(builder, a, b, ""); if (type.floating) { - res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, ""); + res = LLVMBuildBitCast(builder, res, bld->vec_type, ""); } return res; @@ -68,6 +69,7 @@ lp_build_or(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) LLVMValueRef lp_build_and(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef res; @@ -76,14 +78,14 @@ lp_build_and(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) /* can't do bitwise ops on floating-point values */ if (type.floating) { - a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, ""); - b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, ""); + a = LLVMBuildBitCast(builder, a, bld->int_vec_type, ""); + b = LLVMBuildBitCast(builder, b, bld->int_vec_type, ""); } - res = LLVMBuildAnd(bld->builder, a, b, ""); + res = LLVMBuildAnd(builder, a, b, ""); if (type.floating) { - res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, ""); + res = LLVMBuildBitCast(builder, res, bld->vec_type, ""); } return res; @@ -96,6 +98,7 @@ lp_build_and(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) LLVMValueRef lp_build_andnot(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef res; @@ -104,15 +107,15 @@ lp_build_andnot(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) /* can't do bitwise ops on floating-point values */ if (type.floating) { - a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, ""); - b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, ""); + a = LLVMBuildBitCast(builder, a, bld->int_vec_type, ""); + b = LLVMBuildBitCast(builder, b, bld->int_vec_type, ""); } - res = LLVMBuildNot(bld->builder, b, ""); - res = LLVMBuildAnd(bld->builder, a, res, ""); + res = LLVMBuildNot(builder, b, ""); + res = LLVMBuildAnd(builder, a, res, ""); if (type.floating) { - res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, ""); + res = LLVMBuildBitCast(builder, res, bld->vec_type, ""); } return res; @@ -125,6 +128,7 @@ lp_build_andnot(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) LLVMValueRef lp_build_shl(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef res; @@ -133,7 +137,7 @@ lp_build_shl(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) assert(lp_check_value(type, a)); assert(lp_check_value(type, b)); - res = LLVMBuildShl(bld->builder, a, b, ""); + res = LLVMBuildShl(builder, a, b, ""); return res; } @@ -145,6 +149,7 @@ lp_build_shl(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) LLVMValueRef lp_build_shr(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; LLVMValueRef res; @@ -154,9 +159,9 @@ lp_build_shr(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) assert(lp_check_value(type, b)); if (type.sign) { - res = LLVMBuildAShr(bld->builder, a, b, ""); + res = LLVMBuildAShr(builder, a, b, ""); } else { - res = LLVMBuildLShr(bld->builder, a, b, ""); + res = LLVMBuildLShr(builder, a, b, ""); } return res; @@ -169,7 +174,7 @@ lp_build_shr(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) LLVMValueRef lp_build_shl_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm) { - LLVMValueRef b = lp_build_const_int_vec(bld->type, imm); + LLVMValueRef b = lp_build_const_int_vec(bld->gallivm, bld->type, imm); assert(imm <= bld->type.width); return lp_build_shl(bld, a, b); } @@ -181,7 +186,7 @@ lp_build_shl_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm) LLVMValueRef lp_build_shr_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm) { - LLVMValueRef b = lp_build_const_int_vec(bld->type, imm); + LLVMValueRef b = lp_build_const_int_vec(bld->gallivm, bld->type, imm); assert(imm <= bld->type.width); return lp_build_shr(bld, a, b); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.c b/src/gallium/auxiliary/gallivm/lp_bld_const.c index dd839c0bea..6d8b7c26fc 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_const.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_const.c @@ -39,6 +39,7 @@ #include "lp_bld_type.h" #include "lp_bld_const.h" +#include "lp_bld_init.h" unsigned @@ -211,31 +212,31 @@ lp_const_eps(struct lp_type type) LLVMValueRef -lp_build_undef(struct lp_type type) +lp_build_undef(struct gallivm_state *gallivm, struct lp_type type) { - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(gallivm, type); return LLVMGetUndef(vec_type); } LLVMValueRef -lp_build_zero(struct lp_type type) +lp_build_zero(struct gallivm_state *gallivm, struct lp_type type) { if (type.length == 1) { if (type.floating) - return LLVMConstReal(LLVMFloatType(), 0.0); + return lp_build_const_float(gallivm, 0.0); else - return LLVMConstInt(LLVMIntType(type.width), 0, 0); + return LLVMConstInt(LLVMIntTypeInContext(gallivm->context, type.width), 0, 0); } else { - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(gallivm, type); return LLVMConstNull(vec_type); } } LLVMValueRef -lp_build_one(struct lp_type type) +lp_build_one(struct gallivm_state *gallivm, struct lp_type type) { LLVMTypeRef elem_type; LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; @@ -243,7 +244,7 @@ lp_build_one(struct lp_type type) assert(type.length <= LP_MAX_VECTOR_LENGTH); - elem_type = lp_build_elem_type(type); + elem_type = lp_build_elem_type(gallivm, type); if(type.floating) elems[0] = LLVMConstReal(elem_type, 1.0); @@ -283,10 +284,11 @@ lp_build_one(struct lp_type type) * Build constant-valued element from a scalar value. */ LLVMValueRef -lp_build_const_elem(struct lp_type type, +lp_build_const_elem(struct gallivm_state *gallivm, + struct lp_type type, double val) { - LLVMTypeRef elem_type = lp_build_elem_type(type); + LLVMTypeRef elem_type = lp_build_elem_type(gallivm, type); LLVMValueRef elem; if(type.floating) { @@ -306,15 +308,15 @@ lp_build_const_elem(struct lp_type type, * Build constant-valued vector from a scalar value. */ LLVMValueRef -lp_build_const_vec(struct lp_type type, +lp_build_const_vec(struct gallivm_state *gallivm, struct lp_type type, double val) { if (type.length == 1) { - return lp_build_const_elem(type, val); + return lp_build_const_elem(gallivm, type, val); } else { LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; unsigned i; - elems[0] = lp_build_const_elem(type, val); + elems[0] = lp_build_const_elem(gallivm, type, val); for(i = 1; i < type.length; ++i) elems[i] = elems[0]; return LLVMConstVector(elems, type.length); @@ -323,10 +325,10 @@ lp_build_const_vec(struct lp_type type, LLVMValueRef -lp_build_const_int_vec(struct lp_type type, - long long val) +lp_build_const_int_vec(struct gallivm_state *gallivm, struct lp_type type, + long long val) { - LLVMTypeRef elem_type = lp_build_int_elem_type(type); + LLVMTypeRef elem_type = lp_build_int_elem_type(gallivm, type); LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; unsigned i; @@ -343,7 +345,8 @@ lp_build_const_int_vec(struct lp_type type, LLVMValueRef -lp_build_const_aos(struct lp_type type, +lp_build_const_aos(struct gallivm_state *gallivm, + struct lp_type type, double r, double g, double b, double a, const unsigned char *swizzle) { @@ -355,7 +358,7 @@ lp_build_const_aos(struct lp_type type, assert(type.length % 4 == 0); assert(type.length <= LP_MAX_VECTOR_LENGTH); - elem_type = lp_build_elem_type(type); + elem_type = lp_build_elem_type(gallivm, type); if(swizzle == NULL) swizzle = default_swizzle; @@ -386,10 +389,11 @@ lp_build_const_aos(struct lp_type type, * @param mask TGSI_WRITEMASK_xxx */ LLVMValueRef -lp_build_const_mask_aos(struct lp_type type, +lp_build_const_mask_aos(struct gallivm_state *gallivm, + struct lp_type type, unsigned mask) { - LLVMTypeRef elem_type = LLVMIntType(type.width); + LLVMTypeRef elem_type = LLVMIntTypeInContext(gallivm->context, type.width); LLVMValueRef masks[LP_MAX_VECTOR_LENGTH]; unsigned i, j; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.h b/src/gallium/auxiliary/gallivm/lp_bld_const.h index 6b1fc590c1..69718eb4b3 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_const.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_const.h @@ -39,6 +39,7 @@ #include "pipe/p_compiler.h" #include "gallivm/lp_bld.h" +#include "gallivm/lp_bld_init.h" @@ -73,44 +74,71 @@ lp_const_eps(struct lp_type type); LLVMValueRef -lp_build_undef(struct lp_type type); +lp_build_undef(struct gallivm_state *gallivm, struct lp_type type); LLVMValueRef -lp_build_zero(struct lp_type type); +lp_build_zero(struct gallivm_state *gallivm, struct lp_type type); LLVMValueRef -lp_build_one(struct lp_type type); +lp_build_one(struct gallivm_state *gallivm, struct lp_type type); LLVMValueRef -lp_build_const_elem(struct lp_type type, +lp_build_const_elem(struct gallivm_state *gallivm, struct lp_type type, double val); LLVMValueRef -lp_build_const_vec(struct lp_type type, double val); +lp_build_const_vec(struct gallivm_state *gallivm, struct lp_type type, + double val); LLVMValueRef -lp_build_const_int_vec(struct lp_type type, long long val); +lp_build_const_int_vec(struct gallivm_state *gallivm, + struct lp_type type, long long val); LLVMValueRef -lp_build_const_aos(struct lp_type type, +lp_build_const_aos(struct gallivm_state *gallivm, struct lp_type type, double r, double g, double b, double a, const unsigned char *swizzle); LLVMValueRef -lp_build_const_mask_aos(struct lp_type type, +lp_build_const_mask_aos(struct gallivm_state *gallivm, + struct lp_type type, unsigned mask); static INLINE LLVMValueRef -lp_build_const_int32(int i) +lp_build_const_int32(struct gallivm_state *gallivm, int i) { - return LLVMConstInt(LLVMInt32Type(), i, 0); + return LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), i, 0); +} + + +static INLINE LLVMValueRef +lp_build_const_float(struct gallivm_state *gallivm, float x) +{ + return LLVMConstReal(LLVMFloatTypeInContext(gallivm->context), x); +} + + +/** Return constant-valued pointer to int */ +static INLINE LLVMValueRef +lp_build_const_int_pointer(struct gallivm_state *gallivm, const void *ptr) +{ + LLVMTypeRef int_type; + LLVMValueRef v; + + /* int type large enough to hold a pointer */ + int_type = LLVMIntTypeInContext(gallivm->context, 8 * sizeof(void *)); + v = LLVMConstInt(int_type, (uintptr_t) ptr, 0); + v = LLVMBuildIntToPtr(gallivm->builder, v, + LLVMPointerType(int_type, 0), + "cast int to ptr"); + return v; } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.c b/src/gallium/auxiliary/gallivm/lp_bld_conv.c index 6967dd2622..c43ee8ac63 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_conv.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.c @@ -89,12 +89,13 @@ * return { i32, i32, i32, i32 } where each value is in [0, 2^dst_width-1]. */ LLVMValueRef -lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, +lp_build_clamped_float_to_unsigned_norm(struct gallivm_state *gallivm, struct lp_type src_type, unsigned dst_width, LLVMValueRef src) { - LLVMTypeRef int_vec_type = lp_build_int_vec_type(src_type); + LLVMBuilderRef builder = gallivm->builder; + LLVMTypeRef int_vec_type = lp_build_int_vec_type(gallivm, src_type); LLVMValueRef res; unsigned mantissa; @@ -122,10 +123,11 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, scale = (double)mask/ubound; bias = (double)(1ULL << (mantissa - dst_width)); - res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), ""); - res = LLVMBuildFAdd(builder, res, lp_build_const_vec(src_type, bias), ""); + res = LLVMBuildFMul(builder, src, lp_build_const_vec(gallivm, src_type, scale), ""); + res = LLVMBuildFAdd(builder, res, lp_build_const_vec(gallivm, src_type, bias), ""); res = LLVMBuildBitCast(builder, res, int_vec_type, ""); - res = LLVMBuildAnd(builder, res, lp_build_const_int_vec(src_type, mask), ""); + res = LLVMBuildAnd(builder, res, + lp_build_const_int_vec(gallivm, src_type, mask), ""); } else if (dst_width == (mantissa + 1)) { /* @@ -138,7 +140,8 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, scale = (double)((1ULL << dst_width) - 1); - res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), ""); + res = LLVMBuildFMul(builder, src, + lp_build_const_vec(gallivm, src_type, scale), ""); res = LLVMBuildFPToSI(builder, res, int_vec_type, ""); } else { @@ -166,7 +169,8 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, LLVMValueRef lshifted; LLVMValueRef rshifted; - res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), ""); + res = LLVMBuildFMul(builder, src, + lp_build_const_vec(gallivm, src_type, scale), ""); res = LLVMBuildFPToSI(builder, res, int_vec_type, ""); /* @@ -177,7 +181,8 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, */ if (lshift) { lshifted = LLVMBuildShl(builder, res, - lp_build_const_int_vec(src_type, lshift), ""); + lp_build_const_int_vec(gallivm, src_type, + lshift), ""); } else { lshifted = res; } @@ -186,7 +191,8 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, * Align the most significant bit to the right. */ rshifted = LLVMBuildAShr(builder, res, - lp_build_const_int_vec(src_type, rshift), ""); + lp_build_const_int_vec(gallivm, src_type, rshift), + ""); /* * Subtract the MSB to the LSB, therefore re-scaling from @@ -206,13 +212,14 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, * return {float, float, float, float} with values in range [0, 1]. */ LLVMValueRef -lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, +lp_build_unsigned_norm_to_float(struct gallivm_state *gallivm, unsigned src_width, struct lp_type dst_type, LLVMValueRef src) { - LLVMTypeRef vec_type = lp_build_vec_type(dst_type); - LLVMTypeRef int_vec_type = lp_build_int_vec_type(dst_type); + LLVMBuilderRef builder = gallivm->builder; + LLVMTypeRef vec_type = lp_build_vec_type(gallivm, dst_type); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(gallivm, dst_type); LLVMValueRef bias_; LLVMValueRef res; unsigned mantissa; @@ -230,7 +237,8 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, if (src_width == 8) { scale = 1.0/255.0; res = LLVMBuildSIToFP(builder, src, vec_type, ""); - res = LLVMBuildFMul(builder, res, lp_build_const_vec(dst_type, scale), ""); + res = LLVMBuildFMul(builder, res, + lp_build_const_vec(gallivm, dst_type, scale), ""); return res; } @@ -247,10 +255,11 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, if(src_width > mantissa) { int shift = src_width - mantissa; - res = LLVMBuildLShr(builder, res, lp_build_const_int_vec(dst_type, shift), ""); + res = LLVMBuildLShr(builder, res, + lp_build_const_int_vec(gallivm, dst_type, shift), ""); } - bias_ = lp_build_const_vec(dst_type, bias); + bias_ = lp_build_const_vec(gallivm, dst_type, bias); res = LLVMBuildOr(builder, res, @@ -259,7 +268,7 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, res = LLVMBuildBitCast(builder, res, vec_type, ""); res = LLVMBuildFSub(builder, res, bias_, ""); - res = LLVMBuildFMul(builder, res, lp_build_const_vec(dst_type, scale), ""); + res = LLVMBuildFMul(builder, res, lp_build_const_vec(gallivm, dst_type, scale), ""); return res; } @@ -272,12 +281,13 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, * to the lp_type union. */ void -lp_build_conv(LLVMBuilderRef builder, +lp_build_conv(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, const LLVMValueRef *src, unsigned num_srcs, LLVMValueRef *dst, unsigned num_dsts) { + LLVMBuilderRef builder = gallivm->builder; struct lp_type tmp_type; LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH]; unsigned num_tmps; @@ -342,12 +352,12 @@ lp_build_conv(LLVMBuilderRef builder, int32_type.length /= 4; int32_type.sign = 1; - src_vec_type = lp_build_vec_type(src_type); - dst_vec_type = lp_build_vec_type(dst_type); - int16_vec_type = lp_build_vec_type(int16_type); - int32_vec_type = lp_build_vec_type(int32_type); + src_vec_type = lp_build_vec_type(gallivm, src_type); + dst_vec_type = lp_build_vec_type(gallivm, dst_type); + int16_vec_type = lp_build_vec_type(gallivm, int16_type); + int32_vec_type = lp_build_vec_type(gallivm, int32_type); - const_255f = lp_build_const_vec(src_type, 255.0f); + const_255f = lp_build_const_vec(gallivm, src_type, 255.0f); a = LLVMBuildFMul(builder, src[0], const_255f, ""); b = LLVMBuildFMul(builder, src[1], const_255f, ""); @@ -357,14 +367,14 @@ lp_build_conv(LLVMBuilderRef builder, { struct lp_build_context bld; - bld.builder = builder; + bld.gallivm = gallivm; bld.type = src_type; bld.vec_type = src_vec_type; - bld.int_elem_type = lp_build_elem_type(int32_type); + bld.int_elem_type = lp_build_elem_type(gallivm, int32_type); bld.int_vec_type = int32_vec_type; - bld.undef = lp_build_undef(src_type); - bld.zero = lp_build_zero(src_type); - bld.one = lp_build_one(src_type); + bld.undef = lp_build_undef(gallivm, src_type); + bld.zero = lp_build_zero(gallivm, src_type); + bld.one = lp_build_one(gallivm, src_type); src_int0 = lp_build_iround(&bld, a); src_int1 = lp_build_iround(&bld, b); @@ -372,9 +382,9 @@ lp_build_conv(LLVMBuilderRef builder, src_int3 = lp_build_iround(&bld, d); } /* relying on clamping behavior of sse2 intrinsics here */ - lo = lp_build_pack2(builder, int32_type, int16_type, src_int0, src_int1); - hi = lp_build_pack2(builder, int32_type, int16_type, src_int2, src_int3); - dst[i] = lp_build_pack2(builder, int16_type, dst_type, lo, hi); + lo = lp_build_pack2(gallivm, int32_type, int16_type, src_int0, src_int1); + hi = lp_build_pack2(gallivm, int32_type, int16_type, src_int2, src_int3); + dst[i] = lp_build_pack2(gallivm, int16_type, dst_type, lo, hi); } return; } @@ -391,13 +401,13 @@ lp_build_conv(LLVMBuilderRef builder, double dst_max = lp_const_max(dst_type); LLVMValueRef thres; - lp_build_context_init(&bld, builder, tmp_type); + lp_build_context_init(&bld, gallivm, tmp_type); if(src_min < dst_min) { if(dst_min == 0.0) thres = bld.zero; else - thres = lp_build_const_vec(src_type, dst_min); + thres = lp_build_const_vec(gallivm, src_type, dst_min); for(i = 0; i < num_tmps; ++i) tmp[i] = lp_build_max(&bld, tmp[i], thres); } @@ -406,7 +416,7 @@ lp_build_conv(LLVMBuilderRef builder, if(dst_max == 1.0) thres = bld.one; else - thres = lp_build_const_vec(src_type, dst_max); + thres = lp_build_const_vec(gallivm, src_type, dst_max); for(i = 0; i < num_tmps; ++i) tmp[i] = lp_build_min(&bld, tmp[i], thres); } @@ -422,7 +432,7 @@ lp_build_conv(LLVMBuilderRef builder, else if(tmp_type.floating) { if(!dst_type.fixed && !dst_type.sign && dst_type.norm) { for(i = 0; i < num_tmps; ++i) { - tmp[i] = lp_build_clamped_float_to_unsigned_norm(builder, + tmp[i] = lp_build_clamped_float_to_unsigned_norm(gallivm, tmp_type, dst_type.width, tmp[i]); @@ -434,14 +444,14 @@ lp_build_conv(LLVMBuilderRef builder, LLVMTypeRef tmp_vec_type; if (dst_scale != 1.0) { - LLVMValueRef scale = lp_build_const_vec(tmp_type, dst_scale); + LLVMValueRef scale = lp_build_const_vec(gallivm, tmp_type, dst_scale); for(i = 0; i < num_tmps; ++i) tmp[i] = LLVMBuildFMul(builder, tmp[i], scale, ""); } /* Use an equally sized integer for intermediate computations */ tmp_type.floating = FALSE; - tmp_vec_type = lp_build_vec_type(tmp_type); + tmp_vec_type = lp_build_vec_type(gallivm, tmp_type); for(i = 0; i < num_tmps; ++i) { #if 0 if(dst_type.sign) @@ -461,7 +471,8 @@ lp_build_conv(LLVMBuilderRef builder, /* FIXME: compensate different offsets too */ if(src_shift > dst_shift) { - LLVMValueRef shift = lp_build_const_int_vec(tmp_type, src_shift - dst_shift); + LLVMValueRef shift = lp_build_const_int_vec(gallivm, tmp_type, + src_shift - dst_shift); for(i = 0; i < num_tmps; ++i) if(src_type.sign) tmp[i] = LLVMBuildAShr(builder, tmp[i], shift, ""); @@ -485,7 +496,7 @@ lp_build_conv(LLVMBuilderRef builder, new_type.width = dst_type.width; new_type.length = dst_type.length; - lp_build_resize(builder, tmp_type, new_type, tmp, num_srcs, tmp, num_dsts); + lp_build_resize(gallivm, tmp_type, new_type, tmp, num_srcs, tmp, num_dsts); tmp_type = new_type; num_tmps = num_dsts; @@ -501,7 +512,7 @@ lp_build_conv(LLVMBuilderRef builder, else if(!src_type.floating && dst_type.floating) { if(!src_type.fixed && !src_type.sign && src_type.norm) { for(i = 0; i < num_tmps; ++i) { - tmp[i] = lp_build_unsigned_norm_to_float(builder, + tmp[i] = lp_build_unsigned_norm_to_float(gallivm, src_type.width, dst_type, tmp[i]); @@ -515,7 +526,7 @@ lp_build_conv(LLVMBuilderRef builder, /* Use an equally sized integer for intermediate computations */ tmp_type.floating = TRUE; tmp_type.sign = TRUE; - tmp_vec_type = lp_build_vec_type(tmp_type); + tmp_vec_type = lp_build_vec_type(gallivm, tmp_type); for(i = 0; i < num_tmps; ++i) { #if 0 if(dst_type.sign) @@ -529,7 +540,7 @@ lp_build_conv(LLVMBuilderRef builder, } if (src_scale != 1.0) { - LLVMValueRef scale = lp_build_const_vec(tmp_type, 1.0/src_scale); + LLVMValueRef scale = lp_build_const_vec(gallivm, tmp_type, 1.0/src_scale); for(i = 0; i < num_tmps; ++i) tmp[i] = LLVMBuildFMul(builder, tmp[i], scale, ""); } @@ -541,7 +552,7 @@ lp_build_conv(LLVMBuilderRef builder, /* FIXME: compensate different offsets too */ if(src_shift < dst_shift) { - LLVMValueRef shift = lp_build_const_int_vec(tmp_type, dst_shift - src_shift); + LLVMValueRef shift = lp_build_const_int_vec(gallivm, tmp_type, dst_shift - src_shift); for(i = 0; i < num_tmps; ++i) tmp[i] = LLVMBuildShl(builder, tmp[i], shift, ""); } @@ -565,7 +576,7 @@ lp_build_conv(LLVMBuilderRef builder, * This is basically a very trimmed down version of lp_build_conv. */ void -lp_build_conv_mask(LLVMBuilderRef builder, +lp_build_conv_mask(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, const LLVMValueRef *src, unsigned num_srcs, @@ -599,11 +610,11 @@ lp_build_conv_mask(LLVMBuilderRef builder, if(src_type.width > dst_type.width) { assert(num_dsts == 1); - dst[0] = lp_build_pack(builder, src_type, dst_type, TRUE, src, num_srcs); + dst[0] = lp_build_pack(gallivm, src_type, dst_type, TRUE, src, num_srcs); } else if(src_type.width < dst_type.width) { assert(num_srcs == 1); - lp_build_unpack(builder, src_type, dst_type, src[0], dst, num_dsts); + lp_build_unpack(gallivm, src_type, dst_type, src[0], dst, num_dsts); } else { assert(num_srcs == num_dsts); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.h b/src/gallium/auxiliary/gallivm/lp_bld_conv.h index 628831c3ad..cec655980f 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_conv.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.h @@ -44,27 +44,27 @@ struct lp_type; LLVMValueRef -lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, +lp_build_clamped_float_to_unsigned_norm(struct gallivm_state *gallivm, struct lp_type src_type, unsigned dst_width, LLVMValueRef src); LLVMValueRef -lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, +lp_build_unsigned_norm_to_float(struct gallivm_state *gallivm, unsigned src_width, struct lp_type dst_type, LLVMValueRef src); void -lp_build_conv(LLVMBuilderRef builder, +lp_build_conv(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, const LLVMValueRef *srcs, unsigned num_srcs, LLVMValueRef *dsts, unsigned num_dsts); void -lp_build_conv_mask(LLVMBuilderRef builder, +lp_build_conv_mask(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, const LLVMValueRef *src, unsigned num_srcs, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_debug.h b/src/gallium/auxiliary/gallivm/lp_bld_debug.h index eb11dcd4ef..8a58f95b78 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_debug.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_debug.h @@ -42,6 +42,7 @@ #define GALLIVM_DEBUG_NO_OPT (1 << 3) #define GALLIVM_DEBUG_PERF (1 << 4) #define GALLIVM_DEBUG_NO_BRILINEAR (1 << 5) +#define GALLIVM_DEBUG_GC (1 << 6) #ifdef DEBUG diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.c b/src/gallium/auxiliary/gallivm/lp_bld_flow.c index a2cee199a0..a9c9c7af10 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_flow.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.c @@ -34,6 +34,7 @@ #include "util/u_debug.h" #include "util/u_memory.h" +#include "lp_bld_init.h" #include "lp_bld_type.h" #include "lp_bld_flow.h" @@ -51,25 +52,25 @@ * be used elsewhere. */ LLVMBasicBlockRef -lp_build_insert_new_block(LLVMBuilderRef builder, const char *name) +lp_build_insert_new_block(struct gallivm_state *gallivm, const char *name) { LLVMBasicBlockRef current_block; LLVMBasicBlockRef next_block; LLVMBasicBlockRef new_block; /* get current basic block */ - current_block = LLVMGetInsertBlock(builder); + current_block = LLVMGetInsertBlock(gallivm->builder); /* check if there's another block after this one */ next_block = LLVMGetNextBasicBlock(current_block); if (next_block) { /* insert the new block before the next block */ - new_block = LLVMInsertBasicBlock(next_block, name); + new_block = LLVMInsertBasicBlockInContext(gallivm->context, next_block, name); } else { /* append new block after current block */ LLVMValueRef function = LLVMGetBasicBlockParent(current_block); - new_block = LLVMAppendBasicBlock(function, name); + new_block = LLVMAppendBasicBlockInContext(gallivm->context, function, name); } return new_block; @@ -82,12 +83,11 @@ lp_build_insert_new_block(LLVMBuilderRef builder, const char *name) */ void lp_build_flow_skip_begin(struct lp_build_skip_context *skip, - LLVMBuilderRef builder) + struct gallivm_state *gallivm) { - skip->builder = builder; - + skip->gallivm = gallivm; /* create new basic block */ - skip->block = lp_build_insert_new_block(skip->builder, "skip"); + skip->block = lp_build_insert_new_block(gallivm, "skip"); } @@ -101,12 +101,12 @@ lp_build_flow_skip_cond_break(struct lp_build_skip_context *skip, { LLVMBasicBlockRef new_block; - new_block = lp_build_insert_new_block(skip->builder, ""); + new_block = lp_build_insert_new_block(skip->gallivm, ""); /* if cond is true, goto skip->block, else goto new_block */ - LLVMBuildCondBr(skip->builder, cond, skip->block, new_block); + LLVMBuildCondBr(skip->gallivm->builder, cond, skip->block, new_block); - LLVMPositionBuilderAtEnd(skip->builder, new_block); + LLVMPositionBuilderAtEnd(skip->gallivm->builder, new_block); } @@ -114,8 +114,8 @@ void lp_build_flow_skip_end(struct lp_build_skip_context *skip) { /* goto block */ - LLVMBuildBr(skip->builder, skip->block); - LLVMPositionBuilderAtEnd(skip->builder, skip->block); + LLVMBuildBr(skip->gallivm->builder, skip->block); + LLVMPositionBuilderAtEnd(skip->gallivm->builder, skip->block); } @@ -125,7 +125,7 @@ lp_build_flow_skip_end(struct lp_build_skip_context *skip) void lp_build_mask_check(struct lp_build_mask_context *mask) { - LLVMBuilderRef builder = mask->skip.builder; + LLVMBuilderRef builder = mask->skip.gallivm->builder; LLVMValueRef value; LLVMValueRef cond; @@ -152,27 +152,27 @@ lp_build_mask_check(struct lp_build_mask_context *mask) */ void lp_build_mask_begin(struct lp_build_mask_context *mask, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, struct lp_type type, LLVMValueRef value) { memset(mask, 0, sizeof *mask); - mask->reg_type = LLVMIntType(type.width * type.length); - mask->var = lp_build_alloca(builder, - lp_build_int_vec_type(type), + mask->reg_type = LLVMIntTypeInContext(gallivm->context, type.width * type.length); + mask->var = lp_build_alloca(gallivm, + lp_build_int_vec_type(gallivm, type), "execution_mask"); - LLVMBuildStore(builder, value, mask->var); + LLVMBuildStore(gallivm->builder, value, mask->var); - lp_build_flow_skip_begin(&mask->skip, builder); + lp_build_flow_skip_begin(&mask->skip, gallivm); } LLVMValueRef lp_build_mask_value(struct lp_build_mask_context *mask) { - return LLVMBuildLoad(mask->skip.builder, mask->var, ""); + return LLVMBuildLoad(mask->skip.gallivm->builder, mask->var, ""); } @@ -185,10 +185,10 @@ void lp_build_mask_update(struct lp_build_mask_context *mask, LLVMValueRef value) { - value = LLVMBuildAnd(mask->skip.builder, + value = LLVMBuildAnd(mask->skip.gallivm->builder, lp_build_mask_value(mask), value, ""); - LLVMBuildStore(mask->skip.builder, value, mask->var); + LLVMBuildStore(mask->skip.gallivm->builder, value, mask->var); } @@ -205,13 +205,17 @@ lp_build_mask_end(struct lp_build_mask_context *mask) void -lp_build_loop_begin(LLVMBuilderRef builder, - LLVMValueRef start, - struct lp_build_loop_state *state) +lp_build_loop_begin(struct lp_build_loop_state *state, + struct gallivm_state *gallivm, + LLVMValueRef start) + { - state->block = lp_build_insert_new_block(builder, "loop_begin"); + LLVMBuilderRef builder = gallivm->builder; + + state->block = lp_build_insert_new_block(gallivm, "loop_begin"); - state->counter_var = lp_build_alloca(builder, LLVMTypeOf(start), "loop_counter"); + state->counter_var = lp_build_alloca(gallivm, LLVMTypeOf(start), "loop_counter"); + state->gallivm = gallivm; LLVMBuildStore(builder, start, state->counter_var); @@ -224,12 +228,12 @@ lp_build_loop_begin(LLVMBuilderRef builder, void -lp_build_loop_end_cond(LLVMBuilderRef builder, +lp_build_loop_end_cond(struct lp_build_loop_state *state, LLVMValueRef end, LLVMValueRef step, - LLVMIntPredicate llvm_cond, - struct lp_build_loop_state *state) + LLVMIntPredicate llvm_cond) { + LLVMBuilderRef builder = state->gallivm->builder; LLVMValueRef next; LLVMValueRef cond; LLVMBasicBlockRef after_block; @@ -243,7 +247,7 @@ lp_build_loop_end_cond(LLVMBuilderRef builder, cond = LLVMBuildICmp(builder, llvm_cond, next, end, ""); - after_block = lp_build_insert_new_block(builder, "loop_end"); + after_block = lp_build_insert_new_block(state->gallivm, "loop_end"); LLVMBuildCondBr(builder, cond, after_block, state->block); @@ -254,12 +258,11 @@ lp_build_loop_end_cond(LLVMBuilderRef builder, void -lp_build_loop_end(LLVMBuilderRef builder, +lp_build_loop_end(struct lp_build_loop_state *state, LLVMValueRef end, - LLVMValueRef step, - struct lp_build_loop_state *state) + LLVMValueRef step) { - lp_build_loop_end_cond(builder, end, step, LLVMIntNE, state); + lp_build_loop_end_cond(state, end, step, LLVMIntNE); } @@ -296,24 +299,27 @@ lp_build_loop_end(LLVMBuilderRef builder, */ void lp_build_if(struct lp_build_if_state *ifthen, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, LLVMValueRef condition) { - LLVMBasicBlockRef block = LLVMGetInsertBlock(builder); + LLVMBasicBlockRef block = LLVMGetInsertBlock(gallivm->builder); memset(ifthen, 0, sizeof *ifthen); - ifthen->builder = builder; + ifthen->gallivm = gallivm; ifthen->condition = condition; ifthen->entry_block = block; /* create endif/merge basic block for the phi functions */ - ifthen->merge_block = lp_build_insert_new_block(builder, "endif-block"); + ifthen->merge_block = lp_build_insert_new_block(gallivm, "endif-block"); /* create/insert true_block before merge_block */ - ifthen->true_block = LLVMInsertBasicBlock(ifthen->merge_block, "if-true-block"); + ifthen->true_block = + LLVMInsertBasicBlockInContext(gallivm->context, + ifthen->merge_block, + "if-true-block"); /* successive code goes into the true block */ - LLVMPositionBuilderAtEnd(builder, ifthen->true_block); + LLVMPositionBuilderAtEnd(gallivm->builder, ifthen->true_block); } @@ -323,14 +329,19 @@ lp_build_if(struct lp_build_if_state *ifthen, void lp_build_else(struct lp_build_if_state *ifthen) { + LLVMBuilderRef builder = ifthen->gallivm->builder; + /* Append an unconditional Br(anch) instruction on the true_block */ - LLVMBuildBr(ifthen->builder, ifthen->merge_block); + LLVMBuildBr(builder, ifthen->merge_block); /* create/insert false_block before the merge block */ - ifthen->false_block = LLVMInsertBasicBlock(ifthen->merge_block, "if-false-block"); + ifthen->false_block = + LLVMInsertBasicBlockInContext(ifthen->gallivm->context, + ifthen->merge_block, + "if-false-block"); /* successive code goes into the else block */ - LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->false_block); + LLVMPositionBuilderAtEnd(builder, ifthen->false_block); } @@ -340,28 +351,30 @@ lp_build_else(struct lp_build_if_state *ifthen) void lp_build_endif(struct lp_build_if_state *ifthen) { + LLVMBuilderRef builder = ifthen->gallivm->builder; + /* Insert branch to the merge block from current block */ - LLVMBuildBr(ifthen->builder, ifthen->merge_block); + LLVMBuildBr(builder, ifthen->merge_block); /* * Now patch in the various branch instructions. */ /* Insert the conditional branch instruction at the end of entry_block */ - LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->entry_block); + LLVMPositionBuilderAtEnd(builder, ifthen->entry_block); if (ifthen->false_block) { /* we have an else clause */ - LLVMBuildCondBr(ifthen->builder, ifthen->condition, + LLVMBuildCondBr(builder, ifthen->condition, ifthen->true_block, ifthen->false_block); } else { /* no else clause */ - LLVMBuildCondBr(ifthen->builder, ifthen->condition, + LLVMBuildCondBr(builder, ifthen->condition, ifthen->true_block, ifthen->merge_block); } /* Resume building code at end of the ifthen->merge_block */ - LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->merge_block); + LLVMPositionBuilderAtEnd(builder, ifthen->merge_block); } @@ -381,15 +394,16 @@ lp_build_endif(struct lp_build_if_state *ifthen) * - http://www.llvm.org/docs/tutorial/OCamlLangImpl7.html#memory */ LLVMValueRef -lp_build_alloca(LLVMBuilderRef builder, +lp_build_alloca(struct gallivm_state *gallivm, LLVMTypeRef type, const char *name) { + LLVMBuilderRef builder = gallivm->builder; LLVMBasicBlockRef current_block = LLVMGetInsertBlock(builder); LLVMValueRef function = LLVMGetBasicBlockParent(current_block); LLVMBasicBlockRef first_block = LLVMGetEntryBasicBlock(function); LLVMValueRef first_instr = LLVMGetFirstInstruction(first_block); - LLVMBuilderRef first_builder = LLVMCreateBuilder(); + LLVMBuilderRef first_builder = LLVMCreateBuilderInContext(gallivm->context); LLVMValueRef res; if (first_instr) { @@ -422,16 +436,17 @@ lp_build_alloca(LLVMBuilderRef builder, * - http://www.llvm.org/docs/tutorial/OCamlLangImpl7.html#memory */ LLVMValueRef -lp_build_array_alloca(LLVMBuilderRef builder, +lp_build_array_alloca(struct gallivm_state *gallivm, LLVMTypeRef type, LLVMValueRef count, const char *name) { + LLVMBuilderRef builder = gallivm->builder; LLVMBasicBlockRef current_block = LLVMGetInsertBlock(builder); LLVMValueRef function = LLVMGetBasicBlockParent(current_block); LLVMBasicBlockRef first_block = LLVMGetEntryBasicBlock(function); LLVMValueRef first_instr = LLVMGetFirstInstruction(first_block); - LLVMBuilderRef first_builder = LLVMCreateBuilder(); + LLVMBuilderRef first_builder = LLVMCreateBuilderInContext(gallivm->context); LLVMValueRef res; if (first_instr) { diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.h b/src/gallium/auxiliary/gallivm/lp_bld_flow.h index e729ee6eaa..3cd5a9f42a 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_flow.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.h @@ -47,7 +47,7 @@ struct lp_type; */ struct lp_build_skip_context { - LLVMBuilderRef builder; + struct gallivm_state *gallivm; /** Block to skip to */ LLVMBasicBlockRef block; @@ -55,7 +55,7 @@ struct lp_build_skip_context void lp_build_flow_skip_begin(struct lp_build_skip_context *ctx, - LLVMBuilderRef builder); + struct gallivm_state *gallivm); void lp_build_flow_skip_cond_break(struct lp_build_skip_context *ctx, @@ -77,7 +77,7 @@ struct lp_build_mask_context void lp_build_mask_begin(struct lp_build_mask_context *mask, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, struct lp_type type, LLVMValueRef value); @@ -107,31 +107,28 @@ lp_build_mask_end(struct lp_build_mask_context *mask); */ struct lp_build_loop_state { - LLVMBasicBlockRef block; - LLVMValueRef counter_var; - LLVMValueRef counter; + LLVMBasicBlockRef block; + LLVMValueRef counter_var; + LLVMValueRef counter; + struct gallivm_state *gallivm; }; void -lp_build_loop_begin(LLVMBuilderRef builder, - LLVMValueRef start, - struct lp_build_loop_state *state); - +lp_build_loop_begin(struct lp_build_loop_state *state, + struct gallivm_state *gallivm, + LLVMValueRef start); void -lp_build_loop_end(LLVMBuilderRef builder, +lp_build_loop_end(struct lp_build_loop_state *state, LLVMValueRef end, - LLVMValueRef step, - struct lp_build_loop_state *state); + LLVMValueRef step); void -lp_build_loop_end_cond(LLVMBuilderRef builder, +lp_build_loop_end_cond(struct lp_build_loop_state *state, LLVMValueRef end, LLVMValueRef step, - LLVMIntPredicate cond, - struct lp_build_loop_state *state); - + LLVMIntPredicate cond); @@ -140,7 +137,7 @@ lp_build_loop_end_cond(LLVMBuilderRef builder, */ struct lp_build_if_state { - LLVMBuilderRef builder; + struct gallivm_state *gallivm; LLVMValueRef condition; LLVMBasicBlockRef entry_block; LLVMBasicBlockRef true_block; @@ -151,7 +148,7 @@ struct lp_build_if_state void lp_build_if(struct lp_build_if_state *ctx, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, LLVMValueRef condition); void @@ -161,15 +158,15 @@ void lp_build_endif(struct lp_build_if_state *ctx); LLVMBasicBlockRef -lp_build_insert_new_block(LLVMBuilderRef builder, const char *name); +lp_build_insert_new_block(struct gallivm_state *gallivm, const char *name); LLVMValueRef -lp_build_alloca(LLVMBuilderRef builder, +lp_build_alloca(struct gallivm_state *gallivm, LLVMTypeRef type, const char *name); LLVMValueRef -lp_build_array_alloca(LLVMBuilderRef builder, +lp_build_array_alloca(struct gallivm_state *gallivm, LLVMTypeRef type, LLVMValueRef count, const char *name); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format.h b/src/gallium/auxiliary/gallivm/lp_bld_format.h index 60e22d727a..04142d905b 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_format.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_format.h @@ -35,6 +35,7 @@ */ #include "gallivm/lp_bld.h" +#include "gallivm/lp_bld_init.h" #include "pipe/p_format.h" @@ -53,12 +54,12 @@ lp_build_format_swizzle_aos(const struct util_format_description *desc, LLVMValueRef unswizzled); LLVMValueRef -lp_build_pack_rgba_aos(LLVMBuilderRef builder, +lp_build_pack_rgba_aos(struct gallivm_state *gallivm, const struct util_format_description *desc, LLVMValueRef rgba); LLVMValueRef -lp_build_fetch_rgba_aos(LLVMBuilderRef builder, +lp_build_fetch_rgba_aos(struct gallivm_state *gallivm, const struct util_format_description *format_desc, struct lp_type type, LLVMValueRef base_ptr, @@ -78,20 +79,20 @@ lp_build_format_swizzle_soa(const struct util_format_description *format_desc, LLVMValueRef swizzled_out[4]); void -lp_build_unpack_rgba_soa(LLVMBuilderRef builder, +lp_build_unpack_rgba_soa(struct gallivm_state *gallivm, const struct util_format_description *format_desc, struct lp_type type, LLVMValueRef packed, LLVMValueRef rgba_out[4]); void -lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, +lp_build_rgba8_to_f32_soa(struct gallivm_state *gallivm, struct lp_type dst_type, LLVMValueRef packed, LLVMValueRef *rgba); void -lp_build_fetch_rgba_soa(LLVMBuilderRef builder, +lp_build_fetch_rgba_soa(struct gallivm_state *gallivm, const struct util_format_description *format_desc, struct lp_type type, LLVMValueRef base_ptr, @@ -106,7 +107,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, LLVMValueRef -lp_build_fetch_subsampled_rgba_aos(LLVMBuilderRef builder, +lp_build_fetch_subsampled_rgba_aos(struct gallivm_state *gallivm, const struct util_format_description *format_desc, unsigned n, LLVMValueRef base_ptr, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c index 6b9189e1da..82ab19eda1 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c @@ -36,6 +36,7 @@ #include "util/u_format.h" #include "util/u_memory.h" #include "util/u_math.h" +#include "util/u_pointer.h" #include "util/u_string.h" #include "lp_bld_arit.h" @@ -145,10 +146,11 @@ format_matches_type(const struct util_format_description *desc, * @return RGBA in a float[4] or ubyte[4] or ushort[4] vector. */ static INLINE LLVMValueRef -lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder, +lp_build_unpack_arith_rgba_aos(struct gallivm_state *gallivm, const struct util_format_description *desc, LLVMValueRef packed) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef shifted, casted, scaled, masked; LLVMValueRef shifts[4]; LLVMValueRef masks[4]; @@ -167,21 +169,21 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder, /* Do the intermediate integer computations with 32bit integers since it * matches floating point size */ - assert (LLVMTypeOf(packed) == LLVMInt32Type()); + assert (LLVMTypeOf(packed) == LLVMInt32TypeInContext(gallivm->context)); /* Broadcast the packed value to all four channels * before: packed = BGRA * after: packed = {BGRA, BGRA, BGRA, BGRA} */ packed = LLVMBuildInsertElement(builder, - LLVMGetUndef(LLVMVectorType(LLVMInt32Type(), 4)), + LLVMGetUndef(LLVMVectorType(LLVMInt32TypeInContext(gallivm->context), 4)), packed, - LLVMConstNull(LLVMInt32Type()), + LLVMConstNull(LLVMInt32TypeInContext(gallivm->context)), ""); packed = LLVMBuildShuffleVector(builder, packed, - LLVMGetUndef(LLVMVectorType(LLVMInt32Type(), 4)), - LLVMConstNull(LLVMVectorType(LLVMInt32Type(), 4)), + LLVMGetUndef(LLVMVectorType(LLVMInt32TypeInContext(gallivm->context), 4)), + LLVMConstNull(LLVMVectorType(LLVMInt32TypeInContext(gallivm->context), 4)), ""); /* Initialize vector constants */ @@ -194,9 +196,9 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder, unsigned bits = desc->channel[i].size; if (desc->channel[i].type == UTIL_FORMAT_TYPE_VOID) { - shifts[i] = LLVMGetUndef(LLVMInt32Type()); - masks[i] = LLVMConstNull(LLVMInt32Type()); - scales[i] = LLVMConstNull(LLVMFloatType()); + shifts[i] = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context)); + masks[i] = LLVMConstNull(LLVMInt32TypeInContext(gallivm->context)); + scales[i] = LLVMConstNull(LLVMFloatTypeInContext(gallivm->context)); } else { unsigned long long mask = (1ULL << bits) - 1; @@ -207,15 +209,15 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder, needs_uitofp = TRUE; } - shifts[i] = LLVMConstInt(LLVMInt32Type(), shift, 0); - masks[i] = LLVMConstInt(LLVMInt32Type(), mask, 0); + shifts[i] = lp_build_const_int32(gallivm, shift); + masks[i] = lp_build_const_int32(gallivm, mask); if (desc->channel[i].normalized) { - scales[i] = LLVMConstReal(LLVMFloatType(), 1.0/mask); + scales[i] = lp_build_const_float(gallivm, 1.0 / mask); normalized = TRUE; } else - scales[i] = LLVMConstReal(LLVMFloatType(), 1.0); + scales[i] = lp_build_const_float(gallivm, 1.0); } shift += bits; @@ -230,9 +232,9 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder, if (!needs_uitofp) { /* UIToFP can't be expressed in SSE2 */ - casted = LLVMBuildSIToFP(builder, masked, LLVMVectorType(LLVMFloatType(), 4), ""); + casted = LLVMBuildSIToFP(builder, masked, LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4), ""); } else { - casted = LLVMBuildUIToFP(builder, masked, LLVMVectorType(LLVMFloatType(), 4), ""); + casted = LLVMBuildUIToFP(builder, masked, LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4), ""); } /* At this point 'casted' may be a vector of floats such as @@ -258,10 +260,11 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder, * a time is rarely if ever needed. */ LLVMValueRef -lp_build_pack_rgba_aos(LLVMBuilderRef builder, +lp_build_pack_rgba_aos(struct gallivm_state *gallivm, const struct util_format_description *desc, LLVMValueRef rgba) { + LLVMBuilderRef builder = gallivm->builder; LLVMTypeRef type; LLVMValueRef packed = NULL; LLVMValueRef swizzles[4]; @@ -276,7 +279,7 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder, assert(desc->block.width == 1); assert(desc->block.height == 1); - type = LLVMIntType(desc->block.bits); + type = LLVMIntTypeInContext(gallivm->context, desc->block.bits); /* Unswizzle the color components into the source vector. */ for (i = 0; i < 4; ++i) { @@ -285,13 +288,13 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder, break; } if (j < 4) - swizzles[i] = LLVMConstInt(LLVMInt32Type(), j, 0); + swizzles[i] = lp_build_const_int32(gallivm, j); else - swizzles[i] = LLVMGetUndef(LLVMInt32Type()); + swizzles[i] = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context)); } unswizzled = LLVMBuildShuffleVector(builder, rgba, - LLVMGetUndef(LLVMVectorType(LLVMFloatType(), 4)), + LLVMGetUndef(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4)), LLVMConstVector(swizzles, 4), ""); normalized = FALSE; @@ -300,8 +303,8 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder, unsigned bits = desc->channel[i].size; if (desc->channel[i].type == UTIL_FORMAT_TYPE_VOID) { - shifts[i] = LLVMGetUndef(LLVMInt32Type()); - scales[i] = LLVMGetUndef(LLVMFloatType()); + shifts[i] = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context)); + scales[i] = LLVMGetUndef(LLVMFloatTypeInContext(gallivm->context)); } else { unsigned mask = (1 << bits) - 1; @@ -309,14 +312,14 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder, assert(desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED); assert(bits < 32); - shifts[i] = LLVMConstInt(LLVMInt32Type(), shift, 0); + shifts[i] = lp_build_const_int32(gallivm, shift); if (desc->channel[i].normalized) { - scales[i] = LLVMConstReal(LLVMFloatType(), mask); + scales[i] = lp_build_const_float(gallivm, mask); normalized = TRUE; } else - scales[i] = LLVMConstReal(LLVMFloatType(), 1.0); + scales[i] = lp_build_const_float(gallivm, 1.0); } shift += bits; @@ -327,14 +330,15 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder, else scaled = unswizzled; - casted = LLVMBuildFPToSI(builder, scaled, LLVMVectorType(LLVMInt32Type(), 4), ""); + casted = LLVMBuildFPToSI(builder, scaled, LLVMVectorType(LLVMInt32TypeInContext(gallivm->context), 4), ""); shifted = LLVMBuildShl(builder, casted, LLVMConstVector(shifts, 4), ""); /* Bitwise or all components */ for (i = 0; i < 4; ++i) { if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) { - LLVMValueRef component = LLVMBuildExtractElement(builder, shifted, LLVMConstInt(LLVMInt32Type(), i, 0), ""); + LLVMValueRef component = LLVMBuildExtractElement(builder, shifted, + lp_build_const_int32(gallivm, i), ""); if (packed) packed = LLVMBuildOr(builder, packed, component, ""); else @@ -343,7 +347,7 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder, } if (!packed) - packed = LLVMGetUndef(LLVMInt32Type()); + packed = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context)); if (desc->block.bits < 32) packed = LLVMBuildTrunc(builder, packed, type, ""); @@ -364,7 +368,7 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder, * \return a 4 element vector with the pixel's RGBA values. */ LLVMValueRef -lp_build_fetch_rgba_aos(LLVMBuilderRef builder, +lp_build_fetch_rgba_aos(struct gallivm_state *gallivm, const struct util_format_description *format_desc, struct lp_type type, LLVMValueRef base_ptr, @@ -372,13 +376,14 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, LLVMValueRef i, LLVMValueRef j) { + LLVMBuilderRef builder = gallivm->builder; unsigned num_pixels = type.length / 4; struct lp_build_context bld; assert(type.length <= LP_MAX_VECTOR_LENGTH); assert(type.length % 4 == 0); - lp_build_context_init(&bld, builder, type); + lp_build_context_init(&bld, gallivm, type); /* * Trivial case @@ -397,13 +402,14 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, * scaling or converting. */ - packed = lp_build_gather(builder, type.length/4, + packed = lp_build_gather(gallivm, type.length/4, format_desc->block.bits, type.width*4, base_ptr, offset); assert(format_desc->block.bits <= type.width * type.length); - packed = LLVMBuildBitCast(builder, packed, lp_build_vec_type(type), ""); + packed = LLVMBuildBitCast(gallivm->builder, packed, + lp_build_vec_type(gallivm, type), ""); return lp_build_format_swizzle_aos(format_desc, &bld, packed); } @@ -435,11 +441,12 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, for (k = 0; k < num_pixels; ++k) { LLVMValueRef packed; - packed = lp_build_gather_elem(builder, num_pixels, + packed = lp_build_gather_elem(gallivm, num_pixels, format_desc->block.bits, 32, base_ptr, offset, k); - tmps[k] = lp_build_unpack_arith_rgba_aos(builder, format_desc, + tmps[k] = lp_build_unpack_arith_rgba_aos(gallivm, + format_desc, packed); } @@ -455,7 +462,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, __FUNCTION__, format_desc->short_name); } - lp_build_conv(builder, + lp_build_conv(gallivm, lp_float32_vec4_type(), type, tmps, num_pixels, &res, 1); @@ -476,14 +483,14 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, tmp_type.length = num_pixels * 4; tmp_type.norm = TRUE; - tmp = lp_build_fetch_subsampled_rgba_aos(builder, + tmp = lp_build_fetch_subsampled_rgba_aos(gallivm, format_desc, num_pixels, base_ptr, offset, i, j); - lp_build_conv(builder, + lp_build_conv(gallivm, tmp_type, type, &tmp, 1, &tmp, 1); @@ -505,51 +512,52 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, * or incentive to optimize. */ - LLVMModuleRef module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder))); - char name[256]; - LLVMTypeRef i8t = LLVMInt8Type(); + LLVMTypeRef i8t = LLVMInt8TypeInContext(gallivm->context); LLVMTypeRef pi8t = LLVMPointerType(i8t, 0); - LLVMTypeRef i32t = LLVMInt32Type(); + LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context); LLVMValueRef function; LLVMValueRef tmp_ptr; LLVMValueRef tmp; LLVMValueRef res; unsigned k; - util_snprintf(name, sizeof name, "util_format_%s_fetch_rgba_8unorm", - format_desc->short_name); - if (gallivm_debug & GALLIVM_DEBUG_PERF) { - debug_printf("%s: falling back to %s\n", __FUNCTION__, name); + debug_printf("%s: falling back to util_format_%s_fetch_rgba_8unorm\n", + __FUNCTION__, format_desc->short_name); } /* * Declare and bind format_desc->fetch_rgba_8unorm(). */ - function = LLVMGetNamedFunction(module, name); - if (!function) { + { + /* + * Function to call looks like: + * fetch(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j) + */ LLVMTypeRef ret_type; LLVMTypeRef arg_types[4]; LLVMTypeRef function_type; - ret_type = LLVMVoidType(); + ret_type = LLVMVoidTypeInContext(gallivm->context); arg_types[0] = pi8t; arg_types[1] = pi8t; - arg_types[3] = arg_types[2] = LLVMIntType(sizeof(unsigned) * 8); - function_type = LLVMFunctionType(ret_type, arg_types, Elements(arg_types), 0); - function = LLVMAddFunction(module, name, function_type); - - LLVMSetFunctionCallConv(function, LLVMCCallConv); - LLVMSetLinkage(function, LLVMExternalLinkage); - - assert(LLVMIsDeclaration(function)); - - LLVMAddGlobalMapping(lp_build_engine, function, - func_to_pointer((func_pointer)format_desc->fetch_rgba_8unorm)); + arg_types[2] = i32t; + arg_types[3] = i32t; + function_type = LLVMFunctionType(ret_type, arg_types, + Elements(arg_types), 0); + + /* make const pointer for the C fetch_rgba_8unorm function */ + function = lp_build_const_int_pointer(gallivm, + func_to_pointer((func_pointer) format_desc->fetch_rgba_8unorm)); + + /* cast the callee pointer to the function's type */ + function = LLVMBuildBitCast(builder, function, + LLVMPointerType(function_type, 0), + "cast callee"); } - tmp_ptr = lp_build_alloca(builder, i32t, ""); + tmp_ptr = lp_build_alloca(gallivm, i32t, ""); res = LLVMGetUndef(LLVMVectorType(i32t, num_pixels)); @@ -559,11 +567,11 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, */ for (k = 0; k < num_pixels; ++k) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), k, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, k); LLVMValueRef args[4]; args[0] = LLVMBuildBitCast(builder, tmp_ptr, pi8t, ""); - args[1] = lp_build_gather_elem_ptr(builder, num_pixels, + args[1] = lp_build_gather_elem_ptr(gallivm, num_pixels, base_ptr, offset, k); if (num_pixels == 1) { @@ -608,51 +616,58 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, * or incentive to optimize. */ - LLVMModuleRef module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder))); - char name[256]; - LLVMTypeRef f32t = LLVMFloatType(); + LLVMTypeRef f32t = LLVMFloatTypeInContext(gallivm->context); LLVMTypeRef f32x4t = LLVMVectorType(f32t, 4); LLVMTypeRef pf32t = LLVMPointerType(f32t, 0); + LLVMTypeRef pi8t = LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0); + LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context); LLVMValueRef function; LLVMValueRef tmp_ptr; LLVMValueRef tmps[LP_MAX_VECTOR_LENGTH/4]; LLVMValueRef res; unsigned k; - util_snprintf(name, sizeof name, "util_format_%s_fetch_rgba_float", - format_desc->short_name); - if (gallivm_debug & GALLIVM_DEBUG_PERF) { - debug_printf("%s: falling back to %s\n", __FUNCTION__, name); + debug_printf("%s: falling back to util_format_%s_fetch_rgba_float\n", + __FUNCTION__, format_desc->short_name); } /* * Declare and bind format_desc->fetch_rgba_float(). */ - function = LLVMGetNamedFunction(module, name); - if (!function) { + { + /* + * Function to call looks like: + * fetch(float *dst, const uint8_t *src, unsigned i, unsigned j) + */ LLVMTypeRef ret_type; LLVMTypeRef arg_types[4]; LLVMTypeRef function_type; - ret_type = LLVMVoidType(); + ret_type = LLVMVoidTypeInContext(gallivm->context); arg_types[0] = pf32t; - arg_types[1] = LLVMPointerType(LLVMInt8Type(), 0); - arg_types[3] = arg_types[2] = LLVMIntType(sizeof(unsigned) * 8); - function_type = LLVMFunctionType(ret_type, arg_types, Elements(arg_types), 0); - function = LLVMAddFunction(module, name, function_type); + arg_types[1] = pi8t; + arg_types[2] = i32t; + arg_types[3] = i32t; + function_type = LLVMFunctionType(ret_type, arg_types, + Elements(arg_types), 0); - LLVMSetFunctionCallConv(function, LLVMCCallConv); - LLVMSetLinkage(function, LLVMExternalLinkage); + /* Note: we're using this casting here instead of LLVMAddGlobalMapping() + * to work around a bug in LLVM 2.6, and for efficiency/simplicity. + */ - assert(LLVMIsDeclaration(function)); + /* make const pointer for the C fetch_rgba_float function */ + function = lp_build_const_int_pointer(gallivm, + func_to_pointer((func_pointer) format_desc->fetch_rgba_float)); - LLVMAddGlobalMapping(lp_build_engine, function, - func_to_pointer((func_pointer)format_desc->fetch_rgba_float)); + /* cast the callee pointer to the function's type */ + function = LLVMBuildBitCast(builder, function, + LLVMPointerType(function_type, 0), + "cast callee"); } - tmp_ptr = lp_build_alloca(builder, f32x4t, ""); + tmp_ptr = lp_build_alloca(gallivm, f32x4t, ""); /* * Invoke format_desc->fetch_rgba_float() for each pixel and insert the result @@ -663,7 +678,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, LLVMValueRef args[4]; args[0] = LLVMBuildBitCast(builder, tmp_ptr, pf32t, ""); - args[1] = lp_build_gather_elem_ptr(builder, num_pixels, + args[1] = lp_build_gather_elem_ptr(gallivm, num_pixels, base_ptr, offset, k); if (num_pixels == 1) { @@ -671,7 +686,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, args[3] = j; } else { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), k, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, k); args[2] = LLVMBuildExtractElement(builder, i, index, ""); args[3] = LLVMBuildExtractElement(builder, j, index, ""); } @@ -681,7 +696,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, tmps[k] = LLVMBuildLoad(builder, tmp_ptr, ""); } - lp_build_conv(builder, + lp_build_conv(gallivm, lp_float32_vec4_type(), type, tmps, num_pixels, &res, 1); @@ -690,5 +705,5 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder, } assert(0); - return lp_build_undef(type); + return lp_build_undef(gallivm, type); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c index ce7e54afc7..0a57b3ce79 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c @@ -97,12 +97,13 @@ lp_build_format_swizzle_soa(const struct util_format_description *format_desc, * \param rgba_out returns the SoA R,G,B,A vectors */ void -lp_build_unpack_rgba_soa(LLVMBuilderRef builder, +lp_build_unpack_rgba_soa(struct gallivm_state *gallivm, const struct util_format_description *format_desc, struct lp_type type, LLVMValueRef packed, LLVMValueRef rgba_out[4]) { + LLVMBuilderRef builder = gallivm->builder; struct lp_build_context bld; LLVMValueRef inputs[4]; unsigned start; @@ -116,7 +117,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, assert(type.floating); assert(type.width == 32); - lp_build_context_init(&bld, builder, type); + lp_build_context_init(&bld, gallivm, type); /* Decode the input vector components */ start = 0; @@ -129,7 +130,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, switch(format_desc->channel[chan].type) { case UTIL_FORMAT_TYPE_VOID: - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); break; case UTIL_FORMAT_TYPE_UNSIGNED: @@ -138,7 +139,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, */ if (start) { - input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(type, start), ""); + input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(gallivm, type, start), ""); } /* @@ -147,7 +148,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, if (stop < format_desc->block.bits) { unsigned mask = ((unsigned long long)1 << width) - 1; - input = LLVMBuildAnd(builder, input, lp_build_const_int_vec(type, mask), ""); + input = LLVMBuildAnd(builder, input, lp_build_const_int_vec(gallivm, type, mask), ""); } /* @@ -156,14 +157,15 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, if (type.floating) { if(format_desc->channel[chan].normalized) - input = lp_build_unsigned_norm_to_float(builder, width, type, input); + input = lp_build_unsigned_norm_to_float(gallivm, width, type, input); else - input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), ""); + input = LLVMBuildSIToFP(builder, input, + lp_build_vec_type(gallivm, type), ""); } else { /* FIXME */ assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); } break; @@ -175,7 +177,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, if (stop < type.width) { unsigned bits = type.width - stop; - LLVMValueRef bits_val = lp_build_const_int_vec(type, bits); + LLVMValueRef bits_val = lp_build_const_int_vec(gallivm, type, bits); input = LLVMBuildShl(builder, input, bits_val, ""); } @@ -185,7 +187,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, if (format_desc->channel[chan].size < type.width) { unsigned bits = type.width - format_desc->channel[chan].size; - LLVMValueRef bits_val = lp_build_const_int_vec(type, bits); + LLVMValueRef bits_val = lp_build_const_int_vec(gallivm, type, bits); input = LLVMBuildAShr(builder, input, bits_val, ""); } @@ -194,17 +196,17 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, */ if (type.floating) { - input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), ""); + input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(gallivm, type), ""); if (format_desc->channel[chan].normalized) { double scale = 1.0 / ((1 << (format_desc->channel[chan].size - 1)) - 1); - LLVMValueRef scale_val = lp_build_const_vec(type, scale); + LLVMValueRef scale_val = lp_build_const_vec(gallivm, type, scale); input = LLVMBuildFMul(builder, input, scale_val, ""); } } else { /* FIXME */ assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); } break; @@ -214,32 +216,32 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, assert(start == 0); assert(stop == 32); assert(type.width == 32); - input = LLVMBuildBitCast(builder, input, lp_build_vec_type(type), ""); + input = LLVMBuildBitCast(builder, input, lp_build_vec_type(gallivm, type), ""); } else { /* FIXME */ assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); } break; case UTIL_FORMAT_TYPE_FIXED: if (type.floating) { double scale = 1.0 / ((1 << (format_desc->channel[chan].size/2)) - 1); - LLVMValueRef scale_val = lp_build_const_vec(type, scale); - input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), ""); + LLVMValueRef scale_val = lp_build_const_vec(gallivm, type, scale); + input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(gallivm, type), ""); input = LLVMBuildFMul(builder, input, scale_val, ""); } else { /* FIXME */ assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); } break; default: assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); break; } @@ -253,16 +255,17 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, void -lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, +lp_build_rgba8_to_f32_soa(struct gallivm_state *gallivm, struct lp_type dst_type, LLVMValueRef packed, LLVMValueRef *rgba) { - LLVMValueRef mask = lp_build_const_int_vec(dst_type, 0xff); + LLVMBuilderRef builder = gallivm->builder; + LLVMValueRef mask = lp_build_const_int_vec(gallivm, dst_type, 0xff); unsigned chan; packed = LLVMBuildBitCast(builder, packed, - lp_build_int_vec_type(dst_type), ""); + lp_build_int_vec_type(gallivm, dst_type), ""); /* Decode the input vector components */ for (chan = 0; chan < 4; ++chan) { @@ -274,12 +277,12 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, if (start) input = LLVMBuildLShr(builder, input, - lp_build_const_int_vec(dst_type, start), ""); + lp_build_const_int_vec(gallivm, dst_type, start), ""); if (stop < 32) input = LLVMBuildAnd(builder, input, mask, ""); - input = lp_build_unsigned_norm_to_float(builder, 8, dst_type, input); + input = lp_build_unsigned_norm_to_float(gallivm, 8, dst_type, input); rgba[chan] = input; } @@ -303,7 +306,7 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, * be in [0, block_width-1] and j will be in [0, block_height-1]. */ void -lp_build_fetch_rgba_soa(LLVMBuilderRef builder, +lp_build_fetch_rgba_soa(struct gallivm_state *gallivm, const struct util_format_description *format_desc, struct lp_type type, LLVMValueRef base_ptr, @@ -312,6 +315,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, LLVMValueRef j, LLVMValueRef rgba_out[4]) { + LLVMBuilderRef builder = gallivm->builder; if (format_desc->layout == UTIL_FORMAT_LAYOUT_PLAIN && (format_desc->colorspace == UTIL_FORMAT_COLORSPACE_RGB || @@ -334,7 +338,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, * gather the texels from the texture * Ex: packed = {BGRA, BGRA, BGRA, BGRA}. */ - packed = lp_build_gather(builder, + packed = lp_build_gather(gallivm, type.length, format_desc->block.bits, type.width, @@ -343,7 +347,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, /* * convert texels to float rgba */ - lp_build_unpack_rgba_soa(builder, + lp_build_unpack_rgba_soa(gallivm, format_desc, type, packed, rgba_out); @@ -364,10 +368,10 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, tmp_type.length = type.length * 4; tmp_type.norm = TRUE; - tmp = lp_build_fetch_rgba_aos(builder, format_desc, tmp_type, + tmp = lp_build_fetch_rgba_aos(gallivm, format_desc, tmp_type, base_ptr, offset, i, j); - lp_build_rgba8_to_f32_soa(builder, + lp_build_rgba8_to_f32_soa(gallivm, type, tmp, rgba_out); @@ -397,23 +401,24 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, tmp_type.length = 4; for (chan = 0; chan < 4; ++chan) { - rgba_out[chan] = lp_build_undef(type); + rgba_out[chan] = lp_build_undef(gallivm, type); } /* loop over number of pixels */ for(k = 0; k < type.length; ++k) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), k, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, k); LLVMValueRef offset_elem; LLVMValueRef i_elem, j_elem; LLVMValueRef tmp; - offset_elem = LLVMBuildExtractElement(builder, offset, index, ""); + offset_elem = LLVMBuildExtractElement(builder, offset, + index, ""); i_elem = LLVMBuildExtractElement(builder, i, index, ""); j_elem = LLVMBuildExtractElement(builder, j, index, ""); /* Get a single float[4]={R,G,B,A} pixel */ - tmp = lp_build_fetch_rgba_aos(builder, format_desc, tmp_type, + tmp = lp_build_fetch_rgba_aos(gallivm, format_desc, tmp_type, base_ptr, offset_elem, i_elem, j_elem); @@ -422,7 +427,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, * position = 'index'. */ for (chan = 0; chan < 4; ++chan) { - LLVMValueRef chan_val = LLVMConstInt(LLVMInt32Type(), chan, 0), + LLVMValueRef chan_val = lp_build_const_int32(gallivm, chan), tmp_chan = LLVMBuildExtractElement(builder, tmp, chan_val, ""); rgba_out[chan] = LLVMBuildInsertElement(builder, rgba_out[chan], tmp_chan, index, ""); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c b/src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c index 2bce289555..cdf1956c09 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c @@ -43,6 +43,7 @@ #include "lp_bld_conv.h" #include "lp_bld_gather.h" #include "lp_bld_format.h" +#include "lp_bld_init.h" #include "lp_bld_logic.h" /** @@ -51,7 +52,7 @@ * @param i is a <n x i32> vector with the x pixel coordinate (0 or 1) */ static void -uyvy_to_yuv_soa(LLVMBuilderRef builder, +uyvy_to_yuv_soa(struct gallivm_state *gallivm, unsigned n, LLVMValueRef packed, LLVMValueRef i, @@ -59,6 +60,7 @@ uyvy_to_yuv_soa(LLVMBuilderRef builder, LLVMValueRef *u, LLVMValueRef *v) { + LLVMBuilderRef builder = gallivm->builder; struct lp_type type; LLVMValueRef mask; @@ -86,25 +88,25 @@ uyvy_to_yuv_soa(LLVMBuilderRef builder, LLVMValueRef sel, tmp, tmp2; struct lp_build_context bld32; - lp_build_context_init(&bld32, builder, type); + lp_build_context_init(&bld32, gallivm, type); - tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 8), ""); - tmp2 = LLVMBuildLShr(builder, tmp, lp_build_const_int_vec(type, 16), ""); - sel = lp_build_compare(builder, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(type, 0)); + tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 8), ""); + tmp2 = LLVMBuildLShr(builder, tmp, lp_build_const_int_vec(gallivm, type, 16), ""); + sel = lp_build_compare(gallivm, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(gallivm, type, 0)); *y = lp_build_select(&bld32, sel, tmp, tmp2); } else #endif { LLVMValueRef shift; - shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(type, 16), ""); - shift = LLVMBuildAdd(builder, shift, lp_build_const_int_vec(type, 8), ""); + shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(gallivm, type, 16), ""); + shift = LLVMBuildAdd(builder, shift, lp_build_const_int_vec(gallivm, type, 8), ""); *y = LLVMBuildLShr(builder, packed, shift, ""); } *u = packed; - *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 16), ""); + *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 16), ""); - mask = lp_build_const_int_vec(type, 0xff); + mask = lp_build_const_int_vec(gallivm, type, 0xff); *y = LLVMBuildAnd(builder, *y, mask, "y"); *u = LLVMBuildAnd(builder, *u, mask, "u"); @@ -118,7 +120,7 @@ uyvy_to_yuv_soa(LLVMBuilderRef builder, * @param i is a <n x i32> vector with the x pixel coordinate (0 or 1) */ static void -yuyv_to_yuv_soa(LLVMBuilderRef builder, +yuyv_to_yuv_soa(struct gallivm_state *gallivm, unsigned n, LLVMValueRef packed, LLVMValueRef i, @@ -126,6 +128,7 @@ yuyv_to_yuv_soa(LLVMBuilderRef builder, LLVMValueRef *u, LLVMValueRef *v) { + LLVMBuilderRef builder = gallivm->builder; struct lp_type type; LLVMValueRef mask; @@ -153,23 +156,23 @@ yuyv_to_yuv_soa(LLVMBuilderRef builder, LLVMValueRef sel, tmp; struct lp_build_context bld32; - lp_build_context_init(&bld32, builder, type); + lp_build_context_init(&bld32, gallivm, type); - tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 16), ""); - sel = lp_build_compare(builder, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(type, 0)); + tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 16), ""); + sel = lp_build_compare(gallivm, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(gallivm, type, 0)); *y = lp_build_select(&bld32, sel, packed, tmp); } else #endif { LLVMValueRef shift; - shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(type, 16), ""); + shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(gallivm, type, 16), ""); *y = LLVMBuildLShr(builder, packed, shift, ""); } - *u = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 8), ""); - *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 24), ""); + *u = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 8), ""); + *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 24), ""); - mask = lp_build_const_int_vec(type, 0xff); + mask = lp_build_const_int_vec(gallivm, type, 0xff); *y = LLVMBuildAnd(builder, *y, mask, "y"); *u = LLVMBuildAnd(builder, *u, mask, "u"); @@ -178,11 +181,12 @@ yuyv_to_yuv_soa(LLVMBuilderRef builder, static INLINE void -yuv_to_rgb_soa(LLVMBuilderRef builder, +yuv_to_rgb_soa(struct gallivm_state *gallivm, unsigned n, LLVMValueRef y, LLVMValueRef u, LLVMValueRef v, LLVMValueRef *r, LLVMValueRef *g, LLVMValueRef *b) { + LLVMBuilderRef builder = gallivm->builder; struct lp_type type; struct lp_build_context bld; @@ -203,7 +207,7 @@ yuv_to_rgb_soa(LLVMBuilderRef builder, type.width = 32; type.length = n; - lp_build_context_init(&bld, builder, type); + lp_build_context_init(&bld, gallivm, type); assert(lp_check_value(type, y)); assert(lp_check_value(type, u)); @@ -213,17 +217,17 @@ yuv_to_rgb_soa(LLVMBuilderRef builder, * Constants */ - c0 = lp_build_const_int_vec(type, 0); - c8 = lp_build_const_int_vec(type, 8); - c16 = lp_build_const_int_vec(type, 16); - c128 = lp_build_const_int_vec(type, 128); - c255 = lp_build_const_int_vec(type, 255); + c0 = lp_build_const_int_vec(gallivm, type, 0); + c8 = lp_build_const_int_vec(gallivm, type, 8); + c16 = lp_build_const_int_vec(gallivm, type, 16); + c128 = lp_build_const_int_vec(gallivm, type, 128); + c255 = lp_build_const_int_vec(gallivm, type, 255); - cy = lp_build_const_int_vec(type, 298); - cug = lp_build_const_int_vec(type, -100); - cub = lp_build_const_int_vec(type, 516); - cvr = lp_build_const_int_vec(type, 409); - cvg = lp_build_const_int_vec(type, -208); + cy = lp_build_const_int_vec(gallivm, type, 298); + cug = lp_build_const_int_vec(gallivm, type, -100); + cub = lp_build_const_int_vec(gallivm, type, 516); + cvr = lp_build_const_int_vec(gallivm, type, 409); + cvg = lp_build_const_int_vec(gallivm, type, -208); /* * y -= 16; @@ -276,10 +280,11 @@ yuv_to_rgb_soa(LLVMBuilderRef builder, static LLVMValueRef -rgb_to_rgba_aos(LLVMBuilderRef builder, +rgb_to_rgba_aos(struct gallivm_state *gallivm, unsigned n, LLVMValueRef r, LLVMValueRef g, LLVMValueRef b) { + LLVMBuilderRef builder = gallivm->builder; struct lp_type type; LLVMValueRef a; LLVMValueRef rgba; @@ -298,9 +303,9 @@ rgb_to_rgba_aos(LLVMBuilderRef builder, */ r = r; - g = LLVMBuildShl(builder, g, lp_build_const_int_vec(type, 8), ""); - b = LLVMBuildShl(builder, b, lp_build_const_int_vec(type, 16), ""); - a = lp_build_const_int_vec(type, 0xff000000); + g = LLVMBuildShl(builder, g, lp_build_const_int_vec(gallivm, type, 8), ""); + b = LLVMBuildShl(builder, b, lp_build_const_int_vec(gallivm, type, 16), ""); + a = lp_build_const_int_vec(gallivm, type, 0xff000000); rgba = r; rgba = LLVMBuildOr(builder, rgba, g, ""); @@ -308,7 +313,7 @@ rgb_to_rgba_aos(LLVMBuilderRef builder, rgba = LLVMBuildOr(builder, rgba, a, ""); rgba = LLVMBuildBitCast(builder, rgba, - LLVMVectorType(LLVMInt8Type(), 4*n), ""); + LLVMVectorType(LLVMInt8TypeInContext(gallivm->context), 4*n), ""); return rgba; } @@ -318,7 +323,7 @@ rgb_to_rgba_aos(LLVMBuilderRef builder, * Convert from <n x i32> packed UYVY to <4n x i8> RGBA AoS */ static LLVMValueRef -uyvy_to_rgba_aos(LLVMBuilderRef builder, +uyvy_to_rgba_aos(struct gallivm_state *gallivm, unsigned n, LLVMValueRef packed, LLVMValueRef i) @@ -327,9 +332,9 @@ uyvy_to_rgba_aos(LLVMBuilderRef builder, LLVMValueRef r, g, b; LLVMValueRef rgba; - uyvy_to_yuv_soa(builder, n, packed, i, &y, &u, &v); - yuv_to_rgb_soa(builder, n, y, u, v, &r, &g, &b); - rgba = rgb_to_rgba_aos(builder, n, r, g, b); + uyvy_to_yuv_soa(gallivm, n, packed, i, &y, &u, &v); + yuv_to_rgb_soa(gallivm, n, y, u, v, &r, &g, &b); + rgba = rgb_to_rgba_aos(gallivm, n, r, g, b); return rgba; } @@ -339,7 +344,7 @@ uyvy_to_rgba_aos(LLVMBuilderRef builder, * Convert from <n x i32> packed YUYV to <4n x i8> RGBA AoS */ static LLVMValueRef -yuyv_to_rgba_aos(LLVMBuilderRef builder, +yuyv_to_rgba_aos(struct gallivm_state *gallivm, unsigned n, LLVMValueRef packed, LLVMValueRef i) @@ -348,9 +353,9 @@ yuyv_to_rgba_aos(LLVMBuilderRef builder, LLVMValueRef r, g, b; LLVMValueRef rgba; - yuyv_to_yuv_soa(builder, n, packed, i, &y, &u, &v); - yuv_to_rgb_soa(builder, n, y, u, v, &r, &g, &b); - rgba = rgb_to_rgba_aos(builder, n, r, g, b); + yuyv_to_yuv_soa(gallivm, n, packed, i, &y, &u, &v); + yuv_to_rgb_soa(gallivm, n, y, u, v, &r, &g, &b); + rgba = rgb_to_rgba_aos(gallivm, n, r, g, b); return rgba; } @@ -360,7 +365,7 @@ yuyv_to_rgba_aos(LLVMBuilderRef builder, * Convert from <n x i32> packed RG_BG to <4n x i8> RGBA AoS */ static LLVMValueRef -rgbg_to_rgba_aos(LLVMBuilderRef builder, +rgbg_to_rgba_aos(struct gallivm_state *gallivm, unsigned n, LLVMValueRef packed, LLVMValueRef i) @@ -368,8 +373,8 @@ rgbg_to_rgba_aos(LLVMBuilderRef builder, LLVMValueRef r, g, b; LLVMValueRef rgba; - uyvy_to_yuv_soa(builder, n, packed, i, &g, &r, &b); - rgba = rgb_to_rgba_aos(builder, n, r, g, b); + uyvy_to_yuv_soa(gallivm, n, packed, i, &g, &r, &b); + rgba = rgb_to_rgba_aos(gallivm, n, r, g, b); return rgba; } @@ -379,7 +384,7 @@ rgbg_to_rgba_aos(LLVMBuilderRef builder, * Convert from <n x i32> packed GR_GB to <4n x i8> RGBA AoS */ static LLVMValueRef -grgb_to_rgba_aos(LLVMBuilderRef builder, +grgb_to_rgba_aos(struct gallivm_state *gallivm, unsigned n, LLVMValueRef packed, LLVMValueRef i) @@ -387,8 +392,8 @@ grgb_to_rgba_aos(LLVMBuilderRef builder, LLVMValueRef r, g, b; LLVMValueRef rgba; - yuyv_to_yuv_soa(builder, n, packed, i, &g, &r, &b); - rgba = rgb_to_rgba_aos(builder, n, r, g, b); + yuyv_to_yuv_soa(gallivm, n, packed, i, &g, &r, &b); + rgba = rgb_to_rgba_aos(gallivm, n, r, g, b); return rgba; } @@ -401,7 +406,7 @@ grgb_to_rgba_aos(LLVMBuilderRef builder, * @return a <4*n x i8> vector with the pixel RGBA values in AoS */ LLVMValueRef -lp_build_fetch_subsampled_rgba_aos(LLVMBuilderRef builder, +lp_build_fetch_subsampled_rgba_aos(struct gallivm_state *gallivm, const struct util_format_description *format_desc, unsigned n, LLVMValueRef base_ptr, @@ -417,26 +422,26 @@ lp_build_fetch_subsampled_rgba_aos(LLVMBuilderRef builder, assert(format_desc->block.width == 2); assert(format_desc->block.height == 1); - packed = lp_build_gather(builder, n, 32, 32, base_ptr, offset); + packed = lp_build_gather(gallivm, n, 32, 32, base_ptr, offset); (void)j; switch (format_desc->format) { case PIPE_FORMAT_UYVY: - rgba = uyvy_to_rgba_aos(builder, n, packed, i); + rgba = uyvy_to_rgba_aos(gallivm, n, packed, i); break; case PIPE_FORMAT_YUYV: - rgba = yuyv_to_rgba_aos(builder, n, packed, i); + rgba = yuyv_to_rgba_aos(gallivm, n, packed, i); break; case PIPE_FORMAT_R8G8_B8G8_UNORM: - rgba = rgbg_to_rgba_aos(builder, n, packed, i); + rgba = rgbg_to_rgba_aos(gallivm, n, packed, i); break; case PIPE_FORMAT_G8R8_G8B8_UNORM: - rgba = grgb_to_rgba_aos(builder, n, packed, i); + rgba = grgb_to_rgba_aos(gallivm, n, packed, i); break; default: assert(0); - rgba = LLVMGetUndef(LLVMVectorType(LLVMInt8Type(), 4*n)); + rgba = LLVMGetUndef(LLVMVectorType(LLVMInt8TypeInContext(gallivm->context), 4*n)); break; } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_gather.c b/src/gallium/auxiliary/gallivm/lp_bld_gather.c index d60472e065..0dc81b1abb 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_gather.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_gather.c @@ -31,6 +31,7 @@ #include "lp_bld_const.h" #include "lp_bld_format.h" #include "lp_bld_gather.h" +#include "lp_bld_init.h" /** @@ -39,7 +40,7 @@ * @sa lp_build_gather() */ LLVMValueRef -lp_build_gather_elem_ptr(LLVMBuilderRef builder, +lp_build_gather_elem_ptr(struct gallivm_state *gallivm, unsigned length, LLVMValueRef base_ptr, LLVMValueRef offsets, @@ -48,17 +49,17 @@ lp_build_gather_elem_ptr(LLVMBuilderRef builder, LLVMValueRef offset; LLVMValueRef ptr; - assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8Type(), 0)); + assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0)); if (length == 1) { assert(i == 0); offset = offsets; } else { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); - offset = LLVMBuildExtractElement(builder, offsets, index, ""); + LLVMValueRef index = lp_build_const_int32(gallivm, i); + offset = LLVMBuildExtractElement(gallivm->builder, offsets, index, ""); } - ptr = LLVMBuildGEP(builder, base_ptr, &offset, 1, ""); + ptr = LLVMBuildGEP(gallivm->builder, base_ptr, &offset, 1, ""); return ptr; } @@ -70,7 +71,7 @@ lp_build_gather_elem_ptr(LLVMBuilderRef builder, * @sa lp_build_gather() */ LLVMValueRef -lp_build_gather_elem(LLVMBuilderRef builder, +lp_build_gather_elem(struct gallivm_state *gallivm, unsigned length, unsigned src_width, unsigned dst_width, @@ -78,23 +79,23 @@ lp_build_gather_elem(LLVMBuilderRef builder, LLVMValueRef offsets, unsigned i) { - LLVMTypeRef src_type = LLVMIntType(src_width); + LLVMTypeRef src_type = LLVMIntTypeInContext(gallivm->context, src_width); LLVMTypeRef src_ptr_type = LLVMPointerType(src_type, 0); - LLVMTypeRef dst_elem_type = LLVMIntType(dst_width); + LLVMTypeRef dst_elem_type = LLVMIntTypeInContext(gallivm->context, dst_width); LLVMValueRef ptr; LLVMValueRef res; - assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8Type(), 0)); + assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0)); - ptr = lp_build_gather_elem_ptr(builder, length, base_ptr, offsets, i); - ptr = LLVMBuildBitCast(builder, ptr, src_ptr_type, ""); - res = LLVMBuildLoad(builder, ptr, ""); + ptr = lp_build_gather_elem_ptr(gallivm, length, base_ptr, offsets, i); + ptr = LLVMBuildBitCast(gallivm->builder, ptr, src_ptr_type, ""); + res = LLVMBuildLoad(gallivm->builder, ptr, ""); assert(src_width <= dst_width); if (src_width > dst_width) - res = LLVMBuildTrunc(builder, res, dst_elem_type, ""); + res = LLVMBuildTrunc(gallivm->builder, res, dst_elem_type, ""); if (src_width < dst_width) - res = LLVMBuildZExt(builder, res, dst_elem_type, ""); + res = LLVMBuildZExt(gallivm->builder, res, dst_elem_type, ""); return res; } @@ -112,7 +113,7 @@ lp_build_gather_elem(LLVMBuilderRef builder, * @param offsets vector with offsets */ LLVMValueRef -lp_build_gather(LLVMBuilderRef builder, +lp_build_gather(struct gallivm_state *gallivm, unsigned length, unsigned src_width, unsigned dst_width, @@ -123,24 +124,24 @@ lp_build_gather(LLVMBuilderRef builder, if (length == 1) { /* Scalar */ - return lp_build_gather_elem(builder, length, + return lp_build_gather_elem(gallivm, length, src_width, dst_width, base_ptr, offsets, 0); } else { /* Vector */ - LLVMTypeRef dst_elem_type = LLVMIntType(dst_width); + LLVMTypeRef dst_elem_type = LLVMIntTypeInContext(gallivm->context, dst_width); LLVMTypeRef dst_vec_type = LLVMVectorType(dst_elem_type, length); unsigned i; res = LLVMGetUndef(dst_vec_type); for (i = 0; i < length; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, i); LLVMValueRef elem; - elem = lp_build_gather_elem(builder, length, + elem = lp_build_gather_elem(gallivm, length, src_width, dst_width, base_ptr, offsets, i); - res = LLVMBuildInsertElement(builder, res, elem, index, ""); + res = LLVMBuildInsertElement(gallivm->builder, res, elem, index, ""); } } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_gather.h b/src/gallium/auxiliary/gallivm/lp_bld_gather.h index 131af8ea07..5b04131730 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_gather.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_gather.h @@ -34,14 +34,14 @@ LLVMValueRef -lp_build_gather_elem_ptr(LLVMBuilderRef builder, +lp_build_gather_elem_ptr(struct gallivm_state *gallivm, unsigned length, LLVMValueRef base_ptr, LLVMValueRef offsets, unsigned i); LLVMValueRef -lp_build_gather_elem(LLVMBuilderRef builder, +lp_build_gather_elem(struct gallivm_state *gallivm, unsigned length, unsigned src_width, unsigned dst_width, @@ -50,7 +50,7 @@ lp_build_gather_elem(LLVMBuilderRef builder, unsigned i); LLVMValueRef -lp_build_gather(LLVMBuilderRef builder, +lp_build_gather(struct gallivm_state *gallivm, unsigned length, unsigned src_width, unsigned dst_width, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_init.c b/src/gallium/auxiliary/gallivm/lp_bld_init.c index 0b9a6f745f..7504cb5cf2 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_init.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_init.c @@ -29,6 +29,8 @@ #include "pipe/p_compiler.h" #include "util/u_cpu_detect.h" #include "util/u_debug.h" +#include "util/u_memory.h" +#include "util/u_simple_list.h" #include "lp_bld_debug.h" #include "lp_bld_init.h" @@ -45,6 +47,7 @@ static const struct debug_named_value lp_bld_debug_flags[] = { { "nopt", GALLIVM_DEBUG_NO_OPT, NULL }, { "perf", GALLIVM_DEBUG_PERF, NULL }, { "no_brilinear", GALLIVM_DEBUG_NO_BRILINEAR, NULL }, + { "gc", GALLIVM_DEBUG_GC, NULL }, DEBUG_NAMED_VALUE_END }; @@ -52,11 +55,7 @@ DEBUG_GET_ONCE_FLAGS_OPTION(gallivm_debug, "GALLIVM_DEBUG", lp_bld_debug_flags, #endif -LLVMModuleRef lp_build_module = NULL; -LLVMExecutionEngineRef lp_build_engine = NULL; -LLVMModuleProviderRef lp_build_provider = NULL; -LLVMTargetDataRef lp_build_target = NULL; -LLVMPassManagerRef lp_build_pass = NULL; +static boolean gallivm_initialized = FALSE; /* @@ -82,6 +81,19 @@ enum LLVM_CodeGenOpt_Level { }; +/** + * LLVM 2.6 permits only one ExecutionEngine to be created. This is it. + */ +static LLVMExecutionEngineRef GlobalEngine = NULL; + +/** + * Same gallivm state shared by all contexts. + */ +static struct gallivm_state *GlobalGallivm = NULL; + + + + extern void lp_register_oprofile_jit_event_listener(LLVMExecutionEngineRef EE); @@ -89,26 +101,148 @@ extern void lp_set_target_options(void); -void -lp_build_init(void) + +/** + * Create the LLVM (optimization) pass manager and install + * relevant optimization passes. + * \return TRUE for success, FALSE for failure + */ +static boolean +create_pass_manager(struct gallivm_state *gallivm) { -#ifdef DEBUG - gallivm_debug = debug_get_option_gallivm_debug(); + assert(!gallivm->passmgr); + + gallivm->passmgr = LLVMCreateFunctionPassManager(gallivm->provider); + if (!gallivm->passmgr) + return FALSE; + + LLVMAddTargetData(gallivm->target, gallivm->passmgr); + + if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) { + /* These are the passes currently listed in llvm-c/Transforms/Scalar.h, + * but there are more on SVN. + * TODO: Add more passes. + */ + LLVMAddCFGSimplificationPass(gallivm->passmgr); + + if (HAVE_LLVM >= 0x207 && sizeof(void*) == 4) { + /* For LLVM >= 2.7 and 32-bit build, use this order of passes to + * avoid generating bad code. + * Test with piglit glsl-vs-sqrt-zero test. + */ + LLVMAddConstantPropagationPass(gallivm->passmgr); + LLVMAddPromoteMemoryToRegisterPass(gallivm->passmgr); + } + else { + LLVMAddPromoteMemoryToRegisterPass(gallivm->passmgr); + LLVMAddConstantPropagationPass(gallivm->passmgr); + } + + if (util_cpu_caps.has_sse4_1) { + /* FIXME: There is a bug in this pass, whereby the combination + * of fptosi and sitofp (necessary for trunc/floor/ceil/round + * implementation) somehow becomes invalid code. + */ + LLVMAddInstructionCombiningPass(gallivm->passmgr); + } + LLVMAddGVNPass(gallivm->passmgr); + } + else { + /* We need at least this pass to prevent the backends to fail in + * unexpected ways. + */ + LLVMAddPromoteMemoryToRegisterPass(gallivm->passmgr); + } + + return TRUE; +} + + +/** + * Free gallivm object's LLVM allocations, but not the gallivm object itself. + */ +static void +free_gallivm_state(struct gallivm_state *gallivm) +{ +#if HAVE_LLVM >= 0x207 /* XXX or 0x208? */ + /* This leads to crashes w/ some versions of LLVM */ + LLVMModuleRef mod; + char *error; + + if (gallivm->engine && gallivm->provider) + LLVMRemoveModuleProvider(gallivm->engine, gallivm->provider, + &mod, &error); #endif - lp_set_target_options(); +#if 0 + /* XXX this seems to crash with all versions of LLVM */ + if (gallivm->provider) + LLVMDisposeModuleProvider(gallivm->provider); +#endif - LLVMInitializeNativeTarget(); + if (gallivm->passmgr) + LLVMDisposePassManager(gallivm->passmgr); - LLVMLinkInJIT(); +#if HAVE_LLVM >= 0x207 + if (gallivm->module) + LLVMDisposeModule(gallivm->module); +#endif + +#if 0 + /* Don't free the exec engine, it's a global/singleton */ + if (gallivm->engine) + LLVMDisposeExecutionEngine(gallivm->engine); +#endif + +#if 0 + /* Don't free the TargetData, it's owned by the exec engine */ + LLVMDisposeTargetData(gallivm->target); +#endif + + if (gallivm->context) + LLVMContextDispose(gallivm->context); - if (!lp_build_module) - lp_build_module = LLVMModuleCreateWithName("gallivm"); + if (gallivm->builder) + LLVMDisposeBuilder(gallivm->builder); + + gallivm->engine = NULL; + gallivm->target = NULL; + gallivm->module = NULL; + gallivm->provider = NULL; + gallivm->passmgr = NULL; + gallivm->context = NULL; + gallivm->builder = NULL; +} - if (!lp_build_provider) - lp_build_provider = LLVMCreateModuleProviderForExistingModule(lp_build_module); - if (!lp_build_engine) { +/** + * Allocate gallivm LLVM objects. + * \return TRUE for success, FALSE for failure + */ +static boolean +init_gallivm_state(struct gallivm_state *gallivm) +{ + assert(gallivm_initialized); + assert(!gallivm->context); + assert(!gallivm->module); + assert(!gallivm->provider); + + gallivm->context = LLVMContextCreate(); + if (!gallivm->context) + goto fail; + + gallivm->module = LLVMModuleCreateWithNameInContext("gallivm", + gallivm->context); + if (!gallivm->module) + goto fail; + + gallivm->provider = + LLVMCreateModuleProviderForExistingModule(gallivm->module); + if (!gallivm->provider) + goto fail; + + if (!GlobalEngine) { + /* We can only create one LLVMExecutionEngine (w/ LLVM 2.6 anyway) */ enum LLVM_CodeGenOpt_Level optlevel; char *error = NULL; @@ -119,43 +253,153 @@ lp_build_init(void) optlevel = Default; } - if (LLVMCreateJITCompiler(&lp_build_engine, lp_build_provider, - (unsigned)optlevel, &error)) { + if (LLVMCreateJITCompiler(&GlobalEngine, gallivm->provider, + (unsigned) optlevel, &error)) { _debug_printf("%s\n", error); LLVMDisposeMessage(error); - assert(0); + goto fail; } #if defined(DEBUG) || defined(PROFILE) - lp_register_oprofile_jit_event_listener(lp_build_engine); + lp_register_oprofile_jit_event_listener(GlobalEngine); #endif } - if (!lp_build_target) - lp_build_target = LLVMGetExecutionEngineTargetData(lp_build_engine); - - if (!lp_build_pass) { - lp_build_pass = LLVMCreateFunctionPassManager(lp_build_provider); - LLVMAddTargetData(lp_build_target, lp_build_pass); - - if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) { - /* These are the passes currently listed in llvm-c/Transforms/Scalar.h, - * but there are more on SVN. */ - /* TODO: Add more passes */ - LLVMAddCFGSimplificationPass(lp_build_pass); - LLVMAddPromoteMemoryToRegisterPass(lp_build_pass); - LLVMAddConstantPropagationPass(lp_build_pass); - LLVMAddInstructionCombiningPass(lp_build_pass); - LLVMAddGVNPass(lp_build_pass); - } else { - /* We need at least this pass to prevent the backends to fail in - * unexpected ways. - */ - LLVMAddPromoteMemoryToRegisterPass(lp_build_pass); + gallivm->engine = GlobalEngine; + + LLVMAddModuleProvider(gallivm->engine, gallivm->provider);//new + + gallivm->target = LLVMGetExecutionEngineTargetData(gallivm->engine); + if (!gallivm->target) + goto fail; + + if (!create_pass_manager(gallivm)) + goto fail; + + gallivm->builder = LLVMCreateBuilderInContext(gallivm->context); + if (!gallivm->builder) + goto fail; + + return TRUE; + +fail: + free_gallivm_state(gallivm); + return FALSE; +} + + +struct callback +{ + garbage_collect_callback_func func; + void *cb_data; + struct callback *prev, *next; +}; + + +/** list of all garbage collector callbacks */ +static struct callback callback_list = {NULL, NULL, NULL, NULL}; + + +/** + * Register a function with gallivm which will be called when we + * do garbage collection. + */ +void +gallivm_register_garbage_collector_callback(garbage_collect_callback_func func, + void *cb_data) +{ + struct callback *cb; + + if (!callback_list.prev) { + make_empty_list(&callback_list); + } + + /* see if already in list */ + foreach(cb, &callback_list) { + if (cb->func == func && cb->cb_data == cb_data) + return; + } + + /* add to list */ + cb = CALLOC_STRUCT(callback); + if (cb) { + cb->func = func; + cb->cb_data = cb_data; + insert_at_head(&callback_list, cb); + } +} + + +/** + * Remove a callback. + */ +void +gallivm_remove_garbage_collector_callback(garbage_collect_callback_func func, + void *cb_data) +{ + struct callback *cb; + + /* search list */ + foreach(cb, &callback_list) { + if (cb->func == func && cb->cb_data == cb_data) { + /* found, remove it */ + remove_from_list(cb); + return; } } +} + + +/** + * Call the callback functions (which are typically in the + * draw module and llvmpipe driver. + */ +static void +call_garbage_collector_callbacks(void) +{ + struct callback *cb; + foreach(cb, &callback_list) { + cb->func(cb->cb_data); + } +} + + + +/** + * Other gallium components using gallivm should call this periodically + * to let us do garbage collection (or at least try to free memory + * accumulated by the LLVM libraries). + */ +void +gallivm_garbage_collect(struct gallivm_state *gallivm) +{ + if (gallivm->context) { + if (gallivm_debug & GALLIVM_DEBUG_GC) + debug_printf("***** Doing LLVM garbage collection\n"); + + call_garbage_collector_callbacks(); + free_gallivm_state(gallivm); + init_gallivm_state(gallivm); + } +} + + +void +lp_build_init(void) +{ +#ifdef DEBUG + gallivm_debug = debug_get_option_gallivm_debug(); +#endif + + lp_set_target_options(); + + LLVMInitializeNativeTarget(); + + LLVMLinkInJIT(); util_cpu_detect(); + + gallivm_initialized = TRUE; #if 0 /* For simulating less capable machines */ @@ -166,6 +410,39 @@ lp_build_init(void) } + +/** + * Create a new gallivm_state object. + * Note that we return a singleton. + */ +struct gallivm_state * +gallivm_create(void) +{ + if (!GlobalGallivm) { + GlobalGallivm = CALLOC_STRUCT(gallivm_state); + if (GlobalGallivm) { + if (!init_gallivm_state(GlobalGallivm)) { + FREE(GlobalGallivm); + GlobalGallivm = NULL; + } + } + } + return GlobalGallivm; +} + + +/** + * Destroy a gallivm_state object. + */ +void +gallivm_destroy(struct gallivm_state *gallivm) +{ + /* No-op: don't destroy the singleton */ + (void) gallivm; +} + + + /* * Hack to allow the linking of release LLVM static libraries on a debug build. * diff --git a/src/gallium/auxiliary/gallivm/lp_bld_init.h b/src/gallium/auxiliary/gallivm/lp_bld_init.h index 0b4b1ca7d1..f68bf75a85 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_init.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_init.h @@ -30,24 +30,53 @@ #define LP_BLD_INIT_H +#include "pipe/p_compiler.h" #include "lp_bld.h" #include <llvm-c/ExecutionEngine.h> -extern LLVMModuleRef lp_build_module; -extern LLVMExecutionEngineRef lp_build_engine; -extern LLVMModuleProviderRef lp_build_provider; -extern LLVMTargetDataRef lp_build_target; -extern LLVMPassManagerRef lp_build_pass; +struct gallivm_state +{ + LLVMModuleRef module; + LLVMExecutionEngineRef engine; + LLVMModuleProviderRef provider; + LLVMTargetDataRef target; + LLVMPassManagerRef passmgr; + LLVMContextRef context; + LLVMBuilderRef builder; +}; void lp_build_init(void); + extern void lp_func_delete_body(LLVMValueRef func); +void +gallivm_garbage_collect(struct gallivm_state *gallivm); + + +typedef void (*garbage_collect_callback_func)(void *cb_data); + +void +gallivm_register_garbage_collector_callback(garbage_collect_callback_func func, + void *cb_data); + +void +gallivm_remove_garbage_collector_callback(garbage_collect_callback_func func, + void *cb_data); + + +struct gallivm_state * +gallivm_create(void); + +void +gallivm_destroy(struct gallivm_state *gallivm); + + extern LLVMValueRef lp_build_load_volatile(LLVMBuilderRef B, LLVMValueRef PointerVal, const char *Name); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_intr.c b/src/gallium/auxiliary/gallivm/lp_bld_intr.c index 9895749d56..518a01fdb9 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_intr.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_intr.c @@ -46,6 +46,7 @@ #include "util/u_debug.h" +#include "lp_bld_const.h" #include "lp_bld_intr.h" @@ -136,12 +137,13 @@ lp_build_intrinsic_binary(LLVMBuilderRef builder, LLVMValueRef -lp_build_intrinsic_map(LLVMBuilderRef builder, +lp_build_intrinsic_map(struct gallivm_state *gallivm, const char *name, LLVMTypeRef ret_type, LLVMValueRef *args, unsigned num_args) { + LLVMBuilderRef builder = gallivm->builder; LLVMTypeRef ret_elem_type = LLVMGetElementType(ret_type); unsigned n = LLVMGetVectorSize(ret_type); unsigned i, j; @@ -151,7 +153,7 @@ lp_build_intrinsic_map(LLVMBuilderRef builder, res = LLVMGetUndef(ret_type); for(i = 0; i < n; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, i); LLVMValueRef arg_elems[LP_MAX_FUNC_ARGS]; LLVMValueRef res_elem; for(j = 0; j < num_args; ++j) @@ -165,17 +167,17 @@ lp_build_intrinsic_map(LLVMBuilderRef builder, LLVMValueRef -lp_build_intrinsic_map_unary(LLVMBuilderRef builder, +lp_build_intrinsic_map_unary(struct gallivm_state *gallivm, const char *name, LLVMTypeRef ret_type, LLVMValueRef a) { - return lp_build_intrinsic_map(builder, name, ret_type, &a, 1); + return lp_build_intrinsic_map(gallivm, name, ret_type, &a, 1); } LLVMValueRef -lp_build_intrinsic_map_binary(LLVMBuilderRef builder, +lp_build_intrinsic_map_binary(struct gallivm_state *gallivm, const char *name, LLVMTypeRef ret_type, LLVMValueRef a, @@ -186,7 +188,7 @@ lp_build_intrinsic_map_binary(LLVMBuilderRef builder, args[0] = a; args[1] = b; - return lp_build_intrinsic_map(builder, name, ret_type, args, 2); + return lp_build_intrinsic_map(gallivm, name, ret_type, args, 2); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_intr.h b/src/gallium/auxiliary/gallivm/lp_bld_intr.h index 977f767322..b73dd70036 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_intr.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_intr.h @@ -38,6 +38,7 @@ #include "gallivm/lp_bld.h" +#include "gallivm/lp_bld_init.h" /** @@ -77,7 +78,7 @@ lp_build_intrinsic_binary(LLVMBuilderRef builder, LLVMValueRef -lp_build_intrinsic_map(LLVMBuilderRef builder, +lp_build_intrinsic_map(struct gallivm_state *gallivm, const char *name, LLVMTypeRef ret_type, LLVMValueRef *args, @@ -85,14 +86,14 @@ lp_build_intrinsic_map(LLVMBuilderRef builder, LLVMValueRef -lp_build_intrinsic_map_unary(LLVMBuilderRef builder, +lp_build_intrinsic_map_unary(struct gallivm_state *gallivm, const char *name, LLVMTypeRef ret_type, LLVMValueRef a); LLVMValueRef -lp_build_intrinsic_map_binary(LLVMBuilderRef builder, +lp_build_intrinsic_map_binary(struct gallivm_state *gallivm, const char *name, LLVMTypeRef ret_type, LLVMValueRef a, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.c b/src/gallium/auxiliary/gallivm/lp_bld_logic.c index 026b60ac36..f7e6fbaff1 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_logic.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.c @@ -39,6 +39,7 @@ #include "lp_bld_type.h" #include "lp_bld_const.h" +#include "lp_bld_init.h" #include "lp_bld_intr.h" #include "lp_bld_debug.h" #include "lp_bld_logic.h" @@ -70,13 +71,14 @@ * The result values will be 0 for false or ~0 for true. */ LLVMValueRef -lp_build_compare(LLVMBuilderRef builder, +lp_build_compare(struct gallivm_state *gallivm, const struct lp_type type, unsigned func, LLVMValueRef a, LLVMValueRef b) { - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); + LLVMBuilderRef builder = gallivm->builder; + LLVMTypeRef int_vec_type = lp_build_int_vec_type(gallivm, type); LLVMValueRef zeros = LLVMConstNull(int_vec_type); LLVMValueRef ones = LLVMConstAllOnes(int_vec_type); LLVMValueRef cond; @@ -115,7 +117,7 @@ lp_build_compare(LLVMBuilderRef builder, if(type.width * type.length == 128) { if(type.floating && util_cpu_caps.has_sse) { /* float[4] comparison */ - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(gallivm, type); LLVMValueRef args[3]; unsigned cc; boolean swap; @@ -144,7 +146,7 @@ lp_build_compare(LLVMBuilderRef builder, break; default: assert(0); - return lp_build_undef(type); + return lp_build_undef(gallivm, type); } if(swap) { @@ -156,7 +158,7 @@ lp_build_compare(LLVMBuilderRef builder, args[1] = b; } - args[2] = LLVMConstInt(LLVMInt8Type(), cc, 0); + args[2] = LLVMConstInt(LLVMInt8TypeInContext(gallivm->context), cc, 0); res = lp_build_intrinsic(builder, "llvm.x86.sse.cmp.ps", vec_type, @@ -185,7 +187,7 @@ lp_build_compare(LLVMBuilderRef builder, const char *pcmpgt; LLVMValueRef args[2]; LLVMValueRef res; - LLVMTypeRef vec_type = lp_build_vec_type(type); + LLVMTypeRef vec_type = lp_build_vec_type(gallivm, type); switch (type.width) { case 8: @@ -202,14 +204,14 @@ lp_build_compare(LLVMBuilderRef builder, break; default: assert(0); - return lp_build_undef(type); + return lp_build_undef(gallivm, type); } /* There are no unsigned comparison instructions. So flip the sign bit * so that the results match. */ if (table[func].gt && !type.sign) { - LLVMValueRef msb = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1)); + LLVMValueRef msb = lp_build_const_int_vec(gallivm, type, (unsigned long long)1 << (type.width - 1)); a = LLVMBuildXor(builder, a, msb, ""); b = LLVMBuildXor(builder, b, msb, ""); } @@ -270,7 +272,7 @@ lp_build_compare(LLVMBuilderRef builder, break; default: assert(0); - return lp_build_undef(type); + return lp_build_undef(gallivm, type); } #if HAVE_LLVM >= 0x0207 @@ -289,7 +291,7 @@ lp_build_compare(LLVMBuilderRef builder, debug_printf("%s: warning: using slow element-wise float" " vector comparison\n", __FUNCTION__); for (i = 0; i < type.length; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, i); cond = LLVMBuildFCmp(builder, op, LLVMBuildExtractElement(builder, a, index, ""), LLVMBuildExtractElement(builder, b, index, ""), @@ -326,7 +328,7 @@ lp_build_compare(LLVMBuilderRef builder, break; default: assert(0); - return lp_build_undef(type); + return lp_build_undef(gallivm, type); } #if HAVE_LLVM >= 0x0207 @@ -348,7 +350,7 @@ lp_build_compare(LLVMBuilderRef builder, } for(i = 0; i < type.length; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, i); cond = LLVMBuildICmp(builder, op, LLVMBuildExtractElement(builder, a, index, ""), LLVMBuildExtractElement(builder, b, index, ""), @@ -379,7 +381,7 @@ lp_build_cmp(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - return lp_build_compare(bld->builder, bld->type, func, a, b); + return lp_build_compare(bld->gallivm, bld->type, func, a, b); } @@ -392,6 +394,7 @@ lp_build_select_bitwise(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; struct lp_type type = bld->type; LLVMValueRef res; @@ -403,25 +406,25 @@ lp_build_select_bitwise(struct lp_build_context *bld, } if(type.floating) { - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); - a = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); - b = LLVMBuildBitCast(bld->builder, b, int_vec_type, ""); + LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type); + a = LLVMBuildBitCast(builder, a, int_vec_type, ""); + b = LLVMBuildBitCast(builder, b, int_vec_type, ""); } - a = LLVMBuildAnd(bld->builder, a, mask, ""); + a = LLVMBuildAnd(builder, a, mask, ""); /* This often gets translated to PANDN, but sometimes the NOT is * pre-computed and stored in another constant. The best strategy depends * on available registers, so it is not a big deal -- hopefully LLVM does * the right decision attending the rest of the program. */ - b = LLVMBuildAnd(bld->builder, b, LLVMBuildNot(bld->builder, mask, ""), ""); + b = LLVMBuildAnd(builder, b, LLVMBuildNot(builder, mask, ""), ""); - res = LLVMBuildOr(bld->builder, a, b, ""); + res = LLVMBuildOr(builder, a, b, ""); if(type.floating) { - LLVMTypeRef vec_type = lp_build_vec_type(type); - res = LLVMBuildBitCast(bld->builder, res, vec_type, ""); + LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type); + res = LLVMBuildBitCast(builder, res, vec_type, ""); } return res; @@ -440,6 +443,8 @@ lp_build_select(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; + LLVMContextRef lc = bld->gallivm->context; struct lp_type type = bld->type; LLVMValueRef res; @@ -450,8 +455,8 @@ lp_build_select(struct lp_build_context *bld, return a; if (type.length == 1) { - mask = LLVMBuildTrunc(bld->builder, mask, LLVMInt1Type(), ""); - res = LLVMBuildSelect(bld->builder, mask, a, b, ""); + mask = LLVMBuildTrunc(builder, mask, LLVMInt1TypeInContext(lc), ""); + res = LLVMBuildSelect(builder, mask, a, b, ""); } else if (util_cpu_caps.has_sse4_1 && type.width * type.length == 128 && @@ -465,34 +470,34 @@ lp_build_select(struct lp_build_context *bld, if (type.floating && type.width == 64) { intrinsic = "llvm.x86.sse41.blendvpd"; - arg_type = LLVMVectorType(LLVMDoubleType(), 2); + arg_type = LLVMVectorType(LLVMDoubleTypeInContext(lc), 2); } else if (type.floating && type.width == 32) { intrinsic = "llvm.x86.sse41.blendvps"; - arg_type = LLVMVectorType(LLVMFloatType(), 4); + arg_type = LLVMVectorType(LLVMFloatTypeInContext(lc), 4); } else { intrinsic = "llvm.x86.sse41.pblendvb"; - arg_type = LLVMVectorType(LLVMInt8Type(), 16); + arg_type = LLVMVectorType(LLVMInt8TypeInContext(lc), 16); } if (arg_type != bld->int_vec_type) { - mask = LLVMBuildBitCast(bld->builder, mask, arg_type, ""); + mask = LLVMBuildBitCast(builder, mask, arg_type, ""); } if (arg_type != bld->vec_type) { - a = LLVMBuildBitCast(bld->builder, a, arg_type, ""); - b = LLVMBuildBitCast(bld->builder, b, arg_type, ""); + a = LLVMBuildBitCast(builder, a, arg_type, ""); + b = LLVMBuildBitCast(builder, b, arg_type, ""); } args[0] = b; args[1] = a; args[2] = mask; - res = lp_build_intrinsic(bld->builder, intrinsic, + res = lp_build_intrinsic(builder, intrinsic, arg_type, args, Elements(args)); if (arg_type != bld->vec_type) { - res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, ""); + res = LLVMBuildBitCast(builder, res, bld->vec_type, ""); } } else { @@ -514,6 +519,7 @@ lp_build_select_aos(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; const unsigned n = type.length; unsigned i, j; @@ -544,7 +550,7 @@ lp_build_select_aos(struct lp_build_context *bld, /* * Shuffle. */ - LLVMTypeRef elem_type = LLVMInt32Type(); + LLVMTypeRef elem_type = LLVMInt32TypeInContext(bld->gallivm->context); LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH]; for(j = 0; j < n; j += 4) @@ -553,7 +559,7 @@ lp_build_select_aos(struct lp_build_context *bld, (mask & (1 << i) ? 0 : n) + j + i, 0); - return LLVMBuildShuffleVector(bld->builder, a, b, LLVMConstVector(shuffles, n), ""); + return LLVMBuildShuffleVector(builder, a, b, LLVMConstVector(shuffles, n), ""); } else { #if 0 @@ -567,9 +573,9 @@ lp_build_select_aos(struct lp_build_context *bld, cond_vec[j + i] = LLVMConstInt(elem_type, mask & (1 << i) ? 1 : 0, 0); - return LLVMBuildSelect(bld->builder, LLVMConstVector(cond_vec, n), a, b, ""); + return LLVMBuildSelect(builder, LLVMConstVector(cond_vec, n), a, b, ""); #else - LLVMValueRef mask_vec = lp_build_const_mask_aos(type, mask); + LLVMValueRef mask_vec = lp_build_const_mask_aos(bld->gallivm, type, mask); return lp_build_select(bld, mask_vec, a, b); #endif } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.h b/src/gallium/auxiliary/gallivm/lp_bld_logic.h index 141fb92058..ef33a65368 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_logic.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.h @@ -47,7 +47,7 @@ struct lp_build_context; LLVMValueRef -lp_build_compare(LLVMBuilderRef builder, +lp_build_compare(struct gallivm_state *gallivm, const struct lp_type type, unsigned func, LLVMValueRef a, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp b/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp index f56ddee7fd..46dd00d822 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp +++ b/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp @@ -144,6 +144,7 @@ lp_set_target_options(void) llvm::UnsafeFPMath = true; #endif +#if HAVE_LLVM < 0x0209 /* * LLVM will generate MMX instructions for vectors <= 64 bits, leading to * innefficient code, and in 32bit systems, to the corruption of the FPU @@ -162,6 +163,7 @@ lp_set_target_options(void) llvm::cl::ParseCommandLineOptions(2, const_cast<char**>(options)); first = FALSE; } +#endif /* * By default LLVM adds a signal handler to output a pretty stack trace. diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.c b/src/gallium/auxiliary/gallivm/lp_bld_pack.c index f7eb7148ab..fde6bb594f 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_pack.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.c @@ -72,6 +72,7 @@ #include "lp_bld_type.h" #include "lp_bld_const.h" +#include "lp_bld_init.h" #include "lp_bld_intr.h" #include "lp_bld_arit.h" #include "lp_bld_pack.h" @@ -81,7 +82,8 @@ * Build shuffle vectors that match PUNPCKLxx and PUNPCKHxx instructions. */ static LLVMValueRef -lp_build_const_unpack_shuffle(unsigned n, unsigned lo_hi) +lp_build_const_unpack_shuffle(struct gallivm_state *gallivm, + unsigned n, unsigned lo_hi) { LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; unsigned i, j; @@ -92,8 +94,8 @@ lp_build_const_unpack_shuffle(unsigned n, unsigned lo_hi) /* TODO: cache results in a static table */ for(i = 0, j = lo_hi*n/2; i < n; i += 2, ++j) { - elems[i + 0] = LLVMConstInt(LLVMInt32Type(), 0 + j, 0); - elems[i + 1] = LLVMConstInt(LLVMInt32Type(), n + j, 0); + elems[i + 0] = lp_build_const_int32(gallivm, 0 + j); + elems[i + 1] = lp_build_const_int32(gallivm, n + j); } return LLVMConstVector(elems, n); @@ -104,7 +106,7 @@ lp_build_const_unpack_shuffle(unsigned n, unsigned lo_hi) * Build shuffle vectors that match PACKxx instructions. */ static LLVMValueRef -lp_build_const_pack_shuffle(unsigned n) +lp_build_const_pack_shuffle(struct gallivm_state *gallivm, unsigned n) { LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; unsigned i; @@ -112,7 +114,7 @@ lp_build_const_pack_shuffle(unsigned n) assert(n <= LP_MAX_VECTOR_LENGTH); for(i = 0; i < n; ++i) - elems[i] = LLVMConstInt(LLVMInt32Type(), 2*i, 0); + elems[i] = lp_build_const_int32(gallivm, 2*i); return LLVMConstVector(elems, n); } @@ -124,7 +126,7 @@ lp_build_const_pack_shuffle(unsigned n) * Matches the PUNPCKLxx and PUNPCKHxx SSE instructions. */ LLVMValueRef -lp_build_interleave2(LLVMBuilderRef builder, +lp_build_interleave2(struct gallivm_state *gallivm, struct lp_type type, LLVMValueRef a, LLVMValueRef b, @@ -132,9 +134,9 @@ lp_build_interleave2(LLVMBuilderRef builder, { LLVMValueRef shuffle; - shuffle = lp_build_const_unpack_shuffle(type.length, lo_hi); + shuffle = lp_build_const_unpack_shuffle(gallivm, type.length, lo_hi); - return LLVMBuildShuffleVector(builder, a, b, shuffle, ""); + return LLVMBuildShuffleVector(gallivm->builder, a, b, shuffle, ""); } @@ -145,13 +147,14 @@ lp_build_interleave2(LLVMBuilderRef builder, * values themselves. */ void -lp_build_unpack2(LLVMBuilderRef builder, +lp_build_unpack2(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef src, LLVMValueRef *dst_lo, LLVMValueRef *dst_hi) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef msb; LLVMTypeRef dst_vec_type; @@ -162,24 +165,24 @@ lp_build_unpack2(LLVMBuilderRef builder, if(dst_type.sign && src_type.sign) { /* Replicate the sign bit in the most significant bits */ - msb = LLVMBuildAShr(builder, src, lp_build_const_int_vec(src_type, src_type.width - 1), ""); + msb = LLVMBuildAShr(builder, src, lp_build_const_int_vec(gallivm, src_type, src_type.width - 1), ""); } else /* Most significant bits always zero */ - msb = lp_build_zero(src_type); + msb = lp_build_zero(gallivm, src_type); /* Interleave bits */ #ifdef PIPE_ARCH_LITTLE_ENDIAN - *dst_lo = lp_build_interleave2(builder, src_type, src, msb, 0); - *dst_hi = lp_build_interleave2(builder, src_type, src, msb, 1); + *dst_lo = lp_build_interleave2(gallivm, src_type, src, msb, 0); + *dst_hi = lp_build_interleave2(gallivm, src_type, src, msb, 1); #else - *dst_lo = lp_build_interleave2(builder, src_type, msb, src, 0); - *dst_hi = lp_build_interleave2(builder, src_type, msb, src, 1); + *dst_lo = lp_build_interleave2(gallivm, src_type, msb, src, 0); + *dst_hi = lp_build_interleave2(gallivm, src_type, msb, src, 1); #endif /* Cast the result into the new type (twice as wide) */ - dst_vec_type = lp_build_vec_type(dst_type); + dst_vec_type = lp_build_vec_type(gallivm, dst_type); *dst_lo = LLVMBuildBitCast(builder, *dst_lo, dst_vec_type, ""); *dst_hi = LLVMBuildBitCast(builder, *dst_hi, dst_vec_type, ""); @@ -193,7 +196,7 @@ lp_build_unpack2(LLVMBuilderRef builder, * values themselves. */ void -lp_build_unpack(LLVMBuilderRef builder, +lp_build_unpack(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef src, @@ -218,7 +221,7 @@ lp_build_unpack(LLVMBuilderRef builder, tmp_type.length /= 2; for(i = num_tmps; i--; ) { - lp_build_unpack2(builder, src_type, tmp_type, dst[i], &dst[2*i + 0], &dst[2*i + 1]); + lp_build_unpack2(gallivm, src_type, tmp_type, dst[i], &dst[2*i + 0], &dst[2*i + 1]); } src_type = tmp_type; @@ -247,16 +250,17 @@ lp_build_unpack(LLVMBuilderRef builder, * lp_build_packs2 instead. */ LLVMValueRef -lp_build_pack2(LLVMBuilderRef builder, +lp_build_pack2(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef lo, LLVMValueRef hi) { + LLVMBuilderRef builder = gallivm->builder; #if HAVE_LLVM < 0x0207 - LLVMTypeRef src_vec_type = lp_build_vec_type(src_type); + LLVMTypeRef src_vec_type = lp_build_vec_type(gallivm, src_type); #endif - LLVMTypeRef dst_vec_type = lp_build_vec_type(dst_type); + LLVMTypeRef dst_vec_type = lp_build_vec_type(gallivm, dst_type); LLVMValueRef shuffle; LLVMValueRef res = NULL; @@ -318,7 +322,7 @@ lp_build_pack2(LLVMBuilderRef builder, lo = LLVMBuildBitCast(builder, lo, dst_vec_type, ""); hi = LLVMBuildBitCast(builder, hi, dst_vec_type, ""); - shuffle = lp_build_const_pack_shuffle(dst_type.length); + shuffle = lp_build_const_pack_shuffle(gallivm, dst_type.length); res = LLVMBuildShuffleVector(builder, lo, hi, shuffle, ""); @@ -334,7 +338,7 @@ lp_build_pack2(LLVMBuilderRef builder, * destination type. */ LLVMValueRef -lp_build_packs2(LLVMBuilderRef builder, +lp_build_packs2(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef lo, @@ -360,14 +364,14 @@ lp_build_packs2(LLVMBuilderRef builder, if(clamp) { struct lp_build_context bld; unsigned dst_bits = dst_type.sign ? dst_type.width - 1 : dst_type.width; - LLVMValueRef dst_max = lp_build_const_int_vec(src_type, ((unsigned long long)1 << dst_bits) - 1); - lp_build_context_init(&bld, builder, src_type); + LLVMValueRef dst_max = lp_build_const_int_vec(gallivm, src_type, ((unsigned long long)1 << dst_bits) - 1); + lp_build_context_init(&bld, gallivm, src_type); lo = lp_build_min(&bld, lo, dst_max); hi = lp_build_min(&bld, hi, dst_max); /* FIXME: What about lower bound? */ } - return lp_build_pack2(builder, src_type, dst_type, lo, hi); + return lp_build_pack2(gallivm, src_type, dst_type, lo, hi); } @@ -377,13 +381,13 @@ lp_build_packs2(LLVMBuilderRef builder, * TODO: Handle saturation consistently. */ LLVMValueRef -lp_build_pack(LLVMBuilderRef builder, +lp_build_pack(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, boolean clamped, const LLVMValueRef *src, unsigned num_srcs) { - LLVMValueRef (*pack2)(LLVMBuilderRef builder, + LLVMValueRef (*pack2)(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef lo, @@ -419,7 +423,8 @@ lp_build_pack(LLVMBuilderRef builder, num_srcs /= 2; for(i = 0; i < num_srcs; ++i) - tmp[i] = pack2(builder, src_type, tmp_type, tmp[2*i + 0], tmp[2*i + 1]); + tmp[i] = pack2(gallivm, src_type, tmp_type, + tmp[2*i + 0], tmp[2*i + 1]); src_type = tmp_type; } @@ -437,12 +442,13 @@ lp_build_pack(LLVMBuilderRef builder, * intrinsics that do saturation. */ void -lp_build_resize(LLVMBuilderRef builder, +lp_build_resize(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, const LLVMValueRef *src, unsigned num_srcs, LLVMValueRef *dst, unsigned num_dsts) { + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH]; unsigned i; @@ -482,7 +488,7 @@ lp_build_resize(LLVMBuilderRef builder, * Register width remains constant -- use vector packing intrinsics */ - tmp[0] = lp_build_pack(builder, src_type, dst_type, TRUE, src, num_srcs); + tmp[0] = lp_build_pack(gallivm, src_type, dst_type, TRUE, src, num_srcs); } else { /* @@ -490,11 +496,11 @@ lp_build_resize(LLVMBuilderRef builder, */ assert(src_type.length == dst_type.length); - tmp[0] = lp_build_undef(dst_type); + tmp[0] = lp_build_undef(gallivm, dst_type); for (i = 0; i < dst_type.length; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, i); LLVMValueRef val = LLVMBuildExtractElement(builder, src[0], index, ""); - val = LLVMBuildTrunc(builder, val, lp_build_elem_type(dst_type), ""); + val = LLVMBuildTrunc(builder, val, lp_build_elem_type(gallivm, dst_type), ""); tmp[0] = LLVMBuildInsertElement(builder, tmp[0], val, index, ""); } } @@ -510,7 +516,7 @@ lp_build_resize(LLVMBuilderRef builder, /* * Register width remains constant -- use vector unpack intrinsics */ - lp_build_unpack(builder, src_type, dst_type, src[0], tmp, num_dsts); + lp_build_unpack(gallivm, src_type, dst_type, src[0], tmp, num_dsts); } else { /* @@ -518,15 +524,15 @@ lp_build_resize(LLVMBuilderRef builder, */ assert(src_type.length == dst_type.length); - tmp[0] = lp_build_undef(dst_type); + tmp[0] = lp_build_undef(gallivm, dst_type); for (i = 0; i < dst_type.length; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, i); LLVMValueRef val = LLVMBuildExtractElement(builder, src[0], index, ""); if (src_type.sign && dst_type.sign) { - val = LLVMBuildSExt(builder, val, lp_build_elem_type(dst_type), ""); + val = LLVMBuildSExt(builder, val, lp_build_elem_type(gallivm, dst_type), ""); } else { - val = LLVMBuildZExt(builder, val, lp_build_elem_type(dst_type), ""); + val = LLVMBuildZExt(builder, val, lp_build_elem_type(gallivm, dst_type), ""); } tmp[0] = LLVMBuildInsertElement(builder, tmp[0], val, index, ""); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.h b/src/gallium/auxiliary/gallivm/lp_bld_pack.h index e947b90d16..d58da4f01b 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_pack.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.h @@ -46,7 +46,7 @@ struct lp_type; LLVMValueRef -lp_build_interleave2(LLVMBuilderRef builder, +lp_build_interleave2(struct gallivm_state *gallivm, struct lp_type type, LLVMValueRef a, LLVMValueRef b, @@ -54,7 +54,7 @@ lp_build_interleave2(LLVMBuilderRef builder, void -lp_build_unpack2(LLVMBuilderRef builder, +lp_build_unpack2(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef src, @@ -63,7 +63,7 @@ lp_build_unpack2(LLVMBuilderRef builder, void -lp_build_unpack(LLVMBuilderRef builder, +lp_build_unpack(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef src, @@ -71,7 +71,7 @@ lp_build_unpack(LLVMBuilderRef builder, LLVMValueRef -lp_build_packs2(LLVMBuilderRef builder, +lp_build_packs2(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef lo, @@ -79,7 +79,7 @@ lp_build_packs2(LLVMBuilderRef builder, LLVMValueRef -lp_build_pack2(LLVMBuilderRef builder, +lp_build_pack2(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef lo, @@ -87,7 +87,7 @@ lp_build_pack2(LLVMBuilderRef builder, LLVMValueRef -lp_build_pack(LLVMBuilderRef builder, +lp_build_pack(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, boolean clamped, @@ -95,7 +95,7 @@ lp_build_pack(LLVMBuilderRef builder, void -lp_build_resize(LLVMBuilderRef builder, +lp_build_resize(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, const LLVMValueRef *src, unsigned num_srcs, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_printf.c b/src/gallium/auxiliary/gallivm/lp_bld_printf.c index f418e96aff..60cc6094f5 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_printf.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_printf.c @@ -31,6 +31,8 @@ #include "util/u_memory.h" #include "util/u_string.h" #include "lp_bld_const.h" +#include "lp_bld_init.h" +#include "lp_bld_const.h" #include "lp_bld_printf.h" @@ -65,12 +67,14 @@ lp_get_printf_arg_count(const char *fmt) } LLVMValueRef -lp_build_const_string_variable(LLVMModuleRef module, const char *str, int len) +lp_build_const_string_variable(LLVMModuleRef module, + LLVMContextRef context, + const char *str, int len) { - LLVMValueRef string = LLVMAddGlobal(module, LLVMArrayType(LLVMInt8Type(), len + 1), ""); + LLVMValueRef string = LLVMAddGlobal(module, LLVMArrayType(LLVMInt8TypeInContext(context), len + 1), ""); LLVMSetGlobalConstant(string, TRUE); LLVMSetLinkage(string, LLVMInternalLinkage); - LLVMSetInitializer(string, LLVMConstString(str, len + 1, TRUE)); + LLVMSetInitializer(string, LLVMConstStringInContext(context, str, len + 1, TRUE)); return string; } @@ -83,15 +87,18 @@ lp_build_const_string_variable(LLVMModuleRef module, const char *str, int len) * LLVMValueRef. */ LLVMValueRef -lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...) +lp_build_printf(struct gallivm_state *gallivm, const char *fmt, ...) { va_list arglist; int i = 0; int argcount = lp_get_printf_arg_count(fmt); - LLVMModuleRef module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder))); + LLVMBuilderRef builder = gallivm->builder; + LLVMContextRef context = gallivm->context; + LLVMModuleRef module = gallivm->module; LLVMValueRef params[50]; - LLVMValueRef fmtarg = lp_build_const_string_variable(module, fmt, strlen(fmt) + 1); - LLVMValueRef int0 = LLVMConstInt(LLVMInt32Type(), 0, 0); + LLVMValueRef fmtarg = lp_build_const_string_variable(module, context, + fmt, strlen(fmt) + 1); + LLVMValueRef int0 = lp_build_const_int32(gallivm, 0); LLVMValueRef index[2]; LLVMValueRef func_printf = LLVMGetNamedFunction(module, "printf"); @@ -100,7 +107,7 @@ lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...) index[0] = index[1] = int0; if (!func_printf) { - LLVMTypeRef printf_type = LLVMFunctionType(LLVMIntType(32), NULL, 0, 1); + LLVMTypeRef printf_type = LLVMFunctionType(LLVMIntTypeInContext(context, 32), NULL, 0, 1); func_printf = LLVMAddFunction(module, "printf", printf_type); } @@ -113,7 +120,7 @@ lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...) /* printf wants doubles, so lets convert so that * we can actually print them */ if (LLVMGetTypeKind(type) == LLVMFloatTypeKind) - val = LLVMBuildFPExt(builder, val, LLVMDoubleType(), ""); + val = LLVMBuildFPExt(builder, val, LLVMDoubleTypeInContext(context), ""); params[i] = val; } va_end(arglist); @@ -127,16 +134,18 @@ lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...) * Print a float[4] vector. */ LLVMValueRef -lp_build_print_vec4(LLVMBuilderRef builder, const char *msg, LLVMValueRef vec) +lp_build_print_vec4(struct gallivm_state *gallivm, + const char *msg, LLVMValueRef vec) { + LLVMBuilderRef builder = gallivm->builder; char format[1000]; LLVMValueRef x, y, z, w; - x = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(0), ""); - y = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(1), ""); - z = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(2), ""); - w = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(3), ""); + x = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(gallivm, 0), ""); + y = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(gallivm, 1), ""); + z = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(gallivm, 2), ""); + w = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(gallivm, 3), ""); util_snprintf(format, sizeof(format), "%s %%f %%f %%f %%f\n", msg); - return lp_build_printf(builder, format, x, y, z, w); + return lp_build_printf(gallivm, format, x, y, z, w); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_printf.h b/src/gallium/auxiliary/gallivm/lp_bld_printf.h index b6222c62eb..f6bb834869 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_printf.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_printf.h @@ -31,12 +31,19 @@ #include "pipe/p_compiler.h" #include "lp_bld.h" +#include "lp_bld_init.h" -LLVMValueRef lp_build_const_string_variable(LLVMModuleRef module, const char *str, int len); -LLVMValueRef lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...); + +LLVMValueRef lp_build_const_string_variable(LLVMModuleRef module, + LLVMContextRef context, + const char *str, int len); + +LLVMValueRef lp_build_printf(struct gallivm_state *gallivm, + const char *fmt, ...); LLVMValueRef -lp_build_print_vec4(LLVMBuilderRef builder, const char *msg, LLVMValueRef vec); +lp_build_print_vec4(struct gallivm_state *gallivm, + const char *msg, LLVMValueRef vec); #endif diff --git a/src/gallium/auxiliary/gallivm/lp_bld_quad.c b/src/gallium/auxiliary/gallivm/lp_bld_quad.c index c18c8b4710..b0a5bc0267 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_quad.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_quad.c @@ -28,6 +28,7 @@ #include "lp_bld_type.h" #include "lp_bld_arit.h" +#include "lp_bld_const.h" #include "lp_bld_swizzle.h" #include "lp_bld_quad.h" @@ -81,15 +82,15 @@ LLVMValueRef lp_build_scalar_ddx(struct lp_build_context *bld, LLVMValueRef a) { - LLVMTypeRef i32t = LLVMInt32Type(); - LLVMValueRef idx_left = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0); - LLVMValueRef idx_right = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_RIGHT, 0); - LLVMValueRef a_left = LLVMBuildExtractElement(bld->builder, a, idx_left, "left"); - LLVMValueRef a_right = LLVMBuildExtractElement(bld->builder, a, idx_right, "right"); + LLVMBuilderRef builder = bld->gallivm->builder; + LLVMValueRef idx_left = lp_build_const_int32(bld->gallivm, LP_BLD_QUAD_TOP_LEFT); + LLVMValueRef idx_right = lp_build_const_int32(bld->gallivm, LP_BLD_QUAD_TOP_RIGHT); + LLVMValueRef a_left = LLVMBuildExtractElement(builder, a, idx_left, "left"); + LLVMValueRef a_right = LLVMBuildExtractElement(builder, a, idx_right, "right"); if (bld->type.floating) - return LLVMBuildFSub(bld->builder, a_right, a_left, "ddx"); + return LLVMBuildFSub(builder, a_right, a_left, "ddx"); else - return LLVMBuildSub(bld->builder, a_right, a_left, "ddx"); + return LLVMBuildSub(builder, a_right, a_left, "ddx"); } @@ -97,13 +98,13 @@ LLVMValueRef lp_build_scalar_ddy(struct lp_build_context *bld, LLVMValueRef a) { - LLVMTypeRef i32t = LLVMInt32Type(); - LLVMValueRef idx_top = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0); - LLVMValueRef idx_bottom = LLVMConstInt(i32t, LP_BLD_QUAD_BOTTOM_LEFT, 0); - LLVMValueRef a_top = LLVMBuildExtractElement(bld->builder, a, idx_top, "top"); - LLVMValueRef a_bottom = LLVMBuildExtractElement(bld->builder, a, idx_bottom, "bottom"); + LLVMBuilderRef builder = bld->gallivm->builder; + LLVMValueRef idx_top = lp_build_const_int32(bld->gallivm, LP_BLD_QUAD_TOP_LEFT); + LLVMValueRef idx_bottom = lp_build_const_int32(bld->gallivm, LP_BLD_QUAD_BOTTOM_LEFT); + LLVMValueRef a_top = LLVMBuildExtractElement(builder, a, idx_top, "top"); + LLVMValueRef a_bottom = LLVMBuildExtractElement(builder, a, idx_bottom, "bottom"); if (bld->type.floating) - return LLVMBuildFSub(bld->builder, a_bottom, a_top, "ddy"); + return LLVMBuildFSub(builder, a_bottom, a_top, "ddy"); else - return LLVMBuildSub(bld->builder, a_bottom, a_top, "ddy"); + return LLVMBuildSub(builder, a_bottom, a_top, "ddy"); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.c b/src/gallium/auxiliary/gallivm/lp_bld_sample.c index 844d1d935b..8ad34598a9 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.c @@ -134,7 +134,7 @@ lp_sampler_static_state(struct lp_sampler_static_state *state, state->min_img_filter = sampler->min_img_filter; state->mag_img_filter = sampler->mag_img_filter; - if (view->last_level && sampler->max_lod > 0.0f) { + if (view->u.tex.last_level && sampler->max_lod > 0.0f) { state->min_mip_filter = sampler->min_mip_filter; } else { state->min_mip_filter = PIPE_TEX_MIPFILTER_NONE; @@ -155,7 +155,7 @@ lp_sampler_static_state(struct lp_sampler_static_state *state, state->apply_min_lod = 1; } - if (sampler->max_lod < (float)view->last_level) { + if (sampler->max_lod < (float)view->u.tex.last_level) { state->apply_max_lod = 1; } } @@ -190,7 +190,8 @@ lp_build_rho(struct lp_build_sample_context *bld, struct lp_build_context *float_size_bld = &bld->float_size_bld; struct lp_build_context *float_bld = &bld->float_bld; const unsigned dims = bld->dims; - LLVMTypeRef i32t = LLVMInt32Type(); + LLVMBuilderRef builder = bld->gallivm->builder; + LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context); LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0); LLVMValueRef index1 = LLVMConstInt(i32t, 1, 0); LLVMValueRef index2 = LLVMConstInt(i32t, 2, 0); @@ -211,21 +212,21 @@ lp_build_rho(struct lp_build_sample_context *bld, rho_x = float_size_bld->undef; rho_y = float_size_bld->undef; - rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dsdx, index0, ""); - rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dsdy, index0, ""); + rho_x = LLVMBuildInsertElement(builder, rho_x, dsdx, index0, ""); + rho_y = LLVMBuildInsertElement(builder, rho_y, dsdy, index0, ""); dtdx = ddx[1]; dtdy = ddy[1]; - rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dtdx, index1, ""); - rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dtdy, index1, ""); + rho_x = LLVMBuildInsertElement(builder, rho_x, dtdx, index1, ""); + rho_y = LLVMBuildInsertElement(builder, rho_y, dtdy, index1, ""); if (dims >= 3) { drdx = ddx[2]; drdy = ddy[2]; - rho_x = LLVMBuildInsertElement(bld->builder, rho_x, drdx, index2, ""); - rho_y = LLVMBuildInsertElement(bld->builder, rho_y, drdy, index2, ""); + rho_x = LLVMBuildInsertElement(builder, rho_x, drdx, index2, ""); + rho_y = LLVMBuildInsertElement(builder, rho_y, drdy, index2, ""); } } @@ -245,13 +246,13 @@ lp_build_rho(struct lp_build_sample_context *bld, if (dims >= 2) { LLVMValueRef rho_s, rho_t, rho_r; - rho_s = LLVMBuildExtractElement(bld->builder, rho_vec, index0, ""); - rho_t = LLVMBuildExtractElement(bld->builder, rho_vec, index1, ""); + rho_s = LLVMBuildExtractElement(builder, rho_vec, index0, ""); + rho_t = LLVMBuildExtractElement(builder, rho_vec, index1, ""); rho = lp_build_max(float_bld, rho_s, rho_t); if (dims >= 3) { - rho_r = LLVMBuildExtractElement(bld->builder, rho_vec, index0, ""); + rho_r = LLVMBuildExtractElement(builder, rho_vec, index0, ""); rho = lp_build_max(float_bld, rho, rho_r); } } @@ -304,19 +305,19 @@ lp_build_brilinear_lod(struct lp_build_context *bld, double post_offset = 1 - factor; if (0) { - lp_build_printf(bld->builder, "lod = %f\n", lod); + lp_build_printf(bld->gallivm, "lod = %f\n", lod); } lod = lp_build_add(bld, lod, - lp_build_const_vec(bld->type, pre_offset)); + lp_build_const_vec(bld->gallivm, bld->type, pre_offset)); lp_build_ifloor_fract(bld, lod, out_lod_ipart, &lod_fpart); lod_fpart = lp_build_mul(bld, lod_fpart, - lp_build_const_vec(bld->type, factor)); + lp_build_const_vec(bld->gallivm, bld->type, factor)); lod_fpart = lp_build_add(bld, lod_fpart, - lp_build_const_vec(bld->type, post_offset)); + lp_build_const_vec(bld->gallivm, bld->type, post_offset)); /* * It's not necessary to clamp lod_fpart since: @@ -327,8 +328,8 @@ lp_build_brilinear_lod(struct lp_build_context *bld, *out_lod_fpart = lod_fpart; if (0) { - lp_build_printf(bld->builder, "lod_ipart = %i\n", *out_lod_ipart); - lp_build_printf(bld->builder, "lod_fpart = %f\n\n", *out_lod_fpart); + lp_build_printf(bld->gallivm, "lod_ipart = %i\n", *out_lod_ipart); + lp_build_printf(bld->gallivm, "lod_fpart = %f\n\n", *out_lod_fpart); } } @@ -363,7 +364,7 @@ lp_build_brilinear_rho(struct lp_build_context *bld, * part will not need any post adjustments. */ rho = lp_build_mul(bld, rho, - lp_build_const_vec(bld->type, pre_factor)); + lp_build_const_vec(bld->gallivm, bld->type, pre_factor)); /* ipart = ifloor(log2(rho)) */ lod_ipart = lp_build_extract_exponent(bld, rho, 0); @@ -372,10 +373,10 @@ lp_build_brilinear_rho(struct lp_build_context *bld, lod_fpart = lp_build_extract_mantissa(bld, rho); lod_fpart = lp_build_mul(bld, lod_fpart, - lp_build_const_vec(bld->type, factor)); + lp_build_const_vec(bld->gallivm, bld->type, factor)); lod_fpart = lp_build_add(bld, lod_fpart, - lp_build_const_vec(bld->type, post_offset)); + lp_build_const_vec(bld->gallivm, bld->type, post_offset)); /* * Like lp_build_brilinear_lod, it's not necessary to clamp lod_fpart since: @@ -413,6 +414,7 @@ lp_build_lod_selector(struct lp_build_sample_context *bld, LLVMValueRef *out_lod_fpart) { + LLVMBuilderRef builder = bld->gallivm->builder; struct lp_build_context *float_bld = &bld->float_bld; LLVMValueRef lod; @@ -424,17 +426,17 @@ lp_build_lod_selector(struct lp_build_sample_context *bld, * This is hit during mipmap generation. */ LLVMValueRef min_lod = - bld->dynamic_state->min_lod(bld->dynamic_state, bld->builder, unit); + bld->dynamic_state->min_lod(bld->dynamic_state, bld->gallivm, unit); lod = min_lod; } else { LLVMValueRef sampler_lod_bias = - bld->dynamic_state->lod_bias(bld->dynamic_state, bld->builder, unit); - LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0); + bld->dynamic_state->lod_bias(bld->dynamic_state, bld->gallivm, unit); + LLVMValueRef index0 = lp_build_const_int32(bld->gallivm, 0); if (explicit_lod) { - lod = LLVMBuildExtractElement(bld->builder, explicit_lod, + lod = LLVMBuildExtractElement(builder, explicit_lod, index0, ""); } else { @@ -479,27 +481,27 @@ lp_build_lod_selector(struct lp_build_sample_context *bld, /* add shader lod bias */ if (lod_bias) { - lod_bias = LLVMBuildExtractElement(bld->builder, lod_bias, + lod_bias = LLVMBuildExtractElement(builder, lod_bias, index0, ""); - lod = LLVMBuildFAdd(bld->builder, lod, lod_bias, "shader_lod_bias"); + lod = LLVMBuildFAdd(builder, lod, lod_bias, "shader_lod_bias"); } } /* add sampler lod bias */ if (bld->static_state->lod_bias_non_zero) - lod = LLVMBuildFAdd(bld->builder, lod, sampler_lod_bias, "sampler_lod_bias"); + lod = LLVMBuildFAdd(builder, lod, sampler_lod_bias, "sampler_lod_bias"); /* clamp lod */ if (bld->static_state->apply_max_lod) { LLVMValueRef max_lod = - bld->dynamic_state->max_lod(bld->dynamic_state, bld->builder, unit); + bld->dynamic_state->max_lod(bld->dynamic_state, bld->gallivm, unit); lod = lp_build_min(float_bld, lod, max_lod); } if (bld->static_state->apply_min_lod) { LLVMValueRef min_lod = - bld->dynamic_state->min_lod(bld->dynamic_state, bld->builder, unit); + bld->dynamic_state->min_lod(bld->dynamic_state, bld->gallivm, unit); lod = lp_build_max(float_bld, lod, min_lod); } @@ -542,10 +544,10 @@ lp_build_nearest_mip_level(struct lp_build_sample_context *bld, struct lp_build_context *int_bld = &bld->int_bld; LLVMValueRef last_level, level; - LLVMValueRef zero = LLVMConstInt(LLVMInt32Type(), 0, 0); + LLVMValueRef zero = lp_build_const_int32(bld->gallivm, 0); last_level = bld->dynamic_state->last_level(bld->dynamic_state, - bld->builder, unit); + bld->gallivm, unit); /* convert float lod to integer */ level = lod_ipart; @@ -568,7 +570,7 @@ lp_build_linear_mip_levels(struct lp_build_sample_context *bld, LLVMValueRef *level0_out, LLVMValueRef *level1_out) { - LLVMBuilderRef builder = bld->builder; + LLVMBuilderRef builder = bld->gallivm->builder; struct lp_build_context *int_bld = &bld->int_bld; struct lp_build_context *float_bld = &bld->float_bld; LLVMValueRef last_level; @@ -579,7 +581,7 @@ lp_build_linear_mip_levels(struct lp_build_sample_context *bld, *level1_out = lp_build_add(int_bld, lod_ipart, int_bld->one); last_level = bld->dynamic_state->last_level(bld->dynamic_state, - bld->builder, unit); + bld->gallivm, unit); /* * Clamp both lod_ipart and lod_ipart + 1 to [0, last_level], with the @@ -630,11 +632,13 @@ LLVMValueRef lp_build_get_mipmap_level(struct lp_build_sample_context *bld, LLVMValueRef level) { + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef indexes[2], data_ptr; - indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); + + indexes[0] = lp_build_const_int32(bld->gallivm, 0); indexes[1] = level; - data_ptr = LLVMBuildGEP(bld->builder, bld->data_array, indexes, 2, ""); - data_ptr = LLVMBuildLoad(bld->builder, data_ptr, ""); + data_ptr = LLVMBuildGEP(builder, bld->data_array, indexes, 2, ""); + data_ptr = LLVMBuildLoad(builder, data_ptr, ""); return data_ptr; } @@ -643,7 +647,7 @@ LLVMValueRef lp_build_get_const_mipmap_level(struct lp_build_sample_context *bld, int level) { - LLVMValueRef lvl = LLVMConstInt(LLVMInt32Type(), level, 0); + LLVMValueRef lvl = lp_build_const_int32(bld->gallivm, level); return lp_build_get_mipmap_level(bld, lvl); } @@ -657,6 +661,7 @@ lp_build_minify(struct lp_build_context *bld, LLVMValueRef base_size, LLVMValueRef level) { + LLVMBuilderRef builder = bld->gallivm->builder; assert(lp_check_value(bld->type, base_size)); assert(lp_check_value(bld->type, level)); @@ -666,7 +671,7 @@ lp_build_minify(struct lp_build_context *bld, } else { LLVMValueRef size = - LLVMBuildLShr(bld->builder, base_size, level, "minify"); + LLVMBuildLShr(builder, base_size, level, "minify"); assert(bld->type.sign); size = lp_build_max(bld, size, bld->one); return size; @@ -682,11 +687,12 @@ static LLVMValueRef lp_build_get_level_stride_vec(struct lp_build_sample_context *bld, LLVMValueRef stride_array, LLVMValueRef level) { + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef indexes[2], stride; - indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); + indexes[0] = lp_build_const_int32(bld->gallivm, 0); indexes[1] = level; - stride = LLVMBuildGEP(bld->builder, stride_array, indexes, 2, ""); - stride = LLVMBuildLoad(bld->builder, stride, ""); + stride = LLVMBuildGEP(builder, stride_array, indexes, 2, ""); + stride = LLVMBuildLoad(builder, stride, ""); stride = lp_build_broadcast_scalar(&bld->int_coord_bld, stride); return stride; } @@ -747,21 +753,21 @@ lp_build_extract_image_sizes(struct lp_build_sample_context *bld, LLVMValueRef *out_depth) { const unsigned dims = bld->dims; - LLVMTypeRef i32t = LLVMInt32Type(); + LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context); - *out_width = lp_build_extract_broadcast(bld->builder, + *out_width = lp_build_extract_broadcast(bld->gallivm, size_type, coord_type, size, LLVMConstInt(i32t, 0, 0)); if (dims >= 2) { - *out_height = lp_build_extract_broadcast(bld->builder, + *out_height = lp_build_extract_broadcast(bld->gallivm, size_type, coord_type, size, LLVMConstInt(i32t, 1, 0)); if (dims == 3) { - *out_depth = lp_build_extract_broadcast(bld->builder, + *out_depth = lp_build_extract_broadcast(bld->gallivm, size_type, coord_type, size, @@ -812,7 +818,7 @@ static LLVMValueRef lp_build_cube_ima(struct lp_build_context *coord_bld, LLVMValueRef coord) { /* ima = -0.5 / abs(coord); */ - LLVMValueRef negHalf = lp_build_const_vec(coord_bld->type, -0.5); + LLVMValueRef negHalf = lp_build_const_vec(coord_bld->gallivm, coord_bld->type, -0.5); LLVMValueRef absCoord = lp_build_abs(coord_bld, coord); LLVMValueRef ima = lp_build_div(coord_bld, negHalf, absCoord); return ima; @@ -831,7 +837,7 @@ lp_build_cube_coord(struct lp_build_context *coord_bld, LLVMValueRef coord, LLVMValueRef ima) { /* return negate(coord) * ima * sign + 0.5; */ - LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5); + LLVMValueRef half = lp_build_const_vec(coord_bld->gallivm, coord_bld->type, 0.5); LLVMValueRef res; assert(negate_coord == +1 || negate_coord == -1); @@ -859,12 +865,14 @@ lp_build_cube_face(struct lp_build_sample_context *bld, LLVMValueRef major_coord, unsigned pos_face, unsigned neg_face) { - LLVMValueRef cmp = LLVMBuildFCmp(bld->builder, LLVMRealUGE, + struct gallivm_state *gallivm = bld->gallivm; + LLVMBuilderRef builder = gallivm->builder; + LLVMValueRef cmp = LLVMBuildFCmp(builder, LLVMRealUGE, major_coord, bld->float_bld.zero, ""); - LLVMValueRef pos = LLVMConstInt(LLVMInt32Type(), pos_face, 0); - LLVMValueRef neg = LLVMConstInt(LLVMInt32Type(), neg_face, 0); - LLVMValueRef res = LLVMBuildSelect(bld->builder, cmp, pos, neg, ""); + LLVMValueRef pos = lp_build_const_int32(gallivm, pos_face); + LLVMValueRef neg = lp_build_const_int32(gallivm, neg_face); + LLVMValueRef res = LLVMBuildSelect(builder, cmp, pos, neg, ""); return res; } @@ -884,9 +892,10 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld, { struct lp_build_context *float_bld = &bld->float_bld; struct lp_build_context *coord_bld = &bld->coord_bld; + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef rx, ry, rz; LLVMValueRef arx, ary, arz; - LLVMValueRef c25 = LLVMConstReal(LLVMFloatType(), 0.25); + LLVMValueRef c25 = lp_build_const_float(bld->gallivm, 0.25); LLVMValueRef arx_ge_ary, arx_ge_arz; LLVMValueRef ary_ge_arx, ary_ge_arz; LLVMValueRef arx_ge_ary_arz, ary_ge_arx_arz; @@ -911,17 +920,17 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld, /* * Compare sign/magnitude of rx,ry,rz to determine face */ - arx_ge_ary = LLVMBuildFCmp(bld->builder, LLVMRealUGE, arx, ary, ""); - arx_ge_arz = LLVMBuildFCmp(bld->builder, LLVMRealUGE, arx, arz, ""); - ary_ge_arx = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ary, arx, ""); - ary_ge_arz = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ary, arz, ""); + arx_ge_ary = LLVMBuildFCmp(builder, LLVMRealUGE, arx, ary, ""); + arx_ge_arz = LLVMBuildFCmp(builder, LLVMRealUGE, arx, arz, ""); + ary_ge_arx = LLVMBuildFCmp(builder, LLVMRealUGE, ary, arx, ""); + ary_ge_arz = LLVMBuildFCmp(builder, LLVMRealUGE, ary, arz, ""); - arx_ge_ary_arz = LLVMBuildAnd(bld->builder, arx_ge_ary, arx_ge_arz, ""); - ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, ""); + arx_ge_ary_arz = LLVMBuildAnd(builder, arx_ge_ary, arx_ge_arz, ""); + ary_ge_arx_arz = LLVMBuildAnd(builder, ary_ge_arx, ary_ge_arz, ""); - rx_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rx, float_bld->zero, ""); - ry_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ry, float_bld->zero, ""); - rz_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rz, float_bld->zero, ""); + rx_pos = LLVMBuildFCmp(builder, LLVMRealUGE, rx, float_bld->zero, ""); + ry_pos = LLVMBuildFCmp(builder, LLVMRealUGE, ry, float_bld->zero, ""); + rz_pos = LLVMBuildFCmp(builder, LLVMRealUGE, rz, float_bld->zero, ""); { struct lp_build_if_state if_ctx; @@ -929,11 +938,11 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld, LLVMValueRef face_t_var; LLVMValueRef face_var; - face_s_var = lp_build_alloca(bld->builder, bld->coord_bld.vec_type, "face_s_var"); - face_t_var = lp_build_alloca(bld->builder, bld->coord_bld.vec_type, "face_t_var"); - face_var = lp_build_alloca(bld->builder, bld->int_bld.vec_type, "face_var"); + face_s_var = lp_build_alloca(bld->gallivm, bld->coord_bld.vec_type, "face_s_var"); + face_t_var = lp_build_alloca(bld->gallivm, bld->coord_bld.vec_type, "face_t_var"); + face_var = lp_build_alloca(bld->gallivm, bld->int_bld.vec_type, "face_var"); - lp_build_if(&if_ctx, bld->builder, arx_ge_ary_arz); + lp_build_if(&if_ctx, bld->gallivm, arx_ge_ary_arz); { /* +/- X face */ LLVMValueRef sign = lp_build_sgn(float_bld, rx); @@ -943,17 +952,17 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld, *face = lp_build_cube_face(bld, rx, PIPE_TEX_FACE_POS_X, PIPE_TEX_FACE_NEG_X); - LLVMBuildStore(bld->builder, *face_s, face_s_var); - LLVMBuildStore(bld->builder, *face_t, face_t_var); - LLVMBuildStore(bld->builder, *face, face_var); + LLVMBuildStore(builder, *face_s, face_s_var); + LLVMBuildStore(builder, *face_t, face_t_var); + LLVMBuildStore(builder, *face, face_var); } lp_build_else(&if_ctx); { struct lp_build_if_state if_ctx2; - ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, ""); + ary_ge_arx_arz = LLVMBuildAnd(builder, ary_ge_arx, ary_ge_arz, ""); - lp_build_if(&if_ctx2, bld->builder, ary_ge_arx_arz); + lp_build_if(&if_ctx2, bld->gallivm, ary_ge_arx_arz); { /* +/- Y face */ LLVMValueRef sign = lp_build_sgn(float_bld, ry); @@ -963,9 +972,9 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld, *face = lp_build_cube_face(bld, ry, PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y); - LLVMBuildStore(bld->builder, *face_s, face_s_var); - LLVMBuildStore(bld->builder, *face_t, face_t_var); - LLVMBuildStore(bld->builder, *face, face_var); + LLVMBuildStore(builder, *face_s, face_s_var); + LLVMBuildStore(builder, *face_t, face_t_var); + LLVMBuildStore(builder, *face, face_var); } lp_build_else(&if_ctx2); { @@ -977,18 +986,18 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld, *face = lp_build_cube_face(bld, rz, PIPE_TEX_FACE_POS_Z, PIPE_TEX_FACE_NEG_Z); - LLVMBuildStore(bld->builder, *face_s, face_s_var); - LLVMBuildStore(bld->builder, *face_t, face_t_var); - LLVMBuildStore(bld->builder, *face, face_var); + LLVMBuildStore(builder, *face_s, face_s_var); + LLVMBuildStore(builder, *face_t, face_t_var); + LLVMBuildStore(builder, *face, face_var); } lp_build_endif(&if_ctx2); } lp_build_endif(&if_ctx); - *face_s = LLVMBuildLoad(bld->builder, face_s_var, "face_s"); - *face_t = LLVMBuildLoad(bld->builder, face_t_var, "face_t"); - *face = LLVMBuildLoad(bld->builder, face_var, "face"); + *face_s = LLVMBuildLoad(builder, face_s_var, "face_s"); + *face_t = LLVMBuildLoad(builder, face_t_var, "face_t"); + *face = LLVMBuildLoad(builder, face_var, "face"); } } @@ -1011,6 +1020,7 @@ lp_build_sample_partial_offset(struct lp_build_context *bld, LLVMValueRef *out_offset, LLVMValueRef *out_subcoord) { + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef offset; LLVMValueRef subcoord; @@ -1028,14 +1038,14 @@ lp_build_sample_partial_offset(struct lp_build_context *bld, */ #if 0 LLVMValueRef block_width = lp_build_const_int_vec(bld->type, block_length); - subcoord = LLVMBuildURem(bld->builder, coord, block_width, ""); - coord = LLVMBuildUDiv(bld->builder, coord, block_width, ""); + subcoord = LLVMBuildURem(builder, coord, block_width, ""); + coord = LLVMBuildUDiv(builder, coord, block_width, ""); #else unsigned logbase2 = util_unsigned_logbase2(block_length); - LLVMValueRef block_shift = lp_build_const_int_vec(bld->type, logbase2); - LLVMValueRef block_mask = lp_build_const_int_vec(bld->type, block_length - 1); - subcoord = LLVMBuildAnd(bld->builder, coord, block_mask, ""); - coord = LLVMBuildLShr(bld->builder, coord, block_shift, ""); + LLVMValueRef block_shift = lp_build_const_int_vec(bld->gallivm, bld->type, logbase2); + LLVMValueRef block_mask = lp_build_const_int_vec(bld->gallivm, bld->type, block_length - 1); + subcoord = LLVMBuildAnd(builder, coord, block_mask, ""); + coord = LLVMBuildLShr(builder, coord, block_shift, ""); #endif } @@ -1071,7 +1081,8 @@ lp_build_sample_offset(struct lp_build_context *bld, LLVMValueRef x_stride; LLVMValueRef offset; - x_stride = lp_build_const_vec(bld->type, format_desc->block.bits/8); + x_stride = lp_build_const_vec(bld->gallivm, bld->type, + format_desc->block.bits/8); lp_build_sample_partial_offset(bld, format_desc->block.width, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.h b/src/gallium/auxiliary/gallivm/lp_bld_sample.h index ffed27cee8..8c9d5df4e4 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.h @@ -105,64 +105,64 @@ struct lp_sampler_dynamic_state /** Obtain the base texture width (returns int32) */ LLVMValueRef (*width)( const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, unsigned unit); /** Obtain the base texture height (returns int32) */ LLVMValueRef (*height)( const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, unsigned unit); /** Obtain the base texture depth (returns int32) */ LLVMValueRef (*depth)( const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, unsigned unit); /** Obtain the number of mipmap levels minus one (returns int32) */ LLVMValueRef (*last_level)( const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, unsigned unit); /** Obtain stride in bytes between image rows/blocks (returns int32) */ LLVMValueRef (*row_stride)( const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, unsigned unit); /** Obtain stride in bytes between image slices (returns int32) */ LLVMValueRef (*img_stride)( const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, unsigned unit); /** Obtain pointer to array of pointers to mimpap levels */ LLVMValueRef (*data_ptr)( const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, unsigned unit); /** Obtain texture min lod (returns float) */ LLVMValueRef (*min_lod)(const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, unsigned unit); + struct gallivm_state *gallivm, unsigned unit); /** Obtain texture max lod (returns float) */ LLVMValueRef (*max_lod)(const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, unsigned unit); + struct gallivm_state *gallivm, unsigned unit); /** Obtain texture lod bias (returns float) */ LLVMValueRef (*lod_bias)(const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, unsigned unit); + struct gallivm_state *gallivm, unsigned unit); /** Obtain texture border color (returns ptr to float[4]) */ LLVMValueRef (*border_color)(const struct lp_sampler_dynamic_state *state, - LLVMBuilderRef builder, unsigned unit); + struct gallivm_state *gallivm, unsigned unit); }; @@ -171,7 +171,7 @@ struct lp_sampler_dynamic_state */ struct lp_build_sample_context { - LLVMBuilderRef builder; + struct gallivm_state *gallivm; const struct lp_sampler_static_state *static_state; @@ -385,7 +385,7 @@ lp_build_sample_offset(struct lp_build_context *bld, void -lp_build_sample_soa(LLVMBuilderRef builder, +lp_build_sample_soa(struct gallivm_state *gallivm, const struct lp_sampler_static_state *static_state, struct lp_sampler_dynamic_state *dynamic_state, struct lp_type fp_type, @@ -399,7 +399,7 @@ lp_build_sample_soa(LLVMBuilderRef builder, LLVMValueRef texel_out[4]); void -lp_build_sample_nop(struct lp_type type, +lp_build_sample_nop(struct gallivm_state *gallivm, struct lp_type type, LLVMValueRef texel_out[4]); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c index d6831a580b..e61cf9541e 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c @@ -52,6 +52,7 @@ #include "lp_bld_flow.h" #include "lp_bld_gather.h" #include "lp_bld_format.h" +#include "lp_bld_init.h" #include "lp_bld_sample.h" #include "lp_bld_sample_aos.h" #include "lp_bld_quad.h" @@ -82,6 +83,7 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld, LLVMValueRef *out_i) { struct lp_build_context *int_coord_bld = &bld->int_coord_bld; + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef length_minus_one; length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one); @@ -89,12 +91,12 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld, switch(wrap_mode) { case PIPE_TEX_WRAP_REPEAT: if(is_pot) - coord = LLVMBuildAnd(bld->builder, coord, length_minus_one, ""); + coord = LLVMBuildAnd(builder, coord, length_minus_one, ""); else { /* Add a bias to the texcoord to handle negative coords */ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024); - coord = LLVMBuildAdd(bld->builder, coord, bias, ""); - coord = LLVMBuildURem(bld->builder, coord, length, ""); + coord = LLVMBuildAdd(builder, coord, bias, ""); + coord = LLVMBuildURem(builder, coord, length, ""); } break; @@ -147,6 +149,7 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld, LLVMValueRef *i1) { struct lp_build_context *int_coord_bld = &bld->int_coord_bld; + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef length_minus_one; LLVMValueRef lmask, umask, mask; @@ -195,39 +198,39 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld, switch(wrap_mode) { case PIPE_TEX_WRAP_REPEAT: if (is_pot) { - coord0 = LLVMBuildAnd(bld->builder, coord0, length_minus_one, ""); + coord0 = LLVMBuildAnd(builder, coord0, length_minus_one, ""); } else { /* Add a bias to the texcoord to handle negative coords */ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024); - coord0 = LLVMBuildAdd(bld->builder, coord0, bias, ""); - coord0 = LLVMBuildURem(bld->builder, coord0, length, ""); + coord0 = LLVMBuildAdd(builder, coord0, bias, ""); + coord0 = LLVMBuildURem(builder, coord0, length, ""); } - mask = lp_build_compare(bld->builder, int_coord_bld->type, + mask = lp_build_compare(bld->gallivm, int_coord_bld->type, PIPE_FUNC_NOTEQUAL, coord0, length_minus_one); *offset0 = lp_build_mul(int_coord_bld, coord0, stride); - *offset1 = LLVMBuildAnd(bld->builder, + *offset1 = LLVMBuildAnd(builder, lp_build_add(int_coord_bld, *offset0, stride), mask, ""); break; case PIPE_TEX_WRAP_CLAMP_TO_EDGE: - lmask = lp_build_compare(int_coord_bld->builder, int_coord_bld->type, + lmask = lp_build_compare(int_coord_bld->gallivm, int_coord_bld->type, PIPE_FUNC_GEQUAL, coord0, int_coord_bld->zero); - umask = lp_build_compare(int_coord_bld->builder, int_coord_bld->type, + umask = lp_build_compare(int_coord_bld->gallivm, int_coord_bld->type, PIPE_FUNC_LESS, coord0, length_minus_one); coord0 = lp_build_select(int_coord_bld, lmask, coord0, int_coord_bld->zero); coord0 = lp_build_select(int_coord_bld, umask, coord0, length_minus_one); - mask = LLVMBuildAnd(bld->builder, lmask, umask, ""); + mask = LLVMBuildAnd(builder, lmask, umask, ""); *offset0 = lp_build_mul(int_coord_bld, coord0, stride); *offset1 = lp_build_add(int_coord_bld, *offset0, - LLVMBuildAnd(bld->builder, stride, mask, "")); + LLVMBuildAnd(builder, stride, mask, "")); break; case PIPE_TEX_WRAP_CLAMP: @@ -263,7 +266,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, LLVMValueRef *colors_hi) { const unsigned dims = bld->dims; - LLVMBuilderRef builder = bld->builder; + LLVMBuilderRef builder = bld->gallivm->builder; struct lp_build_context i32, h16, u8n; LLVMTypeRef i32_vec_type, h16_vec_type, u8n_vec_type; LLVMValueRef i32_c8; @@ -273,13 +276,13 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, LLVMValueRef x_offset, offset; LLVMValueRef x_subcoord, y_subcoord, z_subcoord; - lp_build_context_init(&i32, builder, lp_type_int_vec(32)); - lp_build_context_init(&h16, builder, lp_type_ufixed(16)); - lp_build_context_init(&u8n, builder, lp_type_unorm(8)); + lp_build_context_init(&i32, bld->gallivm, lp_type_int_vec(32)); + lp_build_context_init(&h16, bld->gallivm, lp_type_ufixed(16)); + lp_build_context_init(&u8n, bld->gallivm, lp_type_unorm(8)); - i32_vec_type = lp_build_vec_type(i32.type); - h16_vec_type = lp_build_vec_type(h16.type); - u8n_vec_type = lp_build_vec_type(u8n.type); + i32_vec_type = lp_build_vec_type(bld->gallivm, i32.type); + h16_vec_type = lp_build_vec_type(bld->gallivm, h16.type); + u8n_vec_type = lp_build_vec_type(bld->gallivm, u8n.type); lp_build_extract_image_sizes(bld, bld->int_size_type, @@ -317,7 +320,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, r = LLVMBuildFPToSI(builder, r, i32_vec_type, ""); /* compute floor (shift right 8) */ - i32_c8 = lp_build_const_int_vec(i32.type, 8); + i32_c8 = lp_build_const_int_vec(bld->gallivm, i32.type, 8); s_ipart = LLVMBuildAShr(builder, s, i32_c8, ""); if (dims >= 2) t_ipart = LLVMBuildAShr(builder, t, i32_c8, ""); @@ -325,7 +328,8 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, r_ipart = LLVMBuildAShr(builder, r, i32_c8, ""); /* get pixel, row, image strides */ - x_stride = lp_build_const_vec(bld->int_coord_bld.type, + x_stride = lp_build_const_vec(bld->gallivm, + bld->int_coord_bld.type, bld->format_desc->block.bits/8); /* Do texcoord wrapping, compute texel offset */ @@ -387,7 +391,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, * Given the format is a rgba8, just read the pixels as is, * without any swizzling. Swizzling will be done later. */ - rgba8 = lp_build_gather(bld->builder, + rgba8 = lp_build_gather(bld->gallivm, bld->texel_type.length, bld->format_desc->block.bits, bld->texel_type.width, @@ -396,7 +400,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, rgba8 = LLVMBuildBitCast(builder, rgba8, u8n_vec_type, ""); } else { - rgba8 = lp_build_fetch_rgba_aos(bld->builder, + rgba8 = lp_build_fetch_rgba_aos(bld->gallivm, bld->format_desc, u8n.type, data_ptr, offset, @@ -405,7 +409,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, } /* Expand one 4*rgba8 to two 2*rgba16 */ - lp_build_unpack2(builder, u8n.type, h16.type, + lp_build_unpack2(bld->gallivm, u8n.type, h16.type, rgba8, colors_lo, colors_hi); } @@ -429,7 +433,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, LLVMValueRef *colors_hi) { const unsigned dims = bld->dims; - LLVMBuilderRef builder = bld->builder; + LLVMBuilderRef builder = bld->gallivm->builder; struct lp_build_context i32, h16, u8n; LLVMTypeRef i32_vec_type, h16_vec_type, u8n_vec_type; LLVMValueRef i32_c8, i32_c128, i32_c255; @@ -450,13 +454,13 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, unsigned i, j, k; unsigned numj, numk; - lp_build_context_init(&i32, builder, lp_type_int_vec(32)); - lp_build_context_init(&h16, builder, lp_type_ufixed(16)); - lp_build_context_init(&u8n, builder, lp_type_unorm(8)); + lp_build_context_init(&i32, bld->gallivm, lp_type_int_vec(32)); + lp_build_context_init(&h16, bld->gallivm, lp_type_ufixed(16)); + lp_build_context_init(&u8n, bld->gallivm, lp_type_unorm(8)); - i32_vec_type = lp_build_vec_type(i32.type); - h16_vec_type = lp_build_vec_type(h16.type); - u8n_vec_type = lp_build_vec_type(u8n.type); + i32_vec_type = lp_build_vec_type(bld->gallivm, i32.type); + h16_vec_type = lp_build_vec_type(bld->gallivm, h16.type); + u8n_vec_type = lp_build_vec_type(bld->gallivm, u8n.type); lp_build_extract_image_sizes(bld, bld->int_size_type, @@ -494,7 +498,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, r = LLVMBuildFPToSI(builder, r, i32_vec_type, ""); /* subtract 0.5 (add -128) */ - i32_c128 = lp_build_const_int_vec(i32.type, -128); + i32_c128 = lp_build_const_int_vec(bld->gallivm, i32.type, -128); s = LLVMBuildAdd(builder, s, i32_c128, ""); if (dims >= 2) { t = LLVMBuildAdd(builder, t, i32_c128, ""); @@ -504,7 +508,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, } /* compute floor (shift right 8) */ - i32_c8 = lp_build_const_int_vec(i32.type, 8); + i32_c8 = lp_build_const_int_vec(bld->gallivm, i32.type, 8); s_ipart = LLVMBuildAShr(builder, s, i32_c8, ""); if (dims >= 2) t_ipart = LLVMBuildAShr(builder, t, i32_c8, ""); @@ -512,7 +516,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, r_ipart = LLVMBuildAShr(builder, r, i32_c8, ""); /* compute fractional part (AND with 0xff) */ - i32_c255 = lp_build_const_int_vec(i32.type, 255); + i32_c255 = lp_build_const_int_vec(bld->gallivm, i32.type, 255); s_fpart = LLVMBuildAnd(builder, s, i32_c255, ""); if (dims >= 2) t_fpart = LLVMBuildAnd(builder, t, i32_c255, ""); @@ -520,7 +524,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, r_fpart = LLVMBuildAnd(builder, r, i32_c255, ""); /* get pixel, row and image strides */ - x_stride = lp_build_const_vec(bld->int_coord_bld.type, + x_stride = lp_build_const_vec(bld->gallivm, bld->int_coord_bld.type, bld->format_desc->block.bits/8); y_stride = row_stride_vec; z_stride = img_stride_vec; @@ -612,7 +616,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, r_fpart = LLVMBuildBitCast(builder, r_fpart, h16_vec_type, ""); { - LLVMTypeRef elem_type = LLVMInt32Type(); + LLVMTypeRef elem_type = LLVMInt32TypeInContext(bld->gallivm->context); LLVMValueRef shuffles_lo[LP_MAX_VECTOR_LENGTH]; LLVMValueRef shuffles_hi[LP_MAX_VECTOR_LENGTH]; LLVMValueRef shuffle_lo; @@ -685,7 +689,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, * Given the format is a rgba8, just read the pixels as is, * without any swizzling. Swizzling will be done later. */ - rgba8 = lp_build_gather(bld->builder, + rgba8 = lp_build_gather(bld->gallivm, bld->texel_type.length, bld->format_desc->block.bits, bld->texel_type.width, @@ -694,7 +698,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, rgba8 = LLVMBuildBitCast(builder, rgba8, u8n_vec_type, ""); } else { - rgba8 = lp_build_fetch_rgba_aos(bld->builder, + rgba8 = lp_build_fetch_rgba_aos(bld->gallivm, bld->format_desc, u8n.type, data_ptr, offset[k][j][i], @@ -703,7 +707,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, } /* Expand one 4*rgba8 to two 2*rgba16 */ - lp_build_unpack2(builder, u8n.type, h16.type, + lp_build_unpack2(bld->gallivm, u8n.type, h16.type, rgba8, &neighbors_lo[k][j][i], &neighbors_hi[k][j][i]); } @@ -790,7 +794,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld, LLVMValueRef colors_lo_var, LLVMValueRef colors_hi_var) { - LLVMBuilderRef builder = bld->builder; + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef size0; LLVMValueRef size1; LLVMValueRef row_stride0_vec; @@ -802,7 +806,6 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld, LLVMValueRef colors0_lo, colors0_hi; LLVMValueRef colors1_lo, colors1_hi; - /* sample the first mipmap level */ lp_build_mipmap_level_sizes(bld, ilevel0, &size0, @@ -829,8 +832,8 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld, LLVMBuildStore(builder, colors0_hi, colors_hi_var); if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { - LLVMValueRef h16_scale = LLVMConstReal(LLVMFloatType(), 256.0); - LLVMTypeRef i32_type = LLVMIntType(32); + LLVMValueRef h16_scale = lp_build_const_float(bld->gallivm, 256.0); + LLVMTypeRef i32_type = LLVMIntTypeInContext(bld->gallivm->context, 32); struct lp_build_if_state if_ctx; LLVMValueRef need_lerp; @@ -842,11 +845,11 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld, lod_fpart, LLVMConstNull(i32_type), "need_lerp"); - lp_build_if(&if_ctx, builder, need_lerp); + lp_build_if(&if_ctx, bld->gallivm, need_lerp); { struct lp_build_context h16_bld; - lp_build_context_init(&h16_bld, builder, lp_type_ufixed(16)); + lp_build_context_init(&h16_bld, bld->gallivm, lp_type_ufixed(16)); /* sample the second mipmap level */ lp_build_mipmap_level_sizes(bld, ilevel1, @@ -885,7 +888,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld, int i; assert(h16_bld.type.length <= Elements(shuffles)); for (i = 0; i < h16_bld.type.length; i++) - shuffles[i] = lp_build_const_int32(2 * (i & 1)); + shuffles[i] = lp_build_const_int32(bld->gallivm, 2 * (i & 1)); shuffle = LLVMConstVector(shuffles, h16_bld.type.length); lod_fpart = LLVMBuildShuffleVector(builder, lod_fpart, lod_fpart, @@ -925,7 +928,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld, LLVMValueRef texel_out[4]) { struct lp_build_context *int_bld = &bld->int_bld; - LLVMBuilderRef builder = bld->builder; + LLVMBuilderRef builder = bld->gallivm->builder; const unsigned mip_filter = bld->static_state->min_mip_filter; const unsigned min_filter = bld->static_state->min_img_filter; const unsigned mag_filter = bld->static_state->mag_img_filter; @@ -936,8 +939,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld, LLVMValueRef unswizzled[4]; LLVMValueRef face_ddx[4], face_ddy[4]; struct lp_build_context h16_bld; - LLVMTypeRef i32t = LLVMInt32Type(); - LLVMValueRef i32t_zero = LLVMConstInt(i32t, 0, 0); + LLVMValueRef i32t_zero = lp_build_const_int32(bld->gallivm, 0); /* we only support the common/simple wrap modes at this time */ assert(lp_is_simple_wrap_mode(bld->static_state->wrap_s)); @@ -948,7 +950,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld, /* make 16-bit fixed-pt builder context */ - lp_build_context_init(&h16_bld, builder, lp_type_ufixed(16)); + lp_build_context_init(&h16_bld, bld->gallivm, lp_type_ufixed(16)); /* cube face selection, compute pre-face coords, etc. */ if (bld->static_state->target == PIPE_TEXTURE_CUBE) { @@ -1026,8 +1028,8 @@ lp_build_sample_aos(struct lp_build_sample_context *bld, * Get/interpolate texture colors. */ - packed_lo = lp_build_alloca(builder, h16_bld.vec_type, "packed_lo"); - packed_hi = lp_build_alloca(builder, h16_bld.vec_type, "packed_hi"); + packed_lo = lp_build_alloca(bld->gallivm, h16_bld.vec_type, "packed_lo"); + packed_hi = lp_build_alloca(bld->gallivm, h16_bld.vec_type, "packed_hi"); if (min_filter == mag_filter) { /* no need to distinquish between minification and magnification */ @@ -1048,7 +1050,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld, minify = LLVMBuildICmp(builder, LLVMIntSGE, lod_ipart, int_bld->zero, ""); - lp_build_if(&if_ctx, builder, minify); + lp_build_if(&if_ctx, bld->gallivm, minify); { /* Use the minification filter */ lp_build_sample_mipmap(bld, @@ -1073,7 +1075,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld, * combine the values stored in 'packed_lo' and 'packed_hi' variables * into 'packed' */ - packed = lp_build_pack2(builder, + packed = lp_build_pack2(bld->gallivm, h16_bld.type, lp_type_unorm(8), LLVMBuildLoad(builder, packed_lo, ""), LLVMBuildLoad(builder, packed_hi, "")); @@ -1081,7 +1083,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld, /* * Convert to SoA and swizzle. */ - lp_build_rgba8_to_f32_soa(builder, + lp_build_rgba8_to_f32_soa(bld->gallivm, bld->texel_type, packed, unswizzled); @@ -1096,6 +1098,4 @@ lp_build_sample_aos(struct lp_build_sample_context *bld, texel_out[2] = unswizzled[2]; texel_out[3] = unswizzled[3]; } - - apply_sampler_swizzle(bld, texel_out); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c index 53cc0c5f34..e685f4b73f 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c @@ -84,6 +84,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, const struct lp_sampler_static_state *static_state = bld->static_state; const unsigned dims = bld->dims; struct lp_build_context *int_coord_bld = &bld->int_coord_bld; + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef offset; LLVMValueRef i, j; LLVMValueRef use_border = NULL; @@ -95,7 +96,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, LLVMValueRef b1, b2; b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, x, int_coord_bld->zero); b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, x, width); - use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2"); + use_border = LLVMBuildOr(builder, b1, b2, "b1_or_b2"); } if (dims >= 2 && @@ -106,11 +107,11 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, y, int_coord_bld->zero); b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, y, height); if (use_border) { - use_border = LLVMBuildOr(bld->builder, use_border, b1, "ub_or_b1"); - use_border = LLVMBuildOr(bld->builder, use_border, b2, "ub_or_b2"); + use_border = LLVMBuildOr(builder, use_border, b1, "ub_or_b1"); + use_border = LLVMBuildOr(builder, use_border, b2, "ub_or_b2"); } else { - use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2"); + use_border = LLVMBuildOr(builder, b1, b2, "b1_or_b2"); } } @@ -122,11 +123,11 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, z, int_coord_bld->zero); b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, z, depth); if (use_border) { - use_border = LLVMBuildOr(bld->builder, use_border, b1, "ub_or_b1"); - use_border = LLVMBuildOr(bld->builder, use_border, b2, "ub_or_b2"); + use_border = LLVMBuildOr(builder, use_border, b1, "ub_or_b1"); + use_border = LLVMBuildOr(builder, use_border, b2, "ub_or_b2"); } else { - use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2"); + use_border = LLVMBuildOr(builder, b1, b2, "b1_or_b2"); } } @@ -148,7 +149,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, offset = lp_build_andnot(&bld->int_coord_bld, offset, use_border); } - lp_build_fetch_rgba_soa(bld->builder, + lp_build_fetch_rgba_soa(bld->gallivm, bld->format_desc, bld->texel_type, data_ptr, offset, @@ -174,20 +175,18 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, /* select texel color or border color depending on use_border */ LLVMValueRef border_color_ptr = bld->dynamic_state->border_color(bld->dynamic_state, - bld->builder, unit); + bld->gallivm, unit); int chan; for (chan = 0; chan < 4; chan++) { LLVMValueRef border_chan = - lp_build_array_get(bld->builder, border_color_ptr, - lp_build_const_int32(chan)); + lp_build_array_get(bld->gallivm, border_color_ptr, + lp_build_const_int32(bld->gallivm, chan)); LLVMValueRef border_chan_vec = lp_build_broadcast_scalar(&bld->float_vec_bld, border_chan); texel_out[chan] = lp_build_select(&bld->texel_bld, use_border, border_chan_vec, texel_out[chan]); } } - - apply_sampler_swizzle(bld, texel_out); } @@ -205,7 +204,7 @@ lp_build_coord_mirror(struct lp_build_sample_context *bld, lp_build_ifloor_fract(coord_bld, coord, &flr, &fract); /* isOdd = flr & 1 */ - isOdd = LLVMBuildAnd(bld->builder, flr, int_coord_bld->one, ""); + isOdd = LLVMBuildAnd(bld->gallivm->builder, flr, int_coord_bld->one, ""); /* make coord positive or negative depending on isOdd */ coord = lp_build_set_sign(coord_bld, fract, isOdd); @@ -239,7 +238,8 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld, { struct lp_build_context *coord_bld = &bld->coord_bld; struct lp_build_context *int_coord_bld = &bld->int_coord_bld; - LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5); + LLVMBuilderRef builder = bld->gallivm->builder; + LLVMValueRef half = lp_build_const_vec(bld->gallivm, coord_bld->type, 0.5); LLVMValueRef length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one); LLVMValueRef coord0, coord1, weight; @@ -253,18 +253,18 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld, /* repeat wrap */ if (is_pot) { coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one); - coord0 = LLVMBuildAnd(bld->builder, coord0, length_minus_one, ""); - coord1 = LLVMBuildAnd(bld->builder, coord1, length_minus_one, ""); + coord0 = LLVMBuildAnd(builder, coord0, length_minus_one, ""); + coord1 = LLVMBuildAnd(builder, coord1, length_minus_one, ""); } else { /* Add a bias to the texcoord to handle negative coords */ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024); LLVMValueRef mask; - coord0 = LLVMBuildAdd(bld->builder, coord0, bias, ""); - coord0 = LLVMBuildURem(bld->builder, coord0, length, ""); - mask = lp_build_compare(bld->builder, int_coord_bld->type, + coord0 = LLVMBuildAdd(builder, coord0, bias, ""); + coord0 = LLVMBuildURem(builder, coord0, length, ""); + mask = lp_build_compare(bld->gallivm, int_coord_bld->type, PIPE_FUNC_NOTEQUAL, coord0, length_minus_one); - coord1 = LLVMBuildAnd(bld->builder, + coord1 = LLVMBuildAnd(builder, lp_build_add(int_coord_bld, coord0, int_coord_bld->one), mask, ""); } @@ -318,7 +318,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld, } /* was: clamp to [-0.5, length + 0.5], then sub 0.5 */ coord = lp_build_sub(coord_bld, coord, half); - min = lp_build_const_vec(coord_bld->type, -1.0F); + min = lp_build_const_vec(bld->gallivm, coord_bld->type, -1.0F); coord = lp_build_clamp(coord_bld, coord, min, length_f); /* convert to int, compute lerp weight */ lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight); @@ -437,6 +437,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld, { struct lp_build_context *coord_bld = &bld->coord_bld; struct lp_build_context *int_coord_bld = &bld->int_coord_bld; + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one); LLVMValueRef icoord; @@ -445,12 +446,12 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld, coord = lp_build_mul(coord_bld, coord, length_f); icoord = lp_build_ifloor(coord_bld, coord); if (is_pot) - icoord = LLVMBuildAnd(bld->builder, icoord, length_minus_one, ""); + icoord = LLVMBuildAnd(builder, icoord, length_minus_one, ""); else { /* Add a bias to the texcoord to handle negative coords */ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024); - icoord = LLVMBuildAdd(bld->builder, icoord, bias, ""); - icoord = LLVMBuildURem(bld->builder, icoord, length, ""); + icoord = LLVMBuildAdd(builder, icoord, bias, ""); + icoord = LLVMBuildURem(builder, icoord, length, ""); } break; @@ -830,7 +831,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld, LLVMValueRef lod_fpart, LLVMValueRef *colors_out) { - LLVMBuilderRef builder = bld->builder; + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef size0; LLVMValueRef size1; LLVMValueRef row_stride0_vec; @@ -878,7 +879,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld, bld->float_bld.zero, "need_lerp"); - lp_build_if(&if_ctx, builder, need_lerp); + lp_build_if(&if_ctx, bld->gallivm, need_lerp); { /* sample the second mipmap level */ lp_build_mipmap_level_sizes(bld, ilevel1, @@ -934,7 +935,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld, LLVMValueRef *colors_out) { struct lp_build_context *int_bld = &bld->int_bld; - LLVMBuilderRef builder = bld->builder; + LLVMBuilderRef builder = bld->gallivm->builder; const unsigned mip_filter = bld->static_state->min_mip_filter; const unsigned min_filter = bld->static_state->min_img_filter; const unsigned mag_filter = bld->static_state->mag_img_filter; @@ -942,8 +943,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld, LLVMValueRef ilevel0, ilevel1 = NULL; LLVMValueRef face_ddx[4], face_ddy[4]; LLVMValueRef texels[4]; - LLVMTypeRef i32t = LLVMInt32Type(); - LLVMValueRef i32t_zero = LLVMConstInt(i32t, 0, 0); + LLVMValueRef i32t_zero = lp_build_const_int32(bld->gallivm, 0); unsigned chan; /* @@ -1030,7 +1030,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld, */ for (chan = 0; chan < 4; ++chan) { - texels[chan] = lp_build_alloca(builder, bld->texel_bld.vec_type, ""); + texels[chan] = lp_build_alloca(bld->gallivm, bld->texel_bld.vec_type, ""); lp_build_name(texels[chan], "sampler%u_texel_%c_var", unit, "xyzw"[chan]); } @@ -1053,7 +1053,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld, minify = LLVMBuildICmp(builder, LLVMIntSGE, lod_ipart, int_bld->zero, ""); - lp_build_if(&if_ctx, builder, minify); + lp_build_if(&if_ctx, bld->gallivm, minify); { /* Use the minification filter */ lp_build_sample_mipmap(bld, unit, @@ -1092,6 +1092,7 @@ lp_build_sample_compare(struct lp_build_sample_context *bld, LLVMValueRef texel[4]) { struct lp_build_context *texel_bld = &bld->texel_bld; + LLVMBuilderRef builder = bld->gallivm->builder; LLVMValueRef res; const unsigned chan = 0; @@ -1100,11 +1101,10 @@ lp_build_sample_compare(struct lp_build_sample_context *bld, /* debug code */ if (0) { - LLVMValueRef indx = lp_build_const_int32(0); - LLVMValueRef coord = LLVMBuildExtractElement(bld->builder, p, indx, ""); - LLVMValueRef tex = LLVMBuildExtractElement(bld->builder, - texel[chan], indx, ""); - lp_build_printf(bld->builder, "shadow compare coord %f to texture %f\n", + LLVMValueRef indx = lp_build_const_int32(bld->gallivm, 0); + LLVMValueRef coord = LLVMBuildExtractElement(builder, p, indx, ""); + LLVMValueRef tex = LLVMBuildExtractElement(builder, texel[chan], indx, ""); + lp_build_printf(bld->gallivm, "shadow compare coord %f to texture %f\n", coord, tex); } @@ -1126,10 +1126,10 @@ lp_build_sample_compare(struct lp_build_sample_context *bld, * For debugging. */ void -lp_build_sample_nop(struct lp_type type, +lp_build_sample_nop(struct gallivm_state *gallivm, struct lp_type type, LLVMValueRef texel_out[4]) { - LLVMValueRef one = lp_build_one(type); + LLVMValueRef one = lp_build_one(gallivm, type); unsigned chan; for (chan = 0; chan < 4; chan++) { @@ -1147,7 +1147,7 @@ lp_build_sample_nop(struct lp_type type, * \param ddy partial derivatives of (s,t,r,q) with respect to y */ void -lp_build_sample_soa(LLVMBuilderRef builder, +lp_build_sample_soa(struct gallivm_state *gallivm, const struct lp_sampler_static_state *static_state, struct lp_sampler_dynamic_state *dynamic_state, struct lp_type type, @@ -1162,8 +1162,8 @@ lp_build_sample_soa(LLVMBuilderRef builder, { unsigned dims = texture_dims(static_state->target); struct lp_build_sample_context bld; - LLVMTypeRef i32t = LLVMInt32Type(); - + LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context); + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef s; LLVMValueRef t; LLVMValueRef r; @@ -1178,7 +1178,7 @@ lp_build_sample_soa(LLVMBuilderRef builder, /* Setup our build context */ memset(&bld, 0, sizeof bld); - bld.builder = builder; + bld.gallivm = gallivm; bld.static_state = static_state; bld.dynamic_state = dynamic_state; bld.format_desc = util_format_description(static_state->format); @@ -1195,22 +1195,22 @@ lp_build_sample_soa(LLVMBuilderRef builder, float_vec_type = lp_type_float_vec(32); - lp_build_context_init(&bld.float_bld, builder, bld.float_type); - lp_build_context_init(&bld.float_vec_bld, builder, float_vec_type); - lp_build_context_init(&bld.int_bld, builder, bld.int_type); - lp_build_context_init(&bld.coord_bld, builder, bld.coord_type); - lp_build_context_init(&bld.int_coord_bld, builder, bld.int_coord_type); - lp_build_context_init(&bld.int_size_bld, builder, bld.int_size_type); - lp_build_context_init(&bld.float_size_bld, builder, bld.float_size_type); - lp_build_context_init(&bld.texel_bld, builder, bld.texel_type); + lp_build_context_init(&bld.float_bld, gallivm, bld.float_type); + lp_build_context_init(&bld.float_vec_bld, gallivm, float_vec_type); + lp_build_context_init(&bld.int_bld, gallivm, bld.int_type); + lp_build_context_init(&bld.coord_bld, gallivm, bld.coord_type); + lp_build_context_init(&bld.int_coord_bld, gallivm, bld.int_coord_type); + lp_build_context_init(&bld.int_size_bld, gallivm, bld.int_size_type); + lp_build_context_init(&bld.float_size_bld, gallivm, bld.float_size_type); + lp_build_context_init(&bld.texel_bld, gallivm, bld.texel_type); /* Get the dynamic state */ - bld.width = dynamic_state->width(dynamic_state, builder, unit); - bld.height = dynamic_state->height(dynamic_state, builder, unit); - bld.depth = dynamic_state->depth(dynamic_state, builder, unit); - bld.row_stride_array = dynamic_state->row_stride(dynamic_state, builder, unit); - bld.img_stride_array = dynamic_state->img_stride(dynamic_state, builder, unit); - bld.data_array = dynamic_state->data_ptr(dynamic_state, builder, unit); + bld.width = dynamic_state->width(dynamic_state, gallivm, unit); + bld.height = dynamic_state->height(dynamic_state, gallivm, unit); + bld.depth = dynamic_state->depth(dynamic_state, gallivm, unit); + bld.row_stride_array = dynamic_state->row_stride(dynamic_state, gallivm, unit); + bld.img_stride_array = dynamic_state->img_stride(dynamic_state, gallivm, unit); + bld.data_array = dynamic_state->data_ptr(dynamic_state, gallivm, unit); /* Note that data_array is an array[level] of pointers to texture images */ s = coords[0]; @@ -1236,7 +1236,7 @@ lp_build_sample_soa(LLVMBuilderRef builder, if (0) { /* For debug: no-op texture sampling */ - lp_build_sample_nop(bld.texel_type, texel_out); + lp_build_sample_nop(gallivm, bld.texel_type, texel_out); } else if (util_format_fits_8unorm(bld.format_desc) && lp_is_simple_wrap_mode(static_state->wrap_s) && @@ -1266,4 +1266,6 @@ lp_build_sample_soa(LLVMBuilderRef builder, } lp_build_sample_compare(&bld, r, texel_out); + + apply_sampler_swizzle(&bld, texel_out); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_struct.c b/src/gallium/auxiliary/gallivm/lp_bld_struct.c index 4693c2de6f..0dc2f24d10 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_struct.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_struct.c @@ -37,12 +37,13 @@ #include "util/u_debug.h" #include "util/u_memory.h" +#include "lp_bld_const.h" #include "lp_bld_debug.h" #include "lp_bld_struct.h" LLVMValueRef -lp_build_struct_get_ptr(LLVMBuilderRef builder, +lp_build_struct_get_ptr(struct gallivm_state *gallivm, LLVMValueRef ptr, unsigned member, const char *name) @@ -51,16 +52,16 @@ lp_build_struct_get_ptr(LLVMBuilderRef builder, LLVMValueRef member_ptr; assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind); assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMStructTypeKind); - indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); - indices[1] = LLVMConstInt(LLVMInt32Type(), member, 0); - member_ptr = LLVMBuildGEP(builder, ptr, indices, Elements(indices), ""); + indices[0] = lp_build_const_int32(gallivm, 0); + indices[1] = lp_build_const_int32(gallivm, member); + member_ptr = LLVMBuildGEP(gallivm->builder, ptr, indices, Elements(indices), ""); lp_build_name(member_ptr, "%s.%s_ptr", LLVMGetValueName(ptr), name); return member_ptr; } LLVMValueRef -lp_build_struct_get(LLVMBuilderRef builder, +lp_build_struct_get(struct gallivm_state *gallivm, LLVMValueRef ptr, unsigned member, const char *name) @@ -69,15 +70,15 @@ lp_build_struct_get(LLVMBuilderRef builder, LLVMValueRef res; assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind); assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMStructTypeKind); - member_ptr = lp_build_struct_get_ptr(builder, ptr, member, name); - res = LLVMBuildLoad(builder, member_ptr, ""); + member_ptr = lp_build_struct_get_ptr(gallivm, ptr, member, name); + res = LLVMBuildLoad(gallivm->builder, member_ptr, ""); lp_build_name(res, "%s.%s", LLVMGetValueName(ptr), name); return res; } LLVMValueRef -lp_build_array_get_ptr(LLVMBuilderRef builder, +lp_build_array_get_ptr(struct gallivm_state *gallivm, LLVMValueRef ptr, LLVMValueRef index) { @@ -85,9 +86,9 @@ lp_build_array_get_ptr(LLVMBuilderRef builder, LLVMValueRef element_ptr; assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind); assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMArrayTypeKind); - indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); + indices[0] = lp_build_const_int32(gallivm, 0); indices[1] = index; - element_ptr = LLVMBuildGEP(builder, ptr, indices, Elements(indices), ""); + element_ptr = LLVMBuildGEP(gallivm->builder, ptr, indices, Elements(indices), ""); #ifdef DEBUG lp_build_name(element_ptr, "&%s[%s]", LLVMGetValueName(ptr), LLVMGetValueName(index)); @@ -97,7 +98,7 @@ lp_build_array_get_ptr(LLVMBuilderRef builder, LLVMValueRef -lp_build_array_get(LLVMBuilderRef builder, +lp_build_array_get(struct gallivm_state *gallivm, LLVMValueRef ptr, LLVMValueRef index) { @@ -105,8 +106,8 @@ lp_build_array_get(LLVMBuilderRef builder, LLVMValueRef res; assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind); assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMArrayTypeKind); - element_ptr = lp_build_array_get_ptr(builder, ptr, index); - res = LLVMBuildLoad(builder, element_ptr, ""); + element_ptr = lp_build_array_get_ptr(gallivm, ptr, index); + res = LLVMBuildLoad(gallivm->builder, element_ptr, ""); #ifdef DEBUG lp_build_name(res, "%s[%s]", LLVMGetValueName(ptr), LLVMGetValueName(index)); #endif @@ -115,7 +116,7 @@ lp_build_array_get(LLVMBuilderRef builder, void -lp_build_array_set(LLVMBuilderRef builder, +lp_build_array_set(struct gallivm_state *gallivm, LLVMValueRef ptr, LLVMValueRef index, LLVMValueRef value) @@ -123,8 +124,8 @@ lp_build_array_set(LLVMBuilderRef builder, LLVMValueRef element_ptr; assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind); assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMArrayTypeKind); - element_ptr = lp_build_array_get_ptr(builder, ptr, index); - LLVMBuildStore(builder, value, element_ptr); + element_ptr = lp_build_array_get_ptr(gallivm, ptr, index); + LLVMBuildStore(gallivm->builder, value, element_ptr); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_struct.h b/src/gallium/auxiliary/gallivm/lp_bld_struct.h index eb87a8eee9..11605c685f 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_struct.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_struct.h @@ -38,7 +38,7 @@ #include "gallivm/lp_bld.h" -#include <llvm-c/Target.h> +#include "gallivm/lp_bld_init.h" #include "util/u_debug.h" #include "util/u_memory.h" @@ -57,7 +57,7 @@ * Get value pointer to a structure member. */ LLVMValueRef -lp_build_struct_get_ptr(LLVMBuilderRef builder, +lp_build_struct_get_ptr(struct gallivm_state *gallivm, LLVMValueRef ptr, unsigned member, const char *name); @@ -66,7 +66,7 @@ lp_build_struct_get_ptr(LLVMBuilderRef builder, * Get the value of a structure member. */ LLVMValueRef -lp_build_struct_get(LLVMBuilderRef builder, +lp_build_struct_get(struct gallivm_state *gallivm, LLVMValueRef ptr, unsigned member, const char *name); @@ -75,7 +75,7 @@ lp_build_struct_get(LLVMBuilderRef builder, * Get value pointer to an array element. */ LLVMValueRef -lp_build_array_get_ptr(LLVMBuilderRef builder, +lp_build_array_get_ptr(struct gallivm_state *gallivm, LLVMValueRef ptr, LLVMValueRef index); @@ -83,7 +83,7 @@ lp_build_array_get_ptr(LLVMBuilderRef builder, * Get the value of an array element. */ LLVMValueRef -lp_build_array_get(LLVMBuilderRef builder, +lp_build_array_get(struct gallivm_state *gallivm, LLVMValueRef ptr, LLVMValueRef index); @@ -91,7 +91,7 @@ lp_build_array_get(LLVMBuilderRef builder, * Set the value of an array element. */ void -lp_build_array_set(LLVMBuilderRef builder, +lp_build_array_set(struct gallivm_state *gallivm, LLVMValueRef ptr, LLVMValueRef index, LLVMValueRef value); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c index 4685a90e41..71693603c1 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c @@ -37,12 +37,13 @@ #include "lp_bld_type.h" #include "lp_bld_const.h" +#include "lp_bld_init.h" #include "lp_bld_logic.h" #include "lp_bld_swizzle.h" LLVMValueRef -lp_build_broadcast(LLVMBuilderRef builder, +lp_build_broadcast(struct gallivm_state *gallivm, LLVMTypeRef vec_type, LLVMValueRef scalar) { @@ -52,8 +53,8 @@ lp_build_broadcast(LLVMBuilderRef builder, res = LLVMGetUndef(vec_type); for(i = 0; i < n; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); - res = LLVMBuildInsertElement(builder, res, scalar, index, ""); + LLVMValueRef index = lp_build_const_int32(gallivm, i); + res = LLVMBuildInsertElement(gallivm->builder, res, scalar, index, ""); } return res; @@ -67,6 +68,7 @@ LLVMValueRef lp_build_broadcast_scalar(struct lp_build_context *bld, LLVMValueRef scalar) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; assert(lp_check_elem_type(type, LLVMTypeOf(scalar))); @@ -82,17 +84,17 @@ lp_build_broadcast_scalar(struct lp_build_context *bld, struct lp_type i32_vec_type = lp_type_int_vec(32); i32_vec_type.length = type.length; - res = LLVMBuildInsertElement(bld->builder, bld->undef, scalar, - LLVMConstInt(LLVMInt32Type(), 0, 0), ""); - res = LLVMBuildShuffleVector(bld->builder, res, bld->undef, - lp_build_const_int_vec(i32_vec_type, 0), ""); + res = LLVMBuildInsertElement(builder, bld->undef, scalar, + lp_build_const_int32(bld->gallivm, 0), ""); + res = LLVMBuildShuffleVector(builder, res, bld->undef, + lp_build_const_int_vec(bld->gallivm, i32_vec_type, 0), ""); #else /* XXX: The above path provokes a bug in LLVM 2.6 */ unsigned i; res = bld->undef; for(i = 0; i < type.length; ++i) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0); - res = LLVMBuildInsertElement(bld->builder, res, scalar, index, ""); + LLVMValueRef index = lp_build_const_int32(bld->gallivm, i); + res = LLVMBuildInsertElement(builder, res, scalar, index, ""); } #endif return res; @@ -104,13 +106,13 @@ lp_build_broadcast_scalar(struct lp_build_context *bld, * Combined extract and broadcast (or a mere shuffle when the two types match) */ LLVMValueRef -lp_build_extract_broadcast(LLVMBuilderRef builder, +lp_build_extract_broadcast(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef vector, LLVMValueRef index) { - LLVMTypeRef i32t = LLVMInt32Type(); + LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context); LLVMValueRef res; assert(src_type.floating == dst_type.floating); @@ -132,8 +134,8 @@ lp_build_extract_broadcast(LLVMBuilderRef builder, * Broadcast scalar -> vector. */ - res = lp_build_broadcast(builder, - lp_build_vec_type(dst_type), + res = lp_build_broadcast(gallivm, + lp_build_vec_type(gallivm, dst_type), vector); } } @@ -144,16 +146,16 @@ lp_build_extract_broadcast(LLVMBuilderRef builder, */ LLVMValueRef shuffle; - shuffle = lp_build_broadcast(builder, + shuffle = lp_build_broadcast(gallivm, LLVMVectorType(i32t, dst_type.length), index); - res = LLVMBuildShuffleVector(builder, vector, - LLVMGetUndef(lp_build_vec_type(dst_type)), + res = LLVMBuildShuffleVector(gallivm->builder, vector, + LLVMGetUndef(lp_build_vec_type(gallivm, dst_type)), shuffle, ""); } else { LLVMValueRef scalar; - scalar = LLVMBuildExtractElement(builder, vector, index, ""); + scalar = LLVMBuildExtractElement(gallivm->builder, vector, index, ""); if (dst_type.length == 1) { /* * Trivial extract scalar from vector. @@ -166,8 +168,8 @@ lp_build_extract_broadcast(LLVMBuilderRef builder, * General case of different sized vectors. */ - res = lp_build_broadcast(builder, - lp_build_vec_type(dst_type), + res = lp_build_broadcast(gallivm, + lp_build_vec_type(gallivm, dst_type), vector); } } @@ -185,6 +187,7 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld, LLVMValueRef a, unsigned channel) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; const unsigned n = type.length; unsigned i, j; @@ -199,14 +202,14 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld, /* * Shuffle. */ - LLVMTypeRef elem_type = LLVMInt32Type(); + LLVMTypeRef elem_type = LLVMInt32TypeInContext(bld->gallivm->context); LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH]; for(j = 0; j < n; j += 4) for(i = 0; i < 4; ++i) shuffles[j + i] = LLVMConstInt(elem_type, j + channel, 0); - return LLVMBuildShuffleVector(bld->builder, a, bld->undef, LLVMConstVector(shuffles, n), ""); + return LLVMBuildShuffleVector(builder, a, bld->undef, LLVMConstVector(shuffles, n), ""); } else { /* @@ -226,8 +229,9 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld, }; unsigned i; - a = LLVMBuildAnd(bld->builder, a, - lp_build_const_mask_aos(type, 1 << channel), ""); + a = LLVMBuildAnd(builder, a, + lp_build_const_mask_aos(bld->gallivm, + type, 1 << channel), ""); /* * Build a type where each element is an integer that cover the four @@ -239,7 +243,7 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld, type4.width *= 4; type4.length /= 4; - a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type4), ""); + a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type4), ""); for(i = 0; i < 2; ++i) { LLVMValueRef tmp = NULL; @@ -250,16 +254,16 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld, #endif if(shift > 0) - tmp = LLVMBuildLShr(bld->builder, a, lp_build_const_int_vec(type4, shift*type.width), ""); + tmp = LLVMBuildLShr(builder, a, lp_build_const_int_vec(bld->gallivm, type4, shift*type.width), ""); if(shift < 0) - tmp = LLVMBuildShl(bld->builder, a, lp_build_const_int_vec(type4, -shift*type.width), ""); + tmp = LLVMBuildShl(builder, a, lp_build_const_int_vec(bld->gallivm, type4, -shift*type.width), ""); assert(tmp); if(tmp) - a = LLVMBuildOr(bld->builder, a, tmp, ""); + a = LLVMBuildOr(builder, a, tmp, ""); } - return LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type), ""); + return LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type), ""); } } @@ -269,6 +273,7 @@ lp_build_swizzle_aos(struct lp_build_context *bld, LLVMValueRef a, const unsigned char swizzles[4]) { + LLVMBuilderRef builder = bld->gallivm->builder; const struct lp_type type = bld->type; const unsigned n = type.length; unsigned i, j; @@ -303,8 +308,8 @@ lp_build_swizzle_aos(struct lp_build_context *bld, /* * Shuffle. */ - LLVMValueRef undef = LLVMGetUndef(lp_build_elem_type(type)); - LLVMTypeRef i32t = LLVMInt32Type(); + LLVMValueRef undef = LLVMGetUndef(lp_build_elem_type(bld->gallivm, type)); + LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context); LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH]; LLVMValueRef aux[LP_MAX_VECTOR_LENGTH]; @@ -326,13 +331,13 @@ lp_build_swizzle_aos(struct lp_build_context *bld, case PIPE_SWIZZLE_ZERO: shuffle = type.length + 0; if (!aux[0]) { - aux[0] = lp_build_const_elem(type, 0.0); + aux[0] = lp_build_const_elem(bld->gallivm, type, 0.0); } break; case PIPE_SWIZZLE_ONE: shuffle = type.length + 1; if (!aux[1]) { - aux[1] = lp_build_const_elem(type, 1.0); + aux[1] = lp_build_const_elem(bld->gallivm, type, 1.0); } break; } @@ -346,7 +351,7 @@ lp_build_swizzle_aos(struct lp_build_context *bld, } } - return LLVMBuildShuffleVector(bld->builder, a, + return LLVMBuildShuffleVector(builder, a, LLVMConstVector(aux, n), LLVMConstVector(shuffles, n), ""); } else { @@ -387,8 +392,8 @@ lp_build_swizzle_aos(struct lp_build_context *bld, type4.width *= 4; type4.length /= 4; - a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type4), ""); - res = LLVMBuildBitCast(bld->builder, res, lp_build_vec_type(type4), ""); + a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type4), ""); + res = LLVMBuildBitCast(builder, res, lp_build_vec_type(bld->gallivm, type4), ""); /* * Mask and shift the channels, trying to group as many channels in the @@ -414,23 +419,24 @@ lp_build_swizzle_aos(struct lp_build_context *bld, if (0) debug_printf("shift = %i, mask = 0x%08llx\n", shift, mask); - masked = LLVMBuildAnd(bld->builder, a, - lp_build_const_int_vec(type4, mask), ""); + masked = LLVMBuildAnd(builder, a, + lp_build_const_int_vec(bld->gallivm, type4, mask), ""); if (shift > 0) { - shifted = LLVMBuildShl(bld->builder, masked, - lp_build_const_int_vec(type4, shift*type.width), ""); + shifted = LLVMBuildShl(builder, masked, + lp_build_const_int_vec(bld->gallivm, type4, shift*type.width), ""); } else if (shift < 0) { - shifted = LLVMBuildLShr(bld->builder, masked, - lp_build_const_int_vec(type4, -shift*type.width), ""); + shifted = LLVMBuildLShr(builder, masked, + lp_build_const_int_vec(bld->gallivm, type4, -shift*type.width), ""); } else { shifted = masked; } - res = LLVMBuildOr(bld->builder, res, shifted, ""); + res = LLVMBuildOr(builder, res, shifted, ""); } } - return LLVMBuildBitCast(bld->builder, res, lp_build_vec_type(type), ""); + return LLVMBuildBitCast(builder, res, + lp_build_vec_type(bld->gallivm, type), ""); } } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h index fdea8442ae..c366a65103 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h @@ -45,7 +45,7 @@ struct lp_build_context; LLVMValueRef -lp_build_broadcast(LLVMBuilderRef builder, +lp_build_broadcast(struct gallivm_state *gallivm, LLVMTypeRef vec_type, LLVMValueRef scalar); @@ -56,7 +56,7 @@ lp_build_broadcast_scalar(struct lp_build_context *bld, LLVMValueRef -lp_build_extract_broadcast(LLVMBuilderRef builder, +lp_build_extract_broadcast(struct gallivm_state *gallivm, struct lp_type src_type, struct lp_type dst_type, LLVMValueRef vector, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h index 694818ccfb..9713d10048 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h @@ -46,6 +46,7 @@ struct tgsi_shader_info; struct lp_type; struct lp_build_context; struct lp_build_mask_context; +struct gallivm_state; enum lp_build_tex_modifier { @@ -141,7 +142,7 @@ struct lp_build_sampler_soa void (*emit_fetch_texel)( const struct lp_build_sampler_soa *sampler, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, struct lp_type type, unsigned unit, unsigned num_coords, @@ -174,7 +175,7 @@ lp_build_tgsi_info(const struct tgsi_token *tokens, void -lp_build_tgsi_soa(LLVMBuilderRef builder, +lp_build_tgsi_soa(struct gallivm_state *gallivm, const struct tgsi_token *tokens, struct lp_type type, struct lp_build_mask_context *mask, @@ -188,7 +189,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, void -lp_build_tgsi_aos(LLVMBuilderRef builder, +lp_build_tgsi_aos(struct gallivm_state *gallivm, const struct tgsi_token *tokens, struct lp_type type, const unsigned char swizzles[4], @@ -200,7 +201,7 @@ lp_build_tgsi_aos(LLVMBuilderRef builder, LLVMValueRef -lp_build_system_values_array(LLVMBuilderRef builder, +lp_build_system_values_array(struct gallivm_state *gallivm, const struct tgsi_shader_info *info, LLVMValueRef instance_id, LLVMValueRef facing); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c index c3c082b2b9..a021efd69f 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c @@ -151,6 +151,7 @@ emit_fetch( const struct tgsi_full_instruction *inst, unsigned src_op) { + LLVMBuilderRef builder = bld->base.gallivm->builder; struct lp_type type = bld->base.type; const struct tgsi_full_src_register *reg = &inst->Src[src_op]; LLVMValueRef res; @@ -175,14 +176,12 @@ emit_fetch( LLVMValueRef scalar; LLVMValueRef swizzle; - index = LLVMConstInt(LLVMInt32Type(), - reg->Register.Index*4 + chan, - 0); + index = lp_build_const_int32(bld->base.gallivm, reg->Register.Index * 4 + chan); - scalar_ptr = LLVMBuildGEP(bld->base.builder, bld->consts_ptr, + scalar_ptr = LLVMBuildGEP(builder, bld->consts_ptr, &index, 1, ""); - scalar = LLVMBuildLoad(bld->base.builder, scalar_ptr, ""); + scalar = LLVMBuildLoad(builder, scalar_ptr, ""); lp_build_name(scalar, "const[%u].%c", reg->Register.Index, "xyzw"[chan]); @@ -190,9 +189,9 @@ emit_fetch( * NOTE: constants array is always assumed to be RGBA */ - swizzle = LLVMConstInt(LLVMInt32Type(), chan, 0); + swizzle = lp_build_const_int32(bld->base.gallivm, chan); - res = LLVMBuildInsertElement(bld->base.builder, res, scalar, swizzle, ""); + res = LLVMBuildInsertElement(builder, res, scalar, swizzle, ""); } /* @@ -206,14 +205,14 @@ emit_fetch( unsigned i; for (chan = 0; chan < 4; ++chan) { - shuffles[chan] = LLVMConstInt(LLVMInt32Type(), chan, 0); + shuffles[chan] = lp_build_const_int32(bld->base.gallivm, chan); } for (i = 4; i < type.length; ++i) { shuffles[i] = shuffles[i % 4]; } - res = LLVMBuildShuffleVector(bld->base.builder, + res = LLVMBuildShuffleVector(builder, res, bld->base.undef, LLVMConstVector(shuffles, type.length), ""); @@ -234,7 +233,7 @@ emit_fetch( { LLVMValueRef temp_ptr; temp_ptr = bld->temps[reg->Register.Index]; - res = LLVMBuildLoad(bld->base.builder, temp_ptr, ""); + res = LLVMBuildLoad(builder, temp_ptr, ""); if (!res) return bld->base.undef; } @@ -281,6 +280,7 @@ emit_store( unsigned index, LLVMValueRef value) { + LLVMBuilderRef builder = bld->base.gallivm->builder; const struct tgsi_full_dst_register *reg = &inst->Dst[index]; LLVMValueRef mask = NULL; LLVMValueRef ptr; @@ -299,7 +299,7 @@ emit_store( break; case TGSI_SAT_MINUS_PLUS_ONE: - value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.type, -1.0)); + value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.gallivm, bld->base.type, -1.0)); value = lp_build_min(&bld->base, value, bld->base.one); break; @@ -344,20 +344,20 @@ emit_store( assert(inst->Predicate.Index < LP_MAX_TGSI_PREDS); - pred = LLVMBuildLoad(bld->base.builder, + pred = LLVMBuildLoad(builder, bld->preds[inst->Predicate.Index], ""); /* * Convert the value to an integer mask. */ - pred = lp_build_compare(bld->base.builder, + pred = lp_build_compare(bld->base.gallivm, bld->base.type, PIPE_FUNC_NOTEQUAL, pred, bld->base.zero); if (inst->Predicate.Negate) { - pred = LLVMBuildNot(bld->base.builder, pred, ""); + pred = LLVMBuildNot(builder, pred, ""); } pred = swizzle_aos(bld, pred, @@ -367,7 +367,7 @@ emit_store( inst->Predicate.SwizzleW); if (mask) { - mask = LLVMBuildAnd(bld->base.builder, mask, pred, ""); + mask = LLVMBuildAnd(builder, mask, pred, ""); } else { mask = pred; } @@ -380,10 +380,11 @@ emit_store( if (reg->Register.WriteMask != TGSI_WRITEMASK_XYZW) { LLVMValueRef writemask; - writemask = lp_build_const_mask_aos(bld->base.type, reg->Register.WriteMask); + writemask = lp_build_const_mask_aos(bld->base.gallivm, bld->base.type, + reg->Register.WriteMask); if (mask) { - mask = LLVMBuildAnd(bld->base.builder, mask, writemask, ""); + mask = LLVMBuildAnd(builder, mask, writemask, ""); } else { mask = writemask; } @@ -392,12 +393,12 @@ emit_store( if (mask) { LLVMValueRef orig_value; - orig_value = LLVMBuildLoad(bld->base.builder, ptr, ""); + orig_value = LLVMBuildLoad(builder, ptr, ""); value = lp_build_select(&bld->base, mask, value, orig_value); } - LLVMBuildStore(bld->base.builder, value, ptr); + LLVMBuildStore(builder, value, ptr); } @@ -454,7 +455,8 @@ emit_declaration( struct lp_build_tgsi_aos_context *bld, const struct tgsi_full_declaration *decl) { - LLVMTypeRef vec_type = lp_build_vec_type(bld->base.type); + struct gallivm_state *gallivm = bld->base.gallivm; + LLVMTypeRef vec_type = lp_build_vec_type(bld->base.gallivm, bld->base.type); unsigned first = decl->Range.First; unsigned last = decl->Range.Last; @@ -465,31 +467,26 @@ emit_declaration( case TGSI_FILE_TEMPORARY: assert(idx < LP_MAX_TGSI_TEMPS); if (bld->indirect_files & (1 << TGSI_FILE_TEMPORARY)) { - LLVMValueRef array_size = LLVMConstInt(LLVMInt32Type(), - last + 1, 0); - bld->temps_array = lp_build_array_alloca(bld->base.builder, + LLVMValueRef array_size = lp_build_const_int32(gallivm, last + 1); + bld->temps_array = lp_build_array_alloca(bld->base.gallivm, vec_type, array_size, ""); } else { - bld->temps[idx] = lp_build_alloca(bld->base.builder, - vec_type, ""); + bld->temps[idx] = lp_build_alloca(gallivm, vec_type, ""); } break; case TGSI_FILE_OUTPUT: - bld->outputs[idx] = lp_build_alloca(bld->base.builder, - vec_type, ""); + bld->outputs[idx] = lp_build_alloca(gallivm, vec_type, ""); break; case TGSI_FILE_ADDRESS: assert(idx < LP_MAX_TGSI_ADDRS); - bld->addr[idx] = lp_build_alloca(bld->base.builder, - vec_type, ""); + bld->addr[idx] = lp_build_alloca(gallivm, vec_type, ""); break; case TGSI_FILE_PREDICATE: assert(idx < LP_MAX_TGSI_PREDS); - bld->preds[idx] = lp_build_alloca(bld->base.builder, - vec_type, ""); + bld->preds[idx] = lp_build_alloca(gallivm, vec_type, ""); break; default: @@ -644,7 +641,7 @@ emit_instruction( src0 = emit_fetch(bld, inst, 0); src1 = emit_fetch(bld, inst, 1); src2 = emit_fetch(bld, inst, 2); - tmp1 = lp_build_const_vec(bld->base.type, 0.5); + tmp1 = lp_build_const_vec(bld->base.gallivm, bld->base.type, 0.5); tmp0 = lp_build_cmp(&bld->base, PIPE_FUNC_GREATER, src2, tmp1); dst0 = lp_build_select(&bld->base, tmp0, src0, src1); break; @@ -1039,7 +1036,7 @@ emit_instruction( void -lp_build_tgsi_aos(LLVMBuilderRef builder, +lp_build_tgsi_aos(struct gallivm_state *gallivm, const struct tgsi_token *tokens, struct lp_type type, const unsigned char swizzles[4], @@ -1058,8 +1055,8 @@ lp_build_tgsi_aos(LLVMBuilderRef builder, /* Setup build context */ memset(&bld, 0, sizeof bld); - lp_build_context_init(&bld.base, builder, type); - lp_build_context_init(&bld.int_bld, builder, lp_int_type(type)); + lp_build_context_init(&bld.base, gallivm, type); + lp_build_context_init(&bld.int_bld, gallivm, lp_int_type(type)); for (chan = 0; chan < 4; ++chan) { bld.swizzles[chan] = swizzles[chan]; @@ -1131,7 +1128,7 @@ lp_build_tgsi_aos(LLVMBuilderRef builder, imm[swizzle] = parse.FullToken.FullImmediate.u[chan].Float; } bld.immediates[num_immediates] = - lp_build_const_aos(type, + lp_build_const_aos(gallivm, type, imm[0], imm[1], imm[2], imm[3], NULL); num_immediates++; @@ -1156,7 +1153,7 @@ lp_build_tgsi_aos(LLVMBuilderRef builder, } if (0) { - LLVMBasicBlockRef block = LLVMGetInsertBlock(builder); + LLVMBasicBlockRef block = LLVMGetInsertBlock(gallivm->builder); LLVMValueRef function = LLVMGetBasicBlockParent(block); debug_printf("11111111111111111111111111111 \n"); tgsi_dump(tokens, 0); @@ -1167,7 +1164,7 @@ lp_build_tgsi_aos(LLVMBuilderRef builder, if (0) { LLVMModuleRef module = LLVMGetGlobalParent( - LLVMGetBasicBlockParent(LLVMGetInsertBlock(bld.base.builder))); + LLVMGetBasicBlockParent(LLVMGetInsertBlock(gallivm->builder))); LLVMDumpModule(module); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c index 3fdfac95a0..d1585c8e2b 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c @@ -51,6 +51,7 @@ #include "lp_bld_arit.h" #include "lp_bld_bitarit.h" #include "lp_bld_gather.h" +#include "lp_bld_init.h" #include "lp_bld_logic.h" #include "lp_bld_swizzle.h" #include "lp_bld_flow.h" @@ -177,22 +178,24 @@ static void lp_exec_mask_init(struct lp_exec_mask *mask, struct lp_build_context mask->loop_stack_size = 0; mask->call_stack_size = 0; - mask->int_vec_type = lp_build_int_vec_type(mask->bld->type); + mask->int_vec_type = lp_build_int_vec_type(bld->gallivm, mask->bld->type); mask->exec_mask = mask->ret_mask = mask->break_mask = mask->cont_mask = mask->cond_mask = LLVMConstAllOnes(mask->int_vec_type); } static void lp_exec_mask_update(struct lp_exec_mask *mask) { + LLVMBuilderRef builder = mask->bld->gallivm->builder; + if (mask->loop_stack_size) { /*for loops we need to update the entire mask at runtime */ LLVMValueRef tmp; assert(mask->break_mask); - tmp = LLVMBuildAnd(mask->bld->builder, + tmp = LLVMBuildAnd(builder, mask->cont_mask, mask->break_mask, "maskcb"); - mask->exec_mask = LLVMBuildAnd(mask->bld->builder, + mask->exec_mask = LLVMBuildAnd(builder, mask->cond_mask, tmp, "maskfull"); @@ -200,7 +203,7 @@ static void lp_exec_mask_update(struct lp_exec_mask *mask) mask->exec_mask = mask->cond_mask; if (mask->call_stack_size) { - mask->exec_mask = LLVMBuildAnd(mask->bld->builder, + mask->exec_mask = LLVMBuildAnd(builder, mask->exec_mask, mask->ret_mask, "callmask"); @@ -214,13 +217,15 @@ static void lp_exec_mask_update(struct lp_exec_mask *mask) static void lp_exec_mask_cond_push(struct lp_exec_mask *mask, LLVMValueRef val) { + LLVMBuilderRef builder = mask->bld->gallivm->builder; + assert(mask->cond_stack_size < LP_MAX_TGSI_NESTING); if (mask->cond_stack_size == 0) { assert(mask->cond_mask == LLVMConstAllOnes(mask->int_vec_type)); } mask->cond_stack[mask->cond_stack_size++] = mask->cond_mask; assert(LLVMTypeOf(val) == mask->int_vec_type); - mask->cond_mask = LLVMBuildAnd(mask->bld->builder, + mask->cond_mask = LLVMBuildAnd(builder, mask->cond_mask, val, ""); @@ -229,6 +234,7 @@ static void lp_exec_mask_cond_push(struct lp_exec_mask *mask, static void lp_exec_mask_cond_invert(struct lp_exec_mask *mask) { + LLVMBuilderRef builder = mask->bld->gallivm->builder; LLVMValueRef prev_mask; LLVMValueRef inv_mask; @@ -238,9 +244,9 @@ static void lp_exec_mask_cond_invert(struct lp_exec_mask *mask) assert(prev_mask == LLVMConstAllOnes(mask->int_vec_type)); } - inv_mask = LLVMBuildNot(mask->bld->builder, mask->cond_mask, ""); + inv_mask = LLVMBuildNot(builder, mask->cond_mask, ""); - mask->cond_mask = LLVMBuildAnd(mask->bld->builder, + mask->cond_mask = LLVMBuildAnd(builder, inv_mask, prev_mask, ""); lp_exec_mask_update(mask); @@ -255,6 +261,8 @@ static void lp_exec_mask_cond_pop(struct lp_exec_mask *mask) static void lp_exec_bgnloop(struct lp_exec_mask *mask) { + LLVMBuilderRef builder = mask->bld->gallivm->builder; + if (mask->loop_stack_size == 0) { assert(mask->loop_block == NULL); assert(mask->cont_mask == LLVMConstAllOnes(mask->int_vec_type)); @@ -270,25 +278,26 @@ static void lp_exec_bgnloop(struct lp_exec_mask *mask) mask->loop_stack[mask->loop_stack_size].break_var = mask->break_var; ++mask->loop_stack_size; - mask->break_var = lp_build_alloca(mask->bld->builder, mask->int_vec_type, ""); - LLVMBuildStore(mask->bld->builder, mask->break_mask, mask->break_var); + mask->break_var = lp_build_alloca(mask->bld->gallivm, mask->int_vec_type, ""); + LLVMBuildStore(builder, mask->break_mask, mask->break_var); - mask->loop_block = lp_build_insert_new_block(mask->bld->builder, "bgnloop"); - LLVMBuildBr(mask->bld->builder, mask->loop_block); - LLVMPositionBuilderAtEnd(mask->bld->builder, mask->loop_block); + mask->loop_block = lp_build_insert_new_block(mask->bld->gallivm, "bgnloop"); + LLVMBuildBr(builder, mask->loop_block); + LLVMPositionBuilderAtEnd(builder, mask->loop_block); - mask->break_mask = LLVMBuildLoad(mask->bld->builder, mask->break_var, ""); + mask->break_mask = LLVMBuildLoad(builder, mask->break_var, ""); lp_exec_mask_update(mask); } static void lp_exec_break(struct lp_exec_mask *mask) { - LLVMValueRef exec_mask = LLVMBuildNot(mask->bld->builder, + LLVMBuilderRef builder = mask->bld->gallivm->builder; + LLVMValueRef exec_mask = LLVMBuildNot(builder, mask->exec_mask, "break"); - mask->break_mask = LLVMBuildAnd(mask->bld->builder, + mask->break_mask = LLVMBuildAnd(builder, mask->break_mask, exec_mask, "break_full"); @@ -297,11 +306,12 @@ static void lp_exec_break(struct lp_exec_mask *mask) static void lp_exec_continue(struct lp_exec_mask *mask) { - LLVMValueRef exec_mask = LLVMBuildNot(mask->bld->builder, + LLVMBuilderRef builder = mask->bld->gallivm->builder; + LLVMValueRef exec_mask = LLVMBuildNot(builder, mask->exec_mask, ""); - mask->cont_mask = LLVMBuildAnd(mask->bld->builder, + mask->cont_mask = LLVMBuildAnd(builder, mask->cont_mask, exec_mask, ""); @@ -309,11 +319,14 @@ static void lp_exec_continue(struct lp_exec_mask *mask) } -static void lp_exec_endloop(struct lp_exec_mask *mask) +static void lp_exec_endloop(struct gallivm_state *gallivm, + struct lp_exec_mask *mask) { + LLVMBuilderRef builder = mask->bld->gallivm->builder; LLVMBasicBlockRef endloop; - LLVMTypeRef reg_type = LLVMIntType(mask->bld->type.width* - mask->bld->type.length); + LLVMTypeRef reg_type = LLVMIntTypeInContext(gallivm->context, + mask->bld->type.width * + mask->bld->type.length); LLVMValueRef i1cond; assert(mask->break_mask); @@ -329,21 +342,21 @@ static void lp_exec_endloop(struct lp_exec_mask *mask) * Unlike the continue mask, the break_mask must be preserved across loop * iterations */ - LLVMBuildStore(mask->bld->builder, mask->break_mask, mask->break_var); + LLVMBuildStore(builder, mask->break_mask, mask->break_var); /* i1cond = (mask == 0) */ i1cond = LLVMBuildICmp( - mask->bld->builder, + builder, LLVMIntNE, - LLVMBuildBitCast(mask->bld->builder, mask->exec_mask, reg_type, ""), + LLVMBuildBitCast(builder, mask->exec_mask, reg_type, ""), LLVMConstNull(reg_type), ""); - endloop = lp_build_insert_new_block(mask->bld->builder, "endloop"); + endloop = lp_build_insert_new_block(mask->bld->gallivm, "endloop"); - LLVMBuildCondBr(mask->bld->builder, + LLVMBuildCondBr(builder, i1cond, mask->loop_block, endloop); - LLVMPositionBuilderAtEnd(mask->bld->builder, endloop); + LLVMPositionBuilderAtEnd(builder, endloop); assert(mask->loop_stack_size); --mask->loop_stack_size; @@ -365,10 +378,12 @@ static void lp_exec_mask_store(struct lp_exec_mask *mask, LLVMValueRef val, LLVMValueRef dst) { + LLVMBuilderRef builder = mask->bld->gallivm->builder; + /* Mix the predicate and execution mask */ if (mask->has_mask) { if (pred) { - pred = LLVMBuildAnd(mask->bld->builder, pred, mask->exec_mask, ""); + pred = LLVMBuildAnd(builder, pred, mask->exec_mask, ""); } else { pred = mask->exec_mask; } @@ -377,14 +392,14 @@ static void lp_exec_mask_store(struct lp_exec_mask *mask, if (pred) { LLVMValueRef real_val, dst_val; - dst_val = LLVMBuildLoad(mask->bld->builder, dst, ""); + dst_val = LLVMBuildLoad(builder, dst, ""); real_val = lp_build_select(mask->bld, pred, val, dst_val); - LLVMBuildStore(mask->bld->builder, real_val, dst); + LLVMBuildStore(builder, real_val, dst); } else - LLVMBuildStore(mask->bld->builder, val, dst); + LLVMBuildStore(builder, val, dst); } static void lp_exec_mask_call(struct lp_exec_mask *mask, @@ -400,6 +415,7 @@ static void lp_exec_mask_call(struct lp_exec_mask *mask, static void lp_exec_mask_ret(struct lp_exec_mask *mask, int *pc) { + LLVMBuilderRef builder = mask->bld->gallivm->builder; LLVMValueRef exec_mask; if (mask->call_stack_size == 0) { @@ -407,11 +423,11 @@ static void lp_exec_mask_ret(struct lp_exec_mask *mask, int *pc) *pc = -1; return; } - exec_mask = LLVMBuildNot(mask->bld->builder, + exec_mask = LLVMBuildNot(builder, mask->exec_mask, "ret"); - mask->ret_mask = LLVMBuildAnd(mask->bld->builder, + mask->ret_mask = LLVMBuildAnd(builder, mask->ret_mask, exec_mask, "ret_full"); @@ -443,10 +459,11 @@ get_temp_ptr(struct lp_build_tgsi_soa_context *bld, unsigned index, unsigned chan) { + LLVMBuilderRef builder = bld->base.gallivm->builder; assert(chan < 4); if (bld->indirect_files & (1 << TGSI_FILE_TEMPORARY)) { - LLVMValueRef lindex = lp_build_const_int32(index * 4 + chan); - return LLVMBuildGEP(bld->base.builder, bld->temps_array, &lindex, 1, ""); + LLVMValueRef lindex = lp_build_const_int32(bld->base.gallivm, index * 4 + chan); + return LLVMBuildGEP(builder, bld->temps_array, &lindex, 1, ""); } else { return bld->temps[index][chan]; @@ -464,10 +481,12 @@ get_output_ptr(struct lp_build_tgsi_soa_context *bld, unsigned index, unsigned chan) { + LLVMBuilderRef builder = bld->base.gallivm->builder; assert(chan < 4); if (bld->indirect_files & (1 << TGSI_FILE_OUTPUT)) { - LLVMValueRef lindex = lp_build_const_int32(index * 4 + chan); - return LLVMBuildGEP(bld->base.builder, bld->outputs_array, &lindex, 1, ""); + LLVMValueRef lindex = lp_build_const_int32(bld->base.gallivm, + index * 4 + chan); + return LLVMBuildGEP(builder, bld->outputs_array, &lindex, 1, ""); } else { return bld->outputs[index][chan]; @@ -484,6 +503,7 @@ build_gather(struct lp_build_tgsi_soa_context *bld, LLVMValueRef base_ptr, LLVMValueRef indexes) { + LLVMBuilderRef builder = bld->base.gallivm->builder; LLVMValueRef res = bld->base.undef; unsigned i; @@ -491,14 +511,14 @@ build_gather(struct lp_build_tgsi_soa_context *bld, * Loop over elements of index_vec, load scalar value, insert it into 'res'. */ for (i = 0; i < bld->base.type.length; i++) { - LLVMValueRef ii = LLVMConstInt(LLVMInt32Type(), i, 0); - LLVMValueRef index = LLVMBuildExtractElement(bld->base.builder, + LLVMValueRef ii = lp_build_const_int32(bld->base.gallivm, i); + LLVMValueRef index = LLVMBuildExtractElement(builder, indexes, ii, ""); - LLVMValueRef scalar_ptr = LLVMBuildGEP(bld->base.builder, base_ptr, + LLVMValueRef scalar_ptr = LLVMBuildGEP(builder, base_ptr, &index, 1, "gather_ptr"); - LLVMValueRef scalar = LLVMBuildLoad(bld->base.builder, scalar_ptr, ""); + LLVMValueRef scalar = LLVMBuildLoad(builder, scalar_ptr, ""); - res = LLVMBuildInsertElement(bld->base.builder, res, scalar, ii, ""); + res = LLVMBuildInsertElement(builder, res, scalar, ii, ""); } return res; @@ -516,13 +536,14 @@ emit_mask_scatter(struct lp_build_tgsi_soa_context *bld, struct lp_exec_mask *mask, LLVMValueRef pred) { - LLVMBuilderRef builder = bld->base.builder; + struct gallivm_state *gallivm = bld->base.gallivm; + LLVMBuilderRef builder = gallivm->builder; unsigned i; /* Mix the predicate and execution mask */ if (mask->has_mask) { if (pred) { - pred = LLVMBuildAnd(mask->bld->builder, pred, mask->exec_mask, ""); + pred = LLVMBuildAnd(builder, pred, mask->exec_mask, ""); } else { pred = mask->exec_mask; @@ -533,7 +554,7 @@ emit_mask_scatter(struct lp_build_tgsi_soa_context *bld, * Loop over elements of index_vec, store scalar value. */ for (i = 0; i < bld->base.type.length; i++) { - LLVMValueRef ii = LLVMConstInt(LLVMInt32Type(), i, 0); + LLVMValueRef ii = lp_build_const_int32(gallivm, i); LLVMValueRef index = LLVMBuildExtractElement(builder, indexes, ii, ""); LLVMValueRef scalar_ptr = LLVMBuildGEP(builder, base_ptr, &index, 1, "scatter_ptr"); LLVMValueRef val = LLVMBuildExtractElement(builder, values, ii, "scatter_val"); @@ -541,7 +562,7 @@ emit_mask_scatter(struct lp_build_tgsi_soa_context *bld, LLVMBuildExtractElement(builder, pred, ii, "scatter_pred") : NULL; if (0) - lp_build_printf(builder, "scatter %d: val %f at %d %p\n", + lp_build_printf(gallivm, "scatter %d: val %f at %d %p\n", ii, val, index, scalar_ptr); if (scalar_pred) { @@ -568,6 +589,7 @@ get_indirect_index(struct lp_build_tgsi_soa_context *bld, unsigned reg_file, unsigned reg_index, const struct tgsi_src_register *indirect_reg) { + LLVMBuilderRef builder = bld->base.gallivm->builder; struct lp_build_context *uint_bld = &bld->uint_bld; /* always use X component of address register */ unsigned swizzle = indirect_reg->SwizzleX; @@ -578,21 +600,22 @@ get_indirect_index(struct lp_build_tgsi_soa_context *bld, assert(bld->indirect_files & (1 << reg_file)); - base = lp_build_const_int_vec(uint_bld->type, reg_index); + base = lp_build_const_int_vec(bld->base.gallivm, uint_bld->type, reg_index); assert(swizzle < 4); - rel = LLVMBuildLoad(bld->base.builder, + rel = LLVMBuildLoad(builder, bld->addr[indirect_reg->Index][swizzle], "load addr reg"); /* for indexing we want integers */ - rel = LLVMBuildFPToSI(bld->base.builder, + rel = LLVMBuildFPToSI(builder, rel, uint_bld->vec_type, ""); index = lp_build_add(uint_bld, base, rel); - max_index = lp_build_const_int_vec(uint_bld->type, + max_index = lp_build_const_int_vec(bld->base.gallivm, + uint_bld->type, bld->info->file_max[reg_file]); assert(!uint_bld->type.sign); @@ -612,6 +635,8 @@ emit_fetch( unsigned src_op, const unsigned chan_index ) { + struct gallivm_state *gallivm = bld->base.gallivm; + LLVMBuilderRef builder = gallivm->builder; struct lp_build_context *uint_bld = &bld->uint_bld; const struct tgsi_full_src_register *reg = &inst->Src[src_op]; const unsigned swizzle = @@ -637,7 +662,7 @@ emit_fetch( case TGSI_FILE_CONSTANT: if (reg->Register.Indirect) { LLVMValueRef swizzle_vec = - lp_build_const_int_vec(uint_bld->type, swizzle); + lp_build_const_int_vec(bld->base.gallivm, uint_bld->type, swizzle); LLVMValueRef index_vec; /* index into the const buffer */ /* index_vec = indirect_index * 4 + swizzle */ @@ -651,11 +676,11 @@ emit_fetch( LLVMValueRef index; /* index into the const buffer */ LLVMValueRef scalar, scalar_ptr; - index = lp_build_const_int32(reg->Register.Index*4 + swizzle); + index = lp_build_const_int32(gallivm, reg->Register.Index*4 + swizzle); - scalar_ptr = LLVMBuildGEP(bld->base.builder, bld->consts_ptr, + scalar_ptr = LLVMBuildGEP(builder, bld->consts_ptr, &index, 1, ""); - scalar = LLVMBuildLoad(bld->base.builder, scalar_ptr, ""); + scalar = LLVMBuildLoad(builder, scalar_ptr, ""); res = lp_build_broadcast_scalar(&bld->base, scalar); } @@ -669,9 +694,9 @@ emit_fetch( case TGSI_FILE_INPUT: if (reg->Register.Indirect) { LLVMValueRef swizzle_vec = - lp_build_const_int_vec(uint_bld->type, swizzle); + lp_build_const_int_vec(gallivm, uint_bld->type, swizzle); LLVMValueRef length_vec = - lp_build_const_int_vec(uint_bld->type, bld->base.type.length); + lp_build_const_int_vec(gallivm, uint_bld->type, bld->base.type.length); LLVMValueRef index_vec; /* index into the const buffer */ LLVMValueRef inputs_array; LLVMTypeRef float4_ptr_type; @@ -682,18 +707,19 @@ emit_fetch( index_vec = lp_build_mul(uint_bld, index_vec, length_vec); /* cast inputs_array pointer to float* */ - float4_ptr_type = LLVMPointerType(LLVMFloatType(), 0); - inputs_array = LLVMBuildBitCast(uint_bld->builder, bld->inputs_array, - float4_ptr_type, ""); + float4_ptr_type = LLVMPointerType(LLVMFloatTypeInContext(gallivm->context), 0); + inputs_array = LLVMBuildBitCast(builder, bld->inputs_array, + float4_ptr_type, ""); /* Gather values from the temporary register array */ res = build_gather(bld, inputs_array, index_vec); } else { if (bld->indirect_files & (1 << TGSI_FILE_INPUT)) { - LLVMValueRef lindex = lp_build_const_int32(reg->Register.Index * 4 + swizzle); - LLVMValueRef input_ptr = LLVMBuildGEP(bld->base.builder, + LLVMValueRef lindex = lp_build_const_int32(gallivm, + reg->Register.Index * 4 + swizzle); + LLVMValueRef input_ptr = LLVMBuildGEP(builder, bld->inputs_array, &lindex, 1, ""); - res = LLVMBuildLoad(bld->base.builder, input_ptr, ""); + res = LLVMBuildLoad(builder, input_ptr, ""); } else { res = bld->inputs[reg->Register.Index][swizzle]; @@ -705,9 +731,10 @@ emit_fetch( case TGSI_FILE_TEMPORARY: if (reg->Register.Indirect) { LLVMValueRef swizzle_vec = - lp_build_const_int_vec(uint_bld->type, swizzle); + lp_build_const_int_vec(bld->base.gallivm, uint_bld->type, swizzle); LLVMValueRef length_vec = - lp_build_const_int_vec(uint_bld->type, bld->base.type.length); + lp_build_const_int_vec(bld->base.gallivm, uint_bld->type, + bld->base.type.length); LLVMValueRef index_vec; /* index into the const buffer */ LLVMValueRef temps_array; LLVMTypeRef float4_ptr_type; @@ -718,8 +745,8 @@ emit_fetch( index_vec = lp_build_mul(uint_bld, index_vec, length_vec); /* cast temps_array pointer to float* */ - float4_ptr_type = LLVMPointerType(LLVMFloatType(), 0); - temps_array = LLVMBuildBitCast(uint_bld->builder, bld->temps_array, + float4_ptr_type = LLVMPointerType(LLVMFloatTypeInContext(bld->base.gallivm->context), 0); + temps_array = LLVMBuildBitCast(builder, bld->temps_array, float4_ptr_type, ""); /* Gather values from the temporary register array */ @@ -728,7 +755,7 @@ emit_fetch( else { LLVMValueRef temp_ptr; temp_ptr = get_temp_ptr(bld, reg->Register.Index, swizzle); - res = LLVMBuildLoad(bld->base.builder, temp_ptr, ""); + res = LLVMBuildLoad(builder, temp_ptr, ""); if (!res) return bld->base.undef; } @@ -740,11 +767,12 @@ emit_fetch( LLVMValueRef index; /* index into the system value array */ LLVMValueRef scalar, scalar_ptr; - index = lp_build_const_int32(reg->Register.Index * 4 + swizzle); + index = lp_build_const_int32(gallivm, + reg->Register.Index * 4 + swizzle); - scalar_ptr = LLVMBuildGEP(bld->base.builder, bld->system_values_array, + scalar_ptr = LLVMBuildGEP(builder, bld->system_values_array, &index, 1, ""); - scalar = LLVMBuildLoad(bld->base.builder, scalar_ptr, ""); + scalar = LLVMBuildLoad(builder, scalar_ptr, ""); res = lp_build_broadcast_scalar(&bld->base, scalar); } @@ -814,6 +842,7 @@ emit_fetch_predicate( const struct tgsi_full_instruction *inst, LLVMValueRef *pred) { + LLVMBuilderRef builder = bld->base.gallivm->builder; unsigned index; unsigned char swizzles[4]; LLVMValueRef unswizzled[4] = {NULL, NULL, NULL, NULL}; @@ -843,7 +872,7 @@ emit_fetch_predicate( * in the swizzles */ if (!unswizzled[swizzle]) { - value = LLVMBuildLoad(bld->base.builder, + value = LLVMBuildLoad(builder, bld->preds[index][swizzle], ""); /* @@ -853,13 +882,13 @@ emit_fetch_predicate( * is needlessly causing two comparisons due to storing the intermediate * result as float vector instead of an integer mask vector. */ - value = lp_build_compare(bld->base.builder, + value = lp_build_compare(bld->base.gallivm, bld->base.type, PIPE_FUNC_NOTEQUAL, value, bld->base.zero); if (inst->Predicate.Negate) { - value = LLVMBuildNot(bld->base.builder, value, ""); + value = LLVMBuildNot(builder, value, ""); } unswizzled[swizzle] = value; @@ -884,6 +913,8 @@ emit_store( LLVMValueRef pred, LLVMValueRef value) { + struct gallivm_state *gallivm = bld->base.gallivm; + LLVMBuilderRef builder = gallivm->builder; const struct tgsi_full_dst_register *reg = &inst->Dst[index]; struct lp_build_context *uint_bld = &bld->uint_bld; LLVMValueRef indirect_index = NULL; @@ -898,7 +929,7 @@ emit_store( break; case TGSI_SAT_MINUS_PLUS_ONE: - value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.type, -1.0)); + value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.gallivm, bld->base.type, -1.0)); value = lp_build_min(&bld->base, value, bld->base.one); break; @@ -918,11 +949,10 @@ emit_store( switch( reg->Register.File ) { case TGSI_FILE_OUTPUT: if (reg->Register.Indirect) { - LLVMBuilderRef builder = bld->base.builder; LLVMValueRef chan_vec = - lp_build_const_int_vec(uint_bld->type, chan_index); + lp_build_const_int_vec(gallivm, uint_bld->type, chan_index); LLVMValueRef length_vec = - lp_build_const_int_vec(uint_bld->type, bld->base.type.length); + lp_build_const_int_vec(gallivm, uint_bld->type, bld->base.type.length); LLVMValueRef index_vec; /* indexes into the temp registers */ LLVMValueRef outputs_array; LLVMValueRef pixel_offsets; @@ -932,7 +962,7 @@ emit_store( /* build pixel offset vector: {0, 1, 2, 3, ...} */ pixel_offsets = uint_bld->undef; for (i = 0; i < bld->base.type.length; i++) { - LLVMValueRef ii = lp_build_const_int32(i); + LLVMValueRef ii = lp_build_const_int32(gallivm, i); pixel_offsets = LLVMBuildInsertElement(builder, pixel_offsets, ii, ii, ""); } @@ -943,7 +973,8 @@ emit_store( index_vec = lp_build_mul(uint_bld, index_vec, length_vec); index_vec = lp_build_add(uint_bld, index_vec, pixel_offsets); - float_ptr_type = LLVMPointerType(LLVMFloatType(), 0); + float_ptr_type = + LLVMPointerType(LLVMFloatTypeInContext(gallivm->context), 0); outputs_array = LLVMBuildBitCast(builder, bld->outputs_array, float_ptr_type, ""); @@ -960,11 +991,11 @@ emit_store( case TGSI_FILE_TEMPORARY: if (reg->Register.Indirect) { - LLVMBuilderRef builder = bld->base.builder; LLVMValueRef chan_vec = - lp_build_const_int_vec(uint_bld->type, chan_index); + lp_build_const_int_vec(gallivm, uint_bld->type, chan_index); LLVMValueRef length_vec = - lp_build_const_int_vec(uint_bld->type, bld->base.type.length); + lp_build_const_int_vec(gallivm, uint_bld->type, + bld->base.type.length); LLVMValueRef index_vec; /* indexes into the temp registers */ LLVMValueRef temps_array; LLVMValueRef pixel_offsets; @@ -974,7 +1005,7 @@ emit_store( /* build pixel offset vector: {0, 1, 2, 3, ...} */ pixel_offsets = uint_bld->undef; for (i = 0; i < bld->base.type.length; i++) { - LLVMValueRef ii = lp_build_const_int32(i); + LLVMValueRef ii = lp_build_const_int32(gallivm, i); pixel_offsets = LLVMBuildInsertElement(builder, pixel_offsets, ii, ii, ""); } @@ -985,7 +1016,8 @@ emit_store( index_vec = lp_build_mul(uint_bld, index_vec, length_vec); index_vec = lp_build_add(uint_bld, index_vec, pixel_offsets); - float_ptr_type = LLVMPointerType(LLVMFloatType(), 0); + float_ptr_type = + LLVMPointerType(LLVMFloatTypeInContext(gallivm->context), 0); temps_array = LLVMBuildBitCast(builder, bld->temps_array, float_ptr_type, ""); @@ -1002,7 +1034,7 @@ emit_store( case TGSI_FILE_ADDRESS: lp_exec_mask_store(&bld->exec_mask, pred, value, - bld->addr[reg->Indirect.Index][chan_index]); + bld->addr[reg->Register.Index][chan_index]); break; case TGSI_FILE_PREDICATE: @@ -1026,6 +1058,7 @@ emit_tex( struct lp_build_tgsi_soa_context *bld, enum lp_build_tex_modifier modifier, LLVMValueRef *texel) { + LLVMBuilderRef builder = bld->base.gallivm->builder; unsigned unit; LLVMValueRef lod_bias, explicit_lod; LLVMValueRef oow = NULL; @@ -1091,13 +1124,12 @@ emit_tex( struct lp_build_tgsi_soa_context *bld, } if (modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV) { - LLVMTypeRef i32t = LLVMInt32Type(); - LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0); + LLVMValueRef index0 = lp_build_const_int32(bld->base.gallivm, 0); for (i = 0; i < num_coords; i++) { LLVMValueRef src1 = emit_fetch( bld, inst, 1, i ); LLVMValueRef src2 = emit_fetch( bld, inst, 2, i ); - ddx[i] = LLVMBuildExtractElement(bld->base.builder, src1, index0, ""); - ddy[i] = LLVMBuildExtractElement(bld->base.builder, src2, index0, ""); + ddx[i] = LLVMBuildExtractElement(builder, src1, index0, ""); + ddy[i] = LLVMBuildExtractElement(builder, src2, index0, ""); } unit = inst->Src[3].Register.Index; } else { @@ -1113,7 +1145,7 @@ emit_tex( struct lp_build_tgsi_soa_context *bld, } bld->sampler->emit_fetch_texel(bld->sampler, - bld->base.builder, + bld->base.gallivm, bld->base.type, unit, num_coords, coords, ddx, ddy, @@ -1168,6 +1200,7 @@ emit_kil( const struct tgsi_full_instruction *inst, int pc) { + LLVMBuilderRef builder = bld->base.gallivm->builder; const struct tgsi_full_src_register *reg = &inst->Src[0]; LLVMValueRef terms[NUM_CHANNELS]; LLVMValueRef mask; @@ -1199,7 +1232,7 @@ emit_kil( chan_mask = lp_build_cmp(&bld->base, PIPE_FUNC_GEQUAL, terms[chan_index], bld->base.zero); if(mask) - mask = LLVMBuildAnd(bld->base.builder, mask, chan_mask, ""); + mask = LLVMBuildAnd(builder, mask, chan_mask, ""); else mask = chan_mask; } @@ -1225,13 +1258,14 @@ emit_kilp(struct lp_build_tgsi_soa_context *bld, const struct tgsi_full_instruction *inst, int pc) { + LLVMBuilderRef builder = bld->base.gallivm->builder; LLVMValueRef mask; /* For those channels which are "alive", disable fragment shader * execution. */ if (bld->exec_mask.has_mask) { - mask = LLVMBuildNot(bld->base.builder, bld->exec_mask.exec_mask, "kilp"); + mask = LLVMBuildNot(builder, bld->exec_mask.exec_mask, "kilp"); } else { LLVMValueRef zero = LLVMConstNull(bld->base.int_vec_type); @@ -1252,38 +1286,39 @@ emit_kilp(struct lp_build_tgsi_soa_context *bld, static void emit_dump_temps(struct lp_build_tgsi_soa_context *bld) { - LLVMBuilderRef builder = bld->base.builder; + struct gallivm_state *gallivm = bld->base.gallivm; + LLVMBuilderRef builder = gallivm->builder; LLVMValueRef temp_ptr; - LLVMValueRef i0 = lp_build_const_int32(0); - LLVMValueRef i1 = lp_build_const_int32(1); - LLVMValueRef i2 = lp_build_const_int32(2); - LLVMValueRef i3 = lp_build_const_int32(3); + LLVMValueRef i0 = lp_build_const_int32(gallivm, 0); + LLVMValueRef i1 = lp_build_const_int32(gallivm, 1); + LLVMValueRef i2 = lp_build_const_int32(gallivm, 2); + LLVMValueRef i3 = lp_build_const_int32(gallivm, 3); int index; int n = bld->info->file_max[TGSI_FILE_TEMPORARY]; for (index = 0; index < n; index++) { - LLVMValueRef idx = lp_build_const_int32(index); + LLVMValueRef idx = lp_build_const_int32(gallivm, index); LLVMValueRef v[4][4], res; int chan; - lp_build_printf(builder, "TEMP[%d]:\n", idx); + lp_build_printf(gallivm, "TEMP[%d]:\n", idx); for (chan = 0; chan < 4; chan++) { temp_ptr = get_temp_ptr(bld, index, chan); - res = LLVMBuildLoad(bld->base.builder, temp_ptr, ""); + res = LLVMBuildLoad(builder, temp_ptr, ""); v[chan][0] = LLVMBuildExtractElement(builder, res, i0, ""); v[chan][1] = LLVMBuildExtractElement(builder, res, i1, ""); v[chan][2] = LLVMBuildExtractElement(builder, res, i2, ""); v[chan][3] = LLVMBuildExtractElement(builder, res, i3, ""); } - lp_build_printf(builder, " X: %f %f %f %f\n", + lp_build_printf(gallivm, " X: %f %f %f %f\n", v[0][0], v[0][1], v[0][2], v[0][3]); - lp_build_printf(builder, " Y: %f %f %f %f\n", + lp_build_printf(gallivm, " Y: %f %f %f %f\n", v[1][0], v[1][1], v[1][2], v[1][3]); - lp_build_printf(builder, " Z: %f %f %f %f\n", + lp_build_printf(gallivm, " Z: %f %f %f %f\n", v[2][0], v[2][1], v[2][2], v[2][3]); - lp_build_printf(builder, " W: %f %f %f %f\n", + lp_build_printf(gallivm, " W: %f %f %f %f\n", v[3][0], v[3][1], v[3][2], v[3][3]); } } @@ -1295,6 +1330,7 @@ emit_declaration( struct lp_build_tgsi_soa_context *bld, const struct tgsi_full_declaration *decl) { + struct gallivm_state *gallivm = bld->base.gallivm; LLVMTypeRef vec_type = bld->base.vec_type; const unsigned first = decl->Range.First; const unsigned last = decl->Range.Last; @@ -1307,15 +1343,14 @@ emit_declaration( assert(idx < LP_MAX_TGSI_TEMPS); if (!(bld->indirect_files & (1 << TGSI_FILE_TEMPORARY))) { for (i = 0; i < NUM_CHANNELS; i++) - bld->temps[idx][i] = lp_build_alloca(bld->base.builder, - vec_type, "temp"); + bld->temps[idx][i] = lp_build_alloca(gallivm, vec_type, "temp"); } break; case TGSI_FILE_OUTPUT: if (!(bld->indirect_files & (1 << TGSI_FILE_OUTPUT))) { for (i = 0; i < NUM_CHANNELS; i++) - bld->outputs[idx][i] = lp_build_alloca(bld->base.builder, + bld->outputs[idx][i] = lp_build_alloca(gallivm, vec_type, "output"); } break; @@ -1323,15 +1358,14 @@ emit_declaration( case TGSI_FILE_ADDRESS: assert(idx < LP_MAX_TGSI_ADDRS); for (i = 0; i < NUM_CHANNELS; i++) - bld->addr[idx][i] = lp_build_alloca(bld->base.builder, - vec_type, "addr"); + bld->addr[idx][i] = lp_build_alloca(gallivm, vec_type, "addr"); break; case TGSI_FILE_PREDICATE: assert(idx < LP_MAX_TGSI_PREDS); for (i = 0; i < NUM_CHANNELS; i++) - bld->preds[idx][i] = lp_build_alloca(bld->base.builder, - vec_type, "predicate"); + bld->preds[idx][i] = lp_build_alloca(gallivm, vec_type, + "predicate"); break; default: @@ -1657,7 +1691,7 @@ emit_instruction( src0 = emit_fetch( bld, inst, 0, chan_index ); src1 = emit_fetch( bld, inst, 1, chan_index ); src2 = emit_fetch( bld, inst, 2, chan_index ); - tmp1 = lp_build_const_vec(bld->base.type, 0.5); + tmp1 = lp_build_const_vec(bld->base.gallivm, bld->base.type, 0.5); tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GREATER, src2, tmp1); dst0[chan_index] = lp_build_select( &bld->base, tmp0, src0, src1 ); } @@ -2169,7 +2203,7 @@ emit_instruction( break; case TGSI_OPCODE_ENDLOOP: - lp_exec_endloop(&bld->exec_mask); + lp_exec_endloop(bld->base.gallivm, &bld->exec_mask); break; case TGSI_OPCODE_ENDSUB: @@ -2302,7 +2336,7 @@ emit_instruction( void -lp_build_tgsi_soa(LLVMBuilderRef builder, +lp_build_tgsi_soa(struct gallivm_state *gallivm, const struct tgsi_token *tokens, struct lp_type type, struct lp_build_mask_context *mask, @@ -2331,9 +2365,9 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, /* Setup build context */ memset(&bld, 0, sizeof bld); - lp_build_context_init(&bld.base, builder, type); - lp_build_context_init(&bld.uint_bld, builder, lp_uint_type(type)); - lp_build_context_init(&bld.elem_bld, builder, lp_elem_type(type)); + lp_build_context_init(&bld.base, gallivm, type); + lp_build_context_init(&bld.uint_bld, gallivm, lp_uint_type(type)); + lp_build_context_init(&bld.elem_bld, gallivm, lp_elem_type(type)); bld.mask = mask; bld.pos = pos; bld.inputs = inputs; @@ -2353,17 +2387,19 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, lp_exec_mask_init(&bld.exec_mask, &bld.base); if (bld.indirect_files & (1 << TGSI_FILE_TEMPORARY)) { - LLVMValueRef array_size = LLVMConstInt(LLVMInt32Type(), - info->file_max[TGSI_FILE_TEMPORARY]*4 + 4, 0); - bld.temps_array = lp_build_array_alloca(bld.base.builder, + LLVMValueRef array_size = + lp_build_const_int32(gallivm, + info->file_max[TGSI_FILE_TEMPORARY] * 4 + 4); + bld.temps_array = lp_build_array_alloca(gallivm, bld.base.vec_type, array_size, "temp_array"); } if (bld.indirect_files & (1 << TGSI_FILE_OUTPUT)) { - LLVMValueRef array_size = LLVMConstInt(LLVMInt32Type(), - info->file_max[TGSI_FILE_OUTPUT]*4 + 4, 0); - bld.outputs_array = lp_build_array_alloca(bld.base.builder, + LLVMValueRef array_size = + lp_build_const_int32(gallivm, + info->file_max[TGSI_FILE_OUTPUT] * 4 + 4); + bld.outputs_array = lp_build_array_alloca(gallivm, bld.base.vec_type, array_size, "output_array"); } @@ -2373,9 +2409,9 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, if (bld.indirect_files & (1 << TGSI_FILE_INPUT)) { unsigned index, chan; LLVMTypeRef vec_type = bld.base.vec_type; - LLVMValueRef array_size = LLVMConstInt(LLVMInt32Type(), - info->file_max[TGSI_FILE_INPUT]*4 + 4, 0); - bld.inputs_array = lp_build_array_alloca(bld.base.builder, + LLVMValueRef array_size = + lp_build_const_int32(gallivm, info->file_max[TGSI_FILE_INPUT]*4 + 4); + bld.inputs_array = lp_build_array_alloca(gallivm, vec_type, array_size, "input_array"); @@ -2383,13 +2419,14 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, for (index = 0; index < info->num_inputs; ++index) { for (chan = 0; chan < NUM_CHANNELS; ++chan) { - LLVMValueRef lindex = lp_build_const_int32(index * 4 + chan); + LLVMValueRef lindex = + lp_build_const_int32(gallivm, index * 4 + chan); LLVMValueRef input_ptr = - LLVMBuildGEP(bld.base.builder, bld.inputs_array, + LLVMBuildGEP(gallivm->builder, bld.inputs_array, &lindex, 1, ""); LLVMValueRef value = bld.inputs[index][chan]; if (value) - LLVMBuildStore(bld.base.builder, value, input_ptr); + LLVMBuildStore(gallivm->builder, value, input_ptr); } } } @@ -2441,7 +2478,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, assert(num_immediates < LP_MAX_TGSI_IMMEDIATES); for( i = 0; i < size; ++i ) bld.immediates[num_immediates][i] = - lp_build_const_vec(type, parse.FullToken.FullImmediate.u[i].Float); + lp_build_const_vec(gallivm, type, parse.FullToken.FullImmediate.u[i].Float); for( i = size; i < 4; ++i ) bld.immediates[num_immediates][i] = bld.base.undef; num_immediates++; @@ -2478,7 +2515,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, } if (0) { - LLVMBasicBlockRef block = LLVMGetInsertBlock(builder); + LLVMBasicBlockRef block = LLVMGetInsertBlock(gallivm->builder); LLVMValueRef function = LLVMGetBasicBlockParent(block); debug_printf("11111111111111111111111111111 \n"); tgsi_dump(tokens, 0); @@ -2489,7 +2526,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, if (0) { LLVMModuleRef module = LLVMGetGlobalParent( - LLVMGetBasicBlockParent(LLVMGetInsertBlock(bld.base.builder))); + LLVMGetBasicBlockParent(LLVMGetInsertBlock(gallivm->builder))); LLVMDumpModule(module); } @@ -2515,24 +2552,25 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, * \return LLVM float array (interpreted as float [][4]) */ LLVMValueRef -lp_build_system_values_array(LLVMBuilderRef builder, +lp_build_system_values_array(struct gallivm_state *gallivm, const struct tgsi_shader_info *info, LLVMValueRef instance_id, LLVMValueRef facing) { - LLVMValueRef size = lp_build_const_int32(4 * info->num_system_values); - LLVMValueRef array = lp_build_array_alloca(builder, LLVMFloatType(), + LLVMValueRef size = lp_build_const_int32(gallivm, 4 * info->num_system_values); + LLVMTypeRef float_t = LLVMFloatTypeInContext(gallivm->context); + LLVMValueRef array = lp_build_array_alloca(gallivm, float_t, size, "sysvals_array"); unsigned i; for (i = 0; i < info->num_system_values; i++) { - LLVMValueRef index = lp_build_const_int32(i * 4); + LLVMValueRef index = lp_build_const_int32(gallivm, i * 4); LLVMValueRef ptr, value; switch (info->system_value_semantic_name[i]) { case TGSI_SEMANTIC_INSTANCEID: /* convert instance ID from int to float */ - value = LLVMBuildSIToFP(builder, instance_id, LLVMFloatType(), + value = LLVMBuildSIToFP(gallivm->builder, instance_id, float_t, "sysval_instanceid"); break; case TGSI_SEMANTIC_FACE: @@ -2541,8 +2579,8 @@ lp_build_system_values_array(LLVMBuilderRef builder, assert(0 && "unexpected semantic in build_system_values_array()"); } - ptr = LLVMBuildGEP(builder, array, &index, 1, ""); - LLVMBuildStore(builder, value, ptr); + ptr = LLVMBuildGEP(gallivm->builder, array, &index, 1, ""); + LLVMBuildStore(gallivm->builder, value, ptr); } return array; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_type.c b/src/gallium/auxiliary/gallivm/lp_bld_type.c index 5205c7ada9..c5cf6d4a6c 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_type.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_type.c @@ -30,34 +30,35 @@ #include "lp_bld_type.h" #include "lp_bld_const.h" +#include "lp_bld_init.h" LLVMTypeRef -lp_build_elem_type(struct lp_type type) +lp_build_elem_type(struct gallivm_state *gallivm, struct lp_type type) { if (type.floating) { switch(type.width) { case 32: - return LLVMFloatType(); + return LLVMFloatTypeInContext(gallivm->context); break; case 64: - return LLVMDoubleType(); + return LLVMDoubleTypeInContext(gallivm->context); break; default: assert(0); - return LLVMFloatType(); + return LLVMFloatTypeInContext(gallivm->context); } } else { - return LLVMIntType(type.width); + return LLVMIntTypeInContext(gallivm->context, type.width); } } LLVMTypeRef -lp_build_vec_type(struct lp_type type) +lp_build_vec_type(struct gallivm_state *gallivm,struct lp_type type) { - LLVMTypeRef elem_type = lp_build_elem_type(type); + LLVMTypeRef elem_type = lp_build_elem_type(gallivm, type); if (type.length == 1) return elem_type; else @@ -149,16 +150,16 @@ lp_check_value(struct lp_type type, LLVMValueRef val) LLVMTypeRef -lp_build_int_elem_type(struct lp_type type) +lp_build_int_elem_type(struct gallivm_state *gallivm, struct lp_type type) { - return LLVMIntType(type.width); + return LLVMIntTypeInContext(gallivm->context, type.width); } LLVMTypeRef -lp_build_int_vec_type(struct lp_type type) +lp_build_int_vec_type(struct gallivm_state *gallivm, struct lp_type type) { - LLVMTypeRef elem_type = lp_build_int_elem_type(type); + LLVMTypeRef elem_type = lp_build_int_elem_type(gallivm, type); if (type.length == 1) return elem_type; else @@ -170,7 +171,7 @@ lp_build_int_vec_type(struct lp_type type) * Build int32[4] vector type */ LLVMTypeRef -lp_build_int32_vec4_type(void) +lp_build_int32_vec4_type(struct gallivm_state *gallivm) { struct lp_type t; LLVMTypeRef type; @@ -182,7 +183,7 @@ lp_build_int32_vec4_type(void) t.width = 32; /* 32-bit int */ t.length = 4; /* 4 elements per vector */ - type = lp_build_int_elem_type(t); + type = lp_build_int_elem_type(gallivm, t); return LLVMVectorType(type, t.length); } @@ -383,15 +384,15 @@ lp_dump_llvmtype(LLVMTypeRef t) void lp_build_context_init(struct lp_build_context *bld, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, struct lp_type type) { - bld->builder = builder; + bld->gallivm = gallivm; bld->type = type; - bld->int_elem_type = lp_build_int_elem_type(type); + bld->int_elem_type = lp_build_int_elem_type(gallivm, type); if (type.floating) - bld->elem_type = lp_build_elem_type(type); + bld->elem_type = lp_build_elem_type(gallivm, type); else bld->elem_type = bld->int_elem_type; @@ -406,5 +407,5 @@ lp_build_context_init(struct lp_build_context *bld, bld->undef = LLVMGetUndef(bld->vec_type); bld->zero = LLVMConstNull(bld->vec_type); - bld->one = lp_build_one(type); + bld->one = lp_build_one(gallivm, type); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_type.h b/src/gallium/auxiliary/gallivm/lp_bld_type.h index a135d0df84..5007e83ac5 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_type.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_type.h @@ -120,7 +120,7 @@ struct lp_type { */ struct lp_build_context { - LLVMBuilderRef builder; + struct gallivm_state *gallivm; /** * This not only describes the input/output LLVM types, but also whether @@ -285,11 +285,11 @@ lp_type_ufixed(unsigned width) LLVMTypeRef -lp_build_elem_type(struct lp_type type); +lp_build_elem_type(struct gallivm_state *gallivm, struct lp_type type); LLVMTypeRef -lp_build_vec_type(struct lp_type type); +lp_build_vec_type(struct gallivm_state *gallivm, struct lp_type type); boolean @@ -305,15 +305,15 @@ lp_check_value(struct lp_type type, LLVMValueRef val); LLVMTypeRef -lp_build_int_elem_type(struct lp_type type); +lp_build_int_elem_type(struct gallivm_state *gallivm, struct lp_type type); LLVMTypeRef -lp_build_int_vec_type(struct lp_type type); +lp_build_int_vec_type(struct gallivm_state *gallivm, struct lp_type type); LLVMTypeRef -lp_build_int32_vec4_type(void); +lp_build_int32_vec4_type(struct gallivm_state *gallivm); static INLINE struct lp_type @@ -394,7 +394,7 @@ lp_dump_llvmtype(LLVMTypeRef t); void lp_build_context_init(struct lp_build_context *bld, - LLVMBuilderRef builder, + struct gallivm_state *gallivm, struct lp_type type); diff --git a/src/gallium/auxiliary/target-helpers/inline_noop_helper.h b/src/gallium/auxiliary/target-helpers/inline_noop_helper.h new file mode 100644 index 0000000000..77c7cfd0c2 --- /dev/null +++ b/src/gallium/auxiliary/target-helpers/inline_noop_helper.h @@ -0,0 +1,51 @@ + +#ifndef INLINE_DEBUG_HELPER_H +#define INLINE_DEBUG_HELPER_H + +#include "pipe/p_compiler.h" +#include "util/u_debug.h" + + +/* Helper function to wrap a screen with + * one or more debug driver: rbug, trace. + */ + +#ifdef GALLIUM_TRACE +#include "trace/tr_public.h" +#endif + +#ifdef GALLIUM_RBUG +#include "rbug/rbug_public.h" +#endif + +#ifdef GALLIUM_GALAHAD +#include "galahad/glhd_public.h" +#endif + +#ifdef GALLIUM_NOOP +#include "noop/noop_public.h" +#endif + +static INLINE struct pipe_screen * +debug_screen_wrap(struct pipe_screen *screen) +{ +#if defined(GALLIUM_RBUG) + screen = rbug_screen_create(screen); +#endif + +#if defined(GALLIUM_TRACE) + screen = trace_screen_create(screen); +#endif + +#if defined(GALLIUM_GALAHAD) + screen = galahad_screen_create(screen); +#endif + +#if defined(GALLIUM_NOOP) + screen = noop_screen_create(screen); +#endif + + return screen; +} + +#endif diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c index 77bde86684..82cd8eaa96 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_dump.c +++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c @@ -138,7 +138,7 @@ static const char *immediate_type_names[] = }; const char * -tgsi_swizzle_names[] = +tgsi_swizzle_names[4] = { "x", "y", @@ -147,7 +147,7 @@ tgsi_swizzle_names[] = }; const char * -tgsi_texture_names[] = +tgsi_texture_names[TGSI_TEXTURE_COUNT] = { "UNKNOWN", "1D", @@ -160,16 +160,17 @@ tgsi_texture_names[] = "SHADOWRECT" }; -static const char *property_names[] = +const char *tgsi_property_names[TGSI_PROPERTY_COUNT] = { "GS_INPUT_PRIMITIVE", "GS_OUTPUT_PRIMITIVE", "GS_MAX_OUTPUT_VERTICES", "FS_COORD_ORIGIN", - "FS_COORD_PIXEL_CENTER" + "FS_COORD_PIXEL_CENTER", + "FS_COLOR0_WRITES_ALL_CBUFS", }; -static const char *primitive_names[] = +const char *tgsi_primitive_names[PIPE_PRIM_MAX] = { "POINTS", "LINES", @@ -187,13 +188,13 @@ static const char *primitive_names[] = "TRIANGLE_STRIP_ADJACENCY" }; -static const char *fs_coord_origin_names[] = +const char *tgsi_fs_coord_origin_names[2] = { "UPPER_LEFT", "LOWER_LEFT" }; -static const char *fs_coord_pixel_center_names[] = +const char *tgsi_fs_coord_pixel_center_names[2] = { "HALF_INTEGER", "INTEGER" @@ -484,10 +485,10 @@ iter_property( int i; struct dump_ctx *ctx = (struct dump_ctx *)iter; - assert(Elements(property_names) == TGSI_PROPERTY_COUNT); + assert(Elements(tgsi_property_names) == TGSI_PROPERTY_COUNT); TXT( "PROPERTY " ); - ENM(prop->Property.PropertyName, property_names); + ENM(prop->Property.PropertyName, tgsi_property_names); if (prop->Property.NrTokens > 1) TXT(" "); @@ -496,13 +497,13 @@ iter_property( switch (prop->Property.PropertyName) { case TGSI_PROPERTY_GS_INPUT_PRIM: case TGSI_PROPERTY_GS_OUTPUT_PRIM: - ENM(prop->u[i].Data, primitive_names); + ENM(prop->u[i].Data, tgsi_primitive_names); break; case TGSI_PROPERTY_FS_COORD_ORIGIN: - ENM(prop->u[i].Data, fs_coord_origin_names); + ENM(prop->u[i].Data, tgsi_fs_coord_origin_names); break; case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER: - ENM(prop->u[i].Data, fs_coord_pixel_center_names); + ENM(prop->u[i].Data, tgsi_fs_coord_pixel_center_names); break; default: SID( prop->u[i].Data ); diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.h b/src/gallium/auxiliary/tgsi/tgsi_dump.h index fc0429ad8d..2491e9198a 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_dump.h +++ b/src/gallium/auxiliary/tgsi/tgsi_dump.h @@ -29,6 +29,7 @@ #define TGSI_DUMP_H #include "pipe/p_compiler.h" +#include "pipe/p_defines.h" #include "pipe/p_shader_tokens.h" #if defined __cplusplus @@ -39,10 +40,22 @@ extern const char * tgsi_file_names[TGSI_FILE_COUNT]; extern const char * -tgsi_swizzle_names[]; +tgsi_swizzle_names[4]; extern const char * -tgsi_texture_names[]; +tgsi_texture_names[TGSI_TEXTURE_COUNT]; + +extern const char * +tgsi_property_names[TGSI_PROPERTY_COUNT]; + +extern const char * +tgsi_primitive_names[PIPE_PRIM_MAX]; + +extern const char * +tgsi_fs_coord_origin_names[2]; + +extern const char * +tgsi_fs_coord_pixel_center_names[2]; void tgsi_dump_str( diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h index d4df585176..2aafa2a6e8 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_parse.h +++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h @@ -136,7 +136,8 @@ tgsi_parse_token( static INLINE unsigned tgsi_num_tokens(const struct tgsi_token *tokens) { - struct tgsi_header header = *(const struct tgsi_header *) tokens; + struct tgsi_header header; + memcpy(&header, tokens, sizeof(header)); return header.HeaderSize + header.BodySize; } diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.h b/src/gallium/auxiliary/tgsi/tgsi_sanity.h index 73f0f414e3..707d11ec6d 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sanity.h +++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.h @@ -28,12 +28,14 @@ #ifndef TGSI_SANITY_H #define TGSI_SANITY_H -#include "pipe/p_shader_tokens.h" - #if defined __cplusplus extern "C" { #endif +#include "pipe/p_compiler.h" + +struct tgsi_token; + /* Check the given token stream for errors and common mistakes. * Diagnostic messages are printed out to the debug output, and is * controlled by the debug option TGSI_PRINT_SANITY (default false). diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.c b/src/gallium/auxiliary/tgsi/tgsi_text.c index 9a38c37979..819b0725a1 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_text.c +++ b/src/gallium/auxiliary/tgsi/tgsi_text.c @@ -36,6 +36,7 @@ #include "tgsi_parse.h" #include "tgsi_sanity.h" #include "tgsi_util.h" +#include "tgsi_dump.h" static boolean is_alpha_underscore( const char *cur ) { @@ -1258,42 +1259,6 @@ static boolean parse_immediate( struct translate_ctx *ctx ) return TRUE; } -static const char *property_names[] = -{ - "GS_INPUT_PRIMITIVE", - "GS_OUTPUT_PRIMITIVE", - "GS_MAX_OUTPUT_VERTICES", - "FS_COORD_ORIGIN", - "FS_COORD_PIXEL_CENTER" -}; - -static const char *primitive_names[] = -{ - "POINTS", - "LINES", - "LINE_LOOP", - "LINE_STRIP", - "TRIANGLES", - "TRIANGLE_STRIP", - "TRIANGLE_FAN", - "QUADS", - "QUAD_STRIP", - "POLYGON" -}; - -static const char *fs_coord_origin_names[] = -{ - "UPPER_LEFT", - "LOWER_LEFT" -}; - -static const char *fs_coord_pixel_center_names[] = -{ - "HALF_INTEGER", - "INTEGER" -}; - - static boolean parse_primitive( const char **pcur, uint *primitive ) { @@ -1302,7 +1267,7 @@ parse_primitive( const char **pcur, uint *primitive ) for (i = 0; i < PIPE_PRIM_MAX; i++) { const char *cur = *pcur; - if (str_match_no_case( &cur, primitive_names[i])) { + if (str_match_no_case( &cur, tgsi_primitive_names[i])) { *primitive = i; *pcur = cur; return TRUE; @@ -1316,10 +1281,10 @@ parse_fs_coord_origin( const char **pcur, uint *fs_coord_origin ) { uint i; - for (i = 0; i < sizeof(fs_coord_origin_names) / sizeof(fs_coord_origin_names[0]); i++) { + for (i = 0; i < Elements(tgsi_fs_coord_origin_names); i++) { const char *cur = *pcur; - if (str_match_no_case( &cur, fs_coord_origin_names[i])) { + if (str_match_no_case( &cur, tgsi_fs_coord_origin_names[i])) { *fs_coord_origin = i; *pcur = cur; return TRUE; @@ -1333,10 +1298,10 @@ parse_fs_coord_pixel_center( const char **pcur, uint *fs_coord_pixel_center ) { uint i; - for (i = 0; i < sizeof(fs_coord_pixel_center_names) / sizeof(fs_coord_pixel_center_names[0]); i++) { + for (i = 0; i < Elements(tgsi_fs_coord_pixel_center_names); i++) { const char *cur = *pcur; - if (str_match_no_case( &cur, fs_coord_pixel_center_names[i])) { + if (str_match_no_case( &cur, tgsi_fs_coord_pixel_center_names[i])) { *fs_coord_pixel_center = i; *pcur = cur; return TRUE; @@ -1364,7 +1329,7 @@ static boolean parse_property( struct translate_ctx *ctx ) } for (property_name = 0; property_name < TGSI_PROPERTY_COUNT; ++property_name) { - if (streq_nocase_uprcase(property_names[property_name], id)) { + if (streq_nocase_uprcase(tgsi_property_names[property_name], id)) { break; } } @@ -1398,6 +1363,7 @@ static boolean parse_property( struct translate_ctx *ctx ) return FALSE; } break; + case TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS: default: if (!parse_uint(&ctx->cur, &values[0] )) { report_error( ctx, "Expected unsigned integer as property!" ); diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.h b/src/gallium/auxiliary/tgsi/tgsi_text.h index 8eeeeef140..5fe53cf007 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_text.h +++ b/src/gallium/auxiliary/tgsi/tgsi_text.h @@ -28,12 +28,14 @@ #ifndef TGSI_TEXT_H #define TGSI_TEXT_H -#include "pipe/p_shader_tokens.h" - #if defined __cplusplus extern "C" { #endif +#include "pipe/p_compiler.h" + +struct tgsi_token; + boolean tgsi_text_translate( const char *text, diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c index 7d13a17bdb..02de12d77d 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c +++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c @@ -148,6 +148,7 @@ struct ureg_program unsigned property_gs_max_vertices; unsigned char property_fs_coord_origin; /* = TGSI_FS_COORD_ORIGIN_* */ unsigned char property_fs_coord_pixel_center; /* = TGSI_FS_COORD_PIXEL_CENTER_* */ + unsigned char property_fs_color0_writes_all_cbufs; /* = TGSI_FS_COLOR0_WRITES_ALL_CBUFS * */ unsigned nr_addrs; unsigned nr_preds; @@ -284,7 +285,12 @@ ureg_property_fs_coord_pixel_center(struct ureg_program *ureg, ureg->property_fs_coord_pixel_center = fs_coord_pixel_center; } - +void +ureg_property_fs_color0_writes_all_cbufs(struct ureg_program *ureg, + unsigned fs_color0_writes_all_cbufs) +{ + ureg->property_fs_color0_writes_all_cbufs = fs_color0_writes_all_cbufs; +} struct ureg_src ureg_DECL_fs_input_cyl_centroid(struct ureg_program *ureg, @@ -1278,6 +1284,14 @@ static void emit_decls( struct ureg_program *ureg ) ureg->property_fs_coord_pixel_center); } + if (ureg->property_fs_color0_writes_all_cbufs) { + assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT); + + emit_property(ureg, + TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS, + ureg->property_fs_color0_writes_all_cbufs); + } + if (ureg->processor == TGSI_PROCESSOR_VERTEX) { for (i = 0; i < UREG_MAX_INPUT; i++) { if (ureg->vs_inputs[i/32] & (1 << (i%32))) { diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h index acc463200a..807128a5e5 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h +++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h @@ -153,6 +153,10 @@ void ureg_property_fs_coord_pixel_center(struct ureg_program *ureg, unsigned fs_coord_pixel_center); +void +ureg_property_fs_color0_writes_all_cbufs(struct ureg_program *ureg, + unsigned fs_color0_writes_all_cbufs); + /*********************************************************************** * Build shader declarations: */ diff --git a/src/gallium/auxiliary/util/u_blit.c b/src/gallium/auxiliary/util/u_blit.c index dfb142b9e1..c11f7d383d 100644 --- a/src/gallium/auxiliary/util/u_blit.c +++ b/src/gallium/auxiliary/util/u_blit.c @@ -291,7 +291,7 @@ regions_overlap(int srcX0, int srcY0, void util_blit_pixels_writemask(struct blit_state *ctx, struct pipe_resource *src_tex, - struct pipe_subresource srcsub, + unsigned src_level, int srcX0, int srcY0, int srcX1, int srcY1, int srcZ0, @@ -316,13 +316,12 @@ util_blit_pixels_writemask(struct blit_state *ctx, assert(filter == PIPE_TEX_MIPFILTER_NEAREST || filter == PIPE_TEX_MIPFILTER_LINEAR); - assert(srcsub.level <= src_tex->last_level); + assert(src_level <= src_tex->last_level); /* do the regions overlap? */ overlap = src_tex == dst->texture && - dst->face == srcsub.face && - dst->level == srcsub.level && - dst->zslice == srcZ0 && + dst->u.tex.level == src_level && + dst->u.tex.first_layer == srcZ0 && regions_overlap(srcX0, srcY0, srcX1, srcY1, dstX0, dstY0, dstX1, dstY1); @@ -339,16 +338,19 @@ util_blit_pixels_writemask(struct blit_state *ctx, (dstX1 - dstX0) == (srcX1 - srcX0) && (dstY1 - dstY0) == (srcY1 - srcY0) && !overlap) { - struct pipe_subresource subdst; - subdst.face = dst->face; - subdst.level = dst->level; + struct pipe_box src_box; + src_box.x = srcX0; + src_box.y = srcY0; + src_box.z = srcZ0; + src_box.width = srcW; + src_box.height = srcH; + src_box.depth = 1; pipe->resource_copy_region(pipe, - dst->texture, subdst, - dstX0, dstY0, dst->zslice,/* dest */ - src_tex, srcsub, - srcX0, srcY0, srcZ0,/* src */ - srcW, srcH); /* size */ - return; + dst->texture, dst->u.tex.level, + dstX0, dstY0, dst->u.tex.first_layer,/* dest */ + src_tex, src_level, + &src_box); + return; } /* Create a temporary texture when src and dest alias or when src @@ -359,16 +361,16 @@ util_blit_pixels_writemask(struct blit_state *ctx, * This can still be improved upon. */ if ((src_tex == dst->texture && - dst->face == srcsub.face && - dst->level == srcsub.level && - dst->zslice == srcZ0) || + dst->u.tex.level == src_level && + dst->u.tex.first_layer == srcZ0) || (src_tex->target != PIPE_TEXTURE_2D && + src_tex->target != PIPE_TEXTURE_2D && src_tex->target != PIPE_TEXTURE_RECT)) { struct pipe_resource texTemp; struct pipe_resource *tex; struct pipe_sampler_view sv_templ; - struct pipe_subresource texsub; + struct pipe_box src_box; const int srcLeft = MIN2(srcX0, srcX1); const int srcTop = MIN2(srcY0, srcY1); @@ -394,19 +396,23 @@ util_blit_pixels_writemask(struct blit_state *ctx, texTemp.width0 = srcW; texTemp.height0 = srcH; texTemp.depth0 = 1; + texTemp.array_size = 1; texTemp.bind = PIPE_BIND_SAMPLER_VIEW; tex = screen->resource_create(screen, &texTemp); if (!tex) return; - texsub.face = 0; - texsub.level = 0; + src_box.x = srcLeft; + src_box.y = srcTop; + src_box.z = srcZ0; + src_box.width = srcW; + src_box.height = srcH; + src_box.depth = 1; /* load temp texture */ pipe->resource_copy_region(pipe, - tex, texsub, 0, 0, 0, /* dest */ - src_tex, srcsub, srcLeft, srcTop, srcZ0, /* src */ - srcW, srcH); /* size */ + tex, 0, 0, 0, 0, /* dest */ + src_tex, src_level, &src_box); normalized = tex->target != PIPE_TEXTURE_RECT; if(normalized) { @@ -433,7 +439,6 @@ util_blit_pixels_writemask(struct blit_state *ctx, } else { u_sampler_view_default_template(&sv_templ, src_tex, src_tex->format); - sv_templ.first_level = sv_templ.last_level = srcsub.level; sampler_view = pipe->create_sampler_view(pipe, src_tex, &sv_templ); if (!sampler_view) { @@ -447,10 +452,10 @@ util_blit_pixels_writemask(struct blit_state *ctx, normalized = sampler_view->texture->target != PIPE_TEXTURE_RECT; if(normalized) { - s0 /= (float)(u_minify(sampler_view->texture->width0, srcsub.level)); - s1 /= (float)(u_minify(sampler_view->texture->width0, srcsub.level)); - t0 /= (float)(u_minify(sampler_view->texture->height0, srcsub.level)); - t1 /= (float)(u_minify(sampler_view->texture->height0, srcsub.level)); + s0 /= (float)(u_minify(sampler_view->texture->width0, src_level)); + s1 /= (float)(u_minify(sampler_view->texture->width0, src_level)); + t0 /= (float)(u_minify(sampler_view->texture->height0, src_level)); + t1 /= (float)(u_minify(sampler_view->texture->height0, src_level)); } } @@ -489,9 +494,8 @@ util_blit_pixels_writemask(struct blit_state *ctx, ctx->sampler.normalized_coords = normalized; ctx->sampler.min_img_filter = filter; ctx->sampler.mag_img_filter = filter; - /* we've limited this already with the sampler view but you never know... */ - ctx->sampler.min_lod = srcsub.level; - ctx->sampler.max_lod = srcsub.level; + ctx->sampler.min_lod = src_level; + ctx->sampler.max_lod = src_level; cso_single_sampler(ctx->cso, 0, &ctx->sampler); cso_single_sampler_done(ctx->cso); @@ -575,7 +579,7 @@ util_blit_pixels_writemask(struct blit_state *ctx, void util_blit_pixels(struct blit_state *ctx, struct pipe_resource *src_tex, - struct pipe_subresource srcsub, + unsigned src_level, int srcX0, int srcY0, int srcX1, int srcY1, int srcZ, @@ -585,7 +589,7 @@ util_blit_pixels(struct blit_state *ctx, float z, uint filter ) { util_blit_pixels_writemask( ctx, src_tex, - srcsub, + src_level, srcX0, srcY0, srcX1, srcY1, srcZ, @@ -662,6 +666,7 @@ util_blit_pixels_tex(struct blit_state *ctx, cso_save_rasterizer(ctx->cso); cso_save_samplers(ctx->cso); cso_save_fragment_sampler_views(ctx->cso); + cso_save_viewport(ctx->cso); cso_save_framebuffer(ctx->cso); cso_save_fragment_shader(ctx->cso); cso_save_vertex_shader(ctx->cso); @@ -729,6 +734,7 @@ util_blit_pixels_tex(struct blit_state *ctx, cso_restore_rasterizer(ctx->cso); cso_restore_samplers(ctx->cso); cso_restore_fragment_sampler_views(ctx->cso); + cso_restore_viewport(ctx->cso); cso_restore_framebuffer(ctx->cso); cso_restore_fragment_shader(ctx->cso); cso_restore_vertex_shader(ctx->cso); diff --git a/src/gallium/auxiliary/util/u_blit.h b/src/gallium/auxiliary/util/u_blit.h index b8a0dfce13..3009e25eca 100644 --- a/src/gallium/auxiliary/util/u_blit.h +++ b/src/gallium/auxiliary/util/u_blit.h @@ -42,7 +42,6 @@ struct cso_context; struct pipe_context; struct pipe_resource; struct pipe_sampler_view; -struct pipe_subresource; struct pipe_surface; @@ -55,7 +54,7 @@ util_destroy_blit(struct blit_state *ctx); extern void util_blit_pixels(struct blit_state *ctx, struct pipe_resource *src_tex, - struct pipe_subresource srcsub, + unsigned src_level, int srcX0, int srcY0, int srcX1, int srcY1, int srcZ0, @@ -67,7 +66,7 @@ util_blit_pixels(struct blit_state *ctx, void util_blit_pixels_writemask(struct blit_state *ctx, struct pipe_resource *src_tex, - struct pipe_subresource srcsub, + unsigned src_level, int srcX0, int srcY0, int srcX1, int srcY1, int srcZ0, diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c index a163f93cb8..545021d264 100644 --- a/src/gallium/auxiliary/util/u_blitter.c +++ b/src/gallium/auxiliary/util/u_blitter.c @@ -63,8 +63,7 @@ struct blitter_context_priv /* Constant state objects. */ /* Vertex shaders. */ - void *vs_col; /**< Vertex shader which passes {pos, color} to the output */ - void *vs_tex; /**< Vertex shader which passes {pos, texcoord} to the output.*/ + void *vs; /**< Vertex shader which passes {pos, generic} to the output.*/ /* Fragment shaders. */ /* The shader at index i outputs color to color buffers 0,1,...,i-1. */ @@ -211,20 +210,12 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe) /* fragment shaders are created on-demand */ - /* vertex shaders */ - { - const uint semantic_names[] = { TGSI_SEMANTIC_POSITION, - TGSI_SEMANTIC_COLOR }; - const uint semantic_indices[] = { 0, 0 }; - ctx->vs_col = - util_make_vertex_passthrough_shader(pipe, 2, semantic_names, - semantic_indices); - } + /* vertex shader */ { const uint semantic_names[] = { TGSI_SEMANTIC_POSITION, TGSI_SEMANTIC_GENERIC }; const uint semantic_indices[] = { 0, 0 }; - ctx->vs_tex = + ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names, semantic_indices); } @@ -257,8 +248,7 @@ void util_blitter_destroy(struct blitter_context *blitter) pipe->delete_depth_stencil_alpha_state(pipe, ctx->dsa_flush_depth_stencil); pipe->delete_rasterizer_state(pipe, ctx->rs_state); - pipe->delete_vs_state(pipe, ctx->vs_col); - pipe->delete_vs_state(pipe, ctx->vs_tex); + pipe->delete_vs_state(pipe, ctx->vs); pipe->delete_vertex_elements_state(pipe, ctx->velem_state); for (i = 0; i < PIPE_MAX_TEXTURE_TYPES; i++) { @@ -419,17 +409,17 @@ static void blitter_set_clear_color(struct blitter_context_priv *ctx, } static void get_texcoords(struct pipe_resource *src, - struct pipe_subresource subsrc, - unsigned x1, unsigned y1, - unsigned x2, unsigned y2, - boolean normalized, float out[4]) + unsigned level, + unsigned x1, unsigned y1, + unsigned x2, unsigned y2, + boolean normalized, float out[4]) { if(normalized) { - out[0] = x1 / (float)u_minify(src->width0, subsrc.level); - out[1] = y1 / (float)u_minify(src->height0, subsrc.level); - out[2] = x2 / (float)u_minify(src->width0, subsrc.level); - out[3] = y2 / (float)u_minify(src->height0, subsrc.level); + out[0] = x1 / (float)u_minify(src->width0, level); + out[1] = y1 / (float)u_minify(src->height0, level); + out[2] = x2 / (float)u_minify(src->width0, level); + out[3] = y2 / (float)u_minify(src->height0, level); } else { @@ -458,14 +448,14 @@ static void set_texcoords_in_vertices(const float coord[4], static void blitter_set_texcoords_2d(struct blitter_context_priv *ctx, struct pipe_resource *src, - struct pipe_subresource subsrc, + unsigned level, unsigned x1, unsigned y1, unsigned x2, unsigned y2) { unsigned i; float coord[4]; - get_texcoords(src, subsrc, x1, y1, x2, y2, TRUE, coord); + get_texcoords(src, level, x1, y1, x2, y2, TRUE, coord); set_texcoords_in_vertices(coord, &ctx->vertices[0][1][0], 8); for (i = 0; i < 4; i++) { @@ -476,15 +466,15 @@ static void blitter_set_texcoords_2d(struct blitter_context_priv *ctx, static void blitter_set_texcoords_3d(struct blitter_context_priv *ctx, struct pipe_resource *src, - struct pipe_subresource subsrc, + unsigned level, unsigned zslice, unsigned x1, unsigned y1, unsigned x2, unsigned y2) { int i; - float r = zslice / (float)u_minify(src->depth0, subsrc.level); + float r = zslice / (float)u_minify(src->depth0, level); - blitter_set_texcoords_2d(ctx, src, subsrc, x1, y1, x2, y2); + blitter_set_texcoords_2d(ctx, src, level, x1, y1, x2, y2); for (i = 0; i < 4; i++) ctx->vertices[i][1][2] = r; /*r*/ @@ -492,7 +482,7 @@ static void blitter_set_texcoords_3d(struct blitter_context_priv *ctx, static void blitter_set_texcoords_cube(struct blitter_context_priv *ctx, struct pipe_resource *src, - struct pipe_subresource subsrc, + unsigned level, unsigned face, unsigned x1, unsigned y1, unsigned x2, unsigned y2) { @@ -500,10 +490,10 @@ static void blitter_set_texcoords_cube(struct blitter_context_priv *ctx, float coord[4]; float st[4][2]; - get_texcoords(src, subsrc, x1, y1, x2, y2, TRUE, coord); + get_texcoords(src, level, x1, y1, x2, y2, TRUE, coord); set_texcoords_in_vertices(coord, &st[0][0], 2); - util_map_texcoords2d_onto_cubemap(subsrc.face, + util_map_texcoords2d_onto_cubemap(face, /* pointer, stride in floats */ &st[0][0], 2, &ctx->vertices[0][1][0], 8); @@ -522,10 +512,13 @@ static void blitter_set_dst_dimensions(struct blitter_context_priv *ctx, static void blitter_draw_quad(struct blitter_context_priv *ctx) { struct pipe_context *pipe = ctx->base.pipe; + struct pipe_box box; /* write vertices and draw them */ - pipe_buffer_write(pipe, ctx->vbuf, - 0, sizeof(ctx->vertices), ctx->vertices); + u_box_1d(0, sizeof(ctx->vertices), &box); + pipe->transfer_inline_write(pipe, ctx->vbuf, 0, + PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, + &box, ctx->vertices, sizeof(ctx->vertices), 0); util_draw_vertex_buffer(pipe, ctx->vbuf, 0, PIPE_PRIM_TRIANGLE_FAN, 4, /* verts */ @@ -566,7 +559,9 @@ void *blitter_get_fs_col(struct blitter_context_priv *ctx, unsigned num_cbufs) if (!ctx->fs_col[num_cbufs]) ctx->fs_col[num_cbufs] = - util_make_fragment_clonecolor_shader(pipe, num_cbufs); + util_make_fragment_cloneinput_shader(pipe, num_cbufs, + TGSI_SEMANTIC_GENERIC, + TGSI_INTERPOLATE_LINEAR); return ctx->fs_col[num_cbufs]; } @@ -697,7 +692,7 @@ void util_blitter_clear(struct blitter_context *blitter, pipe->bind_rasterizer_state(pipe, ctx->rs_state); pipe->bind_vertex_elements_state(pipe, ctx->velem_state); pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, num_cbufs)); - pipe->bind_vs_state(pipe, ctx->vs_col); + pipe->bind_vs_state(pipe, ctx->vs); blitter_set_dst_dimensions(ctx, width, height); blitter->draw_rectangle(blitter, 0, 0, width, height, depth, @@ -714,21 +709,22 @@ boolean is_overlap(unsigned sx1, unsigned sx2, unsigned sy1, unsigned sy2, void util_blitter_copy_region(struct blitter_context *blitter, struct pipe_resource *dst, - struct pipe_subresource subdst, + unsigned dstlevel, unsigned dstx, unsigned dsty, unsigned dstz, struct pipe_resource *src, - struct pipe_subresource subsrc, - unsigned srcx, unsigned srcy, unsigned srcz, - unsigned width, unsigned height, + unsigned srclevel, + const struct pipe_box *srcbox, boolean ignore_stencil) { struct blitter_context_priv *ctx = (struct blitter_context_priv*)blitter; struct pipe_context *pipe = ctx->base.pipe; struct pipe_screen *screen = pipe->screen; - struct pipe_surface *dstsurf; + struct pipe_surface *dstsurf, surf_templ; struct pipe_framebuffer_state fb_state; struct pipe_sampler_view viewTempl, *view; unsigned bind; + unsigned width = srcbox->width; + unsigned height = srcbox->height; boolean is_stencil, is_depth; boolean normalized; @@ -739,12 +735,15 @@ void util_blitter_copy_region(struct blitter_context *blitter, /* Sanity checks. */ if (dst == src) { - assert(!is_overlap(srcx, srcx + width, srcy, srcy + height, + assert(!is_overlap(srcbox->x, srcbox->x + width, srcbox->y, srcbox->y + height, dstx, dstx + width, dsty, dsty + height)); } else { - assert(dst->format == src->format); + assert(util_is_format_compatible(util_format_description(src->format), + util_format_description(dst->format))); } assert(src->target < PIPE_MAX_TEXTURE_TYPES); + /* XXX should handle 3d regions */ + assert(srcbox->depth == 1); /* Is this a ZS format? */ is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0; @@ -762,15 +761,18 @@ void util_blitter_copy_region(struct blitter_context *blitter, dst->nr_samples, bind, 0) || !screen->is_format_supported(screen, src->format, src->target, src->nr_samples, PIPE_BIND_SAMPLER_VIEW, 0)) { - util_resource_copy_region(pipe, dst, subdst, dstx, dsty, dstz, - src, subsrc, srcx, srcy, srcz, width, height); + util_resource_copy_region(pipe, dst, dstlevel, dstx, dsty, dstz, + src, srclevel, srcbox); return; } - /* Get surfaces. */ - dstsurf = screen->get_tex_surface(screen, dst, - subdst.face, subdst.level, dstz, - bind); + /* Get surface. */ + memset(&surf_templ, 0, sizeof(surf_templ)); + u_surface_default_template(&surf_templ, dst, bind); + surf_templ.u.tex.level = dstlevel; + surf_templ.u.tex.first_layer = dstz; + surf_templ.u.tex.last_layer = dstz; + dstsurf = pipe->create_surface(pipe, dst, &surf_templ); /* Check whether the states are properly saved. */ blitter_check_saved_CSOs(ctx); @@ -810,9 +812,9 @@ void util_blitter_copy_region(struct blitter_context *blitter, /* Set rasterizer state, shaders, and textures. */ pipe->bind_rasterizer_state(pipe, ctx->rs_state); - pipe->bind_vs_state(pipe, ctx->vs_tex); + pipe->bind_vs_state(pipe, ctx->vs); pipe->bind_fragment_sampler_states(pipe, 1, - blitter_get_sampler_state(ctx, subsrc.level, normalized)); + blitter_get_sampler_state(ctx, srclevel, normalized)); pipe->bind_vertex_elements_state(pipe, ctx->velem_state); pipe->set_fragment_sampler_views(pipe, 1, &view); pipe->set_framebuffer_state(pipe, &fb_state); @@ -827,8 +829,8 @@ void util_blitter_copy_region(struct blitter_context *blitter, { /* Set texture coordinates. */ float coord[4]; - get_texcoords(src, subsrc, srcx, srcy, - srcx+width, srcy+height, normalized, coord); + get_texcoords(src, srclevel, srcbox->x, srcbox->y, + srcbox->x+width, srcbox->y+height, normalized, coord); /* Draw. */ blitter->draw_rectangle(blitter, dstx, dsty, dstx+width, dsty+height, 0, @@ -841,11 +843,13 @@ void util_blitter_copy_region(struct blitter_context *blitter, case PIPE_TEXTURE_CUBE: /* Set texture coordinates. */ if (src->target == PIPE_TEXTURE_3D) - blitter_set_texcoords_3d(ctx, src, subsrc, srcz, - srcx, srcy, srcx+width, srcy+height); + blitter_set_texcoords_3d(ctx, src, srclevel, srcbox->z, + srcbox->x, srcbox->y, + srcbox->x + width, srcbox->y + height); else - blitter_set_texcoords_cube(ctx, src, subsrc, - srcx, srcy, srcx+width, srcy+height); + blitter_set_texcoords_cube(ctx, src, srclevel, srcbox->z, + srcbox->x, srcbox->y, + srcbox->x + width, srcbox->y + height); /* Draw. */ blitter_set_rectangle(ctx, dstx, dsty, dstx+width, dsty+height, 0); @@ -887,7 +891,7 @@ void util_blitter_clear_render_target(struct blitter_context *blitter, pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_stencil); pipe->bind_rasterizer_state(pipe, ctx->rs_state); pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 1)); - pipe->bind_vs_state(pipe, ctx->vs_col); + pipe->bind_vs_state(pipe, ctx->vs); pipe->bind_vertex_elements_state(pipe, ctx->velem_state); /* set a framebuffer state */ @@ -947,7 +951,7 @@ void util_blitter_clear_depth_stencil(struct blitter_context *blitter, pipe->bind_rasterizer_state(pipe, ctx->rs_state); pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 0)); - pipe->bind_vs_state(pipe, ctx->vs_col); + pipe->bind_vs_state(pipe, ctx->vs); pipe->bind_vertex_elements_state(pipe, ctx->velem_state); /* set a framebuffer state */ @@ -988,7 +992,7 @@ void util_blitter_custom_depth_stencil(struct blitter_context *blitter, pipe->bind_rasterizer_state(pipe, ctx->rs_state); pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 0)); - pipe->bind_vs_state(pipe, ctx->vs_col); + pipe->bind_vs_state(pipe, ctx->vs); pipe->bind_vertex_elements_state(pipe, ctx->velem_state); /* set a framebuffer state */ diff --git a/src/gallium/auxiliary/util/u_blitter.h b/src/gallium/auxiliary/util/u_blitter.h index f9f96f25c7..922a8580ac 100644 --- a/src/gallium/auxiliary/util/u_blitter.h +++ b/src/gallium/auxiliary/util/u_blitter.h @@ -164,12 +164,11 @@ void util_blitter_clear(struct blitter_context *blitter, */ void util_blitter_copy_region(struct blitter_context *blitter, struct pipe_resource *dst, - struct pipe_subresource subdst, + unsigned dstlevel, unsigned dstx, unsigned dsty, unsigned dstz, struct pipe_resource *src, - struct pipe_subresource subsrc, - unsigned srcx, unsigned srcy, unsigned srcz, - unsigned width, unsigned height, + unsigned srclevel, + const struct pipe_box *srcbox, boolean ignore_stencil); /** @@ -319,21 +318,13 @@ util_blitter_save_vertex_buffers(struct blitter_context *blitter, int num_vertex_buffers, struct pipe_vertex_buffer *vertex_buffers) { - unsigned i; assert(num_vertex_buffers <= Elements(blitter->saved_vertex_buffers)); - blitter->saved_num_vertex_buffers = num_vertex_buffers; - - for (i = 0; i < num_vertex_buffers; i++) { - if (vertex_buffers[i].buffer) { - pipe_resource_reference(&blitter->saved_vertex_buffers[i].buffer, - vertex_buffers[i].buffer); - } - } - - memcpy(blitter->saved_vertex_buffers, - vertex_buffers, - num_vertex_buffers * sizeof(struct pipe_vertex_buffer)); + blitter->saved_num_vertex_buffers = 0; + util_copy_vertex_buffers(blitter->saved_vertex_buffers, + (unsigned*)&blitter->saved_num_vertex_buffers, + vertex_buffers, + num_vertex_buffers); } #ifdef __cplusplus diff --git a/src/gallium/auxiliary/util/u_box.h b/src/gallium/auxiliary/util/u_box.h index e9c71743fc..0b28d0f12c 100644 --- a/src/gallium/auxiliary/util/u_box.h +++ b/src/gallium/auxiliary/util/u_box.h @@ -60,7 +60,6 @@ void u_box_2d_zslice( unsigned x, box->depth = 1; } - static INLINE void u_box_3d( unsigned x, unsigned y, @@ -78,15 +77,4 @@ void u_box_3d( unsigned x, box->depth = d; } - -static INLINE -struct pipe_subresource u_subresource( unsigned face, - unsigned level ) -{ - struct pipe_subresource subresource; - subresource.face = face; - subresource.level = level; - return subresource; -} - #endif diff --git a/src/gallium/auxiliary/util/u_debug.c b/src/gallium/auxiliary/util/u_debug.c index 504e6d2a18..f4ad545bee 100644 --- a/src/gallium/auxiliary/util/u_debug.c +++ b/src/gallium/auxiliary/util/u_debug.c @@ -40,12 +40,14 @@ #include "util/u_string.h" #include "util/u_math.h" #include "util/u_tile.h" -#include "util/u_prim.h" +#include "util/u_prim.h" +#include "util/u_surface.h" #include <limits.h> /* CHAR_BIT */ void _debug_vprintf(const char *format, va_list ap) { +#if defined(PIPE_OS_WINDOWS) || defined(PIPE_OS_EMBEDDED) /* We buffer until we find a newline. */ static char buf[4096] = {'\0'}; size_t len = strlen(buf); @@ -54,6 +56,10 @@ void _debug_vprintf(const char *format, va_list ap) os_log_message(buf); buf[0] = '\0'; } +#else + /* Just print as-is to stderr */ + vfprintf(stderr, format, ap); +#endif } @@ -453,9 +459,10 @@ void debug_dump_image(const char *prefix, #endif } +/* FIXME: dump resources, not surfaces... */ void debug_dump_surface(struct pipe_context *pipe, - const char *prefix, - struct pipe_surface *surface) + const char *prefix, + struct pipe_surface *surface) { struct pipe_resource *texture; struct pipe_transfer *transfer; @@ -472,23 +479,23 @@ void debug_dump_surface(struct pipe_context *pipe, */ texture = surface->texture; - transfer = pipe_get_transfer(pipe, texture, surface->face, - surface->level, surface->zslice, - PIPE_TRANSFER_READ, 0, 0, surface->width, - surface->height); - + transfer = pipe_get_transfer(pipe, texture, surface->u.tex.level, + surface->u.tex.first_layer, + PIPE_TRANSFER_READ, + 0, 0, surface->width, surface->height); + data = pipe->transfer_map(pipe, transfer); if(!data) goto error; - - debug_dump_image(prefix, + + debug_dump_image(prefix, texture->format, - util_format_get_blocksize(texture->format), + util_format_get_blocksize(texture->format), util_format_get_nblocksx(texture->format, surface->width), util_format_get_nblocksy(texture->format, surface->height), transfer->stride, data); - + pipe->transfer_unmap(pipe, transfer); error: pipe->transfer_destroy(pipe, transfer); @@ -499,20 +506,18 @@ void debug_dump_texture(struct pipe_context *pipe, const char *prefix, struct pipe_resource *texture) { - struct pipe_surface *surface; - struct pipe_screen *screen; + struct pipe_surface *surface, surf_tmpl; if (!texture) return; - screen = texture->screen; - - /* XXX for now, just dump image for face=0, level=0 */ - surface = screen->get_tex_surface(screen, texture, 0, 0, 0, - PIPE_BIND_SAMPLER_VIEW); + /* XXX for now, just dump image for layer=0, level=0 */ + memset(&surf_tmpl, 0, sizeof(surf_tmpl)); + u_surface_default_template(&surf_tmpl, texture, 0 /* no bind flag - not a surface */); + surface = pipe->create_surface(pipe, texture, &surf_tmpl); if (surface) { debug_dump_surface(pipe, prefix, surface); - screen->tex_surface_destroy(surface); + pipe->surface_destroy(pipe, surface); } } @@ -550,17 +555,16 @@ struct bmp_rgb_quad { void debug_dump_surface_bmp(struct pipe_context *pipe, - const char *filename, + const char *filename, struct pipe_surface *surface) { #ifndef PIPE_SUBSYSTEM_WINDOWS_MINIPORT struct pipe_transfer *transfer; struct pipe_resource *texture = surface->texture; - transfer = pipe_get_transfer(pipe, texture, surface->face, - surface->level, surface->zslice, - PIPE_TRANSFER_READ, 0, 0, surface->width, - surface->height); + transfer = pipe_get_transfer(pipe, texture, surface->u.tex.level, + surface->u.tex.first_layer, PIPE_TRANSFER_READ, + 0, 0, surface->width, surface->height); debug_dump_transfer_bmp(pipe, filename, transfer); diff --git a/src/gallium/auxiliary/util/u_debug_describe.c b/src/gallium/auxiliary/util/u_debug_describe.c index 1c90ff3106..7ed8ee608a 100644 --- a/src/gallium/auxiliary/util/u_debug_describe.c +++ b/src/gallium/auxiliary/util/u_debug_describe.c @@ -69,7 +69,7 @@ debug_describe_surface(char* buf, const struct pipe_surface *ptr) { char res[128]; debug_describe_resource(res, ptr->texture); - util_sprintf(buf, "pipe_surface<%s,%u,%u,%u>", res, ptr->face, ptr->level, ptr->zslice); + util_sprintf(buf, "pipe_surface<%s,%u,%u,%u>", res, ptr->u.tex.level, ptr->u.tex.first_layer, ptr->u.tex.last_layer); } void diff --git a/src/gallium/auxiliary/util/u_debug_stack.c b/src/gallium/auxiliary/util/u_debug_stack.c index 528a1c394b..24e039fd22 100644 --- a/src/gallium/auxiliary/util/u_debug_stack.c +++ b/src/gallium/auxiliary/util/u_debug_stack.c @@ -48,7 +48,10 @@ debug_backtrace_capture(struct debug_stack_frame *backtrace, if(!nr_frames) return; -#if defined(PIPE_CC_GCC) +#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86) + __asm__ __volatile__("mov (%%ebp),%0": "=r" (frame_pointer)); + frame_pointer = (const void **)frame_pointer[0]; +#elif defined(PIPE_CC_GCC) frame_pointer = ((const void **)__builtin_frame_address(1)); #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86) __asm { diff --git a/src/gallium/auxiliary/util/u_debug_symbol.c b/src/gallium/auxiliary/util/u_debug_symbol.c index 332952af88..44d437747a 100644 --- a/src/gallium/auxiliary/util/u_debug_symbol.c +++ b/src/gallium/auxiliary/util/u_debug_symbol.c @@ -40,7 +40,7 @@ #include "u_debug_symbol.h" #include "u_hash_table.h" -#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) && defined(PIPE_ARCH_X86) +#if defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86) #include <windows.h> #include <stddef.h> @@ -165,7 +165,7 @@ debug_symbol_name_glibc(const void *addr, char* buf, unsigned size) void debug_symbol_name(const void *addr, char* buf, unsigned size) { -#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) && defined(PIPE_ARCH_X86) +#if defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86) debug_symbol_name_imagehlp(addr, buf, size); if(buf[0]) return; diff --git a/src/gallium/auxiliary/util/u_dirty_flags.h b/src/gallium/auxiliary/util/u_dirty_flags.h index 7e1be45ad5..40539f0b0e 100644 --- a/src/gallium/auxiliary/util/u_dirty_flags.h +++ b/src/gallium/auxiliary/util/u_dirty_flags.h @@ -24,5 +24,9 @@ #define U_NEW_VERTEX_BUFFER 0x10000 #define U_NEW_QUERY 0x20000 #define U_NEW_DEPTH_STENCIL 0x40000 +#define U_NEW_GS 0x80000 +#define U_NEW_GS_CONSTANTS 0x100000 +#define U_NEW_GS_SAMPLER_VIEW 0x200000 +#define U_NEW_GS_SAMPLER_STATES 0x400000 #endif diff --git a/src/gallium/auxiliary/util/u_dirty_surfaces.h b/src/gallium/auxiliary/util/u_dirty_surfaces.h index fd1bbe5ffd..f3618d9be7 100644 --- a/src/gallium/auxiliary/util/u_dirty_surfaces.h +++ b/src/gallium/auxiliary/util/u_dirty_surfaces.h @@ -77,7 +77,7 @@ util_dirty_surfaces_use_levels_for_sampling(struct pipe_context *pipe, struct ut struct util_dirty_surface *ds = LIST_ENTRY(struct util_dirty_surface, p, dirty_list); next = p->next; - if(ds->base.level >= first && ds->base.level <= last) + if(ds->base.u.tex.level >= first && ds->base.u.tex.level <= last) flush(pipe, &ds->base); } } @@ -86,7 +86,8 @@ static INLINE void util_dirty_surfaces_use_for_sampling_with(struct pipe_context *pipe, struct util_dirty_surfaces *dss, struct pipe_sampler_view *psv, struct pipe_sampler_state *pss, util_dirty_surface_flush_t flush) { if(!LIST_IS_EMPTY(&dss->dirty_list)) - util_dirty_surfaces_use_levels_for_sampling(pipe, dss, (unsigned)pss->min_lod + psv->first_level, MIN2((unsigned)ceilf(pss->max_lod) + psv->first_level, psv->last_level), flush); + util_dirty_surfaces_use_levels_for_sampling(pipe, dss, (unsigned)pss->min_lod + psv->u.tex.first_level, + MIN2((unsigned)ceilf(pss->max_lod) + psv->u.tex.first_level, psv->u.tex.last_level), flush); } static INLINE void diff --git a/src/gallium/auxiliary/util/u_dump_state.c b/src/gallium/auxiliary/util/u_dump_state.c index cda5b8ba51..b471d59eeb 100644 --- a/src/gallium/auxiliary/util/u_dump_state.c +++ b/src/gallium/auxiliary/util/u_dump_state.c @@ -279,6 +279,10 @@ util_dump_template(struct os_stream *stream, const struct pipe_resource *templat util_dump_uint(stream, templat->depth0); util_dump_member_end(stream); + util_dump_member_begin(stream, "array_size"); + util_dump_uint(stream, templat->array_size); + util_dump_member_end(stream); + util_dump_member(stream, uint, templat, last_level); util_dump_member(stream, uint, templat, usage); util_dump_member(stream, uint, templat, bind); @@ -633,14 +637,12 @@ util_dump_surface(struct os_stream *stream, const struct pipe_surface *state) util_dump_member(stream, uint, state, width); util_dump_member(stream, uint, state, height); - util_dump_member(stream, uint, state, layout); - util_dump_member(stream, uint, state, offset); util_dump_member(stream, uint, state, usage); util_dump_member(stream, ptr, state, texture); - util_dump_member(stream, uint, state, face); - util_dump_member(stream, uint, state, level); - util_dump_member(stream, uint, state, zslice); + util_dump_member(stream, uint, state, u.tex.level); + util_dump_member(stream, uint, state, u.tex.first_layer); + util_dump_member(stream, uint, state, u.tex.last_layer); util_dump_struct_end(stream); } @@ -660,7 +662,7 @@ util_dump_transfer(struct os_stream *stream, const struct pipe_transfer *state) /*util_dump_member(stream, uint, state, box);*/ util_dump_member(stream, uint, state, stride); - util_dump_member(stream, uint, state, slice_stride); + util_dump_member(stream, uint, state, layer_stride); /*util_dump_member(stream, ptr, state, data);*/ diff --git a/src/gallium/auxiliary/util/u_format.csv b/src/gallium/auxiliary/util/u_format.csv index 1fbd83841c..26a0eebc54 100644 --- a/src/gallium/auxiliary/util/u_format.csv +++ b/src/gallium/auxiliary/util/u_format.csv @@ -76,6 +76,7 @@ PIPE_FORMAT_B4G4R4X4_UNORM , plain, 1, 1, un4 , un4 , un4 , x4 , zyx1, r PIPE_FORMAT_B5G6R5_UNORM , plain, 1, 1, un5 , un6 , un5 , , zyx1, rgb PIPE_FORMAT_R10G10B10A2_UNORM , plain, 1, 1, un10, un10, un10, un2 , xyzw, rgb PIPE_FORMAT_B10G10R10A2_UNORM , plain, 1, 1, un10, un10, un10, un2 , zyxw, rgb +PIPE_FORMAT_B2G3R3_UNORM , plain, 1, 1, un2 , un3 , un3 , , zyx1, rgb # Luminance/Intensity/Alpha formats PIPE_FORMAT_L8_UNORM , plain, 1, 1, un8 , , , , xxx1, rgb @@ -84,6 +85,9 @@ PIPE_FORMAT_I8_UNORM , plain, 1, 1, un8 , , , , xxxx, r PIPE_FORMAT_L4A4_UNORM , plain, 1, 1, un4 , un4 , , , xxxy, rgb PIPE_FORMAT_L8A8_UNORM , plain, 1, 1, un8 , un8 , , , xxxy, rgb PIPE_FORMAT_L16_UNORM , plain, 1, 1, un16, , , , xxx1, rgb +PIPE_FORMAT_A16_UNORM , plain, 1, 1, un16, , , , 000x, rgb +PIPE_FORMAT_I16_UNORM , plain, 1, 1, un16, , , , xxxx, rgb +PIPE_FORMAT_L16A16_UNORM , plain, 1, 1, un16, un16, , , xxxy, rgb # SRGB formats PIPE_FORMAT_L8_SRGB , plain, 1, 1, un8 , , , , xxx1, srgb diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c index 6a931a9581..d22ae8b375 100644 --- a/src/gallium/auxiliary/util/u_gen_mipmap.c +++ b/src/gallium/auxiliary/util/u_gen_mipmap.c @@ -48,6 +48,8 @@ #include "util/u_simple_shaders.h" #include "util/u_math.h" #include "util/u_texture.h" +#include "util/u_half.h" +#include "util/u_surface.h" #include "cso_cache/cso_context.h" @@ -65,7 +67,7 @@ struct gen_mipmap_state struct pipe_vertex_element velem[2]; void *vs; - void *fs2d, *fsCube; + void *fs1d, *fs2d, *fs3d, *fsCube; struct pipe_resource *vbuf; /**< quad vertices */ unsigned vbuf_slot; @@ -89,24 +91,7 @@ enum dtype }; -typedef ushort half_float; - - -static half_float -float_to_half(float f) -{ - /* XXX fix this */ - return 0; -} - -static float -half_to_float(half_float h) -{ - /* XXX fix this */ - return 0.0f; -} - - +typedef uint16_t half_float; /** @@ -145,7 +130,7 @@ half_to_float(half_float h) rowC[j][e], rowC[k][e], \ rowD[j][e], rowD[k][e]); \ } while(0) - + #define FILTER_F_3D(e) \ do { \ dst[i][e] = (rowA[j][e] + rowA[k][e] \ @@ -156,15 +141,15 @@ half_to_float(half_float h) #define FILTER_HF_3D(e) \ do { \ - const float aj = half_to_float(rowA[j][e]); \ - const float ak = half_to_float(rowA[k][e]); \ - const float bj = half_to_float(rowB[j][e]); \ - const float bk = half_to_float(rowB[k][e]); \ - const float cj = half_to_float(rowC[j][e]); \ - const float ck = half_to_float(rowC[k][e]); \ - const float dj = half_to_float(rowD[j][e]); \ - const float dk = half_to_float(rowD[k][e]); \ - dst[i][e] = float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \ + const float aj = util_half_to_float(rowA[j][e]); \ + const float ak = util_half_to_float(rowA[k][e]); \ + const float bj = util_half_to_float(rowB[j][e]); \ + const float bk = util_half_to_float(rowB[k][e]); \ + const float cj = util_half_to_float(rowC[j][e]); \ + const float ck = util_half_to_float(rowC[k][e]); \ + const float dj = util_half_to_float(rowD[j][e]); \ + const float dk = util_half_to_float(rowD[k][e]); \ + dst[i][e] = util_float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \ * 0.125F); \ } while(0) /*@}*/ @@ -343,8 +328,7 @@ do_row(enum dtype datatype, uint comps, int srcWidth, } } -#if 0 - else if (datatype == HALF_DTYPE_FLOAT && comps == 4) { + else if (datatype == DTYPE_HALF_FLOAT && comps == 4) { uint i, j, k, comp; const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA; const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB; @@ -353,11 +337,11 @@ do_row(enum dtype datatype, uint comps, int srcWidth, i++, j += colStride, k += colStride) { for (comp = 0; comp < 4; comp++) { float aj, ak, bj, bk; - aj = half_to_float(rowA[j][comp]); - ak = half_to_float(rowA[k][comp]); - bj = half_to_float(rowB[j][comp]); - bk = half_to_float(rowB[k][comp]); - dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); + aj = util_half_to_float(rowA[j][comp]); + ak = util_half_to_float(rowA[k][comp]); + bj = util_half_to_float(rowB[j][comp]); + bk = util_half_to_float(rowB[k][comp]); + dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F); } } } @@ -370,11 +354,11 @@ do_row(enum dtype datatype, uint comps, int srcWidth, i++, j += colStride, k += colStride) { for (comp = 0; comp < 3; comp++) { float aj, ak, bj, bk; - aj = half_to_float(rowA[j][comp]); - ak = half_to_float(rowA[k][comp]); - bj = half_to_float(rowB[j][comp]); - bk = half_to_float(rowB[k][comp]); - dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); + aj = util_half_to_float(rowA[j][comp]); + ak = util_half_to_float(rowA[k][comp]); + bj = util_half_to_float(rowB[j][comp]); + bk = util_half_to_float(rowB[k][comp]); + dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F); } } } @@ -387,11 +371,11 @@ do_row(enum dtype datatype, uint comps, int srcWidth, i++, j += colStride, k += colStride) { for (comp = 0; comp < 2; comp++) { float aj, ak, bj, bk; - aj = half_to_float(rowA[j][comp]); - ak = half_to_float(rowA[k][comp]); - bj = half_to_float(rowB[j][comp]); - bk = half_to_float(rowB[k][comp]); - dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); + aj = util_half_to_float(rowA[j][comp]); + ak = util_half_to_float(rowA[k][comp]); + bj = util_half_to_float(rowB[j][comp]); + bk = util_half_to_float(rowB[k][comp]); + dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F); } } } @@ -403,14 +387,13 @@ do_row(enum dtype datatype, uint comps, int srcWidth, for (i = j = 0, k = k0; i < (uint) dstWidth; i++, j += colStride, k += colStride) { float aj, ak, bj, bk; - aj = half_to_float(rowA[j]); - ak = half_to_float(rowA[k]); - bj = half_to_float(rowB[j]); - bk = half_to_float(rowB[k]); - dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F); + aj = util_half_to_float(rowA[j]); + ak = util_half_to_float(rowA[k]); + bj = util_half_to_float(rowB[j]); + bk = util_half_to_float(rowB[k]); + dst[i] = util_float_to_half((aj + ak + bj + bk) * 0.25F); } } -#endif else if (datatype == DTYPE_UINT && comps == 1) { uint i, j, k; @@ -1036,32 +1019,34 @@ reduce_2d(enum pipe_format pformat, static void reduce_3d(enum pipe_format pformat, int srcWidth, int srcHeight, int srcDepth, - int srcRowStride, const ubyte *srcPtr, + int srcRowStride, int srcImageStride, const ubyte *srcPtr, int dstWidth, int dstHeight, int dstDepth, - int dstRowStride, ubyte *dstPtr) + int dstRowStride, int dstImageStride, ubyte *dstPtr) { const int bpt = util_format_get_blocksize(pformat); - const int border = 0; int img, row; - int bytesPerSrcImage, bytesPerDstImage; - int bytesPerSrcRow, bytesPerDstRow; int srcImageOffset, srcRowOffset; enum dtype datatype; uint comps; format_to_type_comps(pformat, &datatype, &comps); - bytesPerSrcImage = srcWidth * srcHeight * bpt; - bytesPerDstImage = dstWidth * dstHeight * bpt; + /* XXX I think we should rather assert those strides */ + if (!srcImageStride) + srcImageStride = srcWidth * srcHeight * bpt; + if (!dstImageStride) + dstImageStride = dstWidth * dstHeight * bpt; - bytesPerSrcRow = srcWidth * bpt; - bytesPerDstRow = dstWidth * bpt; + if (!srcRowStride) + srcRowStride = srcWidth * bpt; + if (!dstRowStride) + dstRowStride = dstWidth * bpt; /* Offset between adjacent src images to be averaged together */ - srcImageOffset = (srcDepth == dstDepth) ? 0 : bytesPerSrcImage; + srcImageOffset = (srcDepth == dstDepth) ? 0 : srcImageStride; /* Offset between adjacent src rows to be averaged together */ - srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt; + srcRowOffset = (srcHeight == dstHeight) ? 0 : srcRowStride; /* * Need to average together up to 8 src pixels for each dest pixel. @@ -1077,16 +1062,13 @@ reduce_3d(enum pipe_format pformat, */ for (img = 0; img < dstDepth; img++) { - /* first source image pointer, skipping border */ + /* first source image pointer */ const ubyte *imgSrcA = srcPtr - + (bytesPerSrcImage + bytesPerSrcRow + border) * bpt * border - + img * (bytesPerSrcImage + srcImageOffset); - /* second source image pointer, skipping border */ + + img * (srcImageStride + srcImageOffset); + /* second source image pointer */ const ubyte *imgSrcB = imgSrcA + srcImageOffset; - /* address of the dest image, skipping border */ - ubyte *imgDst = dstPtr - + (bytesPerDstImage + bytesPerDstRow + border) * bpt * border - + img * bytesPerDstImage; + /* address of the dest image */ + ubyte *imgDst = dstPtr + img * dstImageStride; /* setup the four source row pointers and the dest row pointer */ const ubyte *srcImgARowA = imgSrcA; @@ -1102,11 +1084,11 @@ reduce_3d(enum pipe_format pformat, dstWidth, dstImgRow); /* advance to next rows */ - srcImgARowA += bytesPerSrcRow + srcRowOffset; - srcImgARowB += bytesPerSrcRow + srcRowOffset; - srcImgBRowA += bytesPerSrcRow + srcRowOffset; - srcImgBRowB += bytesPerSrcRow + srcRowOffset; - dstImgRow += bytesPerDstRow; + srcImgARowA += srcRowStride + srcRowOffset; + srcImgARowB += srcRowStride + srcRowOffset; + srcImgBRowA += srcRowStride + srcRowOffset; + srcImgBRowB += srcRowStride + srcRowOffset; + dstImgRow += dstImageStride; } } } @@ -1117,25 +1099,24 @@ reduce_3d(enum pipe_format pformat, static void make_1d_mipmap(struct gen_mipmap_state *ctx, struct pipe_resource *pt, - uint face, uint baseLevel, uint lastLevel) + uint layer, uint baseLevel, uint lastLevel) { struct pipe_context *pipe = ctx->pipe; - const uint zslice = 0; uint dstLevel; for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { const uint srcLevel = dstLevel - 1; struct pipe_transfer *srcTrans, *dstTrans; void *srcMap, *dstMap; - - srcTrans = pipe_get_transfer(pipe, pt, face, srcLevel, zslice, - PIPE_TRANSFER_READ, 0, 0, - u_minify(pt->width0, srcLevel), - u_minify(pt->height0, srcLevel)); - dstTrans = pipe_get_transfer(pipe, pt, face, dstLevel, zslice, - PIPE_TRANSFER_WRITE, 0, 0, - u_minify(pt->width0, dstLevel), - u_minify(pt->height0, dstLevel)); + + srcTrans = pipe_get_transfer(pipe, pt, srcLevel, layer, + PIPE_TRANSFER_READ, 0, 0, + u_minify(pt->width0, srcLevel), + u_minify(pt->height0, srcLevel)); + dstTrans = pipe_get_transfer(pipe, pt, dstLevel, layer, + PIPE_TRANSFER_WRITE, 0, 0, + u_minify(pt->width0, dstLevel), + u_minify(pt->height0, dstLevel)); srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); @@ -1156,12 +1137,11 @@ make_1d_mipmap(struct gen_mipmap_state *ctx, static void make_2d_mipmap(struct gen_mipmap_state *ctx, struct pipe_resource *pt, - uint face, uint baseLevel, uint lastLevel) + uint layer, uint baseLevel, uint lastLevel) { struct pipe_context *pipe = ctx->pipe; - const uint zslice = 0; uint dstLevel; - + assert(util_format_get_blockwidth(pt->format) == 1); assert(util_format_get_blockheight(pt->format) == 1); @@ -1169,15 +1149,15 @@ make_2d_mipmap(struct gen_mipmap_state *ctx, const uint srcLevel = dstLevel - 1; struct pipe_transfer *srcTrans, *dstTrans; ubyte *srcMap, *dstMap; - - srcTrans = pipe_get_transfer(pipe, pt, face, srcLevel, zslice, - PIPE_TRANSFER_READ, 0, 0, - u_minify(pt->width0, srcLevel), - u_minify(pt->height0, srcLevel)); - dstTrans = pipe_get_transfer(pipe, pt, face, dstLevel, zslice, - PIPE_TRANSFER_WRITE, 0, 0, - u_minify(pt->width0, dstLevel), - u_minify(pt->height0, dstLevel)); + + srcTrans = pipe_get_transfer(pipe, pt, srcLevel, layer, + PIPE_TRANSFER_READ, 0, 0, + u_minify(pt->width0, srcLevel), + u_minify(pt->height0, srcLevel)); + dstTrans = pipe_get_transfer(pipe, pt, dstLevel, layer, + PIPE_TRANSFER_WRITE, 0, 0, + u_minify(pt->width0, dstLevel), + u_minify(pt->height0, dstLevel)); srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); @@ -1197,41 +1177,49 @@ make_2d_mipmap(struct gen_mipmap_state *ctx, } +/* XXX looks a bit more like it could work now but need to test */ static void make_3d_mipmap(struct gen_mipmap_state *ctx, struct pipe_resource *pt, uint face, uint baseLevel, uint lastLevel) { -#if 0 struct pipe_context *pipe = ctx->pipe; - struct pipe_screen *screen = pipe->screen; - uint dstLevel, zslice = 0; + uint dstLevel; + struct pipe_box src_box, dst_box; assert(util_format_get_blockwidth(pt->format) == 1); assert(util_format_get_blockheight(pt->format) == 1); + src_box.x = src_box.y = src_box.z = 0; + dst_box.x = dst_box.y = dst_box.z = 0; + for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { const uint srcLevel = dstLevel - 1; struct pipe_transfer *srcTrans, *dstTrans; ubyte *srcMap, *dstMap; - - srcTrans = pipe->get_transfer(pipe, pt, face, srcLevel, zslice, - PIPE_TRANSFER_READ, 0, 0, - u_minify(pt->width0, srcLevel), - u_minify(pt->height0, srcLevel)); - dstTrans = pipe->get_transfer(pipe, pt, face, dstLevel, zslice, - PIPE_TRANSFER_WRITE, 0, 0, - u_minify(pt->width0, dstLevel), - u_minify(pt->height0, dstLevel)); + struct pipe_box src_box, dst_box; + src_box.width = u_minify(pt->width0, srcLevel); + src_box.height = u_minify(pt->height0, srcLevel); + src_box.depth = u_minify(pt->depth0, srcLevel); + dst_box.width = u_minify(pt->width0, dstLevel); + dst_box.height = u_minify(pt->height0, dstLevel); + dst_box.depth = u_minify(pt->depth0, dstLevel); + + srcTrans = pipe->get_transfer(pipe, pt, srcLevel, + PIPE_TRANSFER_READ, + &src_box); + dstTrans = pipe->get_transfer(pipe, pt, dstLevel, + PIPE_TRANSFER_WRITE, + &dst_box); srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); reduce_3d(pt->format, - srcTrans->width, srcTrans->height, - srcTrans->stride, srcMap, - dstTrans->width, dstTrans->height, - dstTrans->stride, dstMap); + srcTrans->box.width, srcTrans->box.height, srcTrans->box.depth, + srcTrans->stride, srcTrans->layer_stride, srcMap, + dstTrans->box.width, dstTrans->box.height, dstTrans->box.depth, + dstTrans->stride, dstTrans->layer_stride, dstMap); pipe->transfer_unmap(pipe, srcTrans); pipe->transfer_unmap(pipe, dstTrans); @@ -1239,28 +1227,25 @@ make_3d_mipmap(struct gen_mipmap_state *ctx, pipe->transfer_destroy(pipe, srcTrans); pipe->transfer_destroy(pipe, dstTrans); } -#else - (void) reduce_3d; -#endif } static void fallback_gen_mipmap(struct gen_mipmap_state *ctx, struct pipe_resource *pt, - uint face, uint baseLevel, uint lastLevel) + uint layer, uint baseLevel, uint lastLevel) { switch (pt->target) { case PIPE_TEXTURE_1D: - make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel); + make_1d_mipmap(ctx, pt, layer, baseLevel, lastLevel); break; case PIPE_TEXTURE_2D: case PIPE_TEXTURE_RECT: case PIPE_TEXTURE_CUBE: - make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel); + make_2d_mipmap(ctx, pt, layer, baseLevel, lastLevel); break; case PIPE_TEXTURE_3D: - make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel); + make_3d_mipmap(ctx, pt, layer, baseLevel, lastLevel); break; default: assert(0); @@ -1328,8 +1313,12 @@ util_create_gen_mipmap(struct pipe_context *pipe, } /* fragment shader */ + ctx->fs1d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_1D, + TGSI_INTERPOLATE_LINEAR); ctx->fs2d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_2D, TGSI_INTERPOLATE_LINEAR); + ctx->fs3d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_3D, + TGSI_INTERPOLATE_LINEAR); ctx->fsCube = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_CUBE, TGSI_INTERPOLATE_LINEAR); @@ -1371,7 +1360,7 @@ get_next_slot(struct gen_mipmap_state *ctx) static unsigned set_vertex_data(struct gen_mipmap_state *ctx, enum pipe_texture_target tex_target, - uint face) + uint layer, float r) { unsigned offset; @@ -1397,26 +1386,26 @@ set_vertex_data(struct gen_mipmap_state *ctx, {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f} }; - util_map_texcoords2d_onto_cubemap(face, &st[0][0], 2, + util_map_texcoords2d_onto_cubemap(layer, &st[0][0], 2, &ctx->vertices[0][1][0], 8); } else { - /* 1D/2D */ + /* 1D/2D/3D */ ctx->vertices[0][1][0] = 0.0f; /*s*/ ctx->vertices[0][1][1] = 0.0f; /*t*/ - ctx->vertices[0][1][2] = 0.0f; /*r*/ + ctx->vertices[0][1][2] = r; /*r*/ ctx->vertices[1][1][0] = 1.0f; ctx->vertices[1][1][1] = 0.0f; - ctx->vertices[1][1][2] = 0.0f; + ctx->vertices[1][1][2] = r; ctx->vertices[2][1][0] = 1.0f; ctx->vertices[2][1][1] = 1.0f; - ctx->vertices[2][1][2] = 0.0f; + ctx->vertices[2][1][2] = r; ctx->vertices[3][1][0] = 0.0f; ctx->vertices[3][1][1] = 1.0f; - ctx->vertices[3][1][2] = 0.0f; + ctx->vertices[3][1][2] = r; } offset = get_next_slot( ctx ); @@ -1437,9 +1426,11 @@ util_destroy_gen_mipmap(struct gen_mipmap_state *ctx) { struct pipe_context *pipe = ctx->pipe; - pipe->delete_vs_state(pipe, ctx->vs); - pipe->delete_fs_state(pipe, ctx->fs2d); pipe->delete_fs_state(pipe, ctx->fsCube); + pipe->delete_fs_state(pipe, ctx->fs3d); + pipe->delete_fs_state(pipe, ctx->fs2d); + pipe->delete_fs_state(pipe, ctx->fs1d); + pipe->delete_vs_state(pipe, ctx->vs); pipe_resource_reference(&ctx->vbuf, NULL); @@ -1478,9 +1469,8 @@ util_gen_mipmap(struct gen_mipmap_state *ctx, struct pipe_screen *screen = pipe->screen; struct pipe_framebuffer_state fb; struct pipe_resource *pt = psv->texture; - void *fs = (pt->target == PIPE_TEXTURE_CUBE) ? ctx->fsCube : ctx->fs2d; + void *fs; uint dstLevel; - uint zslice = 0; uint offset; /* The texture object should have room for the levels which we're @@ -1494,8 +1484,28 @@ util_gen_mipmap(struct gen_mipmap_state *ctx, assert(filter == PIPE_TEX_FILTER_LINEAR || filter == PIPE_TEX_FILTER_NEAREST); + switch (pt->target) { + case PIPE_TEXTURE_1D: + fs = ctx->fs1d; + break; + case PIPE_TEXTURE_2D: + fs = ctx->fs2d; + break; + case PIPE_TEXTURE_3D: + fs = ctx->fs3d; + break; + case PIPE_TEXTURE_CUBE: + fs = ctx->fsCube; + break; + case PIPE_TEXTURE_1D_ARRAY: + case PIPE_TEXTURE_2D_ARRAY: + default: + assert(0); + fs = ctx->fs2d; + } + /* check if we can render in the texture's format */ - if (!screen->is_format_supported(screen, psv->format, PIPE_TEXTURE_2D, + if (!screen->is_format_supported(screen, psv->format, pt->target, pt->nr_samples, PIPE_BIND_RENDER_TARGET, 0)) { fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel); return; @@ -1539,60 +1549,85 @@ util_gen_mipmap(struct gen_mipmap_state *ctx, for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { const uint srcLevel = dstLevel - 1; struct pipe_viewport_state vp; - - struct pipe_surface *surf = - screen->get_tex_surface(screen, pt, face, dstLevel, zslice, - PIPE_BIND_RENDER_TARGET); - - /* - * Setup framebuffer / dest surface - */ - fb.cbufs[0] = surf; - fb.width = u_minify(pt->width0, dstLevel); - fb.height = u_minify(pt->height0, dstLevel); - cso_set_framebuffer(ctx->cso, &fb); - - /* viewport */ - vp.scale[0] = 0.5f * fb.width; - vp.scale[1] = 0.5f * fb.height; - vp.scale[2] = 1.0f; - vp.scale[3] = 1.0f; - vp.translate[0] = 0.5f * fb.width; - vp.translate[1] = 0.5f * fb.height; - vp.translate[2] = 0.0f; - vp.translate[3] = 0.0f; - cso_set_viewport(ctx->cso, &vp); - - /* - * Setup sampler state - * Note: we should only have to set the min/max LOD clamps to ensure - * we grab texels from the right mipmap level. But some hardware - * has trouble with min clamping so we also set the lod_bias to - * try to work around that. - */ - ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel; - ctx->sampler.lod_bias = (float) srcLevel; - cso_single_sampler(ctx->cso, 0, &ctx->sampler); - cso_single_sampler_done(ctx->cso); - - cso_set_fragment_sampler_views(ctx->cso, 1, &psv); - - /* quad coords in clip coords */ - offset = set_vertex_data(ctx, - pt->target, - face); - - util_draw_vertex_buffer(ctx->pipe, - ctx->vbuf, - offset, - PIPE_PRIM_TRIANGLE_FAN, - 4, /* verts */ - 2); /* attribs/vert */ - - pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); - - /* need to signal that the texture has changed _after_ rendering to it */ - pipe_surface_reference( &surf, NULL ); + unsigned nr_layers, layer, i; + float rcoord = 0.0f; + + if (pt->target == PIPE_TEXTURE_3D) + nr_layers = u_minify(pt->depth0, dstLevel); + else + nr_layers = 1; + + for (i = 0; i < nr_layers; i++) { + struct pipe_surface *surf, surf_templ; + if (pt->target == PIPE_TEXTURE_3D) { + /* in theory with geom shaders and driver with full layer support + could do that in one go. */ + layer = i; + offset = 1.0f / (float)(nr_layers * 2); + /* XXX hmm really? */ + rcoord = (float)layer / (float)nr_layers + 1.0f / (float)(nr_layers * 2); + } + else + layer = face; + + memset(&surf_templ, 0, sizeof(surf_templ)); + u_surface_default_template(&surf_templ, pt, PIPE_BIND_RENDER_TARGET); + surf_templ.u.tex.level = dstLevel; + surf_templ.u.tex.first_layer = layer; + surf_templ.u.tex.last_layer = layer; + surf = pipe->create_surface(pipe, pt, &surf_templ); + + /* + * Setup framebuffer / dest surface + */ + fb.cbufs[0] = surf; + fb.width = u_minify(pt->width0, dstLevel); + fb.height = u_minify(pt->height0, dstLevel); + cso_set_framebuffer(ctx->cso, &fb); + + /* viewport */ + vp.scale[0] = 0.5f * fb.width; + vp.scale[1] = 0.5f * fb.height; + vp.scale[2] = 1.0f; + vp.scale[3] = 1.0f; + vp.translate[0] = 0.5f * fb.width; + vp.translate[1] = 0.5f * fb.height; + vp.translate[2] = 0.0f; + vp.translate[3] = 0.0f; + cso_set_viewport(ctx->cso, &vp); + + /* + * Setup sampler state + * Note: we should only have to set the min/max LOD clamps to ensure + * we grab texels from the right mipmap level. But some hardware + * has trouble with min clamping so we also set the lod_bias to + * try to work around that. + */ + ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel; + ctx->sampler.lod_bias = (float) srcLevel; + cso_single_sampler(ctx->cso, 0, &ctx->sampler); + cso_single_sampler_done(ctx->cso); + + cso_set_fragment_sampler_views(ctx->cso, 1, &psv); + + /* quad coords in clip coords */ + offset = set_vertex_data(ctx, + pt->target, + face, + rcoord); + + util_draw_vertex_buffer(ctx->pipe, + ctx->vbuf, + offset, + PIPE_PRIM_TRIANGLE_FAN, + 4, /* verts */ + 2); /* attribs/vert */ + + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); + + /* need to signal that the texture has changed _after_ rendering to it */ + pipe_surface_reference( &surf, NULL ); + } } /* restore state we changed */ diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.h b/src/gallium/auxiliary/util/u_gen_mipmap.h index a7502b9982..a10b6a4aba 100644 --- a/src/gallium/auxiliary/util/u_gen_mipmap.h +++ b/src/gallium/auxiliary/util/u_gen_mipmap.h @@ -60,7 +60,7 @@ util_gen_mipmap_flush( struct gen_mipmap_state *ctx ); extern void util_gen_mipmap(struct gen_mipmap_state *ctx, struct pipe_sampler_view *psv, - uint face, uint baseLevel, uint lastLevel, uint filter); + uint layer, uint baseLevel, uint lastLevel, uint filter); #ifdef __cplusplus diff --git a/src/gallium/auxiliary/util/u_index_modify.c b/src/gallium/auxiliary/util/u_index_modify.c index 65b079ed53..f2c9db3caf 100644 --- a/src/gallium/auxiliary/util/u_index_modify.c +++ b/src/gallium/auxiliary/util/u_index_modify.c @@ -24,26 +24,70 @@ #include "util/u_index_modify.h" #include "util/u_inlines.h" +/* Ubyte indices. */ + +void util_shorten_ubyte_elts_to_userptr(struct pipe_context *context, + struct pipe_resource *elts, + int index_bias, + unsigned start, + unsigned count, + void *out) +{ + struct pipe_transfer *src_transfer; + unsigned char *in_map; + unsigned short *out_map = out; + unsigned i; + + in_map = pipe_buffer_map(context, elts, PIPE_TRANSFER_READ, &src_transfer); + in_map += start; + + for (i = 0; i < count; i++) { + *out_map = (unsigned short)(*in_map + index_bias); + in_map++; + out_map++; + } + + pipe_buffer_unmap(context, src_transfer); +} + void util_shorten_ubyte_elts(struct pipe_context *context, struct pipe_resource **elts, int index_bias, unsigned start, unsigned count) { - struct pipe_screen* screen = context->screen; struct pipe_resource* new_elts; - unsigned char *in_map; unsigned short *out_map; - struct pipe_transfer *src_transfer, *dst_transfer; - unsigned i; + struct pipe_transfer *dst_transfer; - new_elts = pipe_buffer_create(screen, + new_elts = pipe_buffer_create(context->screen, PIPE_BIND_INDEX_BUFFER, 2 * count); - in_map = pipe_buffer_map(context, *elts, PIPE_TRANSFER_READ, &src_transfer); - out_map = pipe_buffer_map(context, new_elts, PIPE_TRANSFER_WRITE, &dst_transfer); + out_map = pipe_buffer_map(context, new_elts, PIPE_TRANSFER_WRITE, + &dst_transfer); + util_shorten_ubyte_elts_to_userptr(context, *elts, index_bias, + start, count, out_map); + pipe_buffer_unmap(context, dst_transfer); + + *elts = new_elts; +} + + +/* Ushort indices. */ +void util_rebuild_ushort_elts_to_userptr(struct pipe_context *context, + struct pipe_resource *elts, + int index_bias, + unsigned start, unsigned count, + void *out) +{ + struct pipe_transfer *in_transfer = NULL; + unsigned short *in_map; + unsigned short *out_map = out; + unsigned i; + + in_map = pipe_buffer_map(context, elts, PIPE_TRANSFER_READ, &in_transfer); in_map += start; for (i = 0; i < count; i++) { @@ -52,10 +96,7 @@ void util_shorten_ubyte_elts(struct pipe_context *context, out_map++; } - pipe_buffer_unmap(context, *elts, src_transfer); - pipe_buffer_unmap(context, new_elts, dst_transfer); - - *elts = new_elts; + pipe_buffer_unmap(context, in_transfer); } void util_rebuild_ushort_elts(struct pipe_context *context, @@ -63,33 +104,47 @@ void util_rebuild_ushort_elts(struct pipe_context *context, int index_bias, unsigned start, unsigned count) { - struct pipe_transfer *in_transfer = NULL; struct pipe_transfer *out_transfer = NULL; struct pipe_resource *new_elts; - unsigned short *in_map; unsigned short *out_map; - unsigned i; new_elts = pipe_buffer_create(context->screen, PIPE_BIND_INDEX_BUFFER, 2 * count); - in_map = pipe_buffer_map(context, *elts, - PIPE_TRANSFER_READ, &in_transfer); out_map = pipe_buffer_map(context, new_elts, PIPE_TRANSFER_WRITE, &out_transfer); + util_rebuild_ushort_elts_to_userptr(context, *elts, index_bias, + start, count, out_map); + pipe_buffer_unmap(context, out_transfer); + + *elts = new_elts; +} + + +/* Uint indices. */ +void util_rebuild_uint_elts_to_userptr(struct pipe_context *context, + struct pipe_resource *elts, + int index_bias, + unsigned start, unsigned count, + void *out) +{ + struct pipe_transfer *in_transfer = NULL; + unsigned int *in_map; + unsigned int *out_map = out; + unsigned i; + + in_map = pipe_buffer_map(context, elts, PIPE_TRANSFER_READ, &in_transfer); in_map += start; + for (i = 0; i < count; i++) { - *out_map = (unsigned short)(*in_map + index_bias); + *out_map = (unsigned int)(*in_map + index_bias); in_map++; out_map++; } - pipe_buffer_unmap(context, *elts, in_transfer); - pipe_buffer_unmap(context, new_elts, out_transfer); - - *elts = new_elts; + pipe_buffer_unmap(context, in_transfer); } void util_rebuild_uint_elts(struct pipe_context *context, @@ -97,31 +152,19 @@ void util_rebuild_uint_elts(struct pipe_context *context, int index_bias, unsigned start, unsigned count) { - struct pipe_transfer *in_transfer = NULL; struct pipe_transfer *out_transfer = NULL; struct pipe_resource *new_elts; - unsigned int *in_map; unsigned int *out_map; - unsigned i; new_elts = pipe_buffer_create(context->screen, PIPE_BIND_INDEX_BUFFER, 2 * count); - in_map = pipe_buffer_map(context, *elts, - PIPE_TRANSFER_READ, &in_transfer); out_map = pipe_buffer_map(context, new_elts, PIPE_TRANSFER_WRITE, &out_transfer); - - in_map += start; - for (i = 0; i < count; i++) { - *out_map = (unsigned int)(*in_map + index_bias); - in_map++; - out_map++; - } - - pipe_buffer_unmap(context, *elts, in_transfer); - pipe_buffer_unmap(context, new_elts, out_transfer); + util_rebuild_uint_elts_to_userptr(context, *elts, index_bias, + start, count, out_map); + pipe_buffer_unmap(context, out_transfer); *elts = new_elts; } diff --git a/src/gallium/auxiliary/util/u_index_modify.h b/src/gallium/auxiliary/util/u_index_modify.h index 01a6cae94f..1e9de3dfac 100644 --- a/src/gallium/auxiliary/util/u_index_modify.h +++ b/src/gallium/auxiliary/util/u_index_modify.h @@ -23,19 +23,46 @@ #ifndef UTIL_INDEX_MODIFY_H #define UTIL_INDEX_MODIFY_H +struct pipe_context; +struct pipe_resource; + +void util_shorten_ubyte_elts_to_userptr(struct pipe_context *context, + struct pipe_resource *elts, + int index_bias, + unsigned start, + unsigned count, + void *out); + void util_shorten_ubyte_elts(struct pipe_context *context, struct pipe_resource **elts, int index_bias, unsigned start, unsigned count); + + +void util_rebuild_ushort_elts_to_userptr(struct pipe_context *context, + struct pipe_resource *elts, + int index_bias, + unsigned start, unsigned count, + void *out); + void util_rebuild_ushort_elts(struct pipe_context *context, struct pipe_resource **elts, int index_bias, unsigned start, unsigned count); + + +void util_rebuild_uint_elts_to_userptr(struct pipe_context *context, + struct pipe_resource *elts, + int index_bias, + unsigned start, unsigned count, + void *out); + void util_rebuild_uint_elts(struct pipe_context *context, struct pipe_resource **elts, int index_bias, unsigned start, unsigned count); + #endif diff --git a/src/gallium/auxiliary/util/u_inlines.h b/src/gallium/auxiliary/util/u_inlines.h index 6ed39561fb..b4870bce98 100644 --- a/src/gallium/auxiliary/util/u_inlines.h +++ b/src/gallium/auxiliary/util/u_inlines.h @@ -109,7 +109,7 @@ pipe_surface_reference(struct pipe_surface **ptr, struct pipe_surface *surf) if (pipe_reference_described(&(*ptr)->reference, &surf->reference, (debug_reference_descriptor)debug_describe_surface)) - old_surf->texture->screen->tex_surface_destroy(old_surf); + old_surf->context->surface_destroy(old_surf->context, old_surf); *ptr = surf; } @@ -136,26 +136,28 @@ pipe_sampler_view_reference(struct pipe_sampler_view **ptr, struct pipe_sampler_ } static INLINE void -pipe_surface_reset(struct pipe_surface* ps, struct pipe_resource *pt, - unsigned face, unsigned level, unsigned zslice, unsigned flags) +pipe_surface_reset(struct pipe_context *ctx, struct pipe_surface* ps, + struct pipe_resource *pt, unsigned level, unsigned layer, + unsigned flags) { pipe_resource_reference(&ps->texture, pt); ps->format = pt->format; ps->width = u_minify(pt->width0, level); ps->height = u_minify(pt->height0, level); ps->usage = flags; - ps->face = face; - ps->level = level; - ps->zslice = zslice; + ps->u.tex.level = level; + ps->u.tex.first_layer = ps->u.tex.last_layer = layer; + ps->context = ctx; } static INLINE void -pipe_surface_init(struct pipe_surface* ps, struct pipe_resource *pt, - unsigned face, unsigned level, unsigned zslice, unsigned flags) +pipe_surface_init(struct pipe_context *ctx, struct pipe_surface* ps, + struct pipe_resource *pt, unsigned level, unsigned layer, + unsigned flags) { ps->texture = 0; pipe_reference_init(&ps->reference, 1); - pipe_surface_reset(ps, pt, face, level, zslice, flags); + pipe_surface_reset(ctx, ps, pt, level, layer, flags); } /* @@ -177,6 +179,7 @@ pipe_buffer_create( struct pipe_screen *screen, buffer.width0 = size; buffer.height0 = 1; buffer.depth0 = 1; + buffer.array_size = 1; return screen->resource_create(screen, &buffer); } @@ -202,15 +205,15 @@ pipe_buffer_map_range(struct pipe_context *pipe, assert(offset < buffer->width0); assert(offset + length <= buffer->width0); assert(length); - + u_box_1d(offset, length, &box); *transfer = pipe->get_transfer( pipe, - buffer, - u_subresource(0, 0), - usage, - &box); - + buffer, + 0, + usage, + &box); + if (*transfer == NULL) return NULL; @@ -231,7 +234,7 @@ static INLINE void * pipe_buffer_map(struct pipe_context *pipe, struct pipe_resource *buffer, unsigned usage, - struct pipe_transfer **transfer) + struct pipe_transfer **transfer) { return pipe_buffer_map_range(pipe, buffer, 0, buffer->width0, usage, transfer); } @@ -239,8 +242,7 @@ pipe_buffer_map(struct pipe_context *pipe, static INLINE void pipe_buffer_unmap(struct pipe_context *pipe, - struct pipe_resource *buf, - struct pipe_transfer *transfer) + struct pipe_transfer *transfer) { if (transfer) { pipe->transfer_unmap(pipe, transfer); @@ -250,7 +252,7 @@ pipe_buffer_unmap(struct pipe_context *pipe, static INLINE void pipe_buffer_flush_mapped_range(struct pipe_context *pipe, - struct pipe_transfer *transfer, + struct pipe_transfer *transfer, unsigned offset, unsigned length) { @@ -266,7 +268,7 @@ pipe_buffer_flush_mapped_range(struct pipe_context *pipe, * mapped range. */ transfer_offset = offset - transfer->box.x; - + u_box_1d(transfer_offset, length, &box); pipe->transfer_flush_region(pipe, transfer, &box); @@ -276,7 +278,7 @@ static INLINE void pipe_buffer_write(struct pipe_context *pipe, struct pipe_resource *buf, unsigned offset, - unsigned size, + unsigned size, const void *data) { struct pipe_box box; @@ -284,13 +286,13 @@ pipe_buffer_write(struct pipe_context *pipe, u_box_1d(offset, size, &box); pipe->transfer_inline_write( pipe, - buf, - u_subresource(0,0), - PIPE_TRANSFER_WRITE, - &box, - data, - size, - 0); + buf, + 0, + PIPE_TRANSFER_WRITE, + &box, + data, + size, + 0); } /** @@ -309,21 +311,21 @@ pipe_buffer_write_nooverlap(struct pipe_context *pipe, u_box_1d(offset, size, &box); - pipe->transfer_inline_write(pipe, - buf, - u_subresource(0,0), - (PIPE_TRANSFER_WRITE | - PIPE_TRANSFER_NOOVERWRITE), - &box, - data, - 0, 0); + pipe->transfer_inline_write(pipe, + buf, + 0, + (PIPE_TRANSFER_WRITE | + PIPE_TRANSFER_NOOVERWRITE), + &box, + data, + 0, 0); } static INLINE void pipe_buffer_read(struct pipe_context *pipe, struct pipe_resource *buf, unsigned offset, - unsigned size, + unsigned size, void *data) { struct pipe_transfer *src_transfer; @@ -338,25 +340,24 @@ pipe_buffer_read(struct pipe_context *pipe, if (map) memcpy(data, map + offset, size); - pipe_buffer_unmap(pipe, buf, src_transfer); + pipe_buffer_unmap(pipe, src_transfer); } static INLINE struct pipe_transfer * pipe_get_transfer( struct pipe_context *context, - struct pipe_resource *resource, - unsigned face, unsigned level, - unsigned zslice, - enum pipe_transfer_usage usage, - unsigned x, unsigned y, - unsigned w, unsigned h) + struct pipe_resource *resource, + unsigned level, unsigned layer, + enum pipe_transfer_usage usage, + unsigned x, unsigned y, + unsigned w, unsigned h) { struct pipe_box box; - u_box_2d_zslice( x, y, zslice, w, h, &box ); + u_box_2d_zslice( x, y, layer, w, h, &box ); return context->get_transfer( context, - resource, - u_subresource(face, level), - usage, - &box ); + resource, + level, + usage, + &box ); } static INLINE void * @@ -376,7 +377,7 @@ pipe_transfer_unmap( struct pipe_context *context, static INLINE void pipe_transfer_destroy( struct pipe_context *context, - struct pipe_transfer *transfer ) + struct pipe_transfer *transfer ) { context->transfer_destroy(context, transfer); } @@ -399,6 +400,34 @@ static INLINE boolean util_get_offset( } } +/** + * This function is used to copy an array of pipe_vertex_buffer structures, + * while properly referencing the pipe_vertex_buffer::buffer member. + * + * \sa util_copy_framebuffer_state + */ +static INLINE void util_copy_vertex_buffers(struct pipe_vertex_buffer *dst, + unsigned *dst_count, + const struct pipe_vertex_buffer *src, + unsigned src_count) +{ + unsigned i; + + /* Reference the buffers of 'src' in 'dst'. */ + for (i = 0; i < src_count; i++) { + pipe_resource_reference(&dst[i].buffer, src[i].buffer); + } + /* Unreference the rest of the buffers in 'dst'. */ + for (; i < *dst_count; i++) { + pipe_resource_reference(&dst[i].buffer, NULL); + } + + /* Update the size of 'dst' and copy over the other members + * of pipe_vertex_buffer. */ + *dst_count = src_count; + memcpy(dst, src, src_count * sizeof(struct pipe_vertex_buffer)); +} + #ifdef __cplusplus } #endif diff --git a/src/gallium/auxiliary/util/u_resource.c b/src/gallium/auxiliary/util/u_resource.c index 9e6474b83d..443c9f8067 100644 --- a/src/gallium/auxiliary/util/u_resource.c +++ b/src/gallium/auxiliary/util/u_resource.c @@ -10,85 +10,85 @@ u_resource( struct pipe_resource *res ) } boolean u_resource_get_handle_vtbl(struct pipe_screen *screen, - struct pipe_resource *resource, - struct winsys_handle *handle) + struct pipe_resource *resource, + struct winsys_handle *handle) { struct u_resource *ur = u_resource(resource); return ur->vtbl->resource_get_handle(screen, resource, handle); } void u_resource_destroy_vtbl(struct pipe_screen *screen, - struct pipe_resource *resource) + struct pipe_resource *resource) { struct u_resource *ur = u_resource(resource); ur->vtbl->resource_destroy(screen, resource); } unsigned u_is_resource_referenced_vtbl( struct pipe_context *pipe, - struct pipe_resource *resource, - unsigned face, unsigned level) + struct pipe_resource *resource, + unsigned level, int layer) { struct u_resource *ur = u_resource(resource); - return ur->vtbl->is_resource_referenced(pipe, resource, face, level); + return ur->vtbl->is_resource_referenced(pipe, resource, level, layer); } struct pipe_transfer *u_get_transfer_vtbl(struct pipe_context *context, - struct pipe_resource *resource, - struct pipe_subresource sr, - enum pipe_transfer_usage usage, - const struct pipe_box *box) + struct pipe_resource *resource, + unsigned level, + enum pipe_transfer_usage usage, + const struct pipe_box *box) { struct u_resource *ur = u_resource(resource); - return ur->vtbl->get_transfer(context, resource, sr, usage, box); + return ur->vtbl->get_transfer(context, resource, level, usage, box); } void u_transfer_destroy_vtbl(struct pipe_context *pipe, - struct pipe_transfer *transfer) + struct pipe_transfer *transfer) { struct u_resource *ur = u_resource(transfer->resource); ur->vtbl->transfer_destroy(pipe, transfer); } void *u_transfer_map_vtbl( struct pipe_context *pipe, - struct pipe_transfer *transfer ) + struct pipe_transfer *transfer ) { struct u_resource *ur = u_resource(transfer->resource); return ur->vtbl->transfer_map(pipe, transfer); } void u_transfer_flush_region_vtbl( struct pipe_context *pipe, - struct pipe_transfer *transfer, - const struct pipe_box *box) + struct pipe_transfer *transfer, + const struct pipe_box *box) { struct u_resource *ur = u_resource(transfer->resource); ur->vtbl->transfer_flush_region(pipe, transfer, box); } void u_transfer_unmap_vtbl( struct pipe_context *pipe, - struct pipe_transfer *transfer ) + struct pipe_transfer *transfer ) { struct u_resource *ur = u_resource(transfer->resource); ur->vtbl->transfer_unmap(pipe, transfer); } void u_transfer_inline_write_vtbl( struct pipe_context *pipe, - struct pipe_resource *resource, - struct pipe_subresource sr, - unsigned usage, - const struct pipe_box *box, - const void *data, - unsigned stride, - unsigned slice_stride) + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box, + const void *data, + unsigned stride, + unsigned layer_stride) { struct u_resource *ur = u_resource(resource); - ur->vtbl->transfer_inline_write(pipe, - resource, - sr, - usage, - box, - data, - stride, - slice_stride); + ur->vtbl->transfer_inline_write(pipe, + resource, + level, + usage, + box, + data, + stride, + layer_stride); } diff --git a/src/gallium/auxiliary/util/u_sampler.c b/src/gallium/auxiliary/util/u_sampler.c index e77f562ea2..bb26099b7e 100644 --- a/src/gallium/auxiliary/util/u_sampler.c +++ b/src/gallium/auxiliary/util/u_sampler.c @@ -40,8 +40,11 @@ default_template(struct pipe_sampler_view *view, */ view->format = format; - view->first_level = 0; - view->last_level = texture->last_level; + view->u.tex.first_level = 0; + view->u.tex.last_level = texture->last_level; + view->u.tex.first_layer = 0; + view->u.tex.last_layer = texture->target == PIPE_TEXTURE_3D ? + texture->depth0 - 1 : texture->array_size - 1; view->swizzle_r = PIPE_SWIZZLE_RED; view->swizzle_g = PIPE_SWIZZLE_GREEN; view->swizzle_b = PIPE_SWIZZLE_BLUE; diff --git a/src/gallium/auxiliary/util/u_simple_screen.h b/src/gallium/auxiliary/util/u_simple_screen.h index b52232f025..7139aaabc5 100644 --- a/src/gallium/auxiliary/util/u_simple_screen.h +++ b/src/gallium/auxiliary/util/u_simple_screen.h @@ -57,7 +57,8 @@ struct pipe_winsys * displayed, eg copy fake frontbuffer. */ void (*flush_frontbuffer)( struct pipe_winsys *ws, - struct pipe_surface *surf, + struct pipe_resource *resource, + unsigned level, unsigned layer, void *context_private ); diff --git a/src/gallium/auxiliary/util/u_simple_shaders.c b/src/gallium/auxiliary/util/u_simple_shaders.c index 58ef68377f..b0f2dd8aa2 100644 --- a/src/gallium/auxiliary/util/u_simple_shaders.c +++ b/src/gallium/auxiliary/util/u_simple_shaders.c @@ -204,7 +204,8 @@ util_make_fragment_tex_shader_writedepth(struct pipe_context *pipe, void * util_make_fragment_passthrough_shader(struct pipe_context *pipe) { - return util_make_fragment_clonecolor_shader(pipe, 1); + return util_make_fragment_cloneinput_shader(pipe, 1, TGSI_SEMANTIC_COLOR, + TGSI_INTERPOLATE_PERSPECTIVE); } @@ -212,7 +213,9 @@ util_make_fragment_passthrough_shader(struct pipe_context *pipe) * Make a fragment shader that copies the input color to N output colors. */ void * -util_make_fragment_clonecolor_shader(struct pipe_context *pipe, int num_cbufs) +util_make_fragment_cloneinput_shader(struct pipe_context *pipe, int num_cbufs, + int input_semantic, + int input_interpolate) { struct ureg_program *ureg; struct ureg_src src; @@ -225,8 +228,8 @@ util_make_fragment_clonecolor_shader(struct pipe_context *pipe, int num_cbufs) if (ureg == NULL) return NULL; - src = ureg_DECL_fs_input( ureg, TGSI_SEMANTIC_COLOR, 0, - TGSI_INTERPOLATE_PERSPECTIVE ); + src = ureg_DECL_fs_input( ureg, input_semantic, 0, + input_interpolate ); for (i = 0; i < num_cbufs; i++) dst[i] = ureg_DECL_output( ureg, TGSI_SEMANTIC_COLOR, i ); diff --git a/src/gallium/auxiliary/util/u_simple_shaders.h b/src/gallium/auxiliary/util/u_simple_shaders.h index 4aa34bc475..1bfec183e3 100644 --- a/src/gallium/auxiliary/util/u_simple_shaders.h +++ b/src/gallium/auxiliary/util/u_simple_shaders.h @@ -71,7 +71,9 @@ util_make_fragment_passthrough_shader(struct pipe_context *pipe); extern void * -util_make_fragment_clonecolor_shader(struct pipe_context *pipe, int num_cbufs); +util_make_fragment_cloneinput_shader(struct pipe_context *pipe, int num_cbufs, + int input_semantic, + int input_interpolate); #ifdef __cplusplus } diff --git a/src/gallium/auxiliary/util/u_mempool.c b/src/gallium/auxiliary/util/u_slab.c index 1f336b39a1..21bf2d735a 100644 --- a/src/gallium/auxiliary/util/u_mempool.c +++ b/src/gallium/auxiliary/util/u_slab.c @@ -20,7 +20,7 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ -#include "util/u_mempool.h" +#include "util/u_slab.h" #include "util/u_math.h" #include "util/u_memory.h" @@ -28,13 +28,13 @@ #include <stdio.h> -#define UTIL_MEMPOOL_MAGIC 0xcafe4321 +#define UTIL_SLAB_MAGIC 0xcafe4321 /* The block is either allocated memory or free space. */ -struct util_mempool_block { +struct util_slab_block { /* The header. */ /* The first next free block. */ - struct util_mempool_block *next_free; + struct util_slab_block *next_free; intptr_t magic; @@ -42,19 +42,19 @@ struct util_mempool_block { * The allocated size is always larger than this structure. */ }; -static struct util_mempool_block * -util_mempool_get_block(struct util_mempool *pool, - struct util_mempool_page *page, unsigned index) +static struct util_slab_block * +util_slab_get_block(struct util_slab_mempool *pool, + struct util_slab_page *page, unsigned index) { - return (struct util_mempool_block*) - ((uint8_t*)page + sizeof(struct util_mempool_page) + + return (struct util_slab_block*) + ((uint8_t*)page + sizeof(struct util_slab_page) + (pool->block_size * index)); } -static void util_mempool_add_new_page(struct util_mempool *pool) +static void util_slab_add_new_page(struct util_slab_mempool *pool) { - struct util_mempool_page *page; - struct util_mempool_block *block; + struct util_slab_page *page; + struct util_slab_block *block; int i; page = MALLOC(pool->page_size); @@ -62,15 +62,15 @@ static void util_mempool_add_new_page(struct util_mempool *pool) /* Mark all blocks as free. */ for (i = 0; i < pool->num_blocks-1; i++) { - block = util_mempool_get_block(pool, page, i); - block->next_free = util_mempool_get_block(pool, page, i+1); - block->magic = UTIL_MEMPOOL_MAGIC; + block = util_slab_get_block(pool, page, i); + block->next_free = util_slab_get_block(pool, page, i+1); + block->magic = UTIL_SLAB_MAGIC; } - block = util_mempool_get_block(pool, page, pool->num_blocks-1); + block = util_slab_get_block(pool, page, pool->num_blocks-1); block->next_free = pool->first_free; - block->magic = UTIL_MEMPOOL_MAGIC; - pool->first_free = util_mempool_get_block(pool, page, 0); + block->magic = UTIL_SLAB_MAGIC; + pool->first_free = util_slab_get_block(pool, page, 0); pool->num_pages++; #if 0 @@ -78,74 +78,74 @@ static void util_mempool_add_new_page(struct util_mempool *pool) #endif } -static void *util_mempool_malloc_st(struct util_mempool *pool) +static void *util_slab_alloc_st(struct util_slab_mempool *pool) { - struct util_mempool_block *block; + struct util_slab_block *block; if (!pool->first_free) - util_mempool_add_new_page(pool); + util_slab_add_new_page(pool); block = pool->first_free; - assert(block->magic == UTIL_MEMPOOL_MAGIC); + assert(block->magic == UTIL_SLAB_MAGIC); pool->first_free = block->next_free; - return (uint8_t*)block + sizeof(struct util_mempool_block); + return (uint8_t*)block + sizeof(struct util_slab_block); } -static void util_mempool_free_st(struct util_mempool *pool, void *ptr) +static void util_slab_free_st(struct util_slab_mempool *pool, void *ptr) { - struct util_mempool_block *block = - (struct util_mempool_block*) - ((uint8_t*)ptr - sizeof(struct util_mempool_block)); + struct util_slab_block *block = + (struct util_slab_block*) + ((uint8_t*)ptr - sizeof(struct util_slab_block)); - assert(block->magic == UTIL_MEMPOOL_MAGIC); + assert(block->magic == UTIL_SLAB_MAGIC); block->next_free = pool->first_free; pool->first_free = block; } -static void *util_mempool_malloc_mt(struct util_mempool *pool) +static void *util_slab_alloc_mt(struct util_slab_mempool *pool) { void *mem; pipe_mutex_lock(pool->mutex); - mem = util_mempool_malloc_st(pool); + mem = util_slab_alloc_st(pool); pipe_mutex_unlock(pool->mutex); return mem; } -static void util_mempool_free_mt(struct util_mempool *pool, void *ptr) +static void util_slab_free_mt(struct util_slab_mempool *pool, void *ptr) { pipe_mutex_lock(pool->mutex); - util_mempool_free_st(pool, ptr); + util_slab_free_st(pool, ptr); pipe_mutex_unlock(pool->mutex); } -void util_mempool_set_thread_safety(struct util_mempool *pool, - enum util_mempool_threading threading) +void util_slab_set_thread_safety(struct util_slab_mempool *pool, + enum util_slab_threading threading) { pool->threading = threading; if (threading) { - pool->malloc = util_mempool_malloc_mt; - pool->free = util_mempool_free_mt; + pool->alloc = util_slab_alloc_mt; + pool->free = util_slab_free_mt; } else { - pool->malloc = util_mempool_malloc_st; - pool->free = util_mempool_free_st; + pool->alloc = util_slab_alloc_st; + pool->free = util_slab_free_st; } } -void util_mempool_create(struct util_mempool *pool, - unsigned item_size, - unsigned num_blocks, - enum util_mempool_threading threading) +void util_slab_create(struct util_slab_mempool *pool, + unsigned item_size, + unsigned num_blocks, + enum util_slab_threading threading) { item_size = align(item_size, sizeof(intptr_t)); pool->num_pages = 0; pool->num_blocks = num_blocks; - pool->block_size = sizeof(struct util_mempool_block) + item_size; + pool->block_size = sizeof(struct util_slab_block) + item_size; pool->block_size = align(pool->block_size, sizeof(intptr_t)); - pool->page_size = sizeof(struct util_mempool_page) + + pool->page_size = sizeof(struct util_slab_page) + num_blocks * pool->block_size; pool->first_free = NULL; @@ -153,12 +153,12 @@ void util_mempool_create(struct util_mempool *pool, pipe_mutex_init(pool->mutex); - util_mempool_set_thread_safety(pool, threading); + util_slab_set_thread_safety(pool, threading); } -void util_mempool_destroy(struct util_mempool *pool) +void util_slab_destroy(struct util_slab_mempool *pool) { - struct util_mempool_page *page, *temp; + struct util_slab_page *page, *temp; foreach_s(page, temp, &pool->list) { remove_from_list(page); diff --git a/src/gallium/auxiliary/util/u_mempool.h b/src/gallium/auxiliary/util/u_slab.h index a5b5d6a9b7..6b9718d08a 100644 --- a/src/gallium/auxiliary/util/u_mempool.h +++ b/src/gallium/auxiliary/util/u_slab.h @@ -22,66 +22,66 @@ /** * @file - * Simple memory pool for equally sized memory allocations. - * util_mempool_malloc and util_mempool_free are in O(1). + * Simple slab allocator for equally sized memory allocations. + * util_slab_alloc and util_slab_free have time complexity in O(1). * * Good for allocations which have very low lifetime and are allocated - * and freed very often. Use a profiler first! + * and freed very often. Use a profiler first to know if it's worth using it! * * Candidates: get_transfer, user_buffer_create * * @author Marek Olšák */ -#ifndef U_MEMPOOL_H -#define U_MEMPOOL_H +#ifndef U_SLAB_H +#define U_SLAB_H #include "os/os_thread.h" -enum util_mempool_threading { - UTIL_MEMPOOL_SINGLETHREADED = FALSE, - UTIL_MEMPOOL_MULTITHREADED = TRUE +enum util_slab_threading { + UTIL_SLAB_SINGLETHREADED = FALSE, + UTIL_SLAB_MULTITHREADED = TRUE }; /* The page is an array of blocks (allocations). */ -struct util_mempool_page { +struct util_slab_page { /* The header (linked-list pointers). */ - struct util_mempool_page *prev, *next; + struct util_slab_page *prev, *next; /* Memory after the last member is dedicated to the page itself. * The allocated size is always larger than this structure. */ }; -struct util_mempool { +struct util_slab_mempool { /* Public members. */ - void *(*malloc)(struct util_mempool *pool); - void (*free)(struct util_mempool *pool, void *ptr); + void *(*alloc)(struct util_slab_mempool *pool); + void (*free)(struct util_slab_mempool *pool, void *ptr); /* Private members. */ - struct util_mempool_block *first_free; + struct util_slab_block *first_free; - struct util_mempool_page list; + struct util_slab_page list; unsigned block_size; unsigned page_size; unsigned num_blocks; unsigned num_pages; - enum util_mempool_threading threading; + enum util_slab_threading threading; pipe_mutex mutex; }; -void util_mempool_create(struct util_mempool *pool, - unsigned item_size, - unsigned num_blocks, - enum util_mempool_threading threading); +void util_slab_create(struct util_slab_mempool *pool, + unsigned item_size, + unsigned num_blocks, + enum util_slab_threading threading); -void util_mempool_destroy(struct util_mempool *pool); +void util_slab_destroy(struct util_slab_mempool *pool); -void util_mempool_set_thread_safety(struct util_mempool *pool, - enum util_mempool_threading threading); +void util_slab_set_thread_safety(struct util_slab_mempool *pool, + enum util_slab_threading threading); -#define util_mempool_malloc(pool) (pool)->malloc(pool) -#define util_mempool_free(pool, ptr) (pool)->free(pool, ptr) +#define util_slab_alloc(pool) (pool)->alloc(pool) +#define util_slab_free(pool, ptr) (pool)->free(pool, ptr) #endif diff --git a/src/gallium/auxiliary/util/u_staging.c b/src/gallium/auxiliary/util/u_staging.c index c5d68f8df8..b6bf241a22 100644 --- a/src/gallium/auxiliary/util/u_staging.c +++ b/src/gallium/auxiliary/util/u_staging.c @@ -41,6 +41,7 @@ util_staging_resource_template(struct pipe_resource *pt, unsigned width, unsigne template->width0 = width; template->height0 = height; template->depth0 = depth; + template->array_size = 1; template->last_level = 0; template->nr_samples = pt->nr_samples; template->bind = 0; @@ -51,7 +52,7 @@ util_staging_resource_template(struct pipe_resource *pt, unsigned width, unsigne struct util_staging_transfer * util_staging_transfer_init(struct pipe_context *pipe, struct pipe_resource *pt, - struct pipe_subresource sr, + unsigned level, unsigned usage, const struct pipe_box *box, bool direct, struct util_staging_transfer *tx) @@ -61,7 +62,7 @@ util_staging_transfer_init(struct pipe_context *pipe, struct pipe_resource staging_resource_template; pipe_resource_reference(&tx->base.resource, pt); - tx->base.sr = sr; + tx->base.level = level; tx->base.usage = usage; tx->base.box = *box; @@ -82,12 +83,20 @@ util_staging_transfer_init(struct pipe_context *pipe, if (usage & PIPE_TRANSFER_READ) { - struct pipe_subresource dstsr; + /* XXX this looks wrong dst is always the same but looping over src z? */ unsigned zi; - dstsr.face = 0; - dstsr.level = 0; - for(zi = 0; zi < box->depth; ++zi) - pipe->resource_copy_region(pipe, tx->staging_resource, dstsr, 0, 0, 0, tx->base.resource, sr, box->x, box->y, box->z + zi, box->width, box->height); + struct pipe_box sbox; + sbox.x = box->x; + sbox.y = box->y; + sbox.z = box->z; + sbox.width = box->width; + sbox.height = box->height; + sbox.depth = 1; + for(zi = 0; zi < box->depth; ++zi) { + sbox.z = sbox.z + zi; + pipe->resource_copy_region(pipe, tx->staging_resource, 0, 0, 0, 0, + tx->base.resource, level, &sbox); + } } return tx; @@ -101,12 +110,18 @@ util_staging_transfer_destroy(struct pipe_context *pipe, struct pipe_transfer *p if (tx->staging_resource != tx->base.resource) { if(tx->base.usage & PIPE_TRANSFER_WRITE) { - struct pipe_subresource srcsr; + /* XXX this looks wrong src is always the same but looping over dst z? */ unsigned zi; - srcsr.face = 0; - srcsr.level = 0; + struct pipe_box sbox; + sbox.x = 0; + sbox.y = 0; + sbox.z = 0; + sbox.width = tx->base.box.width; + sbox.height = tx->base.box.height; + sbox.depth = 1; for(zi = 0; zi < tx->base.box.depth; ++zi) - pipe->resource_copy_region(pipe, tx->base.resource, tx->base.sr, tx->base.box.x, tx->base.box.y, tx->base.box.z + zi, tx->staging_resource, srcsr, 0, 0, 0, tx->base.box.width, tx->base.box.height); + pipe->resource_copy_region(pipe, tx->base.resource, tx->base.level, tx->base.box.x, tx->base.box.y, tx->base.box.z + zi, + tx->staging_resource, 0, &sbox); } pipe_resource_reference(&tx->staging_resource, NULL); diff --git a/src/gallium/auxiliary/util/u_staging.h b/src/gallium/auxiliary/util/u_staging.h index 1aab78cc88..49839d2543 100644 --- a/src/gallium/auxiliary/util/u_staging.h +++ b/src/gallium/auxiliary/util/u_staging.h @@ -52,7 +52,7 @@ struct util_staging_transfer { struct util_staging_transfer * util_staging_transfer_init(struct pipe_context *pipe, struct pipe_resource *pt, - struct pipe_subresource sr, + unsigned level, unsigned usage, const struct pipe_box *box, bool direct, struct util_staging_transfer *tx); diff --git a/src/gallium/auxiliary/util/u_surface.c b/src/gallium/auxiliary/util/u_surface.c index f78b6838a7..4eddd3f519 100644 --- a/src/gallium/auxiliary/util/u_surface.c +++ b/src/gallium/auxiliary/util/u_surface.c @@ -42,6 +42,18 @@ #include "util/u_surface.h" #include "util/u_pack_color.h" +void +u_surface_default_template(struct pipe_surface *view, + const struct pipe_resource *texture, + unsigned bind) +{ + view->format = texture->format; + view->u.tex.level = 0; + view->u.tex.first_layer = 0; + view->u.tex.last_layer = 0; + /* XXX should filter out all non-rt/ds bind flags ? */ + view->usage = bind; +} /** * Helper to quickly create an RGBA rendering surface of a certain size. @@ -50,9 +62,9 @@ * \return TRUE for success, FALSE if failure */ boolean -util_create_rgba_surface(struct pipe_screen *screen, +util_create_rgba_surface(struct pipe_context *pipe, uint width, uint height, - uint bind, + uint bind, struct pipe_resource **textureOut, struct pipe_surface **surfaceOut) { @@ -65,6 +77,8 @@ util_create_rgba_surface(struct pipe_screen *screen, const uint target = PIPE_TEXTURE_2D; enum pipe_format format = PIPE_FORMAT_NONE; struct pipe_resource templ; + struct pipe_surface surf_templ; + struct pipe_screen *screen = pipe->screen; uint i; /* Choose surface format */ @@ -86,17 +100,20 @@ util_create_rgba_surface(struct pipe_screen *screen, templ.width0 = width; templ.height0 = height; templ.depth0 = 1; + templ.array_size = 1; templ.bind = bind; *textureOut = screen->resource_create(screen, &templ); if (!*textureOut) return FALSE; + /* create surface */ + memset(&surf_templ, 0, sizeof(surf_templ)); + u_surface_default_template(&surf_templ, *textureOut, bind); /* create surface / view into texture */ - *surfaceOut = screen->get_tex_surface(screen, - *textureOut, - 0, 0, 0, - bind); + *surfaceOut = pipe->create_surface(pipe, + *textureOut, + &surf_templ); if (!*surfaceOut) { pipe_resource_reference(textureOut, NULL); return FALSE; @@ -126,17 +143,18 @@ util_destroy_rgba_surface(struct pipe_resource *texture, void util_resource_copy_region(struct pipe_context *pipe, struct pipe_resource *dst, - struct pipe_subresource subdst, + unsigned dst_level, unsigned dst_x, unsigned dst_y, unsigned dst_z, struct pipe_resource *src, - struct pipe_subresource subsrc, - unsigned src_x, unsigned src_y, unsigned src_z, - unsigned w, unsigned h) + unsigned src_level, + const struct pipe_box *src_box) { struct pipe_transfer *src_trans, *dst_trans; void *dst_map; const void *src_map; enum pipe_format src_format, dst_format; + unsigned w = src_box->width; + unsigned h = src_box->height; assert(src && dst); if (!src || !dst) @@ -146,20 +164,18 @@ util_resource_copy_region(struct pipe_context *pipe, dst_format = dst->format; src_trans = pipe_get_transfer(pipe, - src, - subsrc.face, - subsrc.level, - src_z, - PIPE_TRANSFER_READ, - src_x, src_y, w, h); + src, + src_level, + src_box->z, + PIPE_TRANSFER_READ, + src_box->x, src_box->y, w, h); dst_trans = pipe_get_transfer(pipe, - dst, - subdst.face, - subdst.level, - src_z, - PIPE_TRANSFER_WRITE, - dst_x, dst_y, w, h); + dst, + dst_level, + dst_z, + PIPE_TRANSFER_WRITE, + dst_x, dst_y, w, h); assert(util_format_get_blocksize(dst_format) == util_format_get_blocksize(src_format)); assert(util_format_get_blockwidth(dst_format) == util_format_get_blockwidth(src_format)); @@ -216,14 +232,13 @@ util_clear_render_target(struct pipe_context *pipe, assert(dst->texture); if (!dst->texture) return; - + /* XXX: should handle multiple layers */ dst_trans = pipe_get_transfer(pipe, - dst->texture, - dst->face, - dst->level, - dst->zslice, - PIPE_TRANSFER_WRITE, - dstx, dsty, width, height); + dst->texture, + dst->u.tex.level, + dst->u.tex.first_layer, + PIPE_TRANSFER_WRITE, + dstx, dsty, width, height); dst_map = pipe->transfer_map(pipe, dst_trans); @@ -271,9 +286,8 @@ util_clear_depth_stencil(struct pipe_context *pipe, return; dst_trans = pipe_get_transfer(pipe, dst->texture, - dst->face, - dst->level, - dst->zslice, + dst->u.tex.level, + dst->u.tex.first_layer, (need_rmw ? PIPE_TRANSFER_READ_WRITE : PIPE_TRANSFER_WRITE), dstx, dsty, width, height); diff --git a/src/gallium/auxiliary/util/u_surface.h b/src/gallium/auxiliary/util/u_surface.h index 6cd12af3a8..6a7cc82b05 100644 --- a/src/gallium/auxiliary/util/u_surface.h +++ b/src/gallium/auxiliary/util/u_surface.h @@ -33,8 +33,18 @@ #include "pipe/p_state.h" +#ifdef __cplusplus +extern "C" { +#endif + + +extern void +u_surface_default_template(struct pipe_surface *view, + const struct pipe_resource *texture, + unsigned bind); + extern boolean -util_create_rgba_surface(struct pipe_screen *screen, +util_create_rgba_surface(struct pipe_context *ctx, uint width, uint height, uint bind, struct pipe_resource **textureOut, struct pipe_surface **surfaceOut); @@ -49,12 +59,11 @@ util_destroy_rgba_surface(struct pipe_resource *texture, extern void util_resource_copy_region(struct pipe_context *pipe, struct pipe_resource *dst, - struct pipe_subresource subdst, + unsigned dst_level, unsigned dst_x, unsigned dst_y, unsigned dst_z, struct pipe_resource *src, - struct pipe_subresource subsrc, - unsigned src_x, unsigned src_y, unsigned src_z, - unsigned w, unsigned h); + unsigned src_level, + const struct pipe_box *src_box); extern void util_clear_render_target(struct pipe_context *pipe, @@ -73,4 +82,9 @@ util_clear_depth_stencil(struct pipe_context *pipe, unsigned width, unsigned height); +#ifdef __cplusplus +} +#endif + + #endif /* U_SURFACE_H */ diff --git a/src/gallium/auxiliary/util/u_surfaces.c b/src/gallium/auxiliary/util/u_surfaces.c index 404e121995..b0cfec2a82 100644 --- a/src/gallium/auxiliary/util/u_surfaces.c +++ b/src/gallium/auxiliary/util/u_surfaces.c @@ -29,8 +29,11 @@ #include "util/u_inlines.h" #include "util/u_memory.h" -struct pipe_surface * -util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, struct pipe_screen *pscreen, struct pipe_resource *pt, unsigned face, unsigned level, unsigned zslice, unsigned flags) +boolean +util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, + struct pipe_context *ctx, struct pipe_resource *pt, + unsigned level, unsigned layer, unsigned flags, + struct pipe_surface **res) { struct pipe_surface *ps; @@ -39,7 +42,7 @@ util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, str if(!us->u.hash) us->u.hash = cso_hash_create(); - ps = cso_hash_iter_data(cso_hash_find(us->u.hash, ((zslice + face) << 8) | level)); + ps = cso_hash_iter_data(cso_hash_find(us->u.hash, (layer << 8) | level)); } else { @@ -48,25 +51,29 @@ util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, str ps = us->u.array[level]; } - if(ps) + if(ps && ps->context == ctx) { p_atomic_inc(&ps->reference.count); - return ps; + *res = ps; + return FALSE; } ps = (struct pipe_surface *)CALLOC(1, surface_struct_size); if(!ps) - return NULL; + { + *res = NULL; + return FALSE; + } - pipe_surface_init(ps, pt, face, level, zslice, flags); - ps->offset = ~0; + pipe_surface_init(ctx, ps, pt, level, layer, flags); if(pt->target == PIPE_TEXTURE_3D || pt->target == PIPE_TEXTURE_CUBE) - cso_hash_insert(us->u.hash, ((zslice + face) << 8) | level, ps); + cso_hash_insert(us->u.hash, (layer << 8) | level, ps); else us->u.array[level] = ps; - return ps; + *res = ps; + return TRUE; } void @@ -75,10 +82,10 @@ util_surfaces_do_detach(struct util_surfaces *us, struct pipe_surface *ps) struct pipe_resource *pt = ps->texture; if(pt->target == PIPE_TEXTURE_3D || pt->target == PIPE_TEXTURE_CUBE) { /* or 2D array */ - cso_hash_erase(us->u.hash, cso_hash_find(us->u.hash, ((ps->zslice + ps->face) << 8) | ps->level)); + cso_hash_erase(us->u.hash, cso_hash_find(us->u.hash, (ps->u.tex.first_layer << 8) | ps->u.tex.level)); } else - us->u.array[ps->level] = 0; + us->u.array[ps->u.tex.level] = 0; } void diff --git a/src/gallium/auxiliary/util/u_surfaces.h b/src/gallium/auxiliary/util/u_surfaces.h index 17d8a5d3a5..9581feda7c 100644 --- a/src/gallium/auxiliary/util/u_surfaces.h +++ b/src/gallium/auxiliary/util/u_surfaces.h @@ -42,33 +42,42 @@ struct util_surfaces } u; }; -struct pipe_surface *util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, struct pipe_screen *pscreen, struct pipe_resource *pt, unsigned face, unsigned level, unsigned zslice, unsigned flags); +/* Return value indicates if the pipe surface result is new */ +boolean +util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, + struct pipe_context *ctx, struct pipe_resource *pt, + unsigned level, unsigned layer, unsigned flags, + struct pipe_surface **res); /* fast inline path for the very common case */ -static INLINE struct pipe_surface * -util_surfaces_get(struct util_surfaces *us, unsigned surface_struct_size, struct pipe_screen *pscreen, struct pipe_resource *pt, unsigned face, unsigned level, unsigned zslice, unsigned flags) +static INLINE boolean +util_surfaces_get(struct util_surfaces *us, unsigned surface_struct_size, + struct pipe_context *ctx, struct pipe_resource *pt, + unsigned level, unsigned layer, unsigned flags, + struct pipe_surface **res) { if(likely((pt->target == PIPE_TEXTURE_2D || pt->target == PIPE_TEXTURE_RECT) && us->u.array)) { struct pipe_surface *ps = us->u.array[level]; - if(ps) + if(ps && ps->context == ctx) { p_atomic_inc(&ps->reference.count); - return ps; + *res = ps; + return FALSE; } } - return util_surfaces_do_get(us, surface_struct_size, pscreen, pt, face, level, zslice, flags); + return util_surfaces_do_get(us, surface_struct_size, ctx, pt, level, layer, flags, res); } static INLINE struct pipe_surface * -util_surfaces_peek(struct util_surfaces *us, struct pipe_resource *pt, unsigned face, unsigned level, unsigned zslice) +util_surfaces_peek(struct util_surfaces *us, struct pipe_resource *pt, unsigned level, unsigned layer) { if(!us->u.pv) return 0; if(unlikely(pt->target == PIPE_TEXTURE_3D || pt->target == PIPE_TEXTURE_CUBE)) - return cso_hash_iter_data(cso_hash_find(us->u.hash, ((zslice + face) << 8) | level)); + return cso_hash_iter_data(cso_hash_find(us->u.hash, (layer << 8) | level)); else return us->u.array[level]; } @@ -80,7 +89,7 @@ util_surfaces_detach(struct util_surfaces *us, struct pipe_surface *ps) { if(likely(ps->texture->target == PIPE_TEXTURE_2D || ps->texture->target == PIPE_TEXTURE_RECT)) { - us->u.array[ps->level] = 0; + us->u.array[ps->u.tex.level] = 0; return; } diff --git a/src/gallium/auxiliary/util/u_transfer.c b/src/gallium/auxiliary/util/u_transfer.c index 69f6fab950..e2828cfd99 100644 --- a/src/gallium/auxiliary/util/u_transfer.c +++ b/src/gallium/auxiliary/util/u_transfer.c @@ -8,22 +8,24 @@ * pointer. XXX: strides?? */ void u_default_transfer_inline_write( struct pipe_context *pipe, - struct pipe_resource *resource, - struct pipe_subresource sr, - unsigned usage, - const struct pipe_box *box, - const void *data, - unsigned stride, - unsigned slice_stride) + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box, + const void *data, + unsigned stride, + unsigned layer_stride) { struct pipe_transfer *transfer = NULL; uint8_t *map = NULL; - - transfer = pipe->get_transfer(pipe, - resource, - sr, - usage, - box ); + const uint8_t *src_data = data; + unsigned i; + + transfer = pipe->get_transfer(pipe, + resource, + level, + usage, + box ); if (transfer == NULL) goto out; @@ -31,17 +33,19 @@ void u_default_transfer_inline_write( struct pipe_context *pipe, if (map == NULL) goto out; - assert(box->depth == 1); /* XXX: fix me */ - - util_copy_rect(map, - resource->format, - transfer->stride, /* bytes */ - 0, 0, - box->width, - box->height, - data, - stride, /* bytes */ - 0, 0); + for (i = 0; i < box->depth; i++) { + util_copy_rect(map, + resource->format, + transfer->stride, /* bytes */ + 0, 0, + box->width, + box->height, + src_data, + stride, /* bytes */ + 0, 0); + map += transfer->layer_stride; + src_data += layer_stride; + } out: if (map) @@ -53,8 +57,8 @@ out: boolean u_default_resource_get_handle(struct pipe_screen *screen, - struct pipe_resource *resource, - struct winsys_handle *handle) + struct pipe_resource *resource, + struct winsys_handle *handle) { return FALSE; } @@ -62,32 +66,32 @@ boolean u_default_resource_get_handle(struct pipe_screen *screen, void u_default_transfer_flush_region( struct pipe_context *pipe, - struct pipe_transfer *transfer, - const struct pipe_box *box) + struct pipe_transfer *transfer, + const struct pipe_box *box) { /* This is a no-op implementation, nothing to do. */ } unsigned u_default_is_resource_referenced( struct pipe_context *pipe, - struct pipe_resource *resource, - unsigned face, unsigned level) + struct pipe_resource *resource, + unsigned level, int layer) { return 0; } struct pipe_transfer * u_default_get_transfer(struct pipe_context *context, - struct pipe_resource *resource, - struct pipe_subresource sr, - unsigned usage, - const struct pipe_box *box) + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box) { struct pipe_transfer *transfer = CALLOC_STRUCT(pipe_transfer); if (transfer == NULL) return NULL; transfer->resource = resource; - transfer->sr = sr; + transfer->level = level; transfer->usage = usage; transfer->box = *box; @@ -98,12 +102,12 @@ struct pipe_transfer * u_default_get_transfer(struct pipe_context *context, } void u_default_transfer_unmap( struct pipe_context *pipe, - struct pipe_transfer *transfer ) + struct pipe_transfer *transfer ) { } void u_default_transfer_destroy(struct pipe_context *pipe, - struct pipe_transfer *transfer) + struct pipe_transfer *transfer) { FREE(transfer); } diff --git a/src/gallium/auxiliary/util/u_transfer.h b/src/gallium/auxiliary/util/u_transfer.h index e3a38730f2..3412e13c3c 100644 --- a/src/gallium/auxiliary/util/u_transfer.h +++ b/src/gallium/auxiliary/util/u_transfer.h @@ -11,37 +11,37 @@ struct pipe_context; struct winsys_handle; boolean u_default_resource_get_handle(struct pipe_screen *screen, - struct pipe_resource *resource, - struct winsys_handle *handle); + struct pipe_resource *resource, + struct winsys_handle *handle); void u_default_transfer_inline_write( struct pipe_context *pipe, - struct pipe_resource *resource, - struct pipe_subresource sr, - unsigned usage, - const struct pipe_box *box, - const void *data, - unsigned stride, - unsigned slice_stride); + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box, + const void *data, + unsigned stride, + unsigned layer_stride); void u_default_transfer_flush_region( struct pipe_context *pipe, - struct pipe_transfer *transfer, - const struct pipe_box *box); + struct pipe_transfer *transfer, + const struct pipe_box *box); unsigned u_default_is_resource_referenced( struct pipe_context *pipe, - struct pipe_resource *resource, - unsigned face, unsigned level); + struct pipe_resource *resource, + unsigned level, int layer); struct pipe_transfer * u_default_get_transfer(struct pipe_context *context, - struct pipe_resource *resource, - struct pipe_subresource sr, - unsigned usage, - const struct pipe_box *box); + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box); void u_default_transfer_unmap( struct pipe_context *pipe, - struct pipe_transfer *transfer ); + struct pipe_transfer *transfer ); void u_default_transfer_destroy(struct pipe_context *pipe, - struct pipe_transfer *transfer); + struct pipe_transfer *transfer); @@ -51,43 +51,43 @@ void u_default_transfer_destroy(struct pipe_context *pipe, struct u_resource_vtbl { boolean (*resource_get_handle)(struct pipe_screen *, - struct pipe_resource *tex, - struct winsys_handle *handle); + struct pipe_resource *tex, + struct winsys_handle *handle); void (*resource_destroy)(struct pipe_screen *, - struct pipe_resource *pt); + struct pipe_resource *pt); unsigned (*is_resource_referenced)(struct pipe_context *pipe, - struct pipe_resource *texture, - unsigned face, unsigned level); + struct pipe_resource *texture, + unsigned level, int layer); struct pipe_transfer *(*get_transfer)(struct pipe_context *, - struct pipe_resource *resource, - struct pipe_subresource, - unsigned usage, - const struct pipe_box *); + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *); void (*transfer_destroy)(struct pipe_context *, - struct pipe_transfer *); + struct pipe_transfer *); void *(*transfer_map)( struct pipe_context *, - struct pipe_transfer *transfer ); + struct pipe_transfer *transfer ); void (*transfer_flush_region)( struct pipe_context *, - struct pipe_transfer *transfer, - const struct pipe_box *); + struct pipe_transfer *transfer, + const struct pipe_box *); void (*transfer_unmap)( struct pipe_context *, - struct pipe_transfer *transfer ); + struct pipe_transfer *transfer ); void (*transfer_inline_write)( struct pipe_context *pipe, - struct pipe_resource *resource, - struct pipe_subresource sr, - unsigned usage, - const struct pipe_box *box, - const void *data, - unsigned stride, - unsigned slice_stride); + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box, + const void *data, + unsigned stride, + unsigned layer_stride); }; @@ -98,43 +98,43 @@ struct u_resource { boolean u_resource_get_handle_vtbl(struct pipe_screen *screen, - struct pipe_resource *resource, - struct winsys_handle *handle); + struct pipe_resource *resource, + struct winsys_handle *handle); void u_resource_destroy_vtbl(struct pipe_screen *screen, - struct pipe_resource *resource); + struct pipe_resource *resource); unsigned u_is_resource_referenced_vtbl( struct pipe_context *pipe, - struct pipe_resource *resource, - unsigned face, unsigned level); + struct pipe_resource *resource, + unsigned level, int layer); struct pipe_transfer *u_get_transfer_vtbl(struct pipe_context *context, - struct pipe_resource *resource, - struct pipe_subresource sr, - unsigned usage, - const struct pipe_box *box); + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box); void u_transfer_destroy_vtbl(struct pipe_context *pipe, - struct pipe_transfer *transfer); + struct pipe_transfer *transfer); void *u_transfer_map_vtbl( struct pipe_context *pipe, - struct pipe_transfer *transfer ); + struct pipe_transfer *transfer ); void u_transfer_flush_region_vtbl( struct pipe_context *pipe, - struct pipe_transfer *transfer, - const struct pipe_box *box); + struct pipe_transfer *transfer, + const struct pipe_box *box); void u_transfer_unmap_vtbl( struct pipe_context *rm_ctx, - struct pipe_transfer *transfer ); + struct pipe_transfer *transfer ); void u_transfer_inline_write_vtbl( struct pipe_context *rm_ctx, - struct pipe_resource *resource, - struct pipe_subresource sr, - unsigned usage, - const struct pipe_box *box, - const void *data, - unsigned stride, - unsigned slice_stride); + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box, + const void *data, + unsigned stride, + unsigned layer_stride); diff --git a/src/gallium/auxiliary/util/u_upload_mgr.c b/src/gallium/auxiliary/util/u_upload_mgr.c index af229e61a0..3b3d5b418f 100644 --- a/src/gallium/auxiliary/util/u_upload_mgr.c +++ b/src/gallium/auxiliary/util/u_upload_mgr.c @@ -41,22 +41,23 @@ struct u_upload_mgr { struct pipe_context *pipe; - unsigned default_size; - unsigned alignment; - unsigned usage; - - /* The active buffer: - */ - struct pipe_resource *buffer; - unsigned size; - unsigned offset; + unsigned default_size; /* Minimum size of the upload buffer, in bytes. */ + unsigned alignment; /* Alignment of each sub-allocation. */ + unsigned bind; /* Bitmask of PIPE_BIND_* flags. */ + + struct pipe_resource *buffer; /* Upload buffer. */ + struct pipe_transfer *transfer; /* Transfer object for the upload buffer. */ + uint8_t *map; /* Pointer to the mapped upload buffer. */ + unsigned size; /* Actual size of the upload buffer. */ + unsigned offset; /* Aligned offset to the upload buffer, pointing + * at the first unused byte. */ }; struct u_upload_mgr *u_upload_create( struct pipe_context *pipe, unsigned default_size, unsigned alignment, - unsigned usage ) + unsigned bind ) { struct u_upload_mgr *upload = CALLOC_STRUCT( u_upload_mgr ); if (!upload) @@ -65,54 +66,12 @@ struct u_upload_mgr *u_upload_create( struct pipe_context *pipe, upload->pipe = pipe; upload->default_size = default_size; upload->alignment = alignment; - upload->usage = usage; + upload->bind = bind; upload->buffer = NULL; return upload; } -/* Slightly specialized version of buffer_write designed to maximize - * chances of the driver consolidating successive writes into a single - * upload. - * - * dirty_size may be slightly greater than size to cope with - * alignment. We don't want to leave holes between succesively mapped - * regions as that may prevent the driver from consolidating uploads. - * - * Note that the 'data' pointer has probably come from the application - * and we cannot read even a byte past its end without risking - * segfaults, or at least complaints from valgrind.. - */ -static INLINE enum pipe_error -my_buffer_write(struct pipe_context *pipe, - struct pipe_resource *buf, - unsigned offset, unsigned size, unsigned dirty_size, - const void *data) -{ - struct pipe_transfer *transfer = NULL; - uint8_t *map; - - assert(offset < buf->width0); - assert(offset + size <= buf->width0); - assert(dirty_size >= size); - assert(size); - - map = pipe_buffer_map_range(pipe, buf, offset, dirty_size, - PIPE_TRANSFER_WRITE | - PIPE_TRANSFER_FLUSH_EXPLICIT | - PIPE_TRANSFER_DISCARD | - PIPE_TRANSFER_UNSYNCHRONIZED, - &transfer); - if (map == NULL) - return PIPE_ERROR_OUT_OF_MEMORY; - - memcpy(map + offset, data, size); - pipe_buffer_flush_mapped_range(pipe, transfer, offset, dirty_size); - pipe_buffer_unmap(pipe, buf, transfer); - - return PIPE_OK; -} - /* Release old buffer. * * This must usually be called prior to firing the command stream @@ -124,6 +83,11 @@ my_buffer_write(struct pipe_context *pipe, */ void u_upload_flush( struct u_upload_mgr *upload ) { + /* Unmap and unreference the upload buffer. */ + if (upload->transfer) { + pipe_transfer_unmap(upload->pipe, upload->transfer); + upload->transfer = NULL; + } pipe_resource_reference( &upload->buffer, NULL ); upload->size = 0; } @@ -142,7 +106,7 @@ u_upload_alloc_buffer( struct u_upload_mgr *upload, { unsigned size; - /* Release old buffer, if present: + /* Release the old buffer, if present: */ u_upload_flush( upload ); @@ -151,10 +115,14 @@ u_upload_alloc_buffer( struct u_upload_mgr *upload, size = align(MAX2(upload->default_size, min_size), 4096); upload->buffer = pipe_buffer_create( upload->pipe->screen, - upload->usage, + upload->bind, size ); if (upload->buffer == NULL) goto fail; + + /* Map the new buffer. */ + upload->map = pipe_buffer_map(upload->pipe, upload->buffer, + PIPE_TRANSFER_WRITE, &upload->transfer); upload->size = size; @@ -168,38 +136,62 @@ fail: return PIPE_ERROR_OUT_OF_MEMORY; } - -enum pipe_error u_upload_data( struct u_upload_mgr *upload, - unsigned size, - const void *data, - unsigned *out_offset, - struct pipe_resource **outbuf ) +enum pipe_error u_upload_alloc( struct u_upload_mgr *upload, + unsigned min_out_offset, + unsigned size, + unsigned *out_offset, + struct pipe_resource **outbuf, + boolean *flushed, + void **ptr ) { unsigned alloc_size = align( size, upload->alignment ); - enum pipe_error ret = PIPE_OK; + unsigned alloc_offset = align(min_out_offset, upload->alignment); + unsigned offset; - if (upload->offset + alloc_size > upload->size) { - ret = u_upload_alloc_buffer( upload, alloc_size ); + /* Make sure we have enough space in the upload buffer + * for the sub-allocation. */ + if (MAX2(upload->offset, alloc_offset) + alloc_size > upload->size) { + enum pipe_error ret = u_upload_alloc_buffer(upload, + alloc_offset + alloc_size); if (ret) return ret; + + *flushed = TRUE; + } else { + *flushed = FALSE; } - /* Copy the data, using map_range if available: - */ - ret = my_buffer_write( upload->pipe, - upload->buffer, - upload->offset, - size, - alloc_size, - data ); + offset = MAX2(upload->offset, alloc_offset); + + assert(offset < upload->buffer->width0); + assert(offset + size <= upload->buffer->width0); + assert(size); + + /* Emit the return values: */ + *ptr = upload->map + offset; + pipe_resource_reference( outbuf, upload->buffer ); + *out_offset = offset; + + upload->offset = offset + alloc_size; + return PIPE_OK; +} + +enum pipe_error u_upload_data( struct u_upload_mgr *upload, + unsigned min_out_offset, + unsigned size, + const void *data, + unsigned *out_offset, + struct pipe_resource **outbuf, + boolean *flushed ) +{ + uint8_t *ptr; + enum pipe_error ret = u_upload_alloc(upload, min_out_offset, size, + out_offset, outbuf, flushed, + (void**)&ptr); if (ret) return ret; - /* Emit the return values: - */ - pipe_resource_reference( outbuf, upload->buffer ); - *out_offset = upload->offset; - upload->offset += alloc_size; + memcpy(ptr, data, size); return PIPE_OK; } @@ -210,11 +202,13 @@ enum pipe_error u_upload_data( struct u_upload_mgr *upload, * renders or DrawElements calls. */ enum pipe_error u_upload_buffer( struct u_upload_mgr *upload, + unsigned min_out_offset, unsigned offset, unsigned size, struct pipe_resource *inbuf, unsigned *out_offset, - struct pipe_resource **outbuf ) + struct pipe_resource **outbuf, + boolean *flushed ) { enum pipe_error ret = PIPE_OK; struct pipe_transfer *transfer = NULL; @@ -233,17 +227,16 @@ enum pipe_error u_upload_buffer( struct u_upload_mgr *upload, if (0) debug_printf("upload ptr %p ofs %d sz %d\n", map, offset, size); - ret = u_upload_data( upload, + ret = u_upload_data( upload, + min_out_offset, size, map + offset, out_offset, - outbuf ); - if (ret) - goto done; + outbuf, flushed ); done: if (map) - pipe_buffer_unmap( upload->pipe, inbuf, transfer ); + pipe_buffer_unmap( upload->pipe, transfer ); return ret; } diff --git a/src/gallium/auxiliary/util/u_upload_mgr.h b/src/gallium/auxiliary/util/u_upload_mgr.h index de016df02e..c9a2ffeb57 100644 --- a/src/gallium/auxiliary/util/u_upload_mgr.h +++ b/src/gallium/auxiliary/util/u_upload_mgr.h @@ -32,15 +32,28 @@ #ifndef U_UPLOAD_MGR_H #define U_UPLOAD_MGR_H +#include "pipe/p_compiler.h" + struct pipe_context; struct pipe_resource; +/** + * Create the upload manager. + * + * \param pipe Pipe driver. + * \param default_size Minimum size of the upload buffer, in bytes. + * \param alignment Alignment of each suballocation in the upload buffer. + * \param bind Bitmask of PIPE_BIND_* flags. + */ struct u_upload_mgr *u_upload_create( struct pipe_context *pipe, unsigned default_size, unsigned alignment, - unsigned usage ); + unsigned bind ); +/** + * Destroy the upload manager. + */ void u_upload_destroy( struct u_upload_mgr *upload ); /* Unmap and release old buffer. @@ -53,20 +66,55 @@ void u_upload_destroy( struct u_upload_mgr *upload ); */ void u_upload_flush( struct u_upload_mgr *upload ); +/** + * Sub-allocate new memory from the upload buffer. + * + * \param upload Upload manager + * \param min_out_offset Minimum offset that should be returned in out_offset. + * \param size Size of the allocation. + * \param out_offset Pointer to where the new buffer offset will be returned. + * \param outbuf Pointer to where the upload buffer will be returned. + * \param flushed Whether the upload buffer was flushed. + * \param ptr Pointer to the allocated memory that is returned. + */ +enum pipe_error u_upload_alloc( struct u_upload_mgr *upload, + unsigned min_out_offset, + unsigned size, + unsigned *out_offset, + struct pipe_resource **outbuf, + boolean *flushed, + void **ptr ); + +/** + * Allocate and write data to the upload buffer. + * + * Same as u_upload_alloc, but in addition to that, it copies "data" + * to the pointer returned from u_upload_alloc. + */ enum pipe_error u_upload_data( struct u_upload_mgr *upload, + unsigned min_out_offset, unsigned size, const void *data, unsigned *out_offset, - struct pipe_resource **outbuf ); + struct pipe_resource **outbuf, + boolean *flushed ); +/** + * Allocate and copy an input buffer to the upload buffer. + * + * Same as u_upload_data, except that the input data comes from a buffer + * instead of a user pointer. + */ enum pipe_error u_upload_buffer( struct u_upload_mgr *upload, + unsigned min_out_offset, unsigned offset, unsigned size, struct pipe_resource *inbuf, unsigned *out_offset, - struct pipe_resource **outbuf ); + struct pipe_resource **outbuf, + boolean *flushed ); |