diff options
Diffstat (limited to 'src/gallium/drivers/llvmpipe')
-rw-r--r-- | src/gallium/drivers/llvmpipe/Makefile | 2 | ||||
-rw-r--r-- | src/gallium/drivers/llvmpipe/README | 65 | ||||
-rw-r--r-- | src/gallium/drivers/llvmpipe/SConscript | 6 | ||||
-rw-r--r-- | src/gallium/drivers/llvmpipe/lp_rast.c | 7 | ||||
-rw-r--r-- | src/gallium/drivers/llvmpipe/lp_setup_line.c | 15 | ||||
-rw-r--r-- | src/gallium/drivers/llvmpipe/lp_setup_tri.c | 44 | ||||
-rw-r--r-- | src/gallium/drivers/llvmpipe/lp_state_fs.c | 9 | ||||
-rw-r--r-- | src/gallium/drivers/llvmpipe/lp_tile_soa.py | 4 |
8 files changed, 65 insertions, 87 deletions
diff --git a/src/gallium/drivers/llvmpipe/Makefile b/src/gallium/drivers/llvmpipe/Makefile index 08da2286b0..669e42e300 100644 --- a/src/gallium/drivers/llvmpipe/Makefile +++ b/src/gallium/drivers/llvmpipe/Makefile @@ -69,7 +69,7 @@ lp_test_sincos.o : sse_mathfun.h PROGS_DEPS := ../../auxiliary/libgallium.a lp_tile_soa.c: lp_tile_soa.py ../../auxiliary/util/u_format_parse.py ../../auxiliary/util/u_format_pack.py ../../auxiliary/util/u_format.csv - python lp_tile_soa.py ../../auxiliary/util/u_format.csv > $@ + $(PYTHON2) $(PYTHON_FLAGS) lp_tile_soa.py ../../auxiliary/util/u_format.csv > $@ LDFLAGS += $(LLVM_LDFLAGS) LIBS += -L../../auxiliary/ -lgallium libllvmpipe.a $(LLVM_LIBS) $(GL_LIB_DEPS) diff --git a/src/gallium/drivers/llvmpipe/README b/src/gallium/drivers/llvmpipe/README index ec30d4d708..e9374cc6ef 100644 --- a/src/gallium/drivers/llvmpipe/README +++ b/src/gallium/drivers/llvmpipe/README @@ -1,53 +1,6 @@ LLVMPIPE -- a fork of softpipe that employs LLVM for code generation. -Status -====== - -Done so far is: - - - the whole fragment pipeline is code generated in a single function - - - input interpolation - - - depth testing - - - texture sampling - - 1D/2D/3D/cube maps supported - - all texture wrap modes supported - - all texture filtering modes supported - - perhaps not all texture formats yet supported - - - fragment shader TGSI translation - - same level of support as the TGSI SSE2 exec machine, with the exception - we don't fallback to TGSI interpretation when an unsupported opcode is - found, but just ignore it - - done in SoA layout - - input interpolation also code generated - - - alpha testing - - - blend (including logic ops) - - both in SoA and AoS layouts, but only the former used for now - - - code is generic - - intermediates can be vectors of floats, ubytes, fixed point, etc, and of - any width and length - - not all operations are implemented for these types yet though - -Most mesa/progs/demos/* work. - -To do (probably by this order): - - - code generate stipple and stencil testing - - - translate TGSI control flow instructions, and all other remaining opcodes - - - integrate with the draw module for VS code generation - - - code generate the triangle setup and rasterization - - Requirements ============ @@ -98,7 +51,7 @@ Building To build everything on Linux invoke scons as: - scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=xlib dri=false + scons build=debug libgl-xlib Alternatively, you can build it with GNU make, if you prefer, by invoking it as @@ -108,19 +61,16 @@ but the rest of these instructions assume that scons is used. For windows is everything the except except the winsys: - scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=gdi dri=false + scons build=debug libgl-gdi Using ===== -On Linux, building will create a drop-in alternative for libGL.so. To use it -set the environment variables: - - export LD_LIBRARY_PATH=$PWD/build/linux-x86_64-debug/lib:$LD_LIBRARY_PATH +On Linux, building will create a drop-in alternative for libGL.so into -or + build/foo/gallium/targets/libgl-xlib/libGL.so - export LD_LIBRARY_PATH=$PWD/build/linux-x86-debug/lib:$LD_LIBRARY_PATH +To use it set the LD_LIBRARY_PATH environment variable accordingly. For performance evaluation pass debug=no to scons, and use the corresponding lib directory without the "-debug" suffix. @@ -136,7 +86,7 @@ Profiling To profile llvmpipe you should pass the options - scons debug=no profile=yes <same-as-before> + scons build=profile <same-as-before> This will ensure that frame pointers are used both in C and JIT functions, and that no tail call optimizations are done by gcc. @@ -200,5 +150,4 @@ Development Notes interfaces very closely, and appear to be complete enough for code generation. See http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html - for a stand-alone example. - See the llvm-c/Core.h file for reference. + for a stand-alone example. See the llvm-c/Core.h file for reference. diff --git a/src/gallium/drivers/llvmpipe/SConscript b/src/gallium/drivers/llvmpipe/SConscript index 49950153a4..26b258b956 100644 --- a/src/gallium/drivers/llvmpipe/SConscript +++ b/src/gallium/drivers/llvmpipe/SConscript @@ -9,8 +9,6 @@ if not env['llvm']: env = env.Clone() -env.Tool('udis86') - env.Append(CPPPATH = ['.']) env.CodeGenerate( @@ -78,6 +76,8 @@ llvmpipe = env.ConvenienceLibrary( lp_tile_soa_os, ]) +env.Alias('llvmpipe', llvmpipe) + if env['platform'] != 'embedded': env = env.Clone() @@ -92,7 +92,7 @@ if env['platform'] != 'embedded': 'sincos', ] - if not msvc: + if not env['msvc']: tests.append('round') for test in tests: diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c index d358a98394..decf3bd449 100644 --- a/src/gallium/drivers/llvmpipe/lp_rast.c +++ b/src/gallium/drivers/llvmpipe/lp_rast.c @@ -485,8 +485,11 @@ static void lp_rast_end_query(struct lp_rasterizer_task *task, const union lp_rast_cmd_arg arg) { - task->query->count[task->thread_index] += task->vis_counter; - task->query = NULL; + assert(task->query); + if (task->query) { + task->query->count[task->thread_index] += task->vis_counter; + task->query = NULL; + } } diff --git a/src/gallium/drivers/llvmpipe/lp_setup_line.c b/src/gallium/drivers/llvmpipe/lp_setup_line.c index 827413bb33..29c231714e 100644 --- a/src/gallium/drivers/llvmpipe/lp_setup_line.c +++ b/src/gallium/drivers/llvmpipe/lp_setup_line.c @@ -569,7 +569,10 @@ try_setup_line( struct lp_setup_context *setup, return TRUE; } - u_rect_find_intersection(&setup->draw_region, &bbox); + /* Can safely discard negative regions: + */ + bbox.x0 = MAX2(bbox.x0, 0); + bbox.y0 = MAX2(bbox.y0, 0); line = lp_setup_alloc_triangle(scene, key->num_inputs, @@ -680,24 +683,26 @@ try_setup_line( struct lp_setup_context *setup, * these planes elsewhere. */ if (nr_planes == 8) { + const struct u_rect *scissor = &setup->scissor; + plane[4].dcdx = -1; plane[4].dcdy = 0; - plane[4].c = 1-bbox.x0; + plane[4].c = 1-scissor->x0; plane[4].eo = 1; plane[5].dcdx = 1; plane[5].dcdy = 0; - plane[5].c = bbox.x1+1; + plane[5].c = scissor->x1+1; plane[5].eo = 0; plane[6].dcdx = 0; plane[6].dcdy = 1; - plane[6].c = 1-bbox.y0; + plane[6].c = 1-scissor->y0; plane[6].eo = 1; plane[7].dcdx = 0; plane[7].dcdy = -1; - plane[7].c = bbox.y1+1; + plane[7].c = scissor->y1+1; plane[7].eo = 0; } diff --git a/src/gallium/drivers/llvmpipe/lp_setup_tri.c b/src/gallium/drivers/llvmpipe/lp_setup_tri.c index 4ab0b72a57..bfb6bf277b 100644 --- a/src/gallium/drivers/llvmpipe/lp_setup_tri.c +++ b/src/gallium/drivers/llvmpipe/lp_setup_tri.c @@ -295,7 +295,12 @@ do_triangle_ccw(struct lp_setup_context *setup, return TRUE; } - u_rect_find_intersection(&setup->draw_region, &bbox); + /* Can safely discard negative regions, but need to keep hold of + * information about when the triangle extends past screen + * boundaries. See trimmed_box in lp_setup_bin_triangle(). + */ + bbox.x0 = MAX2(bbox.x0, 0); + bbox.y0 = MAX2(bbox.y0, 0); tri = lp_setup_alloc_triangle(scene, key->num_inputs, @@ -501,24 +506,26 @@ do_triangle_ccw(struct lp_setup_context *setup, * these planes elsewhere. */ if (nr_planes == 7) { + const struct u_rect *scissor = &setup->scissor; + plane[3].dcdx = -1; plane[3].dcdy = 0; - plane[3].c = 1-bbox.x0; + plane[3].c = 1-scissor->x0; plane[3].eo = 1; plane[4].dcdx = 1; plane[4].dcdy = 0; - plane[4].c = bbox.x1+1; + plane[4].c = scissor->x1+1; plane[4].eo = 0; plane[5].dcdx = 0; plane[5].dcdy = 1; - plane[5].c = 1-bbox.y0; + plane[5].c = 1-scissor->y0; plane[5].eo = 1; plane[6].dcdx = 0; plane[6].dcdy = -1; - plane[6].c = bbox.y1+1; + plane[6].c = scissor->y1+1; plane[6].eo = 0; } @@ -559,6 +566,7 @@ lp_setup_bin_triangle( struct lp_setup_context *setup, int nr_planes ) { struct lp_scene *scene = setup->scene; + struct u_rect trimmed_box = *bbox; int i; /* What is the largest power-of-two boundary this triangle crosses: @@ -572,6 +580,13 @@ lp_setup_bin_triangle( struct lp_setup_context *setup, int sz = floor_pot((bbox->x1 - (bbox->x0 & ~3)) | (bbox->y1 - (bbox->y0 & ~3))); + /* Now apply scissor, etc to the bounding box. Could do this + * earlier, but it confuses the logic for tri-16 and would force + * the rasterizer to also respect scissor, etc, just for the rare + * cases where a small triangle extends beyond the scissor. + */ + u_rect_find_intersection(&setup->draw_region, &trimmed_box); + /* Determine which tile(s) intersect the triangle's bounding box */ if (dx < TILE_SIZE) @@ -626,15 +641,16 @@ lp_setup_bin_triangle( struct lp_setup_context *setup, struct lp_rast_plane *plane = GET_PLANES(tri); int c[MAX_PLANES]; int ei[MAX_PLANES]; + int eo[MAX_PLANES]; int xstep[MAX_PLANES]; int ystep[MAX_PLANES]; int x, y; - int ix0 = bbox->x0 / TILE_SIZE; - int iy0 = bbox->y0 / TILE_SIZE; - int ix1 = bbox->x1 / TILE_SIZE; - int iy1 = bbox->y1 / TILE_SIZE; + int ix0 = trimmed_box.x0 / TILE_SIZE; + int iy0 = trimmed_box.y0 / TILE_SIZE; + int ix1 = trimmed_box.x1 / TILE_SIZE; + int iy1 = trimmed_box.y1 / TILE_SIZE; for (i = 0; i < nr_planes; i++) { c[i] = (plane[i].c + @@ -799,6 +815,16 @@ static void triangle_both( struct lp_setup_context *setup, { float area = calc_area(v0, v1, v2); + if (0) { + assert(!util_is_inf_or_nan(v0[0][0])); + assert(!util_is_inf_or_nan(v0[0][1])); + assert(!util_is_inf_or_nan(v1[0][0])); + assert(!util_is_inf_or_nan(v1[0][1])); + assert(!util_is_inf_or_nan(v2[0][0])); + assert(!util_is_inf_or_nan(v2[0][1])); + assert(!util_is_inf_or_nan(area)); + } + if (area > 0.0f) retry_triangle_ccw( setup, v0, v1, v2, setup->ccw_is_frontface ); else if (area < 0.0f) diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c index 9fbedac165..48971510f2 100644 --- a/src/gallium/drivers/llvmpipe/lp_state_fs.c +++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c @@ -761,11 +761,6 @@ generate_fragment(struct llvmpipe_screen *screen, } } -#ifdef PIPE_ARCH_X86 - /* Avoid corrupting the FPU stack on 32bit OSes. */ - lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0); -#endif - LLVMBuildRetVoid(builder); LLVMDisposeBuilder(builder); @@ -1067,11 +1062,11 @@ llvmpipe_bind_fs_state(struct pipe_context *pipe, void *fs) draw_flush(llvmpipe->draw); + llvmpipe->fs = (struct lp_fragment_shader *) fs; + draw_bind_fragment_shader(llvmpipe->draw, (llvmpipe->fs ? llvmpipe->fs->draw_data : NULL)); - llvmpipe->fs = fs; - llvmpipe->dirty |= LP_NEW_FS; } diff --git a/src/gallium/drivers/llvmpipe/lp_tile_soa.py b/src/gallium/drivers/llvmpipe/lp_tile_soa.py index e49f9c62fe..8df7b236fe 100644 --- a/src/gallium/drivers/llvmpipe/lp_tile_soa.py +++ b/src/gallium/drivers/llvmpipe/lp_tile_soa.py @@ -1,6 +1,6 @@ #!/usr/bin/env python -''' +CopyRight = ''' /************************************************************************** * * Copyright 2009 VMware, Inc. @@ -510,7 +510,7 @@ def main(): print '/* This file is autogenerated by lp_tile_soa.py from u_format.csv. Do not edit directly. */' print # This will print the copyright message on the top of this file - print __doc__.strip() + print CopyRight.strip() print print '#include "pipe/p_compiler.h"' print '#include "util/u_format.h"' |