diff options
Diffstat (limited to 'src/mesa/drivers/dri/i965')
52 files changed, 1357 insertions, 1855 deletions
diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile index 57dcc91586..7758a792fd 100644 --- a/src/mesa/drivers/dri/i965/Makefile +++ b/src/mesa/drivers/dri/i965/Makefile @@ -24,7 +24,6 @@ DRIVER_SOURCES = \ intel_pixel_draw.c \ intel_pixel_read.c \ intel_state.c \ - intel_swapbuffers.c \ intel_syncobj.c \ intel_tex.c \ intel_tex_copy.c \ @@ -96,10 +95,10 @@ ASM_SOURCES = DRIVER_DEFINES = -I../intel -I../intel/server -DRI_LIB_DEPS += -ldrm_intel +INCLUDES += $(INTEL_CFLAGS) +DRI_LIB_DEPS += $(INTEL_LIBS) include ../Makefile.template -symlinks: intel_decode.o: ../intel/intel_decode.c intel_tex_layout.o: ../intel/intel_tex_layout.c diff --git a/src/mesa/drivers/dri/i965/brw_cc.c b/src/mesa/drivers/dri/i965/brw_cc.c index 1088a7a607..bac1c3a49c 100644 --- a/src/mesa/drivers/dri/i965/brw_cc.c +++ b/src/mesa/drivers/dri/i965/brw_cc.c @@ -34,6 +34,7 @@ #include "brw_state.h" #include "brw_defines.h" #include "brw_util.h" +#include "intel_fbo.h" #include "main/macros.h" #include "main/enums.h" @@ -44,17 +45,24 @@ static void prepare_cc_vp( struct brw_context *brw ) memset(&ccv, 0, sizeof(ccv)); - /* _NEW_VIEWPORT */ - ccv.min_depth = ctx->Viewport.Near; - ccv.max_depth = ctx->Viewport.Far; + /* _NEW_TRANSOFORM */ + if (ctx->Transform.DepthClamp) { + /* _NEW_VIEWPORT */ + ccv.min_depth = MIN2(ctx->Viewport.Near, ctx->Viewport.Far); + ccv.max_depth = MAX2(ctx->Viewport.Near, ctx->Viewport.Far); + } else { + ccv.min_depth = 0.0; + ccv.max_depth = 1.0; + } dri_bo_unreference(brw->cc.vp_bo); - brw->cc.vp_bo = brw_cache_data( &brw->cache, BRW_CC_VP, &ccv, NULL, 0 ); + brw->cc.vp_bo = brw_cache_data(&brw->cache, BRW_CC_VP, &ccv, sizeof(ccv), + NULL, 0); } const struct brw_tracked_state brw_cc_vp = { .dirty = { - .mesa = _NEW_VIEWPORT, + .mesa = _NEW_VIEWPORT | _NEW_TRANSFORM, .brw = BRW_NEW_CONTEXT, .cache = 0 }, @@ -82,6 +90,28 @@ struct brw_cc_unit_key { GLenum depth_func; }; +/** + * Modify blend function to force destination alpha to 1.0 + * + * If \c function specifies a blend function that uses destination alpha, + * replace it with a function that hard-wires destination alpha to 1.0. This + * is used when rendering to xRGB targets. + */ +static GLenum +fix_xRGB_alpha(GLenum function) +{ + switch (function) { + case GL_DST_ALPHA: + return GL_ONE; + + case GL_ONE_MINUS_DST_ALPHA: + case GL_SRC_ALPHA_SATURATE: + return GL_ZERO; + } + + return function; +} + static void cc_unit_populate_key(struct brw_context *brw, struct brw_cc_unit_key *key) { @@ -125,6 +155,17 @@ cc_unit_populate_key(struct brw_context *brw, struct brw_cc_unit_key *key) key->blend_dst_rgb = ctx->Color.BlendDstRGB; key->blend_src_a = ctx->Color.BlendSrcA; key->blend_dst_a = ctx->Color.BlendDstA; + + /* If the renderbuffer is XRGB, we have to frob the blend function to + * force the destination alpha to 1.0. This means replacing GL_DST_ALPHA + * with GL_ONE and GL_ONE_MINUS_DST_ALPHA with GL_ZERO. + */ + if (ctx->DrawBuffer->Visual.alphaBits == 0) { + key->blend_src_rgb = fix_xRGB_alpha(key->blend_src_rgb); + key->blend_src_a = fix_xRGB_alpha(key->blend_src_a); + key->blend_dst_rgb = fix_xRGB_alpha(key->blend_dst_rgb); + key->blend_dst_a = fix_xRGB_alpha(key->blend_dst_a); + } } key->alpha_enabled = ctx->Color.AlphaEnabled; diff --git a/src/mesa/drivers/dri/i965/brw_clip.c b/src/mesa/drivers/dri/i965/brw_clip.c index 20a927cf38..af1d975de9 100644 --- a/src/mesa/drivers/dri/i965/brw_clip.c +++ b/src/mesa/drivers/dri/i965/brw_clip.c @@ -50,6 +50,7 @@ static void compile_clip_prog( struct brw_context *brw, struct brw_clip_prog_key *key ) { + struct intel_context *intel = &brw->intel; struct brw_clip_compile c; const GLuint *program; GLuint program_size; @@ -65,27 +66,26 @@ static void compile_clip_prog( struct brw_context *brw, c.func.single_program_flow = 1; c.key = *key; - c.need_ff_sync = BRW_IS_IGDNG(brw); /* Need to locate the two positions present in vertex + header. * These are currently hardcoded: */ c.header_position_offset = ATTR_SIZE; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) delta = 3 * REG_SIZE; else delta = REG_SIZE; for (i = 0; i < VERT_RESULT_MAX; i++) - if (c.key.attrs & (1<<i)) { + if (c.key.attrs & BITFIELD64_BIT(i)) { c.offset[i] = delta; delta += ATTR_SIZE; } c.nr_attrs = brw_count_bits(c.key.attrs); - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) c.nr_regs = (c.nr_attrs + 1) / 2 + 3; /* are vertices packed, or reg-aligned? */ else c.nr_regs = (c.nr_attrs + 1) / 2 + 1; /* are vertices packed, or reg-aligned? */ @@ -143,7 +143,8 @@ static void compile_clip_prog( struct brw_context *brw, */ static void upload_clip_prog(struct brw_context *brw) { - GLcontext *ctx = &brw->intel.ctx; + struct intel_context *intel = &brw->intel; + GLcontext *ctx = &intel->ctx; struct brw_clip_prog_key key; memset(&key, 0, sizeof(key)); @@ -156,10 +157,11 @@ static void upload_clip_prog(struct brw_context *brw) key.attrs = brw->vs.prog_data->outputs_written; /* _NEW_LIGHT */ key.do_flat_shading = (ctx->Light.ShadeModel == GL_FLAT); + key.pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION); /* _NEW_TRANSFORM */ key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled); - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) key.clip_mode = BRW_CLIPMODE_KERNEL_CLIP; else key.clip_mode = BRW_CLIPMODE_NORMAL; diff --git a/src/mesa/drivers/dri/i965/brw_clip.h b/src/mesa/drivers/dri/i965/brw_clip.h index 957df441ab..d71bac7f61 100644 --- a/src/mesa/drivers/dri/i965/brw_clip.h +++ b/src/mesa/drivers/dri/i965/brw_clip.h @@ -42,22 +42,21 @@ * up polygon offset and flatshading at this point: */ struct brw_clip_prog_key { - GLuint attrs:32; + GLbitfield64 attrs; GLuint primitive:4; GLuint nr_userclip:3; GLuint do_flat_shading:1; + GLuint pv_first:1; GLuint do_unfilled:1; GLuint fill_cw:2; /* includes cull information */ GLuint fill_ccw:2; /* includes cull information */ GLuint offset_cw:1; GLuint offset_ccw:1; - GLuint pad0:17; - GLuint copy_bfc_cw:1; GLuint copy_bfc_ccw:1; GLuint clip_mode:3; - GLuint pad1:27; - + GLuint pad0:11; + GLfloat offset_factor; GLfloat offset_units; }; @@ -119,7 +118,6 @@ struct brw_clip_compile { GLuint header_position_offset; GLuint offset[VERT_ATTRIB_MAX]; - GLboolean need_ff_sync; }; #define ATTR_SIZE (4*4) diff --git a/src/mesa/drivers/dri/i965/brw_clip_line.c b/src/mesa/drivers/dri/i965/brw_clip_line.c index 048ca620fa..afc0b11049 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_line.c +++ b/src/mesa/drivers/dri/i965/brw_clip_line.c @@ -46,6 +46,7 @@ static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; GLuint i = 0,j; /* Register usage is static, precompute here: @@ -85,7 +86,7 @@ static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) i++; } - if (c->need_ff_sync) { + if (intel->needs_ff_sync) { c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD); i++; } @@ -126,6 +127,7 @@ static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) static void clip_and_emit_line( struct brw_clip_compile *c ) { struct brw_compile *p = &c->func; + struct brw_context *brw = p->brw; struct brw_indirect vtx0 = brw_indirect(0, 0); struct brw_indirect vtx1 = brw_indirect(1, 0); struct brw_indirect newvtx0 = brw_indirect(2, 0); @@ -152,7 +154,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) brw_clip_init_clipmask(c); /* -ve rhw workaround */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<20)); @@ -189,7 +191,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) * Both can be negative on GM965/G965 due to RHW workaround * if so, this object should be rejected. */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE, c->reg.dp0, brw_imm_f(0.0)); is_neg2 = brw_IF(p, BRW_EXECUTE_1); { @@ -214,7 +216,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) /* If both are positive, do nothing */ /* Only on GM965/G965 */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0)); is_neg2 = brw_IF(p, BRW_EXECUTE_1); } @@ -229,7 +231,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) brw_set_predicate_control(p, BRW_PREDICATE_NONE); } - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_ENDIF(p, is_neg2); } } @@ -269,8 +271,12 @@ void brw_emit_line_clip( struct brw_clip_compile *c ) brw_clip_line_alloc_regs(c); brw_clip_init_ff_sync(c); - if (c->key.do_flat_shading) - brw_clip_copy_colors(c, 0, 1); + if (c->key.do_flat_shading) { + if (c->key.pv_first) + brw_clip_copy_colors(c, 1, 0); + else + brw_clip_copy_colors(c, 0, 1); + } clip_and_emit_line(c); } diff --git a/src/mesa/drivers/dri/i965/brw_clip_state.c b/src/mesa/drivers/dri/i965/brw_clip_state.c index 234b3744bf..c8f24a94e4 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_state.c +++ b/src/mesa/drivers/dri/i965/brw_clip_state.c @@ -74,6 +74,7 @@ static dri_bo * clip_unit_create_from_key(struct brw_context *brw, struct brw_clip_unit_key *key) { + struct intel_context *intel = &brw->intel; struct brw_clip_unit_state clip; dri_bo *bo; @@ -105,7 +106,7 @@ clip_unit_create_from_key(struct brw_context *brw, /* Although up to 16 concurrent Clip threads are allowed on IGDNG, * only 2 threads can output VUEs at a time. */ - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) clip.thread4.max_threads = 16 - 1; else clip.thread4.max_threads = 2 - 1; @@ -130,7 +131,7 @@ clip_unit_create_from_key(struct brw_context *brw, clip.clip5.api_mode = BRW_CLIP_API_OGL; clip.clip5.clip_mode = key->clip_mode; - if (BRW_IS_G4X(brw)) + if (intel->is_g4x) clip.clip5.negative_w_clip_test = 1; clip.clip6.clipper_viewport_state_ptr = 0; diff --git a/src/mesa/drivers/dri/i965/brw_clip_tri.c b/src/mesa/drivers/dri/i965/brw_clip_tri.c index 0efd77225e..cfbb8f2686 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_tri.c +++ b/src/mesa/drivers/dri/i965/brw_clip_tri.c @@ -51,6 +51,7 @@ static void release_tmps( struct brw_clip_compile *c ) void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, GLuint nr_verts ) { + struct intel_context *intel = &c->func.brw->intel; GLuint i = 0,j; /* Register usage is static, precompute here: @@ -78,7 +79,7 @@ void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, for (j = 0; j < 3; j++) { GLuint delta = c->nr_attrs*16 + 32; - if (BRW_IS_IGDNG(c->func.brw)) + if (intel->is_ironlake) delta = c->nr_attrs * 16 + 32 * 3; brw_MOV(&c->func, byte_offset(c->reg.vertex[j], delta), brw_imm_f(0)); @@ -119,7 +120,7 @@ void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, i++; } - if (c->need_ff_sync) { + if (intel->needs_ff_sync) { c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD); i++; } @@ -188,14 +189,20 @@ void brw_clip_tri_flat_shade( struct brw_clip_compile *c ) brw_imm_ud(_3DPRIM_POLYGON)); is_poly = brw_IF(p, BRW_EXECUTE_1); - { + { brw_clip_copy_colors(c, 1, 0); brw_clip_copy_colors(c, 2, 0); } is_poly = brw_ELSE(p, is_poly); { - brw_clip_copy_colors(c, 0, 2); - brw_clip_copy_colors(c, 1, 2); + if (c->key.pv_first) { + brw_clip_copy_colors(c, 1, 0); + brw_clip_copy_colors(c, 2, 0); + } + else { + brw_clip_copy_colors(c, 0, 2); + brw_clip_copy_colors(c, 1, 2); + } } brw_ENDIF(p, is_poly); } @@ -565,6 +572,7 @@ void brw_emit_tri_clip( struct brw_clip_compile *c ) { struct brw_instruction *neg_rhw; struct brw_compile *p = &c->func; + struct brw_context *brw = p->brw; brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6); brw_clip_tri_init_vertices(c); brw_clip_init_clipmask(c); @@ -572,7 +580,7 @@ void brw_emit_tri_clip( struct brw_clip_compile *c ) /* if -ve rhw workaround bit is set, do cliptest */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<20)); diff --git a/src/mesa/drivers/dri/i965/brw_clip_util.c b/src/mesa/drivers/dri/i965/brw_clip_util.c index 5a73abdfee..86fed59fa4 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_util.c +++ b/src/mesa/drivers/dri/i965/brw_clip_util.c @@ -135,6 +135,7 @@ void brw_clip_interp_vertex( struct brw_clip_compile *c, GLboolean force_edgeflag) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; struct brw_reg tmp = get_tmp(c); GLuint i; @@ -142,7 +143,7 @@ void brw_clip_interp_vertex( struct brw_clip_compile *c, */ /* * After CLIP stage, only first 256 bits of the VUE are read - * back on IGDNG, so needn't change it + * back on Ironlake, so needn't change it */ brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1); @@ -151,7 +152,7 @@ void brw_clip_interp_vertex( struct brw_clip_compile *c, for (i = 0; i < c->nr_attrs; i++) { GLuint delta = i*16 + 32; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) delta = i * 16 + 32 * 3; if (delta == c->offset[VERT_RESULT_EDGE]) { @@ -185,7 +186,7 @@ void brw_clip_interp_vertex( struct brw_clip_compile *c, if (i & 1) { GLuint delta = i*16 + 32; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) delta = i * 16 + 32 * 3; brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0)); @@ -359,7 +360,9 @@ void brw_clip_init_clipmask( struct brw_clip_compile *c ) void brw_clip_ff_sync(struct brw_clip_compile *c) { - if (c->need_ff_sync) { + struct intel_context *intel = &c->func.brw->intel; + + if (intel->needs_ff_sync) { struct brw_compile *p = &c->func; struct brw_instruction *need_ff_sync; @@ -388,7 +391,9 @@ void brw_clip_ff_sync(struct brw_clip_compile *c) void brw_clip_init_ff_sync(struct brw_clip_compile *c) { - if (c->need_ff_sync) { + struct intel_context *intel = &c->func.brw->intel; + + if (intel->needs_ff_sync) { struct brw_compile *p = &c->func; brw_MOV(p, c->reg.ff_sync, brw_imm_ud(0)); diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c index c300c33adc..7bb15956b5 100644 --- a/src/mesa/drivers/dri/i965/brw_context.c +++ b/src/mesa/drivers/dri/i965/brw_context.c @@ -77,7 +77,7 @@ static void brwInitDriverFunctions( struct dd_function_table *functions ) } GLboolean brwCreateContext( const __GLcontextModes *mesaVis, - __DRIcontextPrivate *driContextPriv, + __DRIcontext *driContextPriv, void *sharedContextPrivate) { struct dd_function_table functions; @@ -105,11 +105,15 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; + ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; ctx->Const.MaxTextureImageUnits = BRW_MAX_TEX_UNIT; ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ ctx->Const.MaxTextureUnits = MIN2(ctx->Const.MaxTextureCoordUnits, ctx->Const.MaxTextureImageUnits); ctx->Const.MaxVertexTextureImageUnits = 0; /* no vertex shader textures */ + ctx->Const.MaxCombinedTextureImageUnits = + ctx->Const.MaxVertexTextureImageUnits + + ctx->Const.MaxTextureImageUnits; /* Mesa limits textures to 4kx4k; it would be nice to fix that someday */ @@ -154,6 +158,38 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, MIN2(ctx->Const.FragmentProgram.MaxNativeParameters, ctx->Const.FragmentProgram.MaxEnvParams); + if (intel->is_ironlake || intel->is_g4x) { + brw->CMD_VF_STATISTICS = CMD_VF_STATISTICS_GM45; + brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45; + brw->has_surface_tile_offset = GL_TRUE; + brw->has_compr4 = GL_TRUE; + brw->has_aa_line_parameters = GL_TRUE; + } else { + brw->CMD_VF_STATISTICS = CMD_VF_STATISTICS_965; + brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965; + } + + /* WM maximum threads is number of EUs times number of threads per EU. */ + if (intel->is_ironlake) { + brw->urb.size = 1024; + brw->vs_max_threads = 72; + brw->wm_max_threads = 12 * 6; + } else if (intel->is_g4x) { + brw->urb.size = 384; + brw->vs_max_threads = 32; + brw->wm_max_threads = 10 * 5; + } else { + brw->urb.size = 256; + brw->vs_max_threads = 16; + brw->wm_max_threads = 8 * 4; + brw->has_negative_rhw_bug = GL_TRUE; + } + + if (INTEL_DEBUG & DEBUG_SINGLE_THREAD) { + brw->vs_max_threads = 1; + brw->wm_max_threads = 1; + } + brw_init_state( brw ); brw->state.dirty.mesa = ~0; diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index fa3e32c7ff..0dd3087143 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -115,7 +115,7 @@ * Handles blending and (presumably) depth and stencil testing. */ -#define BRW_FALLBACK_TEXTURE 0x1 + #define BRW_MAX_CURBE (32*16) struct brw_context; @@ -172,8 +172,8 @@ struct brw_fragment_program { GLuint id; /**< serial no. to identify frag progs, never re-used */ GLboolean isGLSL; /**< really, any IF/LOOP/CONT/BREAK instructions */ - dri_bo *const_buffer; /** Program constant buffer/surface */ GLboolean use_const_buffer; + dri_bo *const_buffer; /** Program constant buffer/surface */ /** for debugging, which texture units are referenced */ GLbitfield tex_units_used; @@ -231,7 +231,7 @@ struct brw_vs_prog_data { GLuint curb_read_length; GLuint urb_read_length; GLuint total_grf; - GLuint outputs_written; + GLbitfield64 outputs_written; GLuint nr_params; /**< number of float params/constants */ GLuint inputs_read; @@ -252,20 +252,23 @@ struct brw_vs_ouput_sizes { /** Number of texture sampler units */ #define BRW_MAX_TEX_UNIT 16 +/** Max number of render targets in a shader */ +#define BRW_MAX_DRAW_BUFFERS 4 + /** * Size of our surface binding table for the WM. * This contains pointers to the drawing surfaces and current texture * objects and shader constant buffers (+2). */ -#define BRW_WM_MAX_SURF (MAX_DRAW_BUFFERS + BRW_MAX_TEX_UNIT + 1) +#define BRW_WM_MAX_SURF (BRW_MAX_DRAW_BUFFERS + BRW_MAX_TEX_UNIT + 1) /** * Helpers to convert drawing buffers, textures and constant buffers * to surface binding table indexes, for WM. */ #define SURF_INDEX_DRAW(d) (d) -#define SURF_INDEX_FRAG_CONST_BUFFER (MAX_DRAW_BUFFERS) -#define SURF_INDEX_TEXTURE(t) (MAX_DRAW_BUFFERS + 1 + (t)) +#define SURF_INDEX_FRAG_CONST_BUFFER (BRW_MAX_DRAW_BUFFERS) +#define SURF_INDEX_TEXTURE(t) (BRW_MAX_DRAW_BUFFERS + 1 + (t)) /** * Size of surface binding table for the VS. @@ -317,7 +320,6 @@ struct brw_cache_item { GLuint nr_reloc_bufs; dri_bo *bo; - GLuint data_size; struct brw_cache_item *next; }; @@ -330,7 +332,6 @@ struct brw_cache { struct brw_cache_item **items; GLuint size, n_items; - GLuint key_size[BRW_MAX_CACHE]; /* for fixed-size keys */ GLuint aux_size[BRW_MAX_CACHE]; char *name[BRW_MAX_CACHE]; @@ -410,23 +411,6 @@ struct brw_vertex_info { GLuint sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[VERT_ATTRIB_MAX] */ }; - - - -/* Cache for TNL programs. - */ -struct brw_tnl_cache_item { - GLuint hash; - void *key; - void *data; - struct brw_tnl_cache_item *next; -}; - -struct brw_tnl_cache { - struct brw_tnl_cache_item **items; - GLuint size, n_items; -}; - struct brw_query_object { struct gl_query_object Base; @@ -454,9 +438,11 @@ struct brw_context GLuint primitive; GLboolean emit_state_always; - GLboolean tmp_fallback; - GLboolean no_batch_wrap; - + GLboolean has_surface_tile_offset; + GLboolean has_compr4; + GLboolean has_negative_rhw_bug; + GLboolean has_aa_line_parameters; +; struct { struct brw_state_flags dirty; @@ -532,6 +518,12 @@ struct brw_context */ GLuint next_free_page; + /* hw-dependent 3DSTATE_VF_STATISTICS opcode */ + uint32_t CMD_VF_STATISTICS; + /* hw-dependent 3DSTATE_PIPELINE_SELECT opcode */ + uint32_t CMD_PIPELINE_SELECT; + int vs_max_threads; + int wm_max_threads; /* BRW_NEW_URB_ALLOCATIONS: */ @@ -559,6 +551,7 @@ struct brw_context GLuint clip_start; GLuint sf_start; GLuint cs_start; + GLuint size; /* Hardware URB size, in KB. */ } urb; @@ -686,7 +679,7 @@ void brwInitVtbl( struct brw_context *brw ); * brw_context.c */ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, - __DRIcontextPrivate *driContextPriv, + __DRIcontext *driContextPriv, void *sharedContextPrivate); /*====================================================================== @@ -759,9 +752,5 @@ brw_fragment_program_const(const struct gl_fragment_program *p) return (const struct brw_fragment_program *) p; } - - -#define DO_SETUP_BITS ((1<<(FRAG_ATTRIB_MAX)) - 1) - #endif diff --git a/src/mesa/drivers/dri/i965/brw_curbe.c b/src/mesa/drivers/dri/i965/brw_curbe.c index 4be6c77aa1..190310afbb 100644 --- a/src/mesa/drivers/dri/i965/brw_curbe.c +++ b/src/mesa/drivers/dri/i965/brw_curbe.c @@ -130,7 +130,7 @@ static void calculate_curbe_offsets( struct brw_context *brw ) const struct brw_tracked_state brw_curbe_offsets = { .dirty = { .mesa = _NEW_TRANSFORM, - .brw = BRW_NEW_VERTEX_PROGRAM, + .brw = BRW_NEW_VERTEX_PROGRAM | BRW_NEW_CONTEXT, .cache = CACHE_NEW_WM_PROG }, .prepare = calculate_curbe_offsets @@ -340,7 +340,7 @@ static void emit_constant_buffer(struct brw_context *brw) struct intel_context *intel = &brw->intel; GLuint sz = brw->curbe.total_size; - BEGIN_BATCH(2, IGNORE_CLIPRECTS); + BEGIN_BATCH(2); if (sz == 0) { OUT_BATCH((CMD_CONST_BUFFER << 16) | (2 - 2)); OUT_BATCH(0); diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h index 78d457ad2b..ea0d7e05d4 100644 --- a/src/mesa/drivers/dri/i965/brw_defines.h +++ b/src/mesa/drivers/dri/i965/brw_defines.h @@ -673,18 +673,10 @@ #define BRW_SAMPLER_MESSAGE_SIMD8_LD 3 #define BRW_SAMPLER_MESSAGE_SIMD16_LD 3 -#define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_IGDNG 0 -#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_IGDNG 0 -#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_IGDNG 0 -#define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_BIAS_IGDNG 1 -#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_BIAS_IGDNG 1 -#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS_IGDNG 1 -#define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_LOD_IGDNG 2 -#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD_IGDNG 2 -#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_LOD_IGDNG 2 -#define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_COMPARE_IGDNG 3 -#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_COMPARE_IGDNG 3 -#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE_IGDNG 3 +#define BRW_SAMPLER_MESSAGE_SAMPLE_IGDNG 0 +#define BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_IGDNG 1 +#define BRW_SAMPLER_MESSAGE_SAMPLE_LOD_IGDNG 2 +#define BRW_SAMPLER_MESSAGE_SAMPLE_COMPARE_IGDNG 3 /* for IGDNG only */ #define BRW_SAMPLER_SIMD_MODE_SIMD4X2 0 @@ -840,12 +832,4 @@ #include "intel_chipset.h" -#define BRW_IS_G4X(brw) (IS_G4X((brw)->intel.intelScreen->deviceID)) -#define BRW_IS_IGDNG(brw) (IS_IGDNG((brw)->intel.intelScreen->deviceID)) -#define BRW_IS_965(brw) (!(BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw))) -#define CMD_PIPELINE_SELECT(brw) ((BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) ? CMD_PIPELINE_SELECT_GM45 : CMD_PIPELINE_SELECT_965) -#define CMD_VF_STATISTICS(brw) ((BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) ? CMD_VF_STATISTICS_GM45 : CMD_VF_STATISTICS_965) -#define URB_SIZES(brw) (BRW_IS_IGDNG(brw) ? 1024 : \ - (BRW_IS_G4X(brw) ? 384 : 256)) /* 512 bit units */ - #endif diff --git a/src/mesa/drivers/dri/i965/brw_disasm.c b/src/mesa/drivers/dri/i965/brw_disasm.c index 9fef230507..a8f6b993ac 100644 --- a/src/mesa/drivers/dri/i965/brw_disasm.c +++ b/src/mesa/drivers/dri/i965/brw_disasm.c @@ -239,7 +239,7 @@ char *imm_encoding[8] = { [2] = "UW", [3] = "W", [5] = "VF", - [5] = "V", + [6] = "V", [7] = "F" }; @@ -365,6 +365,7 @@ static int format (FILE *f, char *format, ...) va_start (args, format); vsnprintf (buf, sizeof (buf) - 1, format, args); + va_end (args); string (f, buf); return 0; } diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c index 44bb7bd588..df281b27d5 100644 --- a/src/mesa/drivers/dri/i965/brw_draw.c +++ b/src/mesa/drivers/dri/i965/brw_draw.c @@ -145,7 +145,7 @@ static void brw_emit_prim(struct brw_context *brw, prim_packet.base_vert_location = prim->basevertex; /* Can't wrap here, since we rely on the validated state. */ - brw->no_batch_wrap = GL_TRUE; + intel->no_batch_wrap = GL_TRUE; /* If we're set to always flush, do it before and after the primitive emit. * We want to catch both missed flushes that hurt instruction/state cache @@ -153,21 +153,17 @@ static void brw_emit_prim(struct brw_context *brw, * the besides the draw code. */ if (intel->always_flush_cache) { - BEGIN_BATCH(1, IGNORE_CLIPRECTS); - OUT_BATCH(intel->vtbl.flush_cmd()); - ADVANCE_BATCH(); + intel_batchbuffer_emit_mi_flush(intel->batch); } if (prim_packet.verts_per_instance) { intel_batchbuffer_data( brw->intel.batch, &prim_packet, - sizeof(prim_packet), LOOP_CLIPRECTS); + sizeof(prim_packet)); } if (intel->always_flush_cache) { - BEGIN_BATCH(1, IGNORE_CLIPRECTS); - OUT_BATCH(intel->vtbl.flush_cmd()); - ADVANCE_BATCH(); + intel_batchbuffer_emit_mi_flush(intel->batch); } - brw->no_batch_wrap = GL_FALSE; + intel->no_batch_wrap = GL_FALSE; } static void brw_merge_inputs( struct brw_context *brw, @@ -343,13 +339,6 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx, * so can't access it earlier. */ - LOCK_HARDWARE(intel); - - if (!intel->constant_cliprect && intel->driDrawable->numClipRects == 0) { - UNLOCK_HARDWARE(intel); - return GL_TRUE; - } - for (i = 0; i < nr_prims; i++) { uint32_t hw_prim; @@ -360,8 +349,7 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx, * an upper bound of how much we might emit in a single * brw_try_draw_prims(). */ - intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4, - LOOP_CLIPRECTS); + intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4); hw_prim = brw_set_prim(brw, prim[i].mode); @@ -408,7 +396,6 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx, if (intel->always_flush_batch) intel_batchbuffer_flush(intel->batch); out: - UNLOCK_HARDWARE(intel); brw_state_cache_check_size(brw); diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c index 765ae5a2fe..c773b71507 100644 --- a/src/mesa/drivers/dri/i965/brw_draw_upload.c +++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c @@ -243,14 +243,6 @@ static void wrap_buffers( struct brw_context *brw, dri_bo_unreference(brw->vb.upload.bo); brw->vb.upload.bo = dri_bo_alloc(brw->intel.bufmgr, "temporary VBO", size, 1); - - /* Set the internal VBO\ to no-backing-store. We only use them as a - * temporary within a brw_try_draw_prims while the lock is held. - */ - /* DON'T DO THIS AS IF WE HAVE TO RE-ORG MEMORY WE NEED SOMEWHERE WITH - FAKE TO PUSH THIS STUFF */ -// if (!brw->intel.ttm) -// dri_bo_fake_disable_backing_store(brw->vb.upload.bo, NULL, NULL); } static void get_space( struct brw_context *brw, @@ -375,7 +367,7 @@ static void brw_prepare_vertices(struct brw_context *brw) * isn't an issue at this point. */ if (brw->vb.nr_enabled >= BRW_VEP_MAX) { - intel->Fallback = 1; + intel->Fallback = GL_TRUE; /* boolean, not bitfield */ return; } @@ -427,7 +419,7 @@ static void brw_prepare_vertices(struct brw_context *brw) /* Position array not properly enabled: */ if (input->glarray->StrideB == 0) { - intel->Fallback = 1; + intel->Fallback = GL_TRUE; /* boolean, not bitfield */ return; } @@ -502,7 +494,7 @@ static void brw_emit_vertices(struct brw_context *brw) * a VE loads from them. */ if (brw->vb.nr_enabled == 0) { - BEGIN_BATCH(3, IGNORE_CLIPRECTS); + BEGIN_BATCH(3); OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1); OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) | BRW_VE0_VALID | @@ -522,7 +514,7 @@ static void brw_emit_vertices(struct brw_context *brw) * are interleaved or from the same VBO. TBD if this makes a * performance difference. */ - BEGIN_BATCH(1 + brw->vb.nr_enabled * 4, IGNORE_CLIPRECTS); + BEGIN_BATCH(1 + brw->vb.nr_enabled * 4); OUT_BATCH((CMD_VERTEX_BUFFER << 16) | ((1 + brw->vb.nr_enabled * 4) - 2)); @@ -535,24 +527,17 @@ static void brw_emit_vertices(struct brw_context *brw) OUT_RELOC(input->bo, I915_GEM_DOMAIN_VERTEX, 0, input->offset); - if (BRW_IS_IGDNG(brw)) { - if (input->stride) { - OUT_RELOC(input->bo, - I915_GEM_DOMAIN_VERTEX, 0, - input->offset + input->stride * input->count); - } else { - assert(input->count == 1); - OUT_RELOC(input->bo, - I915_GEM_DOMAIN_VERTEX, 0, - input->offset + input->element_size); - } + if (intel->is_ironlake) { + OUT_RELOC(input->bo, + I915_GEM_DOMAIN_VERTEX, 0, + input->bo->size - 1); } else OUT_BATCH(input->stride ? input->count : 0); OUT_BATCH(0); /* Instance data step rate */ } ADVANCE_BATCH(); - BEGIN_BATCH(1 + brw->vb.nr_enabled * 2, IGNORE_CLIPRECTS); + BEGIN_BATCH(1 + brw->vb.nr_enabled * 2); OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + brw->vb.nr_enabled * 2) - 2)); for (i = 0; i < brw->vb.nr_enabled; i++) { struct brw_vertex_element *input = brw->vb.enabled[i]; @@ -578,7 +563,7 @@ static void brw_emit_vertices(struct brw_context *brw) (format << BRW_VE0_FORMAT_SHIFT) | (0 << BRW_VE0_SRC_OFFSET_SHIFT)); - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) | (comp1 << BRW_VE1_COMPONENT_1_SHIFT) | (comp2 << BRW_VE1_COMPONENT_2_SHIFT) | @@ -719,14 +704,14 @@ static void brw_emit_index_buffer(struct brw_context *brw) ib.header.bits.index_format = get_index_type(index_buffer->type); ib.header.bits.cut_index_enable = 0; - BEGIN_BATCH(4, IGNORE_CLIPRECTS); + BEGIN_BATCH(4); OUT_BATCH( ib.header.dword ); OUT_RELOC(brw->ib.bo, I915_GEM_DOMAIN_VERTEX, 0, brw->ib.offset); OUT_RELOC(brw->ib.bo, I915_GEM_DOMAIN_VERTEX, 0, - brw->ib.offset + brw->ib.size); + brw->ib.offset + brw->ib.size - 1); OUT_BATCH( 0 ); ADVANCE_BATCH(); } diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h index 30603bdd0e..39eb88d7c2 100644 --- a/src/mesa/drivers/dri/i965/brw_eu.h +++ b/src/mesa/drivers/dri/i965/brw_eu.h @@ -170,11 +170,11 @@ static INLINE struct brw_reg brw_reg( GLuint file, GLuint writemask ) { struct brw_reg reg; - if (type == BRW_GENERAL_REGISTER_FILE) + if (file == BRW_GENERAL_REGISTER_FILE) assert(nr < BRW_MAX_GRF); - else if (type == BRW_MESSAGE_REGISTER_FILE) - assert(nr < BRW_MAX_MRF); - else if (type == BRW_ARCHITECTURE_REGISTER_FILE) + else if (file == BRW_MESSAGE_REGISTER_FILE) + assert((nr & ~(1 << 7)) < BRW_MAX_MRF); + else if (file == BRW_ARCHITECTURE_REGISTER_FILE) assert(nr <= BRW_ARF_IP); reg.type = type; @@ -538,7 +538,7 @@ static INLINE struct brw_reg brw_mask_reg( GLuint subnr ) static INLINE struct brw_reg brw_message_reg( GLuint nr ) { - assert(nr < BRW_MAX_MRF); + assert((nr & ~(1 << 7)) < BRW_MAX_MRF); return brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, nr, 0); diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c index 241cdc33f8..8d6ac00839 100644 --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c @@ -55,7 +55,8 @@ static void guess_execution_size( struct brw_instruction *insn, static void brw_set_dest( struct brw_instruction *insn, struct brw_reg dest ) { - if (dest.type != BRW_ARCHITECTURE_REGISTER_FILE) + if (dest.file != BRW_ARCHITECTURE_REGISTER_FILE && + dest.file != BRW_MESSAGE_REGISTER_FILE) assert(dest.nr < 128); insn->bits1.da1.dest_reg_file = dest.file; @@ -198,7 +199,7 @@ void brw_set_src1( struct brw_instruction *insn, * in the future: */ assert (reg.address_mode == BRW_ADDRESS_DIRECT); - //assert (reg.file == BRW_GENERAL_REGISTER_FILE); + /* assert (reg.file == BRW_GENERAL_REGISTER_FILE); */ if (insn->header.access_mode == BRW_ALIGN_1) { insn->bits3.da1.src1_subreg_nr = reg.subnr; @@ -251,9 +252,10 @@ static void brw_set_math_message( struct brw_context *brw, GLboolean saturate, GLuint dataType ) { + struct intel_context *intel = &brw->intel; brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.math_igdng.function = function; insn->bits3.math_igdng.int_type = integer_type; insn->bits3.math_igdng.precision = low_precision; @@ -318,9 +320,10 @@ static void brw_set_urb_message( struct brw_context *brw, GLuint offset, GLuint swizzle_control ) { + struct intel_context *intel = &brw->intel; brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.urb_igdng.opcode = 0; /* ? */ insn->bits3.urb_igdng.offset = offset; insn->bits3.urb_igdng.swizzle_control = swizzle_control; @@ -357,9 +360,10 @@ static void brw_set_dp_write_message( struct brw_context *brw, GLuint response_length, GLuint end_of_thread ) { + struct intel_context *intel = &brw->intel; brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.dp_write_igdng.binding_table_index = binding_table_index; insn->bits3.dp_write_igdng.msg_control = msg_control; insn->bits3.dp_write_igdng.pixel_scoreboard_clear = pixel_scoreboard_clear; @@ -394,9 +398,10 @@ static void brw_set_dp_read_message( struct brw_context *brw, GLuint response_length, GLuint end_of_thread ) { + struct intel_context *intel = &brw->intel; brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.dp_read_igdng.binding_table_index = binding_table_index; insn->bits3.dp_read_igdng.msg_control = msg_control; insn->bits3.dp_read_igdng.msg_type = msg_type; @@ -432,10 +437,11 @@ static void brw_set_sampler_message(struct brw_context *brw, GLuint header_present, GLuint simd_mode) { + struct intel_context *intel = &brw->intel; assert(eot == 0); brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.sampler_igdng.binding_table_index = binding_table_index; insn->bits3.sampler_igdng.sampler = sampler; insn->bits3.sampler_igdng.msg_type = msg_type; @@ -446,7 +452,7 @@ static void brw_set_sampler_message(struct brw_context *brw, insn->bits3.sampler_igdng.end_of_thread = eot; insn->bits2.send_igdng.sfid = BRW_MESSAGE_TARGET_SAMPLER; insn->bits2.send_igdng.end_of_thread = eot; - } else if (BRW_IS_G4X(brw)) { + } else if (intel->is_g4x) { insn->bits3.sampler_g4x.binding_table_index = binding_table_index; insn->bits3.sampler_g4x.sampler = sampler; insn->bits3.sampler_g4x.msg_type = msg_type; @@ -647,10 +653,11 @@ struct brw_instruction *brw_IF(struct brw_compile *p, GLuint execute_size) struct brw_instruction *brw_ELSE(struct brw_compile *p, struct brw_instruction *if_insn) { + struct intel_context *intel = &p->brw->intel; struct brw_instruction *insn; GLuint br = 1; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) br = 2; if (p->single_program_flow) { @@ -689,9 +696,10 @@ struct brw_instruction *brw_ELSE(struct brw_compile *p, void brw_ENDIF(struct brw_compile *p, struct brw_instruction *patch_insn) { + struct intel_context *intel = &p->brw->intel; GLuint br = 1; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) br = 2; if (p->single_program_flow) { @@ -802,10 +810,11 @@ struct brw_instruction *brw_DO(struct brw_compile *p, GLuint execute_size) struct brw_instruction *brw_WHILE(struct brw_compile *p, struct brw_instruction *do_insn) { + struct intel_context *intel = &p->brw->intel; struct brw_instruction *insn; GLuint br = 1; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) br = 2; if (p->single_program_flow) @@ -845,14 +854,15 @@ struct brw_instruction *brw_WHILE(struct brw_compile *p, void brw_land_fwd_jump(struct brw_compile *p, struct brw_instruction *jmp_insn) { + struct intel_context *intel = &p->brw->intel; struct brw_instruction *landing = &p->store[p->nr_insn]; GLuint jmpi = 1; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) jmpi = 2; assert(jmp_insn->header.opcode == BRW_OPCODE_JMPI); - assert(jmp_insn->bits1.da1.src1_reg_file = BRW_IMMEDIATE_VALUE); + assert(jmp_insn->bits1.da1.src1_reg_file == BRW_IMMEDIATE_VALUE); jmp_insn->bits3.ud = jmpi * ((landing - jmp_insn) - 1); } diff --git a/src/mesa/drivers/dri/i965/brw_fallback.c b/src/mesa/drivers/dri/i965/brw_fallback.c index d27c6c24ca..fe5c1ae279 100644 --- a/src/mesa/drivers/dri/i965/brw_fallback.c +++ b/src/mesa/drivers/dri/i965/brw_fallback.c @@ -47,7 +47,6 @@ static GLboolean do_check_fallback(struct brw_context *brw) { - struct intel_context *intel = &brw->intel; GLcontext *ctx = &brw->intel.ctx; GLuint i; @@ -86,8 +85,7 @@ static GLboolean do_check_fallback(struct brw_context *brw) } /* _NEW_BUFFERS */ - if (IS_965(intel->intelScreen->deviceID) && - !IS_G4X(intel->intelScreen->deviceID)) { + if (!brw->has_surface_tile_offset) { for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) { struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i]; struct intel_renderbuffer *irb = intel_renderbuffer(rb); @@ -133,7 +131,11 @@ const struct brw_tracked_state brw_check_fallback = { -/* Not used: +/** + * Called by the INTEL_FALLBACK() macro. + * NOTE: this is a no-op for the i965 driver. The brw->intel.Fallback + * field is treated as a boolean, not a bitmask. It's only set in a + * couple of places. */ void intelFallback( struct intel_context *intel, GLuint bit, GLboolean mode ) { diff --git a/src/mesa/drivers/dri/i965/brw_gs.c b/src/mesa/drivers/dri/i965/brw_gs.c index 48c2b9a41c..1bc3eccf49 100644 --- a/src/mesa/drivers/dri/i965/brw_gs.c +++ b/src/mesa/drivers/dri/i965/brw_gs.c @@ -47,6 +47,7 @@ static void compile_gs_prog( struct brw_context *brw, struct brw_gs_prog_key *key ) { + struct intel_context *intel = &brw->intel; struct brw_gs_compile c; const GLuint *program; GLuint program_size; @@ -54,13 +55,12 @@ static void compile_gs_prog( struct brw_context *brw, memset(&c, 0, sizeof(c)); c.key = *key; - c.need_ff_sync = BRW_IS_IGDNG(brw); /* Need to locate the two positions present in vertex + header. * These are currently hardcoded: */ c.nr_attrs = brw_count_bits(c.key.attrs); - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) c.nr_regs = (c.nr_attrs + 1) / 2 + 3; /* are vertices packed, or reg-aligned? */ else c.nr_regs = (c.nr_attrs + 1) / 2 + 1; /* are vertices packed, or reg-aligned? */ @@ -85,10 +85,10 @@ static void compile_gs_prog( struct brw_context *brw, */ switch (key->primitive) { case GL_QUADS: - brw_gs_quads( &c ); + brw_gs_quads( &c, key ); break; case GL_QUAD_STRIP: - brw_gs_quad_strip( &c ); + brw_gs_quad_strip( &c, key ); break; case GL_LINE_LOOP: brw_gs_lines( &c ); @@ -149,6 +149,7 @@ static const GLenum gs_prim[GL_POLYGON+1] = { static void populate_key( struct brw_context *brw, struct brw_gs_prog_key *key ) { + GLcontext *ctx = &brw->intel.ctx; memset(key, 0, sizeof(*key)); /* CACHE_NEW_VS_PROG */ @@ -158,6 +159,9 @@ static void populate_key( struct brw_context *brw, key->primitive = gs_prim[brw->primitive]; key->hint_gs_always = 0; /* debug code? */ + + /* _NEW_LIGHT */ + key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION); key->need_gs_prog = (key->hint_gs_always || brw->primitive == GL_QUADS || @@ -193,7 +197,7 @@ static void prepare_gs_prog(struct brw_context *brw) const struct brw_tracked_state brw_gs_prog = { .dirty = { - .mesa = 0, + .mesa = _NEW_LIGHT, .brw = BRW_NEW_PRIMITIVE, .cache = CACHE_NEW_VS_PROG }, diff --git a/src/mesa/drivers/dri/i965/brw_gs.h b/src/mesa/drivers/dri/i965/brw_gs.h index bbb991ea2e..813b8d447a 100644 --- a/src/mesa/drivers/dri/i965/brw_gs.h +++ b/src/mesa/drivers/dri/i965/brw_gs.h @@ -40,11 +40,12 @@ #define MAX_GS_VERTS (4) struct brw_gs_prog_key { - GLuint attrs:32; + GLbitfield64 attrs; GLuint primitive:4; GLuint hint_gs_always:1; + GLuint pv_first:1; GLuint need_gs_prog:1; - GLuint pad:26; + GLuint pad:25; }; struct brw_gs_compile { @@ -62,13 +63,12 @@ struct brw_gs_compile { GLuint nr_attrs; GLuint nr_regs; GLuint nr_bytes; - GLboolean need_ff_sync; }; #define ATTR_SIZE (4*4) -void brw_gs_quads( struct brw_gs_compile *c ); -void brw_gs_quad_strip( struct brw_gs_compile *c ); +void brw_gs_quads( struct brw_gs_compile *c, struct brw_gs_prog_key *key ); +void brw_gs_quad_strip( struct brw_gs_compile *c, struct brw_gs_prog_key *key ); void brw_gs_tris( struct brw_gs_compile *c ); void brw_gs_lines( struct brw_gs_compile *c ); void brw_gs_points( struct brw_gs_compile *c ); diff --git a/src/mesa/drivers/dri/i965/brw_gs_emit.c b/src/mesa/drivers/dri/i965/brw_gs_emit.c index a9b2aa2eac..a81b972ef4 100644 --- a/src/mesa/drivers/dri/i965/brw_gs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_gs_emit.c @@ -120,38 +120,60 @@ static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim) } -void brw_gs_quads( struct brw_gs_compile *c ) +void brw_gs_quads( struct brw_gs_compile *c, struct brw_gs_prog_key *key ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 4); /* Use polygons for correct edgeflag behaviour. Note that vertex 3 * is the PV for quads, but vertex 0 for polygons: */ - if (c->need_ff_sync) - brw_gs_ff_sync(c, 1); - brw_gs_emit_vue(c, c->reg.vertex[3], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); - brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2)); - brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2)); - brw_gs_emit_vue(c, c->reg.vertex[2], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END)); + if (intel->needs_ff_sync) + brw_gs_ff_sync(c, 1); + if (key->pv_first) { + brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); + brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[2], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[3], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END)); + } + else { + brw_gs_emit_vue(c, c->reg.vertex[3], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); + brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[2], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END)); + } } -void brw_gs_quad_strip( struct brw_gs_compile *c ) +void brw_gs_quad_strip( struct brw_gs_compile *c, struct brw_gs_prog_key *key ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 4); - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); - brw_gs_emit_vue(c, c->reg.vertex[2], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); - brw_gs_emit_vue(c, c->reg.vertex[3], 0, (_3DPRIM_POLYGON << 2)); - brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2)); - brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END)); + if (key->pv_first) { + brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); + brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[2], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[3], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END)); + } + else { + brw_gs_emit_vue(c, c->reg.vertex[2], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); + brw_gs_emit_vue(c, c->reg.vertex[3], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END)); + } } void brw_gs_tris( struct brw_gs_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 3); - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_TRILIST << 2) | R02_PRIM_START)); brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_TRILIST << 2)); @@ -160,9 +182,11 @@ void brw_gs_tris( struct brw_gs_compile *c ) void brw_gs_lines( struct brw_gs_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 2); - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_LINESTRIP << 2) | R02_PRIM_START)); brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_LINESTRIP << 2) | R02_PRIM_END)); @@ -170,9 +194,11 @@ void brw_gs_lines( struct brw_gs_compile *c ) void brw_gs_points( struct brw_gs_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 1); - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); brw_gs_emit_vue(c, c->reg.vertex[0], 1, ((_3DPRIM_POINTLIST << 2) | R02_PRIM_START | R02_PRIM_END)); } diff --git a/src/mesa/drivers/dri/i965/brw_gs_state.c b/src/mesa/drivers/dri/i965/brw_gs_state.c index ed9d2ffe60..1af5790a67 100644 --- a/src/mesa/drivers/dri/i965/brw_gs_state.c +++ b/src/mesa/drivers/dri/i965/brw_gs_state.c @@ -72,6 +72,7 @@ gs_unit_populate_key(struct brw_context *brw, struct brw_gs_unit_key *key) static dri_bo * gs_unit_create_from_key(struct brw_context *brw, struct brw_gs_unit_key *key) { + struct intel_context *intel = &brw->intel; struct brw_gs_unit_state gs; dri_bo *bo; @@ -98,7 +99,7 @@ gs_unit_create_from_key(struct brw_context *brw, struct brw_gs_unit_key *key) else gs.thread4.max_threads = 0; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) gs.thread4.rendering_enable = 1; if (INTEL_DEBUG & DEBUG_STATS) diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c index ea71857548..7b70f787b7 100644 --- a/src/mesa/drivers/dri/i965/brw_misc_state.c +++ b/src/mesa/drivers/dri/i965/brw_misc_state.c @@ -66,7 +66,7 @@ static void upload_blend_constant_color(struct brw_context *brw) const struct brw_tracked_state brw_blend_constant_color = { .dirty = { .mesa = _NEW_COLOR, - .brw = 0, + .brw = BRW_NEW_CONTEXT, .cache = 0 }, .emit = upload_blend_constant_color @@ -78,10 +78,7 @@ static void upload_drawing_rect(struct brw_context *brw) struct intel_context *intel = &brw->intel; GLcontext *ctx = &intel->ctx; - if (!intel->constant_cliprect) - return; - - BEGIN_BATCH(4, NO_LOOP_CLIPRECTS); + BEGIN_BATCH(4); OUT_BATCH(_3DSTATE_DRAWRECT_INFO_I965); OUT_BATCH(0); /* xmin, ymin */ OUT_BATCH(((ctx->DrawBuffer->Width - 1) & 0xffff) | @@ -93,7 +90,7 @@ static void upload_drawing_rect(struct brw_context *brw) const struct brw_tracked_state brw_drawing_rect = { .dirty = { .mesa = _NEW_BUFFERS, - .brw = 0, + .brw = BRW_NEW_CONTEXT, .cache = 0 }, .emit = upload_drawing_rect @@ -116,7 +113,7 @@ static void upload_binding_table_pointers(struct brw_context *brw) { struct intel_context *intel = &brw->intel; - BEGIN_BATCH(6, IGNORE_CLIPRECTS); + BEGIN_BATCH(6); OUT_BATCH(CMD_BINDING_TABLE_PTRS << 16 | (6 - 2)); if (brw->vs.bind_bo != NULL) OUT_RELOC(brw->vs.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* vs */ @@ -150,7 +147,7 @@ static void upload_pipelined_state_pointers(struct brw_context *brw ) { struct intel_context *intel = &brw->intel; - BEGIN_BATCH(7, IGNORE_CLIPRECTS); + BEGIN_BATCH(7); OUT_BATCH(CMD_PIPELINED_STATE_POINTERS << 16 | (7 - 2)); OUT_RELOC(brw->vs.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0); if (brw->gs.prog_active) @@ -212,10 +209,10 @@ static void emit_depthbuffer(struct brw_context *brw) { struct intel_context *intel = &brw->intel; struct intel_region *region = brw->state.depth_region; - unsigned int len = (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) ? 6 : 5; + unsigned int len = (intel->is_g4x || intel->is_ironlake) ? 6 : 5; if (region == NULL) { - BEGIN_BATCH(len, IGNORE_CLIPRECTS); + BEGIN_BATCH(len); OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (len - 2)); OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) | (BRW_SURFACE_NULL << 29)); @@ -223,7 +220,7 @@ static void emit_depthbuffer(struct brw_context *brw) OUT_BATCH(0); OUT_BATCH(0); - if (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) + if (intel->is_g4x || intel->is_ironlake) OUT_BATCH(0); ADVANCE_BATCH(); @@ -247,7 +244,7 @@ static void emit_depthbuffer(struct brw_context *brw) assert(region->tiling != I915_TILING_X); - BEGIN_BATCH(len, IGNORE_CLIPRECTS); + BEGIN_BATCH(len); OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (len - 2)); OUT_BATCH(((region->pitch * region->cpp) - 1) | (format << 18) | @@ -262,7 +259,7 @@ static void emit_depthbuffer(struct brw_context *brw) ((region->height - 1) << 19)); OUT_BATCH(0); - if (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) + if (intel->is_g4x || intel->is_ironlake) OUT_BATCH(0); ADVANCE_BATCH(); @@ -317,7 +314,7 @@ static void upload_polygon_stipple(struct brw_context *brw) const struct brw_tracked_state brw_polygon_stipple = { .dirty = { .mesa = _NEW_POLYGONSTIPPLE, - .brw = 0, + .brw = BRW_NEW_CONTEXT, .cache = 0 }, .emit = upload_polygon_stipple @@ -330,7 +327,7 @@ const struct brw_tracked_state brw_polygon_stipple = { static void upload_polygon_stipple_offset(struct brw_context *brw) { - __DRIdrawablePrivate *dPriv = brw->intel.driDrawable; + __DRIdrawable *dPriv = brw->intel.driDrawable; struct brw_polygon_stipple_offset bpso; memset(&bpso, 0, sizeof(bpso)); @@ -362,7 +359,7 @@ static void upload_polygon_stipple_offset(struct brw_context *brw) const struct brw_tracked_state brw_polygon_stipple_offset = { .dirty = { .mesa = _NEW_WINDOW_POS, - .brw = 0, + .brw = BRW_NEW_CONTEXT, .cache = 0 }, .emit = upload_polygon_stipple_offset @@ -374,8 +371,8 @@ const struct brw_tracked_state brw_polygon_stipple_offset = { static void upload_aa_line_parameters(struct brw_context *brw) { struct brw_aa_line_parameters balp; - - if (BRW_IS_965(brw)) + + if (!brw->has_aa_line_parameters) return; /* use legacy aa line coverage computation */ @@ -425,7 +422,7 @@ static void upload_line_stipple(struct brw_context *brw) const struct brw_tracked_state brw_line_stipple = { .dirty = { .mesa = _NEW_LINE, - .brw = 0, + .brw = BRW_NEW_CONTEXT, .cache = 0 }, .emit = upload_line_stipple @@ -444,7 +441,7 @@ static void upload_invarient_state( struct brw_context *brw ) struct brw_pipeline_select ps; memset(&ps, 0, sizeof(ps)); - ps.header.opcode = CMD_PIPELINE_SELECT(brw); + ps.header.opcode = brw->CMD_PIPELINE_SELECT; ps.header.pipeline_select = 0; BRW_BATCH_STRUCT(brw, &ps); } @@ -480,7 +477,7 @@ static void upload_invarient_state( struct brw_context *brw ) struct brw_vf_statistics vfs; memset(&vfs, 0, sizeof(vfs)); - vfs.opcode = CMD_VF_STATISTICS(brw); + vfs.opcode = brw->CMD_VF_STATISTICS; if (INTEL_DEBUG & DEBUG_STATS) vfs.statistics_enable = 1; @@ -512,8 +509,8 @@ static void upload_state_base_address( struct brw_context *brw ) /* Output the structure (brw_state_base_address) directly to the * batchbuffer, so we can emit relocations inline. */ - if (BRW_IS_IGDNG(brw)) { - BEGIN_BATCH(8, IGNORE_CLIPRECTS); + if (intel->is_ironlake) { + BEGIN_BATCH(8); OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (8 - 2)); OUT_BATCH(1); /* General state base address */ OUT_BATCH(1); /* Surface state base address */ @@ -524,7 +521,7 @@ static void upload_state_base_address( struct brw_context *brw ) OUT_BATCH(1); /* Instruction access upper bound */ ADVANCE_BATCH(); } else { - BEGIN_BATCH(6, IGNORE_CLIPRECTS); + BEGIN_BATCH(6); OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (6 - 2)); OUT_BATCH(1); /* General state base address */ OUT_BATCH(1); /* Surface state base address */ diff --git a/src/mesa/drivers/dri/i965/brw_queryobj.c b/src/mesa/drivers/dri/i965/brw_queryobj.c index a195bc32b0..5399a74244 100644 --- a/src/mesa/drivers/dri/i965/brw_queryobj.c +++ b/src/mesa/drivers/dri/i965/brw_queryobj.c @@ -188,7 +188,7 @@ brw_emit_query_begin(struct brw_context *brw) if (brw->query.active || is_empty_list(&brw->query.active_head)) return; - BEGIN_BATCH(4, IGNORE_CLIPRECTS); + BEGIN_BATCH(4); OUT_BATCH(_3DSTATE_PIPE_CONTROL | PIPE_CONTROL_DEPTH_STALL | PIPE_CONTROL_WRITE_DEPTH_COUNT); @@ -227,7 +227,7 @@ brw_emit_query_end(struct brw_context *brw) if (!brw->query.active) return; - BEGIN_BATCH(4, IGNORE_CLIPRECTS); + BEGIN_BATCH(4); OUT_BATCH(_3DSTATE_PIPE_CONTROL | PIPE_CONTROL_DEPTH_STALL | PIPE_CONTROL_WRITE_DEPTH_COUNT); diff --git a/src/mesa/drivers/dri/i965/brw_sf.c b/src/mesa/drivers/dri/i965/brw_sf.c index e1c2c7777b..968890f7fb 100644 --- a/src/mesa/drivers/dri/i965/brw_sf.c +++ b/src/mesa/drivers/dri/i965/brw_sf.c @@ -61,7 +61,7 @@ static void compile_sf_prog( struct brw_context *brw, c.key = *key; c.nr_attrs = brw_count_bits(c.key.attrs); c.nr_attr_regs = (c.nr_attrs+1)/2; - c.nr_setup_attrs = brw_count_bits(c.key.attrs & DO_SETUP_BITS); + c.nr_setup_attrs = brw_count_bits(c.key.attrs); c.nr_setup_regs = (c.nr_setup_attrs+1)/2; c.prog_data.urb_read_length = c.nr_attr_regs; @@ -70,7 +70,7 @@ static void compile_sf_prog( struct brw_context *brw, /* Construct map from attribute number to position in the vertex. */ for (i = idx = 0; i < VERT_RESULT_MAX; i++) - if (c.key.attrs & (1<<i)) { + if (c.key.attrs & BITFIELD64_BIT(i)) { c.attr_to_idx[i] = idx; c.idx_to_attr[idx] = i; if (i >= VERT_RESULT_TEX0 && i <= VERT_RESULT_TEX7) { @@ -147,7 +147,7 @@ static void upload_sf_prog(struct brw_context *brw) * edgeflag testing here, it is already done in the clip * program. */ - if (key.attrs & (1<<VERT_RESULT_EDGE)) + if (key.attrs & BITFIELD64_BIT(VERT_RESULT_EDGE)) key.primitive = SF_UNFILLED_TRIS; else key.primitive = SF_TRIANGLES; @@ -161,7 +161,7 @@ static void upload_sf_prog(struct brw_context *brw) } key.do_point_sprite = ctx->Point.PointSprite; - key.SpriteOrigin = ctx->Point.SpriteOrigin; + key.sprite_origin_lower_left = (ctx->Point.SpriteOrigin == GL_LOWER_LEFT); /* _NEW_LIGHT */ key.do_flat_shading = (ctx->Light.ShadeModel == GL_FLAT); key.do_twoside_color = (ctx->Light.Enabled && ctx->Light.Model.TwoSide); diff --git a/src/mesa/drivers/dri/i965/brw_sf.h b/src/mesa/drivers/dri/i965/brw_sf.h index 6426b6df9f..0ba731fac9 100644 --- a/src/mesa/drivers/dri/i965/brw_sf.h +++ b/src/mesa/drivers/dri/i965/brw_sf.h @@ -45,19 +45,19 @@ #define SF_UNFILLED_TRIS 3 struct brw_sf_prog_key { - GLuint attrs:32; + GLbitfield64 attrs; GLuint primitive:2; GLuint do_twoside_color:1; GLuint do_flat_shading:1; GLuint frontface_ccw:1; GLuint do_point_sprite:1; GLuint linear_color:1; /**< linear interp vs. perspective interp */ - GLuint pad:25; - GLenum SpriteOrigin; + GLuint sprite_origin_lower_left:1; + GLuint pad:24; }; struct brw_sf_point_tex { - GLboolean CoordReplace; + GLboolean CoordReplace; }; struct brw_sf_compile { diff --git a/src/mesa/drivers/dri/i965/brw_sf_emit.c b/src/mesa/drivers/dri/i965/brw_sf_emit.c index ca8f97f9f9..bb08055e3b 100644 --- a/src/mesa/drivers/dri/i965/brw_sf_emit.c +++ b/src/mesa/drivers/dri/i965/brw_sf_emit.c @@ -56,7 +56,7 @@ static struct brw_reg get_vert_attr(struct brw_sf_compile *c, static GLboolean have_attr(struct brw_sf_compile *c, GLuint attr) { - return (c->key.attrs & (1<<attr)) ? 1 : 0; + return (c->key.attrs & BITFIELD64_BIT(attr)) ? 1 : 0; } /*********************************************************************** @@ -122,8 +122,8 @@ static void do_twoside_color( struct brw_sf_compile *c ) * Flat shading */ -#define VERT_RESULT_COLOR_BITS ((1<<VERT_RESULT_COL0) | \ - (1<<VERT_RESULT_COL1)) +#define VERT_RESULT_COLOR_BITS (BITFIELD64_BIT(VERT_RESULT_COL0) | \ + BITFIELD64_BIT(VERT_RESULT_COL1)) static void copy_colors( struct brw_sf_compile *c, struct brw_reg dst, @@ -149,6 +149,7 @@ static void copy_colors( struct brw_sf_compile *c, static void do_flatshade_triangle( struct brw_sf_compile *c ) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; struct brw_reg ip = brw_ip_reg(); GLuint nr = brw_count_bits(c->key.attrs & VERT_RESULT_COLOR_BITS); GLuint jmpi = 1; @@ -161,7 +162,7 @@ static void do_flatshade_triangle( struct brw_sf_compile *c ) if (c->key.primitive == SF_UNFILLED_TRIS) return; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) jmpi = 2; brw_push_insn_state(p); @@ -187,6 +188,7 @@ static void do_flatshade_triangle( struct brw_sf_compile *c ) static void do_flatshade_line( struct brw_sf_compile *c ) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; struct brw_reg ip = brw_ip_reg(); GLuint nr = brw_count_bits(c->key.attrs & VERT_RESULT_COLOR_BITS); GLuint jmpi = 1; @@ -199,7 +201,7 @@ static void do_flatshade_line( struct brw_sf_compile *c ) if (c->key.primitive == SF_UNFILLED_TRIS) return; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) jmpi = 2; brw_push_insn_state(p); @@ -312,8 +314,8 @@ static GLboolean calculate_masks( struct brw_sf_compile *c, GLushort *pc_linear) { GLboolean is_last_attr = (reg == c->nr_setup_regs - 1); - GLuint persp_mask; - GLuint linear_mask; + GLbitfield64 persp_mask; + GLbitfield64 linear_mask; if (c->key.do_flat_shading || c->key.linear_color) persp_mask = c->key.attrs & ~(FRAG_BIT_WPOS | @@ -331,10 +333,10 @@ static GLboolean calculate_masks( struct brw_sf_compile *c, *pc_linear = 0; *pc = 0xf; - if (persp_mask & (1 << c->idx_to_attr[reg*2])) + if (persp_mask & BITFIELD64_BIT(c->idx_to_attr[reg*2])) *pc_persp = 0xf; - if (linear_mask & (1 << c->idx_to_attr[reg*2])) + if (linear_mask & BITFIELD64_BIT(c->idx_to_attr[reg*2])) *pc_linear = 0xf; /* Maybe only processs one attribute on the final round: @@ -342,10 +344,10 @@ static GLboolean calculate_masks( struct brw_sf_compile *c, if (reg*2+1 < c->nr_setup_attrs) { *pc |= 0xf0; - if (persp_mask & (1 << c->idx_to_attr[reg*2+1])) + if (persp_mask & BITFIELD64_BIT(c->idx_to_attr[reg*2+1])) *pc_persp |= 0xf0; - if (linear_mask & (1 << c->idx_to_attr[reg*2+1])) + if (linear_mask & BITFIELD64_BIT(c->idx_to_attr[reg*2+1])) *pc_linear |= 0xf0; } @@ -551,7 +553,7 @@ void brw_emit_point_sprite_setup( struct brw_sf_compile *c, GLboolean allocate) BRW_MATH_DATA_SCALAR, BRW_MATH_PRECISION_FULL); - if (c->key.SpriteOrigin == GL_LOWER_LEFT) { + if (c->key.sprite_origin_lower_left) { brw_MUL(p, c->m1Cx, c->tmp, c->inv_w[0]); brw_MOV(p, vec1(suboffset(c->m1Cx, 1)), brw_imm_f(0.0)); brw_MUL(p, c->m2Cy, c->tmp, negate(c->inv_w[0])); @@ -570,7 +572,7 @@ void brw_emit_point_sprite_setup( struct brw_sf_compile *c, GLboolean allocate) { brw_set_predicate_control_flag_value(p, pc); if (tex->CoordReplace) { - if (c->key.SpriteOrigin == GL_LOWER_LEFT) { + if (c->key.sprite_origin_lower_left) { brw_MUL(p, c->m3C0, c->inv_w[0], brw_imm_f(1.0)); brw_MOV(p, vec1(suboffset(c->m3C0, 0)), brw_imm_f(0.0)); } diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c index bc0f076073..09223b7cfb 100644 --- a/src/mesa/drivers/dri/i965/brw_sf_state.c +++ b/src/mesa/drivers/dri/i965/brw_sf_state.c @@ -93,7 +93,8 @@ static void upload_sf_vp(struct brw_context *brw) } dri_bo_unreference(brw->sf.vp_bo); - brw->sf.vp_bo = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0 ); + brw->sf.vp_bo = brw_cache_data(&brw->cache, BRW_SF_VP, &sfv, sizeof(sfv), + NULL, 0); } const struct brw_tracked_state brw_sf_vp = { @@ -113,7 +114,8 @@ struct brw_sf_unit_key { unsigned int nr_urb_entries, urb_size, sfsize; - GLenum front_face, cull_face, provoking_vertex; + GLenum front_face, cull_face; + unsigned pv_first:1; unsigned scissor:1; unsigned line_smooth:1; unsigned point_sprite:1; @@ -154,7 +156,7 @@ sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key) key->point_attenuated = ctx->Point._Attenuated; /* _NEW_LIGHT */ - key->provoking_vertex = ctx->Light.ProvokingVertex; + key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION); key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0; } @@ -163,6 +165,7 @@ static dri_bo * sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, dri_bo **reloc_bufs) { + struct intel_context *intel = &brw->intel; struct brw_sf_unit_state sf; dri_bo *bo; int chipset_max_threads; @@ -175,7 +178,7 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, sf.thread3.dispatch_grf_start_reg = 3; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) sf.thread3.urb_entry_read_offset = 3; else sf.thread3.urb_entry_read_offset = 1; @@ -185,10 +188,10 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, sf.thread4.nr_urb_entries = key->nr_urb_entries; sf.thread4.urb_entry_allocation_size = key->sfsize - 1; - /* Each SF thread produces 1 PUE, and there can be up to 24(Pre-IGDNG) or - * 48(IGDNG) threads + /* Each SF thread produces 1 PUE, and there can be up to 24 (Pre-Ironlake) or + * 48 (Ironlake) threads. */ - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) chipset_max_threads = 48; else chipset_max_threads = 24; @@ -287,7 +290,7 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons: */ - if (key->provoking_vertex == GL_LAST_VERTEX_CONVENTION) { + if (!key->pv_first) { sf.sf7.trifan_pv = 2; sf.sf7.linestrip_pv = 1; sf.sf7.tristrip_pv = 2; diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h index d639656b9d..9c9d145c4b 100644 --- a/src/mesa/drivers/dri/i965/brw_state.h +++ b/src/mesa/drivers/dri/i965/brw_state.h @@ -35,7 +35,7 @@ #include "brw_context.h" -static inline void +static INLINE void brw_add_validated_bo(struct brw_context *brw, dri_bo *bo) { assert(brw->state.validated_bo_count < ARRAY_SIZE(brw->state.validated_bos)); @@ -112,6 +112,7 @@ void brw_validate_state(struct brw_context *brw); void brw_upload_state(struct brw_context *brw); void brw_init_state(struct brw_context *brw); void brw_destroy_state(struct brw_context *brw); +void brw_clear_validated_bos(struct brw_context *brw); /*********************************************************************** * brw_state_cache.c @@ -119,16 +120,10 @@ void brw_destroy_state(struct brw_context *brw); dri_bo *brw_cache_data(struct brw_cache *cache, enum brw_cache_id cache_id, const void *data, + GLuint size, dri_bo **reloc_bufs, GLuint nr_reloc_bufs); -dri_bo *brw_cache_data_sz(struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *data, - GLuint data_size, - dri_bo **reloc_bufs, - GLuint nr_reloc_bufs); - dri_bo *brw_upload_cache( struct brw_cache *cache, enum brw_cache_id cache_id, const void *key, @@ -156,7 +151,7 @@ void brw_state_cache_bo_delete(struct brw_cache *cache, dri_bo *bo); /*********************************************************************** * brw_state_batch.c */ -#define BRW_BATCH_STRUCT(brw, s) intel_batchbuffer_data( brw->intel.batch, (s), sizeof(*(s)), IGNORE_CLIPRECTS) +#define BRW_BATCH_STRUCT(brw, s) intel_batchbuffer_data( brw->intel.batch, (s), sizeof(*(s))) #define BRW_CACHED_BATCH_STRUCT(brw, s) brw_cached_batch_struct( brw, (s), sizeof(*(s)) ) GLboolean brw_cached_batch_struct( struct brw_context *brw, diff --git a/src/mesa/drivers/dri/i965/brw_state_batch.c b/src/mesa/drivers/dri/i965/brw_state_batch.c index 7821898cf9..ed8120d617 100644 --- a/src/mesa/drivers/dri/i965/brw_state_batch.c +++ b/src/mesa/drivers/dri/i965/brw_state_batch.c @@ -48,7 +48,7 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw, struct header *newheader = (struct header *)data; if (brw->emit_state_always) { - intel_batchbuffer_data(brw->intel.batch, data, sz, IGNORE_CLIPRECTS); + intel_batchbuffer_data(brw->intel.batch, data, sz); return GL_TRUE; } @@ -75,7 +75,7 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw, emit: memcpy(item->header, newheader, sz); - intel_batchbuffer_data(brw->intel.batch, data, sz, IGNORE_CLIPRECTS); + intel_batchbuffer_data(brw->intel.batch, data, sz); return GL_TRUE; } diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c index f8e46aacf7..e4c9ba7d87 100644 --- a/src/mesa/drivers/dri/i965/brw_state_cache.c +++ b/src/mesa/drivers/dri/i965/brw_state_cache.c @@ -245,7 +245,6 @@ brw_upload_cache( struct brw_cache *cache, item->bo = bo; dri_bo_reference(bo); - item->data_size = data_size; if (cache->n_items > cache->size * 1.5) rehash(cache); @@ -275,15 +274,22 @@ brw_upload_cache( struct brw_cache *cache, /** - * This doesn't really work with aux data. Use search/upload instead + * Wrapper around brw_cache_data_sz using the cache_id's canonical key size. + * + * If nr_reloc_bufs is nonzero, brw_search_cache()/brw_upload_cache() would be + * better to use, as the potentially changing offsets in the data-used-as-key + * will result in excessive cache misses. + * + * If aux data is involved, use search/upload instead. + */ dri_bo * -brw_cache_data_sz(struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *data, - GLuint data_size, - dri_bo **reloc_bufs, - GLuint nr_reloc_bufs) +brw_cache_data(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *data, + GLuint data_size, + dri_bo **reloc_bufs, + GLuint nr_reloc_bufs) { dri_bo *bo; struct brw_cache_item *item; @@ -306,25 +312,6 @@ brw_cache_data_sz(struct brw_cache *cache, return bo; } - -/** - * Wrapper around brw_cache_data_sz using the cache_id's canonical key size. - * - * If nr_reloc_bufs is nonzero, brw_search_cache()/brw_upload_cache() would be - * better to use, as the potentially changing offsets in the data-used-as-key - * will result in excessive cache misses. - */ -dri_bo * -brw_cache_data(struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *data, - dri_bo **reloc_bufs, - GLuint nr_reloc_bufs) -{ - return brw_cache_data_sz(cache, cache_id, data, cache->key_size[cache_id], - reloc_bufs, nr_reloc_bufs); -} - enum pool_type { DW_SURFACE_STATE, DW_GENERAL_STATE @@ -335,11 +322,9 @@ static void brw_init_cache_id(struct brw_cache *cache, const char *name, enum brw_cache_id id, - GLuint key_size, GLuint aux_size) { cache->name[id] = strdup(name); - cache->key_size[id] = key_size; cache->aux_size[id] = aux_size; } @@ -359,91 +344,76 @@ brw_init_non_surface_cache(struct brw_context *brw) brw_init_cache_id(cache, "CC_VP", BRW_CC_VP, - sizeof(struct brw_cc_viewport), 0); brw_init_cache_id(cache, "CC_UNIT", BRW_CC_UNIT, - sizeof(struct brw_cc_unit_state), 0); brw_init_cache_id(cache, "WM_PROG", BRW_WM_PROG, - sizeof(struct brw_wm_prog_key), sizeof(struct brw_wm_prog_data)); brw_init_cache_id(cache, "SAMPLER_DEFAULT_COLOR", BRW_SAMPLER_DEFAULT_COLOR, - sizeof(struct brw_sampler_default_color), 0); brw_init_cache_id(cache, "SAMPLER", BRW_SAMPLER, - 0, /* variable key/data size */ 0); brw_init_cache_id(cache, "WM_UNIT", BRW_WM_UNIT, - sizeof(struct brw_wm_unit_state), 0); brw_init_cache_id(cache, "SF_PROG", BRW_SF_PROG, - sizeof(struct brw_sf_prog_key), sizeof(struct brw_sf_prog_data)); brw_init_cache_id(cache, "SF_VP", BRW_SF_VP, - sizeof(struct brw_sf_viewport), 0); brw_init_cache_id(cache, "SF_UNIT", BRW_SF_UNIT, - sizeof(struct brw_sf_unit_state), 0); brw_init_cache_id(cache, "VS_UNIT", BRW_VS_UNIT, - sizeof(struct brw_vs_unit_state), 0); brw_init_cache_id(cache, "VS_PROG", BRW_VS_PROG, - sizeof(struct brw_vs_prog_key), sizeof(struct brw_vs_prog_data)); brw_init_cache_id(cache, "CLIP_UNIT", BRW_CLIP_UNIT, - sizeof(struct brw_clip_unit_state), 0); brw_init_cache_id(cache, "CLIP_PROG", BRW_CLIP_PROG, - sizeof(struct brw_clip_prog_key), sizeof(struct brw_clip_prog_data)); brw_init_cache_id(cache, "GS_UNIT", BRW_GS_UNIT, - sizeof(struct brw_gs_unit_state), 0); brw_init_cache_id(cache, "GS_PROG", BRW_GS_PROG, - sizeof(struct brw_gs_prog_key), sizeof(struct brw_gs_prog_data)); } @@ -463,13 +433,11 @@ brw_init_surface_cache(struct brw_context *brw) brw_init_cache_id(cache, "SS_SURFACE", BRW_SS_SURFACE, - sizeof(struct brw_surface_state), 0); brw_init_cache_id(cache, "SS_SURF_BIND", BRW_SS_SURF_BIND, - 0, 0); } @@ -534,14 +502,9 @@ brw_state_cache_bo_delete(struct brw_cache *cache, dri_bo *bo) for (i = 0; i < cache->size; i++) { for (prev = &cache->items[i]; *prev;) { struct brw_cache_item *c = *prev; - int j; - - for (j = 0; j < c->nr_reloc_bufs; j++) { - if (c->reloc_bufs[j] == bo) - break; - } - if (j != c->nr_reloc_bufs) { + if (drm_intel_bo_references(c->bo, bo)) { + int j; *prev = c->next; @@ -551,17 +514,8 @@ brw_state_cache_bo_delete(struct brw_cache *cache, dri_bo *bo) free((void *)c->key); free(c); cache->n_items--; - - /* Delete up the tree. Notably we're trying to get from - * a request to delete the surface, to deleting the surface state - * object, to deleting the binding table. We're slack and restart - * the deletion process when we do this because the other delete - * may kill our *prev. - */ - brw_state_cache_bo_delete(cache, c->bo); - prev = &cache->items[i]; } else { - prev = &(*prev)->next; + prev = &c->next; } } } diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c index b817b741e7..af8dfb4c15 100644 --- a/src/mesa/drivers/dri/i965/brw_state_upload.c +++ b/src/mesa/drivers/dri/i965/brw_state_upload.c @@ -34,6 +34,7 @@ #include "brw_context.h" #include "brw_state.h" #include "intel_batchbuffer.h" +#include "intel_buffers.h" /* This is used to initialize brw->state.atoms[]. We could use this * list directly except for a single atom, brw_constant_buffer, which @@ -142,7 +143,7 @@ static void xor_states( struct brw_state_flags *result, result->cache = a->cache ^ b->cache; } -static void +void brw_clear_validated_bos(struct brw_context *brw) { int i; @@ -308,7 +309,7 @@ void brw_validate_state( struct brw_context *brw ) if (brw->state.dirty.brw & BRW_NEW_CONTEXT) brw_clear_batch_cache(brw); - brw->intel.Fallback = 0; + brw->intel.Fallback = GL_FALSE; /* boolean, not bitfield */ /* do prepare stage for all atoms */ for (i = 0; i < Elements(atoms); i++) { @@ -324,6 +325,8 @@ void brw_validate_state( struct brw_context *brw ) } } + intel_check_front_buffer_rendering(intel); + /* Make sure that the textures which are referenced by the current * brw fragment program are actually present/valid. * If this fails, we can experience GPU lock-ups. diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c index 5986cbffad..64a9535282 100644 --- a/src/mesa/drivers/dri/i965/brw_tex_layout.c +++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c @@ -49,7 +49,7 @@ GLboolean brw_miptree_layout(struct intel_context *intel, switch (mt->target) { case GL_TEXTURE_CUBE_MAP: - if (IS_IGDNG(intel->intelScreen->deviceID)) { + if (intel->is_ironlake) { GLuint align_h = 2, align_w = 4; GLuint level; GLuint x = 0; @@ -86,10 +86,10 @@ GLboolean brw_miptree_layout(struct intel_context *intel, mt->pitch = intel_miptree_pitch_align(intel, mt, tiling, mt->pitch); if (mt->compressed) { - qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * mt->pitch * mt->cpp; + qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4; mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * 6; } else { - qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * mt->pitch * mt->cpp; + qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h); mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * 6; } @@ -102,7 +102,8 @@ GLboolean brw_miptree_layout(struct intel_context *intel, height, 1); for (q = 0; q < nr_images; q++) - intel_miptree_set_image_offset_ex(mt, level, q, x, y, q * qpitch); + intel_miptree_set_image_offset(mt, level, q, + x, y + q * qpitch); if (mt->compressed) img_height = MAX2(1, height/4); diff --git a/src/mesa/drivers/dri/i965/brw_urb.c b/src/mesa/drivers/dri/i965/brw_urb.c index 8c6f4355a6..f2cdb203b8 100644 --- a/src/mesa/drivers/dri/i965/brw_urb.c +++ b/src/mesa/drivers/dri/i965/brw_urb.c @@ -105,7 +105,8 @@ static GLboolean check_urb_layout( struct brw_context *brw ) brw->urb.sf_start = brw->urb.clip_start + brw->urb.nr_clip_entries * brw->urb.vsize; brw->urb.cs_start = brw->urb.sf_start + brw->urb.nr_sf_entries * brw->urb.sfsize; - return brw->urb.cs_start + brw->urb.nr_cs_entries * brw->urb.csize <= URB_SIZES(brw); + return brw->urb.cs_start + brw->urb.nr_cs_entries * + brw->urb.csize <= brw->urb.size; } /* Most minimal update, forces re-emit of URB fence packet after GS @@ -113,6 +114,7 @@ static GLboolean check_urb_layout( struct brw_context *brw ) */ static void recalculate_urb_fence( struct brw_context *brw ) { + struct intel_context *intel = &brw->intel; GLuint csize = brw->curbe.total_size; GLuint vsize = brw->vs.prog_data->urb_entry_size; GLuint sfsize = brw->sf.prog_data->urb_entry_size; @@ -146,7 +148,7 @@ static void recalculate_urb_fence( struct brw_context *brw ) brw->urb.constrained = 0; - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { brw->urb.nr_vs_entries = 128; brw->urb.nr_sf_entries = 48; if (check_urb_layout(brw)) { @@ -156,7 +158,7 @@ static void recalculate_urb_fence( struct brw_context *brw ) brw->urb.nr_vs_entries = limits[VS].preferred_nr_entries; brw->urb.nr_sf_entries = limits[SF].preferred_nr_entries; } - } else if (BRW_IS_G4X(brw)) { + } else if (intel->is_g4x) { brw->urb.nr_vs_entries = 64; if (check_urb_layout(brw)) { goto done; @@ -200,7 +202,7 @@ done: brw->urb.clip_start, brw->urb.sf_start, brw->urb.cs_start, - URB_SIZES(brw)); + brw->urb.size); brw->state.dirty.brw |= BRW_NEW_URB_FENCE; } @@ -244,7 +246,7 @@ void brw_upload_urb_fence(struct brw_context *brw) uf.bits0.gs_fence = brw->urb.clip_start; uf.bits0.clp_fence = brw->urb.sf_start; uf.bits1.sf_fence = brw->urb.cs_start; - uf.bits1.cs_fence = URB_SIZES(brw); + uf.bits1.cs_fence = brw->urb.size; BRW_BATCH_STRUCT(brw, &uf); } diff --git a/src/mesa/drivers/dri/i965/brw_util.c b/src/mesa/drivers/dri/i965/brw_util.c index ce21aa4869..bba9249d1b 100644 --- a/src/mesa/drivers/dri/i965/brw_util.c +++ b/src/mesa/drivers/dri/i965/brw_util.c @@ -35,7 +35,7 @@ #include "brw_util.h" #include "brw_defines.h" -GLuint brw_count_bits( GLuint val ) +GLuint brw_count_bits(uint64_t val) { GLuint i; for (i = 0; val ; val >>= 1) diff --git a/src/mesa/drivers/dri/i965/brw_util.h b/src/mesa/drivers/dri/i965/brw_util.h index 33e7cd87e4..04f3175d3e 100644 --- a/src/mesa/drivers/dri/i965/brw_util.h +++ b/src/mesa/drivers/dri/i965/brw_util.h @@ -35,7 +35,7 @@ #include "main/mtypes.h" -extern GLuint brw_count_bits( GLuint val ); +extern GLuint brw_count_bits(uint64_t val); extern GLuint brw_parameter_list_state_flags(struct gl_program_parameter_list *paramList); extern GLuint brw_translate_blend_factor( GLenum factor ); extern GLuint brw_translate_blend_equation( GLenum mode ); diff --git a/src/mesa/drivers/dri/i965/brw_vs.c b/src/mesa/drivers/dri/i965/brw_vs.c index f0c79efbd9..fd055e225e 100644 --- a/src/mesa/drivers/dri/i965/brw_vs.c +++ b/src/mesa/drivers/dri/i965/brw_vs.c @@ -56,7 +56,7 @@ static void do_vs_prog( struct brw_context *brw, c.prog_data.inputs_read = vp->program.Base.InputsRead; if (c.key.copy_edgeflag) { - c.prog_data.outputs_written |= 1<<VERT_RESULT_EDGE; + c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_EDGE); c.prog_data.inputs_read |= 1<<VERT_ATTRIB_EDGEFLAG; } diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c index 1638ef8111..1b84dd505f 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c @@ -67,6 +67,7 @@ static void release_tmps( struct brw_vs_compile *c ) */ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; GLuint i, reg = 0, mrf; int attributes_in_vue; @@ -141,13 +142,13 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) c->first_output = reg; c->first_overflow_output = 0; - if (BRW_IS_IGDNG(c->func.brw)) + if (intel->is_ironlake) mrf = 8; else mrf = 4; for (i = 0; i < VERT_RESULT_MAX; i++) { - if (c->prog_data.outputs_written & (1 << i)) { + if (c->prog_data.outputs_written & BITFIELD64_BIT(i)) { c->nr_outputs++; assert(i < Elements(c->regs[PROGRAM_OUTPUT])); if (i == VERT_RESULT_HPOS) { @@ -238,7 +239,7 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) */ attributes_in_vue = MAX2(c->nr_outputs, c->nr_inputs); - if (BRW_IS_IGDNG(c->func.brw)) + if (intel->is_ironlake) c->prog_data.urb_entry_size = (attributes_in_vue + 6 + 3) / 4; else c->prog_data.urb_entry_size = (attributes_in_vue + 2 + 3) / 4; @@ -331,63 +332,76 @@ static void unalias3( struct brw_vs_compile *c, } } -static void emit_sop( struct brw_compile *p, +static void emit_sop( struct brw_vs_compile *c, struct brw_reg dst, struct brw_reg arg0, struct brw_reg arg1, GLuint cond) { + struct brw_compile *p = &c->func; + brw_MOV(p, dst, brw_imm_f(0.0f)); brw_CMP(p, brw_null_reg(), cond, arg0, arg1); brw_MOV(p, dst, brw_imm_f(1.0f)); brw_set_predicate_control_flag_value(p, 0xff); } -static void emit_seq( struct brw_compile *p, +static void emit_seq( struct brw_vs_compile *c, struct brw_reg dst, struct brw_reg arg0, struct brw_reg arg1 ) { - emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_EQ); + emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_EQ); } -static void emit_sne( struct brw_compile *p, +static void emit_sne( struct brw_vs_compile *c, struct brw_reg dst, struct brw_reg arg0, struct brw_reg arg1 ) { - emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_NEQ); + emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_NEQ); } -static void emit_slt( struct brw_compile *p, +static void emit_slt( struct brw_vs_compile *c, struct brw_reg dst, struct brw_reg arg0, struct brw_reg arg1 ) { - emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_L); + emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_L); } -static void emit_sle( struct brw_compile *p, +static void emit_sle( struct brw_vs_compile *c, struct brw_reg dst, struct brw_reg arg0, struct brw_reg arg1 ) { - emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_LE); + emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_LE); } -static void emit_sgt( struct brw_compile *p, +static void emit_sgt( struct brw_vs_compile *c, struct brw_reg dst, struct brw_reg arg0, struct brw_reg arg1 ) { - emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_G); + emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_G); } -static void emit_sge( struct brw_compile *p, +static void emit_sge( struct brw_vs_compile *c, struct brw_reg dst, struct brw_reg arg0, struct brw_reg arg1 ) { - emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_GE); + emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_GE); +} + +static void emit_cmp( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1, + struct brw_reg arg2 ) +{ + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, brw_imm_f(0)); + brw_SEL(p, dst, arg1, arg2); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); } static void emit_max( struct brw_compile *p, @@ -912,6 +926,7 @@ get_src_reg( struct brw_vs_compile *c, case PROGRAM_CONSTANT: case PROGRAM_UNIFORM: case PROGRAM_ENV_PARAM: + case PROGRAM_LOCAL_PARAM: if (c->vp->use_const_buffer) { return get_constant(c, inst, argIndex); } @@ -930,7 +945,6 @@ get_src_reg( struct brw_vs_compile *c, /* this is a normal case since we loop over all three src args */ return brw_null_reg(); - case PROGRAM_LOCAL_PARAM: case PROGRAM_WRITE_ONLY: default: assert(0); @@ -1100,6 +1114,8 @@ static void emit_swz( struct brw_vs_compile *c, static void emit_vertex_write( struct brw_vs_compile *c) { struct brw_compile *p = &c->func; + struct brw_context *brw = p->brw; + struct intel_context *intel = &brw->intel; struct brw_reg m0 = brw_message_reg(0); struct brw_reg pos = c->regs[PROGRAM_OUTPUT][VERT_RESULT_HPOS]; struct brw_reg ndc; @@ -1122,8 +1138,8 @@ static void emit_vertex_write( struct brw_vs_compile *c) /* Update the header for point size, user clipping flags, and -ve rhw * workaround. */ - if ((c->prog_data.outputs_written & (1<<VERT_RESULT_PSIZ)) || - c->key.nr_userclip || BRW_IS_965(p->brw)) + if ((c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_PSIZ)) || + c->key.nr_userclip || brw->has_negative_rhw_bug) { struct brw_reg header1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); GLuint i; @@ -1132,7 +1148,7 @@ static void emit_vertex_write( struct brw_vs_compile *c) brw_set_access_mode(p, BRW_ALIGN_16); - if (c->prog_data.outputs_written & (1<<VERT_RESULT_PSIZ)) { + if (c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_PSIZ)) { struct brw_reg psiz = c->regs[PROGRAM_OUTPUT][VERT_RESULT_PSIZ]; brw_MUL(p, brw_writemask(header1, WRITEMASK_W), brw_swizzle1(psiz, 0), brw_imm_f(1<<11)); brw_AND(p, brw_writemask(header1, WRITEMASK_W), header1, brw_imm_ud(0x7ff<<8)); @@ -1154,7 +1170,7 @@ static void emit_vertex_write( struct brw_vs_compile *c) * Later, clipping will detect ucp[6] and ensure the primitive is * clipped against all fixed planes. */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_CMP(p, vec8(brw_null_reg()), BRW_CONDITIONAL_L, @@ -1182,8 +1198,8 @@ static void emit_vertex_write( struct brw_vs_compile *c) brw_set_access_mode(p, BRW_ALIGN_1); brw_MOV(p, offset(m0, 2), ndc); - if (BRW_IS_IGDNG(p->brw)) { - /* There are 20 DWs (D0-D19) in VUE vertex header on IGDNG */ + if (intel->is_ironlake) { + /* There are 20 DWs (D0-D19) in VUE vertex header on Ironlake */ brw_MOV(p, offset(m0, 3), pos); /* a portion of vertex header */ /* m4, m5 contain the distances from vertex to the user clip planeXXX. * Seems it is useless for us. @@ -1222,7 +1238,7 @@ static void emit_vertex_write( struct brw_vs_compile *c) */ GLuint i, mrf = 0; for (i = c->first_overflow_output; i < VERT_RESULT_MAX; i++) { - if (c->prog_data.outputs_written & (1 << i)) { + if (c->prog_data.outputs_written & BITFIELD64_BIT(i)) { /* move from GRF to MRF */ brw_MOV(p, brw_message_reg(4+mrf), c->regs[PROGRAM_OUTPUT][i]); mrf++; @@ -1269,6 +1285,38 @@ post_vs_emit( struct brw_vs_compile *c, } } +static GLboolean +accumulator_contains(struct brw_vs_compile *c, struct brw_reg val) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *prev_insn = &p->store[p->nr_insn - 1]; + + if (p->nr_insn == 0) + return GL_FALSE; + + if (val.address_mode != BRW_ADDRESS_DIRECT) + return GL_FALSE; + + switch (prev_insn->header.opcode) { + case BRW_OPCODE_MOV: + case BRW_OPCODE_MAC: + case BRW_OPCODE_MUL: + if (prev_insn->header.access_mode == BRW_ALIGN_16 && + prev_insn->header.execution_size == val.width && + prev_insn->bits1.da1.dest_reg_file == val.file && + prev_insn->bits1.da1.dest_reg_type == val.type && + prev_insn->bits1.da1.dest_address_mode == val.address_mode && + prev_insn->bits1.da1.dest_reg_nr == val.nr && + prev_insn->bits1.da16.dest_subreg_nr == val.subnr / 16 && + prev_insn->bits1.da16.dest_writemask == 0xf) + return GL_TRUE; + else + return GL_FALSE; + default: + return GL_FALSE; + } +} + static uint32_t get_predicate(const struct prog_instruction *inst) { @@ -1314,6 +1362,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) #define MAX_LOOP_DEPTH 32 struct brw_compile *p = &c->func; struct brw_context *brw = p->brw; + struct intel_context *intel = &brw->intel; const GLuint nr_insns = c->vp->program.Base.NumInstructions; GLuint insn, if_depth = 0, loop_depth = 0; GLuint end_offset = 0; @@ -1447,9 +1496,13 @@ void brw_vs_emit(struct brw_vs_compile *c ) unalias3(c, dst, args[0], args[1], args[2], emit_lrp_noalias); break; case OPCODE_MAD: - brw_MOV(p, brw_acc_reg(), args[2]); + if (!accumulator_contains(c, args[2])) + brw_MOV(p, brw_acc_reg(), args[2]); brw_MAC(p, dst, args[0], args[1]); break; + case OPCODE_CMP: + emit_cmp(p, dst, args[0], args[1], args[2]); + break; case OPCODE_MAX: emit_max(p, dst, args[0], args[1]); break; @@ -1473,25 +1526,25 @@ void brw_vs_emit(struct brw_vs_compile *c ) break; case OPCODE_SEQ: - emit_seq(p, dst, args[0], args[1]); + unalias2(c, dst, args[0], args[1], emit_seq); break; case OPCODE_SIN: emit_math1(c, BRW_MATH_FUNCTION_SIN, dst, args[0], BRW_MATH_PRECISION_FULL); break; case OPCODE_SNE: - emit_sne(p, dst, args[0], args[1]); + unalias2(c, dst, args[0], args[1], emit_sne); break; case OPCODE_SGE: - emit_sge(p, dst, args[0], args[1]); + unalias2(c, dst, args[0], args[1], emit_sge); break; case OPCODE_SGT: - emit_sgt(p, dst, args[0], args[1]); + unalias2(c, dst, args[0], args[1], emit_sgt); break; case OPCODE_SLT: - emit_slt(p, dst, args[0], args[1]); + unalias2(c, dst, args[0], args[1], emit_slt); break; case OPCODE_SLE: - emit_sle(p, dst, args[0], args[1]); + unalias2(c, dst, args[0], args[1], emit_sle); break; case OPCODE_SUB: brw_ADD(p, dst, args[0], negate(args[1])); @@ -1543,7 +1596,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) loop_depth--; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) br = 2; inst0 = inst1 = brw_WHILE(p, loop_inst[loop_depth]); diff --git a/src/mesa/drivers/dri/i965/brw_vs_state.c b/src/mesa/drivers/dri/i965/brw_vs_state.c index d790ab6555..345ffa7ee1 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_state.c +++ b/src/mesa/drivers/dri/i965/brw_vs_state.c @@ -82,9 +82,9 @@ vs_unit_populate_key(struct brw_context *brw, struct brw_vs_unit_key *key) static dri_bo * vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) { + struct intel_context *intel = &brw->intel; struct brw_vs_unit_state vs; dri_bo *bo; - int chipset_max_threads; memset(&vs, 0, sizeof(vs)); @@ -98,7 +98,7 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) */ vs.thread1.single_program_flow = 0; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) vs.thread1.binding_table_entry_count = 0; /* hardware requirement */ else vs.thread1.binding_table_entry_count = key->nr_surfaces; @@ -109,24 +109,44 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) vs.thread3.urb_entry_read_offset = 0; vs.thread3.const_urb_entry_read_offset = key->curbe_offset * 2; - if (BRW_IS_IGDNG(brw)) - vs.thread4.nr_urb_entries = key->nr_urb_entries >> 2; - else - vs.thread4.nr_urb_entries = key->nr_urb_entries; + if (intel->is_ironlake) { + switch (key->nr_urb_entries) { + case 8: + case 12: + case 16: + case 32: + case 64: + case 96: + case 128: + case 168: + case 192: + case 224: + case 256: + vs.thread4.nr_urb_entries = key->nr_urb_entries >> 2; + break; + default: + assert(0); + } + } else { + switch (key->nr_urb_entries) { + case 8: + case 12: + case 16: + case 32: + break; + case 64: + assert(intel->is_g4x); + break; + default: + assert(0); + } + vs.thread4.nr_urb_entries = key->nr_urb_entries; + } vs.thread4.urb_entry_allocation_size = key->urb_size - 1; - if (BRW_IS_IGDNG(brw)) - chipset_max_threads = 72; - else if (BRW_IS_G4X(brw)) - chipset_max_threads = 32; - else - chipset_max_threads = 16; vs.thread4.max_threads = CLAMP(key->nr_urb_entries / 2, - 1, chipset_max_threads) - 1; - - if (INTEL_DEBUG & DEBUG_SINGLE_THREAD) - vs.thread4.max_threads = 0; + 1, brw->vs_max_threads) - 1; /* No samplers for ARB_vp programs: */ diff --git a/src/mesa/drivers/dri/i965/brw_vs_surface_state.c b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c index 89f47522a1..3bc9840a97 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c @@ -30,7 +30,6 @@ */ #include "main/mtypes.h" -#include "main/texformat.h" #include "main/texstore.h" #include "shader/prog_parameter.h" @@ -53,6 +52,7 @@ brw_vs_update_constant_buffer(struct brw_context *brw) const struct gl_program_parameter_list *params = vp->program.Base.Parameters; const int size = params->NumParameters * 4 * sizeof(GLfloat); drm_intel_bo *const_buffer; + int i; /* BRW_NEW_VERTEX_PROGRAM */ if (!vp->use_const_buffer) @@ -62,7 +62,19 @@ brw_vs_update_constant_buffer(struct brw_context *brw) size, 64); /* _NEW_PROGRAM_CONSTANTS */ - dri_bo_subdata(const_buffer, 0, size, params->ParameterValues); + + /* Updates the ParamaterValues[i] pointers for all parameters of the + * basic type of PROGRAM_STATE_VAR. + */ + _mesa_load_state_parameters(&brw->intel.ctx, vp->program.Base.Parameters); + + intel_bo_map_gtt_preferred(intel, const_buffer, GL_TRUE); + for (i = 0; i < params->NumParameters; i++) { + memcpy(const_buffer->virtual + i * 4 * sizeof(float), + params->ParameterValues[i], + 4 * sizeof(float)); + } + intel_bo_unmap_gtt_preferred(intel, const_buffer); return const_buffer; } diff --git a/src/mesa/drivers/dri/i965/brw_vtbl.c b/src/mesa/drivers/dri/i965/brw_vtbl.c index 124fde25fe..72749b3859 100644 --- a/src/mesa/drivers/dri/i965/brw_vtbl.c +++ b/src/mesa/drivers/dri/i965/brw_vtbl.c @@ -46,7 +46,7 @@ #include "brw_state.h" #include "brw_fallback.h" #include "brw_vs.h" - +#include "brw_wm.h" static void dri_bo_release(dri_bo **bo) @@ -66,8 +66,14 @@ static void brw_destroy_context( struct intel_context *intel ) brw_destroy_state(brw); brw_draw_destroy( brw ); - - _mesa_free(brw->wm.compile_data); + brw_clear_validated_bos(brw); + if (brw->wm.compile_data) { + _mesa_free(brw->wm.compile_data->instruction); + _mesa_free(brw->wm.compile_data->vreg); + _mesa_free(brw->wm.compile_data->refs); + _mesa_free(brw->wm.compile_data->prog_instructions); + _mesa_free(brw->wm.compile_data); + } for (i = 0; i < brw->state.nr_color_regions; i++) intel_region_release(&brw->state.color_regions[i]); @@ -144,9 +150,6 @@ static void brw_new_batch( struct intel_context *intel ) { struct brw_context *brw = brw_context(&intel->ctx); - /* Check that we didn't just wrap our batchbuffer at a bad time. */ - assert(!brw->no_batch_wrap); - brw->curbe.need_new_bo = GL_TRUE; /* Mark all context state as needing to be re-emitted. @@ -175,20 +178,6 @@ static void brw_note_fence( struct intel_context *intel, GLuint fence ) brw_context(&intel->ctx)->state.dirty.brw |= BRW_NEW_FENCE; } -/* called from intelWaitForIdle() and intelFlush() - * - * For now, just flush everything. Could be smarter later. - */ -static GLuint brw_flush_cmd( void ) -{ - struct brw_mi_flush flush; - flush.opcode = CMD_MI_FLUSH; - flush.pad = 0; - flush.flags = BRW_FLUSH_STATE_CACHE; - return *(GLuint *)&flush; -} - - static void brw_invalidate_state( struct intel_context *intel, GLuint new_state ) { /* nothing */ @@ -209,6 +198,5 @@ void brwInitVtbl( struct brw_context *brw ) brw->intel.vtbl.finish_batch = brw_finish_batch; brw->intel.vtbl.destroy = brw_destroy_context; brw->intel.vtbl.set_draw_region = brw_set_draw_region; - brw->intel.vtbl.flush_cmd = brw_flush_cmd; brw->intel.vtbl.debug_batch = brw_debug_batch; } diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c index 2292de94c4..6895f64410 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.c +++ b/src/mesa/drivers/dri/i965/brw_wm.c @@ -29,7 +29,6 @@ * Keith Whitwell <keith@tungstengraphics.com> */ -#include "main/texformat.h" #include "brw_context.h" #include "brw_util.h" #include "brw_wm.h" @@ -153,8 +152,21 @@ static void do_wm_prog( struct brw_context *brw, */ return; } + c->instruction = _mesa_calloc(BRW_WM_MAX_INSN * sizeof(*c->instruction)); + c->prog_instructions = _mesa_calloc(BRW_WM_MAX_INSN * + sizeof(*c->prog_instructions)); + c->vreg = _mesa_calloc(BRW_WM_MAX_VREG * sizeof(*c->vreg)); + c->refs = _mesa_calloc(BRW_WM_MAX_REF * sizeof(*c->refs)); } else { + void *instruction = c->instruction; + void *prog_instructions = c->prog_instructions; + void *vreg = c->vreg; + void *refs = c->refs; memset(c, 0, sizeof(*brw->wm.compile_data)); + c->instruction = instruction; + c->prog_instructions = prog_instructions; + c->vreg = vreg; + c->refs = refs; } memcpy(&c->key, key, sizeof(*key)); @@ -218,7 +230,7 @@ static void brw_wm_populate_key( struct brw_context *brw, ctx->Color.AlphaEnabled) lookup |= IZ_PS_KILL_ALPHATEST_BIT; - if (fp->program.Base.OutputsWritten & (1<<FRAG_RESULT_DEPTH)) + if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) lookup |= IZ_PS_COMPUTES_DEPTH_BIT; /* _NEW_DEPTH */ @@ -288,7 +300,7 @@ static void brw_wm_populate_key( struct brw_context *brw, const struct gl_texture_image *img = t->Image[0][t->BaseLevel]; if (img->InternalFormat == GL_YCBCR_MESA) { key->yuvtex_mask |= 1 << i; - if (img->TexFormat->MesaFormat == MESA_FORMAT_YCBCR) + if (img->TexFormat == MESA_FORMAT_YCBCR) key->yuvtex_swap_mask |= 1 << i; } @@ -309,6 +321,9 @@ static void brw_wm_populate_key( struct brw_context *brw, * from the incoming screen origin relative position we get as part of our * payload. * + * This is only needed for the WM_WPOSXY opcode when the fragment program + * uses the gl_FragCoord input. + * * We could avoid recompiling by including this as a constant referenced by * our program, but if we were to do that it would also be nice to handle * getting that constant updated at batchbuffer submit time (when we @@ -317,17 +332,21 @@ static void brw_wm_populate_key( struct brw_context *brw, * just avoid using this as key data if the program doesn't use * fragment.position. * - * This pretty much becomes moot with DRI2 and redirected buffers anyway, - * as our origins will always be zero then. + * For DRI2 the origin_x/y will always be (0,0) but we still need the + * drawable height in order to invert the Y axis. */ - if (brw->intel.driDrawable != NULL) { - key->origin_x = brw->intel.driDrawable->x; - key->origin_y = brw->intel.driDrawable->y; - key->drawable_height = brw->intel.driDrawable->h; + if (fp->program.Base.InputsRead & FRAG_BIT_WPOS) { + if (brw->intel.driDrawable != NULL) { + key->origin_x = brw->intel.driDrawable->x; + key->origin_y = brw->intel.driDrawable->y; + key->drawable_height = brw->intel.driDrawable->h; + } } + key->nr_color_regions = brw->state.nr_color_regions; + /* CACHE_NEW_VS_PROG */ - key->vp_outputs_written = brw->vs.prog_data->outputs_written & DO_SETUP_BITS; + key->vp_outputs_written = brw->vs.prog_data->outputs_written; /* The unique fragment program ID */ key->program_string_id = fp->id; diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h index 872b1f3ecf..b9b987ea70 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.h +++ b/src/mesa/drivers/dri/i965/brw_wm.h @@ -67,18 +67,19 @@ struct brw_wm_prog_key { GLuint flat_shade:1; GLuint linear_color:1; /**< linear interpolation vs perspective interp */ GLuint runtime_check_aads_emit:1; + GLuint nr_color_regions:2; GLbitfield proj_attrib_mask; /**< one bit per fragment program attribute */ GLuint shadowtex_mask:16; GLuint yuvtex_mask:16; GLuint yuvtex_swap_mask:16; /* UV swaped */ - GLuint tex_swizzles[BRW_MAX_TEX_UNIT]; + GLushort tex_swizzles[BRW_MAX_TEX_UNIT]; + GLushort origin_x, origin_y; + GLushort drawable_height; + GLbitfield64 vp_outputs_written; GLuint program_string_id:32; - GLuint origin_x, origin_y; - GLuint drawable_height; - GLuint vp_outputs_written; }; @@ -153,15 +154,16 @@ struct brw_wm_instruction { }; -#define BRW_WM_MAX_INSN (MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS*3 + FRAG_ATTRIB_MAX + 3) +#define BRW_WM_MAX_INSN (MAX_PROGRAM_INSTRUCTIONS*3 + FRAG_ATTRIB_MAX + 3) #define BRW_WM_MAX_GRF 128 /* hardware limit */ #define BRW_WM_MAX_VREG (BRW_WM_MAX_INSN * 4) #define BRW_WM_MAX_REF (BRW_WM_MAX_INSN * 12) #define BRW_WM_MAX_PARAM 256 #define BRW_WM_MAX_CONST 256 -#define BRW_WM_MAX_KILLS MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS #define BRW_WM_MAX_SUBROUTINE 16 +/* used in masks next to WRITEMASK_*. */ +#define SATURATE (1<<5) /* New opcodes to track internal operations required for WM unit. @@ -200,7 +202,7 @@ struct brw_wm_compile { * simplifying and adding instructions for interpolation and * framebuffer writes. */ - struct prog_instruction prog_instructions[BRW_WM_MAX_INSN]; + struct prog_instruction *prog_instructions; GLuint nr_fp_insns; GLuint fp_temp; GLuint fp_interp_emitted; @@ -211,7 +213,7 @@ struct brw_wm_compile { struct prog_src_register pixel_w; - struct brw_wm_value vreg[BRW_WM_MAX_VREG]; + struct brw_wm_value *vreg; GLuint nr_vreg; struct brw_wm_value creg[BRW_WM_MAX_PARAM]; @@ -228,10 +230,10 @@ struct brw_wm_compile { struct brw_wm_ref undef_ref; struct brw_wm_value undef_value; - struct brw_wm_ref refs[BRW_WM_MAX_REF]; + struct brw_wm_ref *refs; GLuint nr_refs; - struct brw_wm_instruction instruction[BRW_WM_MAX_INSN]; + struct brw_wm_instruction *instruction; GLuint nr_insns; struct brw_wm_constref constref[BRW_WM_MAX_CONST]; @@ -271,6 +273,12 @@ struct brw_wm_compile { }; +/** Bits for prog_instruction::Aux field */ +#define INST_AUX_EOT 0x1 +#define INST_AUX_TARGET(T) (T << 1) +#define INST_AUX_GET_TARGET(AUX) ((AUX) >> 1) + + GLuint brw_wm_nr_args( GLuint opcode ); GLuint brw_wm_is_scalar_result( GLuint opcode ); @@ -300,10 +308,141 @@ void brw_wm_lookup_iz( GLuint line_aa, GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp); void brw_wm_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c); +/* brw_wm_emit.c */ +void emit_alu1(struct brw_compile *p, + struct brw_instruction *(*func)(struct brw_compile *, + struct brw_reg, + struct brw_reg), + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0); +void emit_alu2(struct brw_compile *p, + struct brw_instruction *(*func)(struct brw_compile *, + struct brw_reg, + struct brw_reg, + struct brw_reg), + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1); +void emit_cinterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0); void emit_ddxy(struct brw_compile *p, const struct brw_reg *dst, GLuint mask, GLboolean is_ddx, const struct brw_reg *arg0); +void emit_delta_xy(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0); +void emit_dp3(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1); +void emit_dp4(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1); +void emit_dph(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1); +void emit_fb_write(struct brw_wm_compile *c, + struct brw_reg *arg0, + struct brw_reg *arg1, + struct brw_reg *arg2, + GLuint target, + GLuint eot); +void emit_frontfacing(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask); +void emit_linterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas); +void emit_lrp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1, + const struct brw_reg *arg2); +void emit_mad(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1, + const struct brw_reg *arg2); +void emit_math1(struct brw_wm_compile *c, + GLuint function, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0); +void emit_math2(struct brw_wm_compile *c, + GLuint function, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1); +void emit_min(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1); +void emit_max(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1); +void emit_pinterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas, + const struct brw_reg *w); +void emit_pixel_xy(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask); +void emit_pixel_w(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas); +void emit_sop(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + GLuint cond, + const struct brw_reg *arg0, + const struct brw_reg *arg1); +void emit_tex(struct brw_wm_compile *c, + struct brw_reg *dst, + GLuint dst_flags, + struct brw_reg *arg, + struct brw_reg depth_payload, + GLuint tex_idx, + GLuint sampler, + GLboolean shadow); +void emit_txb(struct brw_wm_compile *c, + struct brw_reg *dst, + GLuint dst_flags, + struct brw_reg *arg, + struct brw_reg depth_payload, + GLuint tex_idx, + GLuint sampler); +void emit_wpos_xy(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0); +void emit_xpd(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1); #endif diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c index bf80a2942a..cc1052f757 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_emit.c +++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c @@ -44,6 +44,7 @@ static INLINE struct brw_reg sechalf( struct brw_reg reg ) return reg; } + /* Payload R0: * * R0.0 -- pixel mask, one bit for each of 4 pixels in 4 tiles, @@ -60,42 +61,50 @@ static INLINE struct brw_reg sechalf( struct brw_reg reg ) * R1.8 -- ? */ - -static void emit_pixel_xy(struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask) +void emit_pixel_xy(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask) { + struct brw_compile *p = &c->func; struct brw_reg r1 = brw_vec1_grf(1, 0); struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW); + struct brw_reg dst0_uw, dst1_uw; + brw_push_insn_state(p); brw_set_compression_control(p, BRW_COMPRESSION_NONE); + if (c->dispatch_width == 16) { + dst0_uw = vec16(retype(dst[0], BRW_REGISTER_TYPE_UW)); + dst1_uw = vec16(retype(dst[1], BRW_REGISTER_TYPE_UW)); + } else { + dst0_uw = vec8(retype(dst[0], BRW_REGISTER_TYPE_UW)); + dst1_uw = vec8(retype(dst[1], BRW_REGISTER_TYPE_UW)); + } + /* Calculate pixel centers by adding 1 or 0 to each of the * micro-tile coordinates passed in r1. */ if (mask & WRITEMASK_X) { brw_ADD(p, - vec16(retype(dst[0], BRW_REGISTER_TYPE_UW)), + dst0_uw, stride(suboffset(r1_uw, 4), 2, 4, 0), brw_imm_v(0x10101010)); } if (mask & WRITEMASK_Y) { brw_ADD(p, - vec16(retype(dst[1], BRW_REGISTER_TYPE_UW)), + dst1_uw, stride(suboffset(r1_uw,5), 2, 4, 0), brw_imm_v(0x11001100)); } - - brw_set_compression_control(p, BRW_COMPRESSION_COMPRESSED); + brw_pop_insn_state(p); } - -static void emit_delta_xy(struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0) +void emit_delta_xy(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { struct brw_reg r1 = brw_vec1_grf(1, 0); @@ -118,10 +127,10 @@ static void emit_delta_xy(struct brw_compile *p, } } -static void emit_wpos_xy(struct brw_wm_compile *c, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0) +void emit_wpos_xy(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { struct brw_compile *p = &c->func; @@ -146,12 +155,14 @@ static void emit_wpos_xy(struct brw_wm_compile *c, } -static void emit_pixel_w( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *deltas) +void emit_pixel_w(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas) { + struct brw_compile *p = &c->func; + /* Don't need this if all you are doing is interpolating color, for * instance. */ @@ -165,21 +176,29 @@ static void emit_pixel_w( struct brw_compile *p, brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), deltas[1]); /* Calc w */ - brw_math_16( p, dst[3], - BRW_MATH_FUNCTION_INV, - BRW_MATH_SATURATE_NONE, - 2, brw_null_reg(), - BRW_MATH_PRECISION_FULL); + if (c->dispatch_width == 16) { + brw_math_16(p, dst[3], + BRW_MATH_FUNCTION_INV, + BRW_MATH_SATURATE_NONE, + 2, brw_null_reg(), + BRW_MATH_PRECISION_FULL); + } else { + brw_math(p, dst[3], + BRW_MATH_FUNCTION_INV, + BRW_MATH_SATURATE_NONE, + 2, brw_null_reg(), + BRW_MATH_DATA_VECTOR, + BRW_MATH_PRECISION_FULL); + } } } - -static void emit_linterp( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *deltas ) +void emit_linterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas) { struct brw_reg interp[4]; GLuint nr = arg0[0].nr; @@ -199,12 +218,12 @@ static void emit_linterp( struct brw_compile *p, } -static void emit_pinterp( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *deltas, - const struct brw_reg *w) +void emit_pinterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas, + const struct brw_reg *w) { struct brw_reg interp[4]; GLuint nr = arg0[0].nr; @@ -229,10 +248,10 @@ static void emit_pinterp( struct brw_compile *p, } -static void emit_cinterp( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0 ) +void emit_cinterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { struct brw_reg interp[4]; GLuint nr = arg0[0].nr; @@ -251,9 +270,9 @@ static void emit_cinterp( struct brw_compile *p, } /* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */ -static void emit_frontfacing( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask ) +void emit_frontfacing(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask) { struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD); GLuint i; @@ -352,13 +371,13 @@ void emit_ddxy(struct brw_compile *p, brw_set_saturate(p, 0); } -static void emit_alu1( struct brw_compile *p, - struct brw_instruction *(*func)(struct brw_compile *, - struct brw_reg, - struct brw_reg), - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0 ) +void emit_alu1(struct brw_compile *p, + struct brw_instruction *(*func)(struct brw_compile *, + struct brw_reg, + struct brw_reg), + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { GLuint i; @@ -376,15 +395,15 @@ static void emit_alu1( struct brw_compile *p, } -static void emit_alu2( struct brw_compile *p, - struct brw_instruction *(*func)(struct brw_compile *, - struct brw_reg, - struct brw_reg, - struct brw_reg), - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1 ) +void emit_alu2(struct brw_compile *p, + struct brw_instruction *(*func)(struct brw_compile *, + struct brw_reg, + struct brw_reg, + struct brw_reg), + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { GLuint i; @@ -402,12 +421,12 @@ static void emit_alu2( struct brw_compile *p, } -static void emit_mad( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1, - const struct brw_reg *arg2 ) +void emit_mad(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1, + const struct brw_reg *arg2) { GLuint i; @@ -422,26 +441,12 @@ static void emit_mad( struct brw_compile *p, } } -static void emit_trunc( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0) -{ - GLuint i; - - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - brw_RNDZ(p, dst[i], arg0[i]); - } - } -} - -static void emit_lrp( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1, - const struct brw_reg *arg2 ) +void emit_lrp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1, + const struct brw_reg *arg2) { GLuint i; @@ -461,21 +466,24 @@ static void emit_lrp( struct brw_compile *p, } } -static void emit_sop( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - GLuint cond, - const struct brw_reg *arg0, - const struct brw_reg *arg1 ) +void emit_sop(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + GLuint cond, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { GLuint i; for (i = 0; i < 4; i++) { if (mask & (1<<i)) { - brw_MOV(p, dst[i], brw_imm_f(0)); + brw_push_insn_state(p); brw_CMP(p, brw_null_reg(), cond, arg0[i], arg1[i]); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_MOV(p, dst[i], brw_imm_f(0)); + brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); brw_MOV(p, dst[i], brw_imm_f(1.0)); - brw_set_predicate_control_flag_value(p, 0xff); + brw_pop_insn_state(p); } } } @@ -559,11 +567,11 @@ static void emit_cmp( struct brw_compile *p, } } -static void emit_max( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1 ) +void emit_max(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { GLuint i; @@ -583,11 +591,11 @@ static void emit_max( struct brw_compile *p, } } -static void emit_min( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1 ) +void emit_min(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { GLuint i; @@ -608,11 +616,11 @@ static void emit_min( struct brw_compile *p, } -static void emit_dp3( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1 ) +void emit_dp3(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; @@ -630,11 +638,11 @@ static void emit_dp3( struct brw_compile *p, } -static void emit_dp4( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1 ) +void emit_dp4(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; @@ -653,11 +661,11 @@ static void emit_dp4( struct brw_compile *p, } -static void emit_dph( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1 ) +void emit_dph(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { const int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; @@ -676,11 +684,11 @@ static void emit_dph( struct brw_compile *p, } -static void emit_xpd( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1 ) +void emit_xpd(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { GLuint i; @@ -701,41 +709,68 @@ static void emit_xpd( struct brw_compile *p, } -static void emit_math1( struct brw_compile *p, - GLuint function, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0 ) +void emit_math1(struct brw_wm_compile *c, + GLuint function, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { + struct brw_compile *p = &c->func; int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + GLuint saturate = ((mask & SATURATE) ? + BRW_MATH_SATURATE_SATURATE : + BRW_MATH_SATURATE_NONE); if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ assert(is_power_of_two(mask & WRITEMASK_XYZW)); + /* If compressed, this will write message reg 2,3 from arg0.x's 16 + * channels. + */ brw_MOV(p, brw_message_reg(2), arg0[0]); /* Send two messages to perform all 16 operations: */ - brw_math_16(p, - dst[dst_chan], + brw_push_insn_state(p); + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + brw_math(p, + dst[dst_chan], + function, + saturate, + 2, + brw_null_reg(), + BRW_MATH_DATA_VECTOR, + BRW_MATH_PRECISION_FULL); + + if (c->dispatch_width == 16) { + brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); + brw_math(p, + offset(dst[dst_chan],1), function, - (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, - 2, + saturate, + 3, brw_null_reg(), + BRW_MATH_DATA_VECTOR, BRW_MATH_PRECISION_FULL); + } + brw_pop_insn_state(p); } -static void emit_math2( struct brw_compile *p, - GLuint function, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1) +void emit_math2(struct brw_wm_compile *c, + GLuint function, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { + struct brw_compile *p = &c->func; int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + GLuint saturate = ((mask & SATURATE) ? + BRW_MATH_SATURATE_SATURATE : + BRW_MATH_SATURATE_NONE); if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ @@ -746,173 +781,233 @@ static void emit_math2( struct brw_compile *p, brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_MOV(p, brw_message_reg(2), arg0[0]); - brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); - brw_MOV(p, brw_message_reg(4), sechalf(arg0[0])); + if (c->dispatch_width == 16) { + brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); + brw_MOV(p, brw_message_reg(4), sechalf(arg0[0])); + } brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_MOV(p, brw_message_reg(3), arg1[0]); - brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); - brw_MOV(p, brw_message_reg(5), sechalf(arg1[0])); + if (c->dispatch_width == 16) { + brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); + brw_MOV(p, brw_message_reg(5), sechalf(arg1[0])); + } - - /* Send two messages to perform all 16 operations: - */ brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_math(p, dst[dst_chan], function, - (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, + saturate, 2, brw_null_reg(), BRW_MATH_DATA_VECTOR, BRW_MATH_PRECISION_FULL); - brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); - brw_math(p, - offset(dst[dst_chan],1), - function, - (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, - 4, - brw_null_reg(), - BRW_MATH_DATA_VECTOR, - BRW_MATH_PRECISION_FULL); - + /* Send two messages to perform all 16 operations: + */ + if (c->dispatch_width == 16) { + brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); + brw_math(p, + offset(dst[dst_chan],1), + function, + saturate, + 4, + brw_null_reg(), + BRW_MATH_DATA_VECTOR, + BRW_MATH_PRECISION_FULL); + } brw_pop_insn_state(p); } - -static void emit_tex( struct brw_wm_compile *c, - const struct brw_wm_instruction *inst, - struct brw_reg *dst, - GLuint dst_flags, - struct brw_reg *arg ) +void emit_tex(struct brw_wm_compile *c, + struct brw_reg *dst, + GLuint dst_flags, + struct brw_reg *arg, + struct brw_reg depth_payload, + GLuint tex_idx, + GLuint sampler, + GLboolean shadow) { struct brw_compile *p = &c->func; - GLuint msgLength, responseLength; - GLuint i, nr; + struct intel_context *intel = &p->brw->intel; + struct brw_reg dst_retyped; + GLuint cur_mrf = 2, response_length; + GLuint i, nr_texcoords; GLuint emit; GLuint msg_type; + GLuint mrf_per_channel; + GLuint simd_mode; + + if (c->dispatch_width == 16) { + mrf_per_channel = 2; + response_length = 8; + dst_retyped = retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW); + simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD16; + } else { + mrf_per_channel = 1; + response_length = 4; + dst_retyped = retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW); + simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD8; + } /* How many input regs are there? */ - switch (inst->tex_idx) { + switch (tex_idx) { case TEXTURE_1D_INDEX: emit = WRITEMASK_X; - nr = 1; + nr_texcoords = 1; break; case TEXTURE_2D_INDEX: case TEXTURE_RECT_INDEX: emit = WRITEMASK_XY; - nr = 2; + nr_texcoords = 2; break; case TEXTURE_3D_INDEX: case TEXTURE_CUBE_INDEX: emit = WRITEMASK_XYZ; - nr = 3; + nr_texcoords = 3; break; default: /* unexpected target */ abort(); } - if (inst->tex_shadow) { - nr = 4; - emit |= WRITEMASK_W; - } + /* Pre-Ironlake, the 8-wide sampler always took u,v,r. */ + if (!intel->is_ironlake && c->dispatch_width == 8) + nr_texcoords = 3; - msgLength = 1; + /* For shadow comparisons, we have to supply u,v,r. */ + if (shadow) + nr_texcoords = 3; - for (i = 0; i < nr; i++) { - static const GLuint swz[4] = {0,1,2,2}; - if (emit & (1<<i)) - brw_MOV(p, brw_message_reg(msgLength+1), arg[swz[i]]); + /* Emit the texcoords. */ + for (i = 0; i < nr_texcoords; i++) { + if (emit & (1<<i)) + brw_MOV(p, brw_message_reg(cur_mrf), arg[i]); else - brw_MOV(p, brw_message_reg(msgLength+1), brw_imm_f(0)); - msgLength += 2; + brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0)); + cur_mrf += mrf_per_channel; } - responseLength = 8; /* always */ + /* Fill in the shadow comparison reference value. */ + if (shadow) { + if (intel->is_ironlake) { + /* Fill in the cube map array index value. */ + brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0)); + cur_mrf += mrf_per_channel; + } else if (c->dispatch_width == 8) { + /* Fill in the LOD bias value. */ + brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0)); + cur_mrf += mrf_per_channel; + } + brw_MOV(p, brw_message_reg(cur_mrf), arg[2]); + cur_mrf += mrf_per_channel; + } - if (BRW_IS_IGDNG(p->brw)) { - if (inst->tex_shadow) - msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE_IGDNG; - else - msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_IGDNG; + if (intel->is_ironlake) { + if (shadow) + msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_COMPARE_IGDNG; + else + msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_IGDNG; } else { - if (inst->tex_shadow) - msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE; - else - msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE; + /* Note that G45 and older determines shadow compare and dispatch width + * from message length for most messages. + */ + if (c->dispatch_width == 16 && shadow) + msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE; + else + msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE; } - brw_SAMPLE(p, - retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW), + brw_SAMPLE(p, + dst_retyped, 1, - retype(c->payload.depth[0].hw_reg, BRW_REGISTER_TYPE_UW), - SURF_INDEX_TEXTURE(inst->tex_unit), - inst->tex_unit, /* sampler */ - inst->writemask, - msg_type, - responseLength, - msgLength, - 0, + retype(depth_payload, BRW_REGISTER_TYPE_UW), + SURF_INDEX_TEXTURE(sampler), + sampler, + dst_flags & WRITEMASK_XYZW, + msg_type, + response_length, + cur_mrf - 1, + 0, 1, - BRW_SAMPLER_SIMD_MODE_SIMD16); + simd_mode); } -static void emit_txb( struct brw_wm_compile *c, - const struct brw_wm_instruction *inst, - struct brw_reg *dst, - GLuint dst_flags, - struct brw_reg *arg ) +void emit_txb(struct brw_wm_compile *c, + struct brw_reg *dst, + GLuint dst_flags, + struct brw_reg *arg, + struct brw_reg depth_payload, + GLuint tex_idx, + GLuint sampler) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; GLuint msgLength; GLuint msg_type; - /* Shadow ignored for txb. + GLuint mrf_per_channel; + GLuint response_length; + struct brw_reg dst_retyped; + + /* The G45 and older chipsets don't support 8-wide dispatch for LOD biased + * samples, so we'll use the 16-wide instruction, leave the second halves + * undefined, and trust the execution mask to keep the undefined pixels + * from mattering. */ - switch (inst->tex_idx) { + if (c->dispatch_width == 16 || !intel->is_ironlake) { + if (intel->is_ironlake) + msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_IGDNG; + else + msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS; + mrf_per_channel = 2; + dst_retyped = retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW); + response_length = 8; + } else { + msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_IGDNG; + mrf_per_channel = 1; + dst_retyped = retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW); + response_length = 4; + } + + /* Shadow ignored for txb. */ + switch (tex_idx) { case TEXTURE_1D_INDEX: - brw_MOV(p, brw_message_reg(2), arg[0]); - brw_MOV(p, brw_message_reg(4), brw_imm_f(0)); - brw_MOV(p, brw_message_reg(6), brw_imm_f(0)); + brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]); + brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), brw_imm_f(0)); + brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), brw_imm_f(0)); break; case TEXTURE_2D_INDEX: case TEXTURE_RECT_INDEX: - brw_MOV(p, brw_message_reg(2), arg[0]); - brw_MOV(p, brw_message_reg(4), arg[1]); - brw_MOV(p, brw_message_reg(6), brw_imm_f(0)); + brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]); + brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), arg[1]); + brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), brw_imm_f(0)); break; case TEXTURE_3D_INDEX: case TEXTURE_CUBE_INDEX: - brw_MOV(p, brw_message_reg(2), arg[0]); - brw_MOV(p, brw_message_reg(4), arg[1]); - brw_MOV(p, brw_message_reg(6), arg[2]); + brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]); + brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), arg[1]); + brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), arg[2]); break; default: /* unexpected target */ abort(); } - brw_MOV(p, brw_message_reg(8), arg[3]); - msgLength = 9; - - if (BRW_IS_IGDNG(p->brw)) - msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS_IGDNG; - else - msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS; + brw_MOV(p, brw_message_reg(2 + 3 * mrf_per_channel), arg[3]); + msgLength = 2 + 4 * mrf_per_channel - 1; brw_SAMPLE(p, - retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW), + dst_retyped, 1, - retype(c->payload.depth[0].hw_reg, BRW_REGISTER_TYPE_UW), - SURF_INDEX_TEXTURE(inst->tex_unit), - inst->tex_unit, /* sampler */ - inst->writemask, + retype(depth_payload, BRW_REGISTER_TYPE_UW), + SURF_INDEX_TEXTURE(sampler), + sampler, + dst_flags & WRITEMASK_XYZW, msg_type, - 8, /* responseLength */ + response_length, msgLength, 0, 1, @@ -920,11 +1015,13 @@ static void emit_txb( struct brw_wm_compile *c, } -static void emit_lit( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0 ) +static void emit_lit(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { + struct brw_compile *p = &c->func; + assert((mask & WRITEMASK_XW) == 0); if (mask & WRITEMASK_Y) { @@ -934,7 +1031,7 @@ static void emit_lit( struct brw_compile *p, } if (mask & WRITEMASK_Z) { - emit_math2(p, BRW_MATH_FUNCTION_POW, + emit_math2(c, BRW_MATH_FUNCTION_POW, &dst[2], WRITEMASK_X | (mask & SATURATE), &arg0[1], @@ -989,7 +1086,7 @@ static void emit_kil_nv( struct brw_wm_compile *c ) brw_push_insn_state(p); brw_set_mask_control(p, BRW_MASK_DISABLE); - brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK + brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); /* IMASK */ brw_AND(p, r0uw, c->emit_mask_reg, r0uw); brw_pop_insn_state(p); } @@ -1001,7 +1098,13 @@ static void fire_fb_write( struct brw_wm_compile *c, GLuint eot ) { struct brw_compile *p = &c->func; - + struct brw_reg dst; + + if (c->dispatch_width == 16) + dst = retype(vec16(brw_null_reg()), BRW_REGISTER_TYPE_UW); + else + dst = retype(vec8(brw_null_reg()), BRW_REGISTER_TYPE_UW); + /* Pass through control information: */ /* mov (8) m1.0<1>:ud r1.0<8;8,1>:ud { Align1 NoMask } */ @@ -1018,7 +1121,7 @@ static void fire_fb_write( struct brw_wm_compile *c, /* Send framebuffer write message: */ /* send (16) null.0<1>:uw m0 r0.0<8;8,1>:uw 0x85a04000:ud { Align1 EOT } */ brw_fb_WRITE(p, - retype(vec16(brw_null_reg()), BRW_REGISTER_TYPE_UW), + dst, base_reg, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW), target, @@ -1050,14 +1153,15 @@ static void emit_aa( struct brw_wm_compile *c, * \param arg1 the pass-through depth value * \param arg2 the shader-computed depth value */ -static void emit_fb_write( struct brw_wm_compile *c, - struct brw_reg *arg0, - struct brw_reg *arg1, - struct brw_reg *arg2, - GLuint target, - GLuint eot) +void emit_fb_write(struct brw_wm_compile *c, + struct brw_reg *arg0, + struct brw_reg *arg1, + struct brw_reg *arg2, + GLuint target, + GLuint eot) { struct brw_compile *p = &c->func; + struct brw_context *brw = p->brw; GLuint nr = 2; GLuint channel; @@ -1069,30 +1173,37 @@ static void emit_fb_write( struct brw_wm_compile *c, /* I don't really understand how this achieves the color interleave * (ie RGBARGBA) in the result: [Do the saturation here] */ - { - brw_push_insn_state(p); - - for (channel = 0; channel < 4; channel++) { + brw_push_insn_state(p); + + for (channel = 0; channel < 4; channel++) { + if (c->dispatch_width == 16 && brw->has_compr4) { + /* By setting the high bit of the MRF register number, we indicate + * that we want COMPR4 mode - instead of doing the usual destination + * + 1 for the second half we get destination + 4. + */ + brw_MOV(p, + brw_message_reg(nr + channel + (1 << 7)), + arg0[channel]); + } else { /* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */ /* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */ - brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_MOV(p, brw_message_reg(nr + channel), arg0[channel]); - - brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); - brw_MOV(p, - brw_message_reg(nr + channel + 4), - sechalf(arg0[channel])); - } - /* skip over the regs populated above: - */ - nr += 8; - - brw_pop_insn_state(p); + if (c->dispatch_width == 16) { + brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); + brw_MOV(p, + brw_message_reg(nr + channel + 4), + sechalf(arg0[channel])); + } + } } + /* skip over the regs populated above: + */ + nr += 8; + brw_pop_insn_state(p); if (c->key.source_depth_to_render_target) { @@ -1142,7 +1253,7 @@ static void emit_fb_write( struct brw_wm_compile *c, get_element_ud(brw_vec8_grf(1,0), 6), brw_imm_ud(1<<26)); - jmp = brw_JMPI(p, ip, ip, brw_imm_d(0)); + jmp = brw_JMPI(p, ip, ip, brw_imm_w(0)); { emit_aa(c, arg1, 2); fire_fb_write(c, 0, nr, target, eot); @@ -1156,7 +1267,6 @@ static void emit_fb_write( struct brw_wm_compile *c, } } - /** * Move a GPR to scratch memory. */ @@ -1294,7 +1404,7 @@ void brw_wm_emit( struct brw_wm_compile *c ) /* Generated instructions for calculating triangle interpolants: */ case WM_PIXELXY: - emit_pixel_xy(p, dst, dst_flags); + emit_pixel_xy(c, dst, dst_flags); break; case WM_DELTAXY: @@ -1306,7 +1416,7 @@ void brw_wm_emit( struct brw_wm_compile *c ) break; case WM_PIXELW: - emit_pixel_w(p, dst, dst_flags, args[0], args[1]); + emit_pixel_w(c, dst, dst_flags, args[0], args[1]); break; case WM_LINTERP: @@ -1364,7 +1474,7 @@ void brw_wm_emit( struct brw_wm_compile *c ) break; case OPCODE_TRUNC: - emit_trunc(p, dst, dst_flags, args[0]); + emit_alu1(p, brw_RNDZ, dst, dst_flags, args[0]); break; case OPCODE_LRP: @@ -1391,27 +1501,27 @@ void brw_wm_emit( struct brw_wm_compile *c ) /* Higher math functions: */ case OPCODE_RCP: - emit_math1(p, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]); + emit_math1(c, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]); break; case OPCODE_RSQ: - emit_math1(p, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]); + emit_math1(c, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]); break; case OPCODE_SIN: - emit_math1(p, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]); + emit_math1(c, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]); break; case OPCODE_COS: - emit_math1(p, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]); + emit_math1(c, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]); break; case OPCODE_EX2: - emit_math1(p, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]); + emit_math1(c, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]); break; case OPCODE_LG2: - emit_math1(p, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]); + emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]); break; case OPCODE_SCS: @@ -1419,13 +1529,13 @@ void brw_wm_emit( struct brw_wm_compile *c ) * fixup for 16-element execution. */ if (dst_flags & WRITEMASK_X) - emit_math1(p, BRW_MATH_FUNCTION_COS, dst, (dst_flags&SATURATE)|WRITEMASK_X, args[0]); + emit_math1(c, BRW_MATH_FUNCTION_COS, dst, (dst_flags&SATURATE)|WRITEMASK_X, args[0]); if (dst_flags & WRITEMASK_Y) - emit_math1(p, BRW_MATH_FUNCTION_SIN, dst+1, (dst_flags&SATURATE)|WRITEMASK_X, args[0]); + emit_math1(c, BRW_MATH_FUNCTION_SIN, dst+1, (dst_flags&SATURATE)|WRITEMASK_X, args[0]); break; case OPCODE_POW: - emit_math2(p, BRW_MATH_FUNCTION_POW, dst, dst_flags, args[0], args[1]); + emit_math2(c, BRW_MATH_FUNCTION_POW, dst, dst_flags, args[0], args[1]); break; /* Comparisons: @@ -1463,17 +1573,20 @@ void brw_wm_emit( struct brw_wm_compile *c ) break; case OPCODE_LIT: - emit_lit(p, dst, dst_flags, args[0]); + emit_lit(c, dst, dst_flags, args[0]); break; /* Texturing operations: */ case OPCODE_TEX: - emit_tex(c, inst, dst, dst_flags, args[0]); + emit_tex(c, dst, dst_flags, args[0], c->payload.depth[0].hw_reg, + inst->tex_idx, inst->tex_unit, + inst->tex_shadow); break; case OPCODE_TXB: - emit_txb(c, inst, dst, dst_flags, args[0]); + emit_txb(c, dst, dst_flags, args[0], c->payload.depth[0].hw_reg, + inst->tex_idx, inst->tex_unit); break; case OPCODE_KIL: diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c index 4e3edfbbff..3737faf26f 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_fp.c +++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c @@ -138,7 +138,6 @@ static struct prog_dst_register dst_reg(GLuint file, GLuint idx) reg.CondMask = COND_TR; reg.CondSwizzle = 0; reg.CondSrc = 0; - reg.pad = 0; return reg; } @@ -181,6 +180,9 @@ static void release_temp( struct brw_wm_compile *c, struct prog_dst_register tem static struct prog_instruction *get_fp_inst(struct brw_wm_compile *c) { + assert(c->nr_fp_insns < BRW_WM_MAX_INSN); + memset(&c->prog_instructions[c->nr_fp_insns], 0, + sizeof(*c->prog_instructions)); return &c->prog_instructions[c->nr_fp_insns++]; } @@ -447,7 +449,6 @@ static void emit_interp( struct brw_wm_compile *c, break; case FRAG_ATTRIB_FACE: - /* XXX review/test this case */ emit_op(c, WM_FRONTFACING, dst_mask(dst, WRITEMASK_X), @@ -956,7 +957,7 @@ static void precalc_txp( struct brw_wm_compile *c, -static void emit_fb_write( struct brw_wm_compile *c ) +static void emit_render_target_writes( struct brw_wm_compile *c ) { struct prog_src_register payload_r0_depth = src_reg(PROGRAM_PAYLOAD, PAYLOAD_DEPTH); struct prog_src_register outdepth = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DEPTH); @@ -964,36 +965,34 @@ static void emit_fb_write( struct brw_wm_compile *c ) GLuint i; struct prog_instruction *inst, *last_inst; - struct brw_context *brw = c->func.brw; /* The inst->Aux field is used for FB write target and the EOT marker */ - if (brw->state.nr_color_regions > 1) { - for (i = 0 ; i < brw->state.nr_color_regions; i++) { + if (c->key.nr_color_regions > 1) { + for (i = 0 ; i < c->key.nr_color_regions; i++) { outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DATA0 + i); - last_inst = inst = emit_op(c, - WM_FB_WRITE, dst_mask(dst_undef(),0), 0, - outcolor, payload_r0_depth, outdepth); - inst->Aux = (i<<1); + last_inst = inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(), 0), + 0, outcolor, payload_r0_depth, outdepth); + inst->Aux = INST_AUX_TARGET(i); if (c->fp_fragcolor_emitted) { outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR); - last_inst = inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(),0), + last_inst = inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(), 0), 0, outcolor, payload_r0_depth, outdepth); - inst->Aux = (i<<1); + inst->Aux = INST_AUX_TARGET(i); } } - last_inst->Aux |= 1; //eot + last_inst->Aux |= INST_AUX_EOT; } else { /* if gl_FragData[0] is written, use it, else use gl_FragColor */ - if (c->fp->program.Base.OutputsWritten & (1 << FRAG_RESULT_DATA0)) + if (c->fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DATA0)) outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DATA0); else outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR); inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(),0), 0, outcolor, payload_r0_depth, outdepth); - inst->Aux = 1|(0<<1); + inst->Aux = INST_AUX_EOT | INST_AUX_TARGET(0); } } @@ -1155,7 +1154,7 @@ void brw_wm_pass_fp( struct brw_wm_compile *c ) out->DstReg.WriteMask = 0; break; case OPCODE_END: - emit_fb_write(c); + emit_render_target_writes(c); break; case OPCODE_PRINT: break; diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index c9fe1dd8ad..fde83eea62 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -371,7 +371,7 @@ static void prealloc_reg(struct brw_wm_compile *c) for (j = 0; j < 4; j++) set_reg(c, PROGRAM_PAYLOAD, fp_input, j, reg); } - if (c->key.vp_outputs_written & (1 << i)) { + if (c->key.vp_outputs_written & BITFIELD64_BIT(i)) { reg_index += 2; } } @@ -550,42 +550,6 @@ static struct brw_reg get_src_reg(struct brw_wm_compile *c, } } - -/** - * Same as \sa get_src_reg() but if the register is a literal, emit - * a brw_reg encoding the literal. - * Note that a brw instruction only allows one src operand to be a literal. - * For instructions with more than one operand, only the second can be a - * literal. This means that we treat some literals as constants/uniforms - * (which why PROGRAM_CONSTANT is checked in fetch_constants()). - * - */ -static struct brw_reg get_src_reg_imm(struct brw_wm_compile *c, - const struct prog_instruction *inst, - GLuint srcRegIndex, GLuint channel) -{ - const struct prog_src_register *src = &inst->SrcReg[srcRegIndex]; - if (src->File == PROGRAM_CONSTANT) { - /* a literal */ - const int component = GET_SWZ(src->Swizzle, channel); - const GLfloat *param = - c->fp->program.Base.Parameters->ParameterValues[src->Index]; - GLfloat value = param[component]; - if (src->Negate & (1 << channel)) - value = -value; - if (src->Abs) - value = FABSF(value); -#if 0 - printf(" form immed value %f for chan %d\n", value, channel); -#endif - return brw_imm_f(value); - } - else { - return get_src_reg(c, inst, srcRegIndex, channel); - } -} - - /** * Subroutines are minimal support for resusable instruction sequences. * They are implemented as simply as possible to minimise overhead: there @@ -650,542 +614,110 @@ static void invoke_subroutine( struct brw_wm_compile *c, } } -static void emit_trunc( struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - int i; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF); - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - struct brw_reg src, dst; - dst = get_dst_reg(c, inst, i); - src = get_src_reg(c, inst, 0, i); - brw_RNDZ(p, dst, src); - } - } - brw_set_saturate(p, 0); -} - -static void emit_mov( struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - int i; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF); - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - struct brw_reg src, dst; - dst = get_dst_reg(c, inst, i); - /* XXX some moves from immediate value don't work reliably!!! */ - /*src = get_src_reg_imm(c, inst, 0, i);*/ - src = get_src_reg(c, inst, 0, i); - brw_MOV(p, dst, src); - } - } - brw_set_saturate(p, 0); -} - -static void emit_pixel_xy(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_reg r1 = brw_vec1_grf(1, 0); - struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW); - - struct brw_reg dst0, dst1; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - - dst0 = get_dst_reg(c, inst, 0); - dst1 = get_dst_reg(c, inst, 1); - /* Calculate pixel centers by adding 1 or 0 to each of the - * micro-tile coordinates passed in r1. - */ - if (mask & WRITEMASK_X) { - brw_ADD(p, - vec8(retype(dst0, BRW_REGISTER_TYPE_UW)), - stride(suboffset(r1_uw, 4), 2, 4, 0), - brw_imm_v(0x10101010)); - } - - if (mask & WRITEMASK_Y) { - brw_ADD(p, - vec8(retype(dst1, BRW_REGISTER_TYPE_UW)), - stride(suboffset(r1_uw, 5), 2, 4, 0), - brw_imm_v(0x11001100)); - } -} - -static void emit_delta_xy(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_reg r1 = brw_vec1_grf(1, 0); - struct brw_reg dst0, dst1, src0, src1; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - - dst0 = get_dst_reg(c, inst, 0); - dst1 = get_dst_reg(c, inst, 1); - src0 = get_src_reg(c, inst, 0, 0); - src1 = get_src_reg(c, inst, 0, 1); - /* Calc delta X,Y by subtracting origin in r1 from the pixel - * centers. - */ - if (mask & WRITEMASK_X) { - brw_ADD(p, - dst0, - retype(src0, BRW_REGISTER_TYPE_UW), - negate(r1)); - } - - if (mask & WRITEMASK_Y) { - brw_ADD(p, - dst1, - retype(src1, BRW_REGISTER_TYPE_UW), - negate(suboffset(r1,1))); - - } -} - -static void fire_fb_write( struct brw_wm_compile *c, - GLuint base_reg, - GLuint nr, - GLuint target, - GLuint eot) -{ - struct brw_compile *p = &c->func; - /* Pass through control information: - */ - /* mov (8) m1.0<1>:ud r1.0<8;8,1>:ud { Align1 NoMask } */ - { - brw_push_insn_state(p); - brw_set_mask_control(p, BRW_MASK_DISABLE); /* ? */ - brw_MOV(p, - brw_message_reg(base_reg + 1), - brw_vec8_grf(1, 0)); - brw_pop_insn_state(p); - } - /* Send framebuffer write message: */ - brw_fb_WRITE(p, - retype(vec8(brw_null_reg()), BRW_REGISTER_TYPE_UW), - base_reg, - retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW), - target, - nr, - 0, - eot); -} - -static void emit_fb_write(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - int nr = 2; - int channel; - GLuint target, eot; - struct brw_reg src0; - - /* Reserve a space for AA - may not be needed: - */ - if (c->key.aa_dest_stencil_reg) - nr += 1; - - brw_push_insn_state(p); - for (channel = 0; channel < 4; channel++) { - src0 = get_src_reg(c, inst, 0, channel); - /* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */ - /* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */ - brw_MOV(p, brw_message_reg(nr + channel), src0); - } - /* skip over the regs populated above: */ - nr += 8; - brw_pop_insn_state(p); - - if (c->key.source_depth_to_render_target) { - if (c->key.computes_depth) { - src0 = get_src_reg(c, inst, 2, 2); - brw_MOV(p, brw_message_reg(nr), src0); - } - else { - src0 = get_src_reg(c, inst, 1, 1); - brw_MOV(p, brw_message_reg(nr), src0); - } - - nr += 2; - } - - if (c->key.dest_depth_reg) { - const GLuint comp = c->key.dest_depth_reg / 2; - const GLuint off = c->key.dest_depth_reg % 2; - - if (off != 0) { - /* XXX this code needs review/testing */ - struct brw_reg arg1_0 = get_src_reg(c, inst, 1, comp); - struct brw_reg arg1_1 = get_src_reg(c, inst, 1, comp+1); - - brw_push_insn_state(p); - brw_set_compression_control(p, BRW_COMPRESSION_NONE); - - brw_MOV(p, brw_message_reg(nr), offset(arg1_0, 1)); - /* 2nd half? */ - brw_MOV(p, brw_message_reg(nr+1), arg1_1); - brw_pop_insn_state(p); - } - else - { - struct brw_reg src = get_src_reg(c, inst, 1, 1); - brw_MOV(p, brw_message_reg(nr), src); - } - nr += 2; - } - - target = inst->Aux >> 1; - eot = inst->Aux & 1; - fire_fb_write(c, 0, nr, target, eot); -} - -static void emit_pixel_w( struct brw_wm_compile *c, - const struct prog_instruction *inst) +/* Workaround for using brw_wm_emit.c's emit functions, which expect + * destination regs to be uniquely written. Moves arguments out to + * temporaries as necessary for instructions which use their destination as + * a temporary. + */ +static void +unalias3(struct brw_wm_compile *c, + void (*func)(struct brw_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1, + const struct brw_reg *arg2), + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1, + const struct brw_reg *arg2) { struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - if (mask & WRITEMASK_W) { - struct brw_reg dst, src0, delta0, delta1; - struct brw_reg interp3; - - dst = get_dst_reg(c, inst, 3); - src0 = get_src_reg(c, inst, 0, 0); - delta0 = get_src_reg(c, inst, 1, 0); - delta1 = get_src_reg(c, inst, 1, 1); - - interp3 = brw_vec1_grf(src0.nr+1, 4); - /* Calc 1/w - just linterp wpos[3] optimized by putting the - * result straight into a message reg. - */ - brw_LINE(p, brw_null_reg(), interp3, delta0); - brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), delta1); - - /* Calc w */ - brw_math_16( p, dst, - BRW_MATH_FUNCTION_INV, - BRW_MATH_SATURATE_NONE, - 2, brw_null_reg(), - BRW_MATH_PRECISION_FULL); - } -} + struct brw_reg tmp_arg0[4], tmp_arg1[4], tmp_arg2[4]; + int i, j; + int mark = mark_tmps(c); -static void emit_linterp(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - struct brw_reg interp[4]; - struct brw_reg dst, delta0, delta1; - struct brw_reg src0; - GLuint nr, i; - - src0 = get_src_reg(c, inst, 0, 0); - delta0 = get_src_reg(c, inst, 1, 0); - delta1 = get_src_reg(c, inst, 1, 1); - nr = src0.nr; - - interp[0] = brw_vec1_grf(nr, 0); - interp[1] = brw_vec1_grf(nr, 4); - interp[2] = brw_vec1_grf(nr+1, 0); - interp[3] = brw_vec1_grf(nr+1, 4); - - for(i = 0; i < 4; i++ ) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - brw_LINE(p, brw_null_reg(), interp[i], delta0); - brw_MAC(p, dst, suboffset(interp[i],1), delta1); - } + for (j = 0; j < 4; j++) { + tmp_arg0[j] = arg0[j]; + tmp_arg1[j] = arg1[j]; + tmp_arg2[j] = arg2[j]; } -} - -static void emit_cinterp(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - - struct brw_reg interp[4]; - struct brw_reg dst, src0; - GLuint nr, i; - src0 = get_src_reg(c, inst, 0, 0); - nr = src0.nr; - - interp[0] = brw_vec1_grf(nr, 0); - interp[1] = brw_vec1_grf(nr, 4); - interp[2] = brw_vec1_grf(nr+1, 0); - interp[3] = brw_vec1_grf(nr+1, 4); - - for(i = 0; i < 4; i++ ) { + for (i = 0; i < 4; i++) { if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - brw_MOV(p, dst, suboffset(interp[i],3)); + for (j = 0; j < 4; j++) { + if (arg0[j].file == dst[i].file && + dst[i].nr == arg0[j].nr) { + tmp_arg0[j] = alloc_tmp(c); + brw_MOV(p, tmp_arg0[j], arg0[j]); + } + if (arg1[j].file == dst[i].file && + dst[i].nr == arg1[j].nr) { + tmp_arg1[j] = alloc_tmp(c); + brw_MOV(p, tmp_arg1[j], arg1[j]); + } + if (arg2[j].file == dst[i].file && + dst[i].nr == arg2[j].nr) { + tmp_arg2[j] = alloc_tmp(c); + brw_MOV(p, tmp_arg2[j], arg2[j]); + } + } } } -} - -static void emit_pinterp(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - struct brw_reg interp[4]; - struct brw_reg dst, delta0, delta1; - struct brw_reg src0, w; - GLuint nr, i; + func(p, dst, mask, tmp_arg0, tmp_arg1, tmp_arg2); - src0 = get_src_reg(c, inst, 0, 0); - delta0 = get_src_reg(c, inst, 1, 0); - delta1 = get_src_reg(c, inst, 1, 1); - w = get_src_reg(c, inst, 2, 3); - nr = src0.nr; - - interp[0] = brw_vec1_grf(nr, 0); - interp[1] = brw_vec1_grf(nr, 4); - interp[2] = brw_vec1_grf(nr+1, 0); - interp[3] = brw_vec1_grf(nr+1, 4); - - for(i = 0; i < 4; i++ ) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - brw_LINE(p, brw_null_reg(), interp[i], delta0); - brw_MAC(p, dst, suboffset(interp[i],1), - delta1); - brw_MUL(p, dst, dst, w); - } - } + release_tmps(c, mark); } -/* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */ -static void emit_frontfacing(struct brw_wm_compile *c, - const struct prog_instruction *inst) +/* Workaround for using brw_wm_emit.c's emit functions, which expect + * destination regs to be uniquely written. Moves arguments out to + * temporaries as necessary for instructions which use their destination as + * a temporary. + */ +static void +unalias2(struct brw_wm_compile *c, + void (*func)(struct brw_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1), + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *arg1) { struct brw_compile *p = &c->func; - struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD); - struct brw_reg dst; - GLuint mask = inst->DstReg.WriteMask; - int i; - - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - brw_MOV(p, dst, brw_imm_f(0.0)); - } - } + struct brw_reg tmp_arg0[4], tmp_arg1[4]; + int i, j; + int mark = mark_tmps(c); - /* bit 31 is "primitive is back face", so checking < (1 << 31) gives - * us front face - */ - brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, r1_6ud, brw_imm_ud(1 << 31)); - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - brw_MOV(p, dst, brw_imm_f(1.0)); - } + for (j = 0; j < 4; j++) { + tmp_arg0[j] = arg0[j]; + tmp_arg1[j] = arg1[j]; } - brw_set_predicate_control_flag_value(p, 0xff); -} -static void emit_xpd(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - int i; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; for (i = 0; i < 4; i++) { - GLuint i2 = (i+2)%3; - GLuint i1 = (i+1)%3; if (mask & (1<<i)) { - struct brw_reg src0, src1, dst; - dst = get_dst_reg(c, inst, i); - src0 = negate(get_src_reg(c, inst, 0, i2)); - src1 = get_src_reg_imm(c, inst, 1, i1); - brw_MUL(p, brw_null_reg(), src0, src1); - src0 = get_src_reg(c, inst, 0, i1); - src1 = get_src_reg_imm(c, inst, 1, i2); - brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF); - brw_MAC(p, dst, src0, src1); - brw_set_saturate(p, 0); + for (j = 0; j < 4; j++) { + if (arg0[j].file == dst[i].file && + dst[i].nr == arg0[j].nr) { + tmp_arg0[j] = alloc_tmp(c); + brw_MOV(p, tmp_arg0[j], arg0[j]); + } + if (arg1[j].file == dst[i].file && + dst[i].nr == arg1[j].nr) { + tmp_arg1[j] = alloc_tmp(c); + brw_MOV(p, tmp_arg1[j], arg1[j]); + } + } } } - brw_set_saturate(p, 0); -} -static void emit_dp3(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_reg src0[3], src1[3], dst; - int i; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; - - if (!(mask & WRITEMASK_XYZW)) - return; - - assert(is_power_of_two(mask & WRITEMASK_XYZW)); - - for (i = 0; i < 3; i++) { - src0[i] = get_src_reg(c, inst, 0, i); - src1[i] = get_src_reg_imm(c, inst, 1, i); - } - - dst = get_dst_reg(c, inst, dst_chan); - brw_MUL(p, brw_null_reg(), src0[0], src1[0]); - brw_MAC(p, brw_null_reg(), src0[1], src1[1]); - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - brw_MAC(p, dst, src0[2], src1[2]); - brw_set_saturate(p, 0); -} - -static void emit_dp4(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_reg src0[4], src1[4], dst; - int i; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; - - if (!(mask & WRITEMASK_XYZW)) - return; - - assert(is_power_of_two(mask & WRITEMASK_XYZW)); - - for (i = 0; i < 4; i++) { - src0[i] = get_src_reg(c, inst, 0, i); - src1[i] = get_src_reg_imm(c, inst, 1, i); - } - dst = get_dst_reg(c, inst, dst_chan); - brw_MUL(p, brw_null_reg(), src0[0], src1[0]); - brw_MAC(p, brw_null_reg(), src0[1], src1[1]); - brw_MAC(p, brw_null_reg(), src0[2], src1[2]); - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - brw_MAC(p, dst, src0[3], src1[3]); - brw_set_saturate(p, 0); -} - -static void emit_dph(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_reg src0[4], src1[4], dst; - int i; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; - - if (!(mask & WRITEMASK_XYZW)) - return; - - assert(is_power_of_two(mask & WRITEMASK_XYZW)); - - for (i = 0; i < 4; i++) { - src0[i] = get_src_reg(c, inst, 0, i); - src1[i] = get_src_reg_imm(c, inst, 1, i); - } - dst = get_dst_reg(c, inst, dst_chan); - brw_MUL(p, brw_null_reg(), src0[0], src1[0]); - brw_MAC(p, brw_null_reg(), src0[1], src1[1]); - brw_MAC(p, dst, src0[2], src1[2]); - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - brw_ADD(p, dst, dst, src1[3]); - brw_set_saturate(p, 0); -} - -/** - * Emit a scalar instruction, like RCP, RSQ, LOG, EXP. - * Note that the result of the function is smeared across the dest - * register's X, Y, Z and W channels (subject to writemasking of course). - */ -static void emit_math1(struct brw_wm_compile *c, - const struct prog_instruction *inst, GLuint func) -{ - struct brw_compile *p = &c->func; - struct brw_reg src0, dst; - GLuint mask = inst->DstReg.WriteMask; - int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; - - if (!(mask & WRITEMASK_XYZW)) - return; - - assert(is_power_of_two(mask & WRITEMASK_XYZW)); - - /* Get first component of source register */ - dst = get_dst_reg(c, inst, dst_chan); - src0 = get_src_reg(c, inst, 0, 0); - - brw_MOV(p, brw_message_reg(2), src0); - brw_math(p, - dst, - func, - (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, - 2, - brw_null_reg(), - BRW_MATH_DATA_VECTOR, - BRW_MATH_PRECISION_FULL); -} - -static void emit_rcp(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_math1(c, inst, BRW_MATH_FUNCTION_INV); -} - -static void emit_rsq(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_math1(c, inst, BRW_MATH_FUNCTION_RSQ); -} - -static void emit_sin(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_math1(c, inst, BRW_MATH_FUNCTION_SIN); -} + func(p, dst, mask, tmp_arg0, tmp_arg1); -static void emit_cos(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_math1(c, inst, BRW_MATH_FUNCTION_COS); -} - -static void emit_ex2(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_math1(c, inst, BRW_MATH_FUNCTION_EXP); -} - -static void emit_lg2(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_math1(c, inst, BRW_MATH_FUNCTION_LOG); -} - -static void emit_add(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg src0, src1, dst; - GLuint mask = inst->DstReg.WriteMask; - int i; - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - for (i = 0 ; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg(c, inst, 0, i); - src1 = get_src_reg_imm(c, inst, 1, i); - brw_ADD(p, dst, src0, src1); - } - } - brw_set_saturate(p, 0); + release_tmps(c, mark); } static void emit_arl(struct brw_wm_compile *c, @@ -1201,180 +733,6 @@ static void emit_arl(struct brw_wm_compile *c, brw_set_saturate(p, 0); } - -static void emit_mul(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg src0, src1, dst; - GLuint mask = inst->DstReg.WriteMask; - int i; - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - for (i = 0 ; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg(c, inst, 0, i); - src1 = get_src_reg_imm(c, inst, 1, i); - brw_MUL(p, dst, src0, src1); - } - } - brw_set_saturate(p, 0); -} - -static void emit_frc(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg src0, dst; - GLuint mask = inst->DstReg.WriteMask; - int i; - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - for (i = 0 ; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg_imm(c, inst, 0, i); - brw_FRC(p, dst, src0); - } - } - if (inst->SaturateMode != SATURATE_OFF) - brw_set_saturate(p, 0); -} - -static void emit_flr(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg src0, dst; - GLuint mask = inst->DstReg.WriteMask; - int i; - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - for (i = 0 ; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg_imm(c, inst, 0, i); - brw_RNDD(p, dst, src0); - } - } - brw_set_saturate(p, 0); -} - - -static void emit_min_max(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - const GLuint mask = inst->DstReg.WriteMask; - const int mark = mark_tmps(c); - int i; - brw_push_insn_state(p); - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - struct brw_reg real_dst = get_dst_reg(c, inst, i); - struct brw_reg src0 = get_src_reg(c, inst, 0, i); - struct brw_reg src1 = get_src_reg(c, inst, 1, i); - struct brw_reg dst; - /* if dst==src0 or dst==src1 we need to use a temp reg */ - GLboolean use_temp = brw_same_reg(dst, src0) || - brw_same_reg(dst, src1); - if (use_temp) - dst = alloc_tmp(c); - else - dst = real_dst; - - /* - printf(" Min/max: dst %d src0 %d src1 %d\n", - dst.nr, src0.nr, src1.nr); - */ - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - brw_MOV(p, dst, src0); - brw_set_saturate(p, 0); - - if (inst->Opcode == OPCODE_MIN) - brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, src1, src0); - else - brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, src1, src0); - - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); - brw_MOV(p, dst, src1); - brw_set_saturate(p, 0); - brw_set_predicate_control_flag_value(p, 0xff); - if (use_temp) - brw_MOV(p, real_dst, dst); - } - } - brw_pop_insn_state(p); - release_tmps(c, mark); -} - -static void emit_pow(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg dst, src0, src1; - GLuint mask = inst->DstReg.WriteMask; - int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; - - if (!(mask & WRITEMASK_XYZW)) - return; - - assert(is_power_of_two(mask & WRITEMASK_XYZW)); - - dst = get_dst_reg(c, inst, dst_chan); - src0 = get_src_reg_imm(c, inst, 0, 0); - src1 = get_src_reg_imm(c, inst, 1, 0); - - brw_MOV(p, brw_message_reg(2), src0); - brw_MOV(p, brw_message_reg(3), src1); - - brw_math(p, - dst, - BRW_MATH_FUNCTION_POW, - (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, - 2, - brw_null_reg(), - BRW_MATH_DATA_VECTOR, - BRW_MATH_PRECISION_FULL); -} - -static void emit_lrp(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - struct brw_reg dst, tmp1, tmp2, src0, src1, src2; - int i; - int mark = mark_tmps(c); - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg(c, inst, 0, i); - - src1 = get_src_reg_imm(c, inst, 1, i); - - if (src1.nr == dst.nr) { - tmp1 = alloc_tmp(c); - brw_MOV(p, tmp1, src1); - } else - tmp1 = src1; - - src2 = get_src_reg(c, inst, 2, i); - if (src2.nr == dst.nr) { - tmp2 = alloc_tmp(c); - brw_MOV(p, tmp2, src2); - } else - tmp2 = src2; - - brw_ADD(p, dst, negate(src0), brw_imm_f(1.0)); - brw_MUL(p, brw_null_reg(), dst, tmp2); - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - brw_MAC(p, dst, src0, tmp1); - brw_set_saturate(p, 0); - } - release_tmps(c, mark); - } -} - /** * For GLSL shaders, this KIL will be unconditional. * It may be contained inside an IF/ENDIF structure of course. @@ -1385,94 +743,11 @@ static void emit_kil(struct brw_wm_compile *c) struct brw_reg depth = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW); brw_push_insn_state(p); brw_set_mask_control(p, BRW_MASK_DISABLE); - brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK + brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); /* IMASK */ brw_AND(p, depth, c->emit_mask_reg, depth); brw_pop_insn_state(p); } -static void emit_mad(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - struct brw_reg dst, src0, src1, src2; - int i; - - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg(c, inst, 0, i); - src1 = get_src_reg_imm(c, inst, 1, i); - src2 = get_src_reg_imm(c, inst, 2, i); - brw_MUL(p, dst, src0, src1); - - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - brw_ADD(p, dst, dst, src2); - brw_set_saturate(p, 0); - } - } -} - -static void emit_sop(struct brw_wm_compile *c, - const struct prog_instruction *inst, GLuint cond) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - struct brw_reg dst, src0, src1; - int i; - - for (i = 0; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg(c, inst, 0, i); - src1 = get_src_reg_imm(c, inst, 1, i); - brw_push_insn_state(p); - brw_CMP(p, brw_null_reg(), cond, src0, src1); - brw_set_predicate_control(p, BRW_PREDICATE_NONE); - brw_MOV(p, dst, brw_imm_f(0.0)); - brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); - brw_MOV(p, dst, brw_imm_f(1.0)); - brw_pop_insn_state(p); - } - } -} - -static void emit_slt(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_sop(c, inst, BRW_CONDITIONAL_L); -} - -static void emit_sle(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_sop(c, inst, BRW_CONDITIONAL_LE); -} - -static void emit_sgt(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_sop(c, inst, BRW_CONDITIONAL_G); -} - -static void emit_sge(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_sop(c, inst, BRW_CONDITIONAL_GE); -} - -static void emit_seq(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_sop(c, inst, BRW_CONDITIONAL_EQ); -} - -static void emit_sne(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - emit_sop(c, inst, BRW_CONDITIONAL_NEQ); -} - static INLINE struct brw_reg high_words( struct brw_reg reg ) { return stride( suboffset( retype( reg, BRW_REGISTER_TYPE_W ), 1 ), @@ -2525,196 +1800,6 @@ static void emit_noise4( struct brw_wm_compile *c, release_tmps( c, mark ); } - -static void emit_wpos_xy(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - struct brw_reg src0[2], dst[2]; - - dst[0] = get_dst_reg(c, inst, 0); - dst[1] = get_dst_reg(c, inst, 1); - - src0[0] = get_src_reg(c, inst, 0, 0); - src0[1] = get_src_reg(c, inst, 0, 1); - - /* Calculate the pixel offset from window bottom left into destination - * X and Y channels. - */ - if (mask & WRITEMASK_X) { - /* X' = X - origin_x */ - brw_ADD(p, - dst[0], - retype(src0[0], BRW_REGISTER_TYPE_W), - brw_imm_d(0 - c->key.origin_x)); - } - - if (mask & WRITEMASK_Y) { - /* Y' = height - (Y - origin_y) = height + origin_y - Y */ - brw_ADD(p, - dst[1], - negate(retype(src0[1], BRW_REGISTER_TYPE_W)), - brw_imm_d(c->key.origin_y + c->key.drawable_height - 1)); - } -} - -/* TODO - BIAS on SIMD8 not working yet... - */ -static void emit_txb(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg dst[4], src[4], payload_reg; - /* Note: TexSrcUnit was already looked up through SamplerTextures[] */ - const GLuint unit = inst->TexSrcUnit; - GLuint i; - GLuint msg_type; - - assert(unit < BRW_MAX_TEX_UNIT); - - payload_reg = get_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, 0, 1, 0, 0); - - for (i = 0; i < 4; i++) - dst[i] = get_dst_reg(c, inst, i); - for (i = 0; i < 4; i++) - src[i] = get_src_reg(c, inst, 0, i); - - switch (inst->TexSrcTarget) { - case TEXTURE_1D_INDEX: - brw_MOV(p, brw_message_reg(2), src[0]); /* s coord */ - brw_MOV(p, brw_message_reg(3), brw_imm_f(0)); /* t coord */ - brw_MOV(p, brw_message_reg(4), brw_imm_f(0)); /* r coord */ - break; - case TEXTURE_2D_INDEX: - case TEXTURE_RECT_INDEX: - brw_MOV(p, brw_message_reg(2), src[0]); - brw_MOV(p, brw_message_reg(3), src[1]); - brw_MOV(p, brw_message_reg(4), brw_imm_f(0)); - break; - case TEXTURE_3D_INDEX: - case TEXTURE_CUBE_INDEX: - brw_MOV(p, brw_message_reg(2), src[0]); - brw_MOV(p, brw_message_reg(3), src[1]); - brw_MOV(p, brw_message_reg(4), src[2]); - break; - default: - /* invalid target */ - abort(); - } - brw_MOV(p, brw_message_reg(5), src[3]); /* bias */ - brw_MOV(p, brw_message_reg(6), brw_imm_f(0)); /* ref (unused?) */ - - if (BRW_IS_IGDNG(p->brw)) { - msg_type = BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_BIAS_IGDNG; - } else { - /* Does it work well on SIMD8? */ - msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS; - } - - brw_SAMPLE(p, - retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW), /* dest */ - 1, /* msg_reg_nr */ - retype(payload_reg, BRW_REGISTER_TYPE_UW), /* src0 */ - SURF_INDEX_TEXTURE(unit), - unit, /* sampler */ - inst->DstReg.WriteMask, /* writemask */ - msg_type, /* msg_type */ - 4, /* response_length */ - 4, /* msg_length */ - 0, /* eot */ - 1, - BRW_SAMPLER_SIMD_MODE_SIMD8); -} - - -static void emit_tex(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg dst[4], src[4], payload_reg; - /* Note: TexSrcUnit was already looked up through SamplerTextures[] */ - const GLuint unit = inst->TexSrcUnit; - GLuint msg_len; - GLuint i, nr; - GLuint emit; - GLboolean shadow = (c->key.shadowtex_mask & (1<<unit)) ? 1 : 0; - GLuint msg_type; - - assert(unit < BRW_MAX_TEX_UNIT); - - payload_reg = get_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, 0, 1, 0, 0); - - for (i = 0; i < 4; i++) - dst[i] = get_dst_reg(c, inst, i); - for (i = 0; i < 4; i++) - src[i] = get_src_reg(c, inst, 0, i); - - switch (inst->TexSrcTarget) { - case TEXTURE_1D_INDEX: - emit = WRITEMASK_X; - nr = 1; - break; - case TEXTURE_2D_INDEX: - case TEXTURE_RECT_INDEX: - emit = WRITEMASK_XY; - nr = 2; - break; - case TEXTURE_3D_INDEX: - case TEXTURE_CUBE_INDEX: - emit = WRITEMASK_XYZ; - nr = 3; - break; - default: - /* invalid target */ - abort(); - } - msg_len = 1; - - /* move/load S, T, R coords */ - for (i = 0; i < nr; i++) { - static const GLuint swz[4] = {0,1,2,2}; - if (emit & (1<<i)) - brw_MOV(p, brw_message_reg(msg_len+1), src[swz[i]]); - else - brw_MOV(p, brw_message_reg(msg_len+1), brw_imm_f(0)); - msg_len += 1; - } - - if (shadow) { - brw_MOV(p, brw_message_reg(5), brw_imm_f(0)); /* lod / bias */ - brw_MOV(p, brw_message_reg(6), src[2]); /* ref value / R coord */ - } - - if (BRW_IS_IGDNG(p->brw)) { - if (shadow) - msg_type = BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_COMPARE_IGDNG; - else - msg_type = BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_IGDNG; - } else { - /* Does it work for shadow on SIMD8 ? */ - msg_type = BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE; - } - - brw_SAMPLE(p, - retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW), /* dest */ - 1, /* msg_reg_nr */ - retype(payload_reg, BRW_REGISTER_TYPE_UW), /* src0 */ - SURF_INDEX_TEXTURE(unit), - unit, /* sampler */ - inst->DstReg.WriteMask, /* writemask */ - msg_type, /* msg_type */ - 4, /* response_length */ - shadow ? 6 : 4, /* msg_length */ - 0, /* eot */ - 1, - BRW_SAMPLER_SIMD_MODE_SIMD8); - - if (shadow) - brw_MOV(p, dst[3], brw_imm_f(1.0)); -} - /** * Resolve subroutine calls after code emit is done. @@ -2741,6 +1826,7 @@ get_argument_regs(struct brw_wm_compile *c, static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) { + struct intel_context *intel = &brw->intel; #define MAX_IF_DEPTH 32 #define MAX_LOOP_DEPTH 32 struct brw_instruction *if_inst[MAX_IF_DEPTH], *loop_inst[MAX_LOOP_DEPTH]; @@ -2771,137 +1857,153 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) if (c->fp->use_const_buffer) fetch_constants(c, inst); - if (inst->CondUpdate) - brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); - else - brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE); + if (inst->Opcode != OPCODE_ARL) { + for (j = 0; j < 4; j++) { + if (inst->DstReg.WriteMask & (1 << j)) + dst[j] = get_dst_reg(c, inst, j); + else + dst[j] = brw_null_reg(); + } + } + for (j = 0; j < brw_wm_nr_args(inst->Opcode); j++) + get_argument_regs(c, inst, j, args[j], WRITEMASK_XYZW); dst_flags = inst->DstReg.WriteMask; if (inst->SaturateMode == SATURATE_ZERO_ONE) dst_flags |= SATURATE; + if (inst->CondUpdate) + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + else + brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE); + switch (inst->Opcode) { case WM_PIXELXY: - emit_pixel_xy(c, inst); + emit_pixel_xy(c, dst, dst_flags); break; case WM_DELTAXY: - emit_delta_xy(c, inst); + emit_delta_xy(p, dst, dst_flags, args[0]); break; case WM_PIXELW: - emit_pixel_w(c, inst); + emit_pixel_w(c, dst, dst_flags, args[0], args[1]); break; case WM_LINTERP: - emit_linterp(c, inst); + emit_linterp(p, dst, dst_flags, args[0], args[1]); break; case WM_PINTERP: - emit_pinterp(c, inst); + emit_pinterp(p, dst, dst_flags, args[0], args[1], args[2]); break; case WM_CINTERP: - emit_cinterp(c, inst); + emit_cinterp(p, dst, dst_flags, args[0]); break; case WM_WPOSXY: - emit_wpos_xy(c, inst); + emit_wpos_xy(c, dst, dst_flags, args[0]); break; case WM_FB_WRITE: - emit_fb_write(c, inst); + emit_fb_write(c, args[0], args[1], args[2], + INST_AUX_GET_TARGET(inst->Aux), + inst->Aux & INST_AUX_EOT); break; case WM_FRONTFACING: - emit_frontfacing(c, inst); + emit_frontfacing(p, dst, dst_flags); break; case OPCODE_ADD: - emit_add(c, inst); + emit_alu2(p, brw_ADD, dst, dst_flags, args[0], args[1]); break; case OPCODE_ARL: emit_arl(c, inst); break; case OPCODE_FRC: - emit_frc(c, inst); + emit_alu1(p, brw_FRC, dst, dst_flags, args[0]); break; case OPCODE_FLR: - emit_flr(c, inst); + emit_alu1(p, brw_RNDD, dst, dst_flags, args[0]); break; case OPCODE_LRP: - emit_lrp(c, inst); + unalias3(c, emit_lrp, + dst, dst_flags, args[0], args[1], args[2]); break; case OPCODE_TRUNC: - emit_trunc(c, inst); + emit_alu1(p, brw_RNDZ, dst, dst_flags, args[0]); break; case OPCODE_MOV: case OPCODE_SWZ: - emit_mov(c, inst); + emit_alu1(p, brw_MOV, dst, dst_flags, args[0]); break; case OPCODE_DP3: - emit_dp3(c, inst); + emit_dp3(p, dst, dst_flags, args[0], args[1]); break; case OPCODE_DP4: - emit_dp4(c, inst); + emit_dp4(p, dst, dst_flags, args[0], args[1]); break; case OPCODE_XPD: - emit_xpd(c, inst); + emit_xpd(p, dst, dst_flags, args[0], args[1]); break; case OPCODE_DPH: - emit_dph(c, inst); + emit_dph(p, dst, dst_flags, args[0], args[1]); break; case OPCODE_RCP: - emit_rcp(c, inst); + emit_math1(c, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]); break; case OPCODE_RSQ: - emit_rsq(c, inst); + emit_math1(c, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]); break; case OPCODE_SIN: - emit_sin(c, inst); + emit_math1(c, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]); break; case OPCODE_COS: - emit_cos(c, inst); + emit_math1(c, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]); break; case OPCODE_EX2: - emit_ex2(c, inst); + emit_math1(c, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]); break; case OPCODE_LG2: - emit_lg2(c, inst); + emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]); break; case OPCODE_MIN: + unalias2(c, emit_min, dst, dst_flags, args[0], args[1]); + break; case OPCODE_MAX: - emit_min_max(c, inst); + unalias2(c, emit_max, dst, dst_flags, args[0], args[1]); break; case OPCODE_DDX: case OPCODE_DDY: - for (j = 0; j < 4; j++) { - if (inst->DstReg.WriteMask & (1 << j)) - dst[j] = get_dst_reg(c, inst, j); - else - dst[j] = brw_null_reg(); - } - get_argument_regs(c, inst, 0, args[0], WRITEMASK_XYZW); emit_ddxy(p, dst, dst_flags, (inst->Opcode == OPCODE_DDX), args[0]); break; case OPCODE_SLT: - emit_slt(c, inst); + emit_sop(p, dst, dst_flags, + BRW_CONDITIONAL_L, args[0], args[1]); break; case OPCODE_SLE: - emit_sle(c, inst); + emit_sop(p, dst, dst_flags, + BRW_CONDITIONAL_LE, args[0], args[1]); break; case OPCODE_SGT: - emit_sgt(c, inst); + emit_sop(p, dst, dst_flags, + BRW_CONDITIONAL_G, args[0], args[1]); break; case OPCODE_SGE: - emit_sge(c, inst); + emit_sop(p, dst, dst_flags, + BRW_CONDITIONAL_GE, args[0], args[1]); break; case OPCODE_SEQ: - emit_seq(c, inst); + emit_sop(p, dst, dst_flags, + BRW_CONDITIONAL_EQ, args[0], args[1]); break; case OPCODE_SNE: - emit_sne(c, inst); + emit_sop(p, dst, dst_flags, + BRW_CONDITIONAL_NEQ, args[0], args[1]); break; case OPCODE_MUL: - emit_mul(c, inst); + emit_alu2(p, brw_MUL, dst, dst_flags, args[0], args[1]); break; case OPCODE_POW: - emit_pow(c, inst); + emit_math2(c, BRW_MATH_FUNCTION_POW, + dst, dst_flags, args[0], args[1]); break; case OPCODE_MAD: - emit_mad(c, inst); + emit_mad(p, dst, dst_flags, args[0], args[1], args[2]); break; case OPCODE_NOISE1: emit_noise1(c, inst); @@ -2916,10 +2018,19 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) emit_noise4(c, inst); break; case OPCODE_TEX: - emit_tex(c, inst); + emit_tex(c, dst, dst_flags, args[0], + get_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, + 0, 1, 0, 0), + inst->TexSrcTarget, + inst->TexSrcUnit, + (c->key.shadowtex_mask & (1 << inst->TexSrcUnit)) != 0); break; case OPCODE_TXB: - emit_txb(c, inst); + emit_txb(c, dst, dst_flags, args[0], + get_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, + 0, 1, 0, 0), + inst->TexSrcTarget, + c->fp->program.Base.SamplerUnits[inst->TexSrcUnit]); break; case OPCODE_KIL_NV: emit_kil(c); @@ -2929,6 +2040,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) if_inst[if_depth++] = brw_IF(p, BRW_EXECUTE_8); break; case OPCODE_ELSE: + assert(if_depth > 0); if_inst[if_depth-1] = brw_ELSE(p, if_inst[if_depth-1]); break; case OPCODE_ENDIF: @@ -2982,9 +2094,10 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) struct brw_instruction *inst0, *inst1; GLuint br = 1; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) br = 2; - + + assert(loop_depth > 0); loop_depth--; inst0 = inst1 = brw_WHILE(p, loop_inst[loop_depth]); /* patch all the BREAK/CONT instructions from last BGNLOOP */ diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass0.c b/src/mesa/drivers/dri/i965/brw_wm_pass0.c index 6279258339..ff4c082d5e 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass0.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass0.c @@ -42,12 +42,14 @@ static struct brw_wm_ref *get_ref( struct brw_wm_compile *c ) { assert(c->nr_refs < BRW_WM_MAX_REF); + memset(&c->refs[c->nr_refs], 0, sizeof(*c->refs)); return &c->refs[c->nr_refs++]; } static struct brw_wm_value *get_value( struct brw_wm_compile *c) { assert(c->nr_refs < BRW_WM_MAX_VREG); + memset(&c->vreg[c->nr_vreg], 0, sizeof(*c->vreg)); return &c->vreg[c->nr_vreg++]; } @@ -55,6 +57,7 @@ static struct brw_wm_value *get_value( struct brw_wm_compile *c) static struct brw_wm_instruction *get_instruction( struct brw_wm_compile *c ) { assert(c->nr_insns < BRW_WM_MAX_INSN); + memset(&c->instruction[c->nr_insns], 0, sizeof(*c->instruction)); return &c->instruction[c->nr_insns++]; } @@ -322,8 +325,8 @@ translate_insn(struct brw_wm_compile *c, out->tex_unit = inst->TexSrcUnit; out->tex_idx = inst->TexSrcTarget; out->tex_shadow = inst->TexShadow; - out->eot = inst->Aux & 1; - out->target = inst->Aux >> 1; + out->eot = inst->Aux & INST_AUX_EOT; + out->target = INST_AUX_GET_TARGET(inst->Aux); /* Args: */ diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass2.c b/src/mesa/drivers/dri/i965/brw_wm_pass2.c index 6faea018fb..31303febf0 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass2.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass2.c @@ -82,8 +82,8 @@ static void init_registers( struct brw_wm_compile *c ) for (j = 0; j < c->nr_creg; j++) prealloc_reg(c, &c->creg[j], i++); - for (j = 0; j < FRAG_ATTRIB_MAX; j++) { - if (c->key.vp_outputs_written & (1<<j)) { + for (j = 0; j < VERT_RESULT_MAX; j++) { + if (c->key.vp_outputs_written & BITFIELD64_BIT(j)) { int fp_index; if (j >= VERT_RESULT_VAR0) diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c index dff466587a..ad267a4e6a 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c @@ -66,19 +66,6 @@ static GLuint translate_wrap_mode( GLenum wrap ) } } - -static GLuint U_FIXED(GLfloat value, GLuint frac_bits) -{ - value *= (1<<frac_bits); - return value < 0 ? 0 : value; -} - -static GLint S_FIXED(GLfloat value, GLuint frac_bits) -{ - return value * (1<<frac_bits); -} - - static dri_bo *upload_default_color( struct brw_context *brw, const GLfloat *color ) { @@ -86,8 +73,8 @@ static dri_bo *upload_default_color( struct brw_context *brw, COPY_4V(sdc.color, color); - return brw_cache_data( &brw->cache, BRW_SAMPLER_DEFAULT_COLOR, &sdc, - NULL, 0 ); + return brw_cache_data(&brw->cache, BRW_SAMPLER_DEFAULT_COLOR, + &sdc, sizeof(sdc), NULL, 0); } @@ -228,8 +215,8 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key, */ sampler->ss0.base_level = U_FIXED(0, 1); - sampler->ss1.max_lod = U_FIXED(MIN2(MAX2(key->maxlod, 0), 13), 6); - sampler->ss1.min_lod = U_FIXED(MIN2(MAX2(key->minlod, 0), 13), 6); + sampler->ss1.max_lod = U_FIXED(CLAMP(key->maxlod, 0, 13), 6); + sampler->ss1.min_lod = U_FIXED(CLAMP(key->minlod, 0, 13), 6); sampler->ss2.default_color_pointer = sdc_bo->offset >> 5; /* reloc */ } @@ -275,10 +262,10 @@ brw_wm_sampler_populate_key(struct brw_context *brw, dri_bo_unreference(brw->wm.sdc_bo[unit]); if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) { float bordercolor[4] = { - texObj->BorderColor[0], - texObj->BorderColor[0], - texObj->BorderColor[0], - texObj->BorderColor[0] + texObj->BorderColor.f[0], + texObj->BorderColor.f[0], + texObj->BorderColor.f[0], + texObj->BorderColor.f[0] }; /* GL specs that border color for depth textures is taken from the * R channel, while the hardware uses A. Spam R into all the @@ -287,7 +274,7 @@ brw_wm_sampler_populate_key(struct brw_context *brw, brw->wm.sdc_bo[unit] = upload_default_color(brw, bordercolor); } else { brw->wm.sdc_bo[unit] = upload_default_color(brw, - texObj->BorderColor); + texObj->BorderColor.f); } key->sampler_count = unit + 1; } diff --git a/src/mesa/drivers/dri/i965/brw_wm_state.c b/src/mesa/drivers/dri/i965/brw_wm_state.c index 361f91292b..d3373ea79e 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_state.c @@ -49,8 +49,6 @@ struct brw_wm_unit_key { unsigned int curbe_offset; unsigned int urb_size; - unsigned int max_threads; - unsigned int nr_surfaces, sampler_count; GLboolean uses_depth, computes_depth, uses_kill, is_glsl; GLboolean polygon_stipple, stats_wm, line_stipple, offset_enable; @@ -67,18 +65,6 @@ wm_unit_populate_key(struct brw_context *brw, struct brw_wm_unit_key *key) memset(key, 0, sizeof(*key)); - if (INTEL_DEBUG & DEBUG_SINGLE_THREAD) - key->max_threads = 1; - else { - /* WM maximum threads is number of EUs times number of threads per EU. */ - if (BRW_IS_IGDNG(brw)) - key->max_threads = 12 * 6; - else if (BRW_IS_G4X(brw)) - key->max_threads = 10 * 5; - else - key->max_threads = 8 * 4; - } - /* CACHE_NEW_WM_PROG */ key->total_grf = brw->wm.prog_data->total_grf; key->urb_entry_read_length = brw->wm.prog_data->urb_read_length; @@ -106,7 +92,7 @@ wm_unit_populate_key(struct brw_context *brw, struct brw_wm_unit_key *key) /* as far as we can tell */ key->computes_depth = - (fp->Base.OutputsWritten & (1 << FRAG_RESULT_DEPTH)) != 0; + (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) != 0; /* BRW_NEW_DEPTH_BUFFER * Override for NULL depthbuffer case, required by the Pixel Shader Computed * Depth field. @@ -140,6 +126,7 @@ static dri_bo * wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, dri_bo **reloc_bufs) { + struct intel_context *intel = &brw->intel; struct brw_wm_unit_state wm; dri_bo *bo; @@ -150,7 +137,7 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, wm.thread1.depth_coef_urb_read_offset = 1; wm.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) wm.thread1.binding_table_entry_count = 0; /* hardware requirement */ else wm.thread1.binding_table_entry_count = key->nr_surfaces; @@ -170,7 +157,7 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, wm.thread3.const_urb_entry_read_length = key->curb_entry_read_length; wm.thread3.const_urb_entry_read_offset = key->curbe_offset * 2; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) wm.wm4.sampler_count = 0; /* hardware requirement */ else wm.wm4.sampler_count = (key->sampler_count + 1) / 4; @@ -191,7 +178,7 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, else wm.wm5.enable_16_pix = 1; - wm.wm5.max_threads = key->max_threads - 1; + wm.wm5.max_threads = brw->wm_max_threads - 1; wm.wm5.thread_dispatch_enable = 1; /* AKA: color_write */ wm.wm5.legacy_line_rast = 0; wm.wm5.legacy_global_depth_bias = 0; @@ -268,7 +255,7 @@ static void upload_wm_unit( struct brw_context *brw ) */ assert(key.total_scratch <= 12 * 1024); if (key.total_scratch) { - GLuint total = key.total_scratch * key.max_threads; + GLuint total = key.total_scratch * brw->wm_max_threads; if (brw->wm.scratch_bo && total > brw->wm.scratch_bo->size) { dri_bo_unreference(brw->wm.scratch_bo); diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c index 9c28a22a29..f26cfabb7d 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c @@ -31,7 +31,6 @@ #include "main/mtypes.h" -#include "main/texformat.h" #include "main/texstore.h" #include "shader/prog_parameter.h" @@ -70,7 +69,8 @@ static GLuint translate_tex_target( GLenum target ) } -static GLuint translate_tex_format( GLuint mesa_format, GLenum internal_format, +static GLuint translate_tex_format( gl_format mesa_format, + GLenum internal_format, GLenum depth_mode ) { switch( mesa_format ) { @@ -86,21 +86,22 @@ static GLuint translate_tex_format( GLuint mesa_format, GLenum internal_format, case MESA_FORMAT_AL88: return BRW_SURFACEFORMAT_L8A8_UNORM; + case MESA_FORMAT_AL1616: + return BRW_SURFACEFORMAT_L16A16_UNORM; + case MESA_FORMAT_RGB888: assert(0); /* not supported for sampling */ return BRW_SURFACEFORMAT_R8G8B8_UNORM; case MESA_FORMAT_ARGB8888: - if (internal_format == GL_RGB) - return BRW_SURFACEFORMAT_B8G8R8X8_UNORM; - else - return BRW_SURFACEFORMAT_B8G8R8A8_UNORM; + return BRW_SURFACEFORMAT_B8G8R8A8_UNORM; + + case MESA_FORMAT_XRGB8888: + return BRW_SURFACEFORMAT_B8G8R8X8_UNORM; case MESA_FORMAT_RGBA8888_REV: - if (internal_format == GL_RGB) - return BRW_SURFACEFORMAT_R8G8B8X8_UNORM; - else - return BRW_SURFACEFORMAT_R8G8B8A8_UNORM; + _mesa_problem(NULL, "unexpected format in i965:translate_tex_format()"); + return BRW_SURFACEFORMAT_R8G8B8A8_UNORM; case MESA_FORMAT_RGB565: return BRW_SURFACEFORMAT_B5G6R5_UNORM; @@ -287,7 +288,7 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit ) key.bo = NULL; key.offset = intelObj->textureOffset; } else { - key.format = firstImage->TexFormat->MesaFormat; + key.format = firstImage->TexFormat; key.internal_format = firstImage->InternalFormat; key.pitch = intelObj->mt->pitch; key.depth = firstImage->Depth; @@ -354,7 +355,10 @@ brw_create_constant_surface( struct brw_context *brw, NULL, NULL); if (key->bo) { - /* Emit relocation to surface contents */ + /* Emit relocation to surface contents. Section 5.1.1 of the gen4 + * bspec ("Data Cache") says that the data cache does not exist as + * a separate cache and is just the sampler cache. + */ dri_bo_emit_reloc(bo, I915_GEM_DOMAIN_SAMPLER, 0, 0, @@ -507,7 +511,8 @@ brw_update_renderbuffer_surface(struct brw_context *brw, struct gl_renderbuffer *rb, unsigned int unit) { - GLcontext *ctx = &brw->intel.ctx; + struct intel_context *intel = &brw->intel;; + GLcontext *ctx = &intel->ctx; dri_bo *region_bo = NULL; struct intel_renderbuffer *irb = intel_renderbuffer(rb); struct intel_region *region = irb ? irb->region : NULL; @@ -518,7 +523,8 @@ brw_update_renderbuffer_surface(struct brw_context *brw, GLubyte color_mask[4]; GLboolean color_blend; uint32_t tiling; - uint32_t draw_offset; + uint32_t draw_x; + uint32_t draw_y; } key; memset(&key, 0, sizeof(key)); @@ -527,8 +533,15 @@ brw_update_renderbuffer_surface(struct brw_context *brw, region_bo = region->buffer; key.surface_type = BRW_SURFACE_2D; - switch (irb->texformat->MesaFormat) { + switch (irb->Base.Format) { + /* XRGB and ARGB are treated the same here because the chips in this + * family cannot render to XRGB targets. This means that we have to + * mask writes to alpha (ala glColorMask) and reconfigure the alpha + * blending hardware to use GL_ONE (or GL_ZERO) for cases where + * GL_DST_ALPHA (or GL_ONE_MINUS_DST_ALPHA) is used. + */ case MESA_FORMAT_ARGB8888: + case MESA_FORMAT_XRGB8888: key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM; break; case MESA_FORMAT_RGB565: @@ -541,8 +554,7 @@ brw_update_renderbuffer_surface(struct brw_context *brw, key.surface_format = BRW_SURFACEFORMAT_B4G4R4A4_UNORM; break; default: - _mesa_problem(ctx, "Bad renderbuffer format: %d\n", - irb->texformat->MesaFormat); + _mesa_problem(ctx, "Bad renderbuffer format: %d\n", irb->Base.Format); } key.tiling = region->tiling; if (brw->intel.intelScreen->driScrnPriv->dri2.enabled) { @@ -554,7 +566,8 @@ brw_update_renderbuffer_surface(struct brw_context *brw, } key.pitch = region->pitch; key.cpp = region->cpp; - key.draw_offset = region->draw_offset; /* cur 3d or cube face offset */ + key.draw_x = region->draw_x; + key.draw_y = region->draw_y; } else { key.surface_type = BRW_SURFACE_NULL; key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM; @@ -562,10 +575,19 @@ brw_update_renderbuffer_surface(struct brw_context *brw, key.width = 1; key.height = 1; key.cpp = 4; - key.draw_offset = 0; + key.draw_x = 0; + key.draw_y = 0; } - memcpy(key.color_mask, ctx->Color.ColorMask, + /* _NEW_COLOR */ + memcpy(key.color_mask, ctx->Color.ColorMask[0], sizeof(key.color_mask)); + + /* As mentioned above, disable writes to the alpha component when the + * renderbuffer is XRGB. + */ + if (ctx->DrawBuffer->Visual.alphaBits == 0) + key.color_mask[3] = GL_FALSE; + key.color_blend = (!ctx->Color._LogicOpEnabled && ctx->Color.BlendEnabled); @@ -584,25 +606,32 @@ brw_update_renderbuffer_surface(struct brw_context *brw, surf.ss0.surface_format = key.surface_format; surf.ss0.surface_type = key.surface_type; if (key.tiling == I915_TILING_NONE) { - surf.ss1.base_addr = key.draw_offset; + surf.ss1.base_addr = (key.draw_x + key.draw_y * key.pitch) * key.cpp; } else { - uint32_t tile_offset = key.draw_offset % 4096; - - surf.ss1.base_addr = key.draw_offset - tile_offset; - - assert(BRW_IS_G4X(brw) || tile_offset == 0); - if (BRW_IS_G4X(brw)) { - if (key.tiling == I915_TILING_X) { - /* Note that the low bits of these fields are missing, so - * there's the possibility of getting in trouble. - */ - surf.ss5.x_offset = (tile_offset % 512) / key.cpp / 4; - surf.ss5.y_offset = tile_offset / 512 / 2; - } else { - surf.ss5.x_offset = (tile_offset % 128) / key.cpp / 4; - surf.ss5.y_offset = tile_offset / 128 / 2; - } + uint32_t tile_base, tile_x, tile_y; + uint32_t pitch = key.pitch * key.cpp; + + if (key.tiling == I915_TILING_X) { + tile_x = key.draw_x % (512 / key.cpp); + tile_y = key.draw_y % 8; + tile_base = ((key.draw_y / 8) * (8 * pitch)); + tile_base += (key.draw_x - tile_x) / (512 / key.cpp) * 4096; + } else { + /* Y */ + tile_x = key.draw_x % (128 / key.cpp); + tile_y = key.draw_y % 32; + tile_base = ((key.draw_y / 32) * (32 * pitch)); + tile_base += (key.draw_x - tile_x) / (128 / key.cpp) * 4096; } + assert(intel->is_g4x || (tile_x == 0 && tile_y == 0)); + assert(tile_x % 4 == 0); + assert(tile_y % 2 == 0); + /* Note that the low bits of these fields are missing, so + * there's the possibility of getting in trouble. + */ + surf.ss1.base_addr = tile_base; + surf.ss5.x_offset = tile_x / 4; + surf.ss5.y_offset = tile_y / 2; } if (region_bo != NULL) surf.ss1.base_addr += region_bo->offset; /* reloc */ @@ -693,11 +722,10 @@ brw_wm_get_binding_table(struct brw_context *brw) static void prepare_wm_surfaces(struct brw_context *brw ) { GLcontext *ctx = &brw->intel.ctx; - struct intel_context *intel = &brw->intel; GLuint i; int old_nr_surfaces; - /* _NEW_BUFFERS */ + /* _NEW_BUFFERS | _NEW_COLOR */ /* Update surfaces for drawing buffers */ if (ctx->DrawBuffer->_NumColorDrawBuffers >= 1) { for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) { @@ -710,7 +738,7 @@ static void prepare_wm_surfaces(struct brw_context *brw ) } old_nr_surfaces = brw->wm.nr_surfaces; - brw->wm.nr_surfaces = MAX_DRAW_BUFFERS; + brw->wm.nr_surfaces = BRW_MAX_DRAW_BUFFERS; if (brw->wm.surf_bo[SURF_INDEX_FRAG_CONST_BUFFER] != NULL) brw->wm.nr_surfaces = SURF_INDEX_FRAG_CONST_BUFFER + 1; diff --git a/src/mesa/drivers/dri/i965/intel_swapbuffers.c b/src/mesa/drivers/dri/i965/intel_swapbuffers.c deleted file mode 120000 index 148d5215aa..0000000000 --- a/src/mesa/drivers/dri/i965/intel_swapbuffers.c +++ /dev/null @@ -1 +0,0 @@ -../intel/intel_swapbuffers.c
\ No newline at end of file |