OSDN Git Service
Dave Airlie [Mon, 30 Aug 2010 05:19:20 +0000 (15:19 +1000)]
r600g: add initial relative support to assembler
passes another ~20 piglits.
/me starts to run out low hanging fruit around now.
Ian Romanick [Mon, 30 Aug 2010 05:07:49 +0000 (22:07 -0700)]
linker: Treat sized and unsized array types as the same
If two shaders contain variables declared with array types that have
the same base type but one is sized and the other is not, linking
should succeed. I'm not super pleased with the way this is
implemented, and I am more convinced than ever that we need more
linker tests. We especially need "negative" tests.
Fixes bugzilla #29697 and piglit test glsl-link-array-01.
Dave Airlie [Mon, 30 Aug 2010 04:41:09 +0000 (14:41 +1000)]
r600g: add KILP support
passes glsl1-discard tests
Dave Airlie [Mon, 30 Aug 2010 04:13:01 +0000 (14:13 +1000)]
r600g: fix SSG and op3 neg writing
8 more piglits, mainly the two SSG tests.
Dave Airlie [Fri, 27 Aug 2010 06:08:55 +0000 (16:08 +1000)]
r600g : add basic loop support.
Adds BGNLOOP, BRK, CONT, ENDLOOP support, ported from r600c.
17 piglits more on r300g.tests.
Bas Nieuwenhuizen [Sun, 29 Aug 2010 09:19:23 +0000 (11:19 +0200)]
r600g: use the values from the correct literals
Created an array for literals as we should not always use the last declared literal.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Bas Nieuwenhuizen [Sun, 29 Aug 2010 09:19:22 +0000 (11:19 +0200)]
r600g: added literals where needed for POW instruction
Fixes size calculation for the bytecode buffer.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Richard Li [Sun, 29 Aug 2010 23:27:46 +0000 (19:27 -0400)]
evergreen : fix shader const allocation and instruction bugs.
Vinson Lee [Sun, 29 Aug 2010 21:05:07 +0000 (14:05 -0700)]
glsl: Initialize data in ast_function_expression::hir.
Completely initialize data that is passed to ir_constant constructor.
Fixes piglit glsl-orangebook-ch06-bump valgrind uninitialized variable
error on softpipe and llvmpipe.
Vinson Lee [Sun, 29 Aug 2010 20:15:56 +0000 (13:15 -0700)]
glsl: Completely initialize value member in ir_constant constructor.
The
ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
did not completely initialize the entire value member.
Fixes piglit glsl-fs-sampler-numbering-2 valgrind uninitialized value
error in softpipe and llvmpipe.
Vinson Lee [Sun, 29 Aug 2010 19:19:57 +0000 (12:19 -0700)]
glsl: Initialize variable in ir_swizzle::constant_expression_value.
Complete initialize data passed to ir_constant constructor.
Fixes piglit glsl-mat-from-int-ctor-02 valgrind unintialized variable
error with softpipe and llvmpipe.
Vinson Lee [Sun, 29 Aug 2010 18:48:02 +0000 (11:48 -0700)]
glsl: Initialize data in read_constant.
Completely initialize data that is passed into a ir_constant constructor.
Fixes piglit glsl-fs-mix valgrind uninitialized variable error on
softpipe and llvmpipe.
Vinson Lee [Sun, 29 Aug 2010 18:31:33 +0000 (11:31 -0700)]
ir_to_mesa: Initialize variable in ir_to_mesa_visitor::visit.
Fixes piglit glsl-fs-loop valgrind uninitialized value error on softpipe
and llvmpipe.
Dave Airlie [Sat, 28 Aug 2010 09:01:49 +0000 (19:01 +1000)]
r600g: drop file I added by mistake in a previous commit
Vinson Lee [Sun, 29 Aug 2010 06:55:51 +0000 (23:55 -0700)]
glsl: Initialize the rest of values of ir_constant::value.
Fixes valgrind uninitialized value errors in the piglit shader tests for
softpipe and llvmpipe.
Vinson Lee [Sun, 29 Aug 2010 05:15:37 +0000 (22:15 -0700)]
gallium: Remove unnecessary header from p_state.h.
Remove p_screen.h.
Vinson Lee [Sun, 29 Aug 2010 05:12:55 +0000 (22:12 -0700)]
nvfx: Remove util_is_pot in favor of util_is_power_of_two.
This is a follow up to commit
89b2897220acfacdc431f138377fbcec9f0ea812.
Vinson Lee [Sun, 29 Aug 2010 04:46:41 +0000 (21:46 -0700)]
util: Include missing header in u_draw.h.
Include p_state.h for complete type to pipe_draw_info.
Vinson Lee [Sun, 29 Aug 2010 04:42:28 +0000 (21:42 -0700)]
util: Add forward declaration in u_transfer.h.
Marek Olšák [Sun, 29 Aug 2010 04:08:24 +0000 (06:08 +0200)]
r300g,u_blitter: use u_framebuffer
Removing another function duplication in u_blitter.
Marek Olšák [Sun, 29 Aug 2010 04:03:39 +0000 (06:03 +0200)]
util: remove util_is_pot in favor of util_is_power_of_two
The function was duplicated.
Eric Anholt [Sat, 28 Aug 2010 21:43:50 +0000 (14:43 -0700)]
i965: Add initial support for texturing to the new FS backend.
Fixes 11 piglit tests.
Eric Anholt [Sun, 29 Aug 2010 02:55:53 +0000 (19:55 -0700)]
i965: Add disasm for gen5 sampler messages.
Cedric Vivier [Sun, 29 Aug 2010 03:01:46 +0000 (20:01 -0700)]
i965: Move libdrm/C++ hack introduced in
fa2deb3d to intel_context.h
Fixes build on Linux/GCC 4.4 as libdrm includes are also used by other
brw_fs_*.cpp files.
Bug #29855
Vinson Lee [Sat, 28 Aug 2010 21:24:42 +0000 (14:24 -0700)]
st/mesa: Include missing header in st_mesa_to_tgsi.c.
Include p_screen.h for complete type to pipe_screen.
Vinson Lee [Sat, 28 Aug 2010 21:21:28 +0000 (14:21 -0700)]
softpipe: Include missing header in sp_flush.c.
Include p_screen.h for complete type to pipe_screen.
Vinson Lee [Sat, 28 Aug 2010 21:18:57 +0000 (14:18 -0700)]
draw: Include missing header in draw_vs_llvm.c.
Include p_screen.h for completely type to pipe_screen.
Vinson Lee [Sat, 28 Aug 2010 21:14:33 +0000 (14:14 -0700)]
llvmpipe: Include missing header in lp_flush.c.
Include p_screen.h for complete type to pipe_screen.
Marek Olšák [Sat, 28 Aug 2010 05:54:36 +0000 (07:54 +0200)]
r300g: fix blitting between 2D NPOT mipmaps
Even though MIP filtering is not supported, we can bind an arbitrary mipmap
as the zero mipmap level.
NPOT textures now follow GL_TEXTURE_BASE_LEVEL and GL_TEXTURE_MIN_LOD.
This fixes piglit/fbo-copyteximage.
Mathias Fröhlich [Sat, 28 Aug 2010 16:16:41 +0000 (18:16 +0200)]
r300g: fix min/max lod computation
Marek Olšák [Sat, 28 Aug 2010 05:51:55 +0000 (07:51 +0200)]
r300g: set the correct value in PVS_NUM_CNTLRS
As per docs.
Vinson Lee [Sat, 28 Aug 2010 07:29:02 +0000 (00:29 -0700)]
llvmpipe: Remove unnecessary header.
Ian Romanick [Thu, 19 Aug 2010 00:53:47 +0000 (17:53 -0700)]
glsl2: Decompose matrix comparison into vector operations
Jakob Bornecrantz [Fri, 27 Aug 2010 23:30:00 +0000 (00:30 +0100)]
trace: Don't try to dump the rgba array if null
Eric Anholt [Fri, 27 Aug 2010 22:34:42 +0000 (15:34 -0700)]
glsl: Rename a couple of common variable names in mat_op_to_vec.
It was easy while typing implementations to accidentally overwrite the
original expression or assignment variables.
Eric Anholt [Fri, 27 Aug 2010 22:32:59 +0000 (15:32 -0700)]
glsl: Make mat_op_to_vec allocate out of the IR's parent.
This will reduce memory consumption of compiled shaders by not
dragging optimized-out children around.
Eric Anholt [Fri, 27 Aug 2010 23:04:16 +0000 (16:04 -0700)]
i965: Set the pop count on BRK/CONT inside of an if statement in the FS.
This is the same as
8de8c97275e9555183a7e8f2238143657bbe60b2 for the
VS, and fixes glsl-fs-if-nested-loop and the mandelbrot demo.
Bug #29498
Eric Anholt [Fri, 27 Aug 2010 22:24:23 +0000 (15:24 -0700)]
glsl: Regenerate for double destroy fix.
Eric Anholt [Fri, 27 Aug 2010 22:23:29 +0000 (15:23 -0700)]
glsl: Protect against double compiler-destroy.
DRI was doing teardown when we close the last screen, then an atexit()
was added to call it as well.
Eric Anholt [Fri, 27 Aug 2010 21:43:39 +0000 (14:43 -0700)]
dri: Get prototype for _mesa_destroy_compiler().
Bug #29665.
Eric Anholt [Fri, 27 Aug 2010 21:15:42 +0000 (14:15 -0700)]
i965: Don't strip negate/abs flags when assigning uniform locations.
Fixes glsl-algebraic-sub-zero-4.
Eric Anholt [Fri, 27 Aug 2010 21:09:05 +0000 (14:09 -0700)]
i965: Add missing handling for BRW_OPCODE_SEL.
Fixes 4 piglit tests about min, max, and clamp.
Eric Anholt [Fri, 27 Aug 2010 20:49:09 +0000 (13:49 -0700)]
i965: Mask out higher bits of the result of BRW_CMP producing a boolean.
When it says it sets the LSB, that's not just a hint as to where the
result goes. Only the LSB is modified. Fixes 20 piglit cases.
Eric Anholt [Fri, 27 Aug 2010 20:47:37 +0000 (13:47 -0700)]
i965: Fix the types of immediate integer values.
When we're trying to do integer ops, handing a float in doesn't help.
Eric Anholt [Fri, 27 Aug 2010 20:29:45 +0000 (13:29 -0700)]
i965: Add translation for RNDD and RNDZ.
Fixes:
glsl-fs-any.
glsl1-integer division with uniform var
Eric Anholt [Fri, 27 Aug 2010 20:24:41 +0000 (13:24 -0700)]
i965: Add support for ir_binop_mod using do_mod_to_fract.
Fixes glsl-fs-mod.
Jerome Glisse [Fri, 27 Aug 2010 20:52:29 +0000 (16:52 -0400)]
r600g: fix vbo size
Silence the kernel, vbo size is size - 1.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Brian Paul [Fri, 27 Aug 2010 20:35:52 +0000 (14:35 -0600)]
mesa: move null ptr check (fdo bug 29843)
Nick Bowler [Fri, 27 Aug 2010 00:52:30 +0000 (17:52 -0700)]
mesa: Use the format info to identify packed depth/stencil formats.
Due to a misunderstanding of the Z24_X8 and X8_Z24 formats, the earlier
patch created depth/stencil wrappers for them. This broke swrast.
Use the format info instead, which only identifies Z24_S8 and S8_Z24 as
packed depth/stencil. It also has the advantage of being nicer code.
Signed-off-by: Nick Bowler <nbowler@draconx.ca>
Signed-off-by: Brian Paul <brianp@vmware.com>
Brian Paul [Fri, 27 Aug 2010 17:54:32 +0000 (11:54 -0600)]
mesa: use atexit() handler to release GLSL compiler memory
This releases a bunch of memory that was showing up as leaks with
valgrind.
If atexit() isn't widely supported we may need to add some #ifdef
tests around the call.
Brian Paul [Fri, 27 Aug 2010 17:53:19 +0000 (11:53 -0600)]
glsl2: restructure header file for C++ and C inclusion
As it was, the header could not be cleanly #included by a C source.
Brian Paul [Fri, 27 Aug 2010 17:15:03 +0000 (11:15 -0600)]
glsl2: remove 'extern' keyword in .c file
Eric Anholt [Fri, 27 Aug 2010 20:19:28 +0000 (13:19 -0700)]
i965: Fix swapped instructions in ir_unop_abs and ir_unop_neg.
Fixes glsl-fs-neg and 5 other tests.
Eric Anholt [Fri, 27 Aug 2010 20:14:15 +0000 (13:14 -0700)]
i965: Add generate() handling for AND, OR, XOR.
10 more piglit tests pass.
Eric Anholt [Fri, 27 Aug 2010 19:54:12 +0000 (12:54 -0700)]
i965: Add support for if instructions in the new FS backend.
20 more piglit tests pass.
Eric Anholt [Fri, 27 Aug 2010 19:53:48 +0000 (12:53 -0700)]
i965: When encountering an unknown opcode in new FS backend, print its name.
Eric Anholt [Fri, 27 Aug 2010 19:19:30 +0000 (12:19 -0700)]
i965: Fix the maximum grf counting in the new FS backend.
glsl-algebraic-rcp-rsq managed to use 33 registers, and we claimed to
only use 32, so the write to g32 would go stomping over the precious
g0 of some other thread.
Eric Anholt [Fri, 27 Aug 2010 19:02:53 +0000 (12:02 -0700)]
i965: Validate the IR tree after doing our custom optimization passes.
This wouldn't catch the last failure fixed in them, because we don't
validate assignments well (due to the fact that we've got a pretty
glaring inconsistency in how we handle assignment writemasking), but
it could catch other failure we may produce.
Eric Anholt [Fri, 27 Aug 2010 17:44:04 +0000 (10:44 -0700)]
i965: Add a bit of support for matrices to the new FS.
Eric Anholt [Fri, 27 Aug 2010 18:05:47 +0000 (11:05 -0700)]
i965: Fix destination writemasking in the new FS.
Eric Anholt [Fri, 27 Aug 2010 19:00:29 +0000 (12:00 -0700)]
i965: Fix swizzling in vector splitting for the new FS backend.
We weren't smearing a component of a split RHS out to reach an unsplit
LHS's writemask, so gl_FragColor (always unsplit) would often get
uninitialized values.
Fixes: glsl-algebraic-add-add-1 (and probably many others).
Zack Rusin [Fri, 27 Aug 2010 17:09:38 +0000 (13:09 -0400)]
graw: fix the build (missing header)
Brian Paul [Fri, 27 Aug 2010 16:51:47 +0000 (10:51 -0600)]
mesa: free the fallback texture object in free_shared_state()
Brian Paul [Fri, 27 Aug 2010 16:21:59 +0000 (10:21 -0600)]
mesa: fix double-underscore naming
Keith Whitwell [Fri, 27 Aug 2010 12:40:23 +0000 (13:40 +0100)]
llvmpipe: use util_iround in place of round
Fix mingw build.
Keith Whitwell [Fri, 27 Aug 2010 12:29:00 +0000 (13:29 +0100)]
util: fix typo in MAX4
Thanks to Michal for spotting it.
Keith Whitwell [Fri, 27 Aug 2010 10:03:58 +0000 (11:03 +0100)]
llvmpipe: eliminate tri->dx, tri->dy values
Use an internal struct for line setup information.
Hui Qi Tay [Tue, 10 Aug 2010 10:41:32 +0000 (11:41 +0100)]
llvmpipe: point sprites rasterization
Point sprites now done in the rasterizer setup code instead of
going through the draw module.
Hui Qi Tay [Fri, 27 Aug 2010 09:46:19 +0000 (10:46 +0100)]
llvmpipe: native point rasterization with better pixel rasterization
A few subpixel_snap and fixed width changes.
Conflicts:
src/gallium/drivers/llvmpipe/lp_setup_point.c
Hui Qi Tay [Fri, 27 Aug 2010 09:37:09 +0000 (10:37 +0100)]
llvmpipe: native point rasterization
Conflicts:
src/gallium/drivers/llvmpipe/lp_setup_context.h
src/gallium/drivers/llvmpipe/lp_setup_line.c
src/gallium/drivers/llvmpipe/lp_setup_tri.c
Keith Whitwell [Thu, 26 Aug 2010 19:09:22 +0000 (20:09 +0100)]
llvmpipe: update line rasterization code to current master
Keith Whitwell [Thu, 26 Aug 2010 19:03:03 +0000 (20:03 +0100)]
util: add MIN4, MAX4
Hui Qi Tay [Wed, 4 Aug 2010 16:13:39 +0000 (17:13 +0100)]
llvmpipe: native line rasterization with correct pixel rasterization
Line rasterization that follows diamond exit rule.
Can still optimize logic for start/endpoints.
Hui Qi Tay [Mon, 19 Jul 2010 14:23:09 +0000 (15:23 +0100)]
llvmpipe: native rasterization for lines
Rasterize lines directly by treating them as 4-sided polygons.
Still need to check the exact pixel rasteration.
Keith Whitwell [Fri, 27 Aug 2010 10:18:11 +0000 (11:18 +0100)]
llvmpipe: add lp_setup_coef to makefile
Keith Whitwell [Sun, 22 Aug 2010 09:57:12 +0000 (10:57 +0100)]
llvmpipe: intrinsics version of triangle coeficient calculation
Looks nice, but makes almost no impact on performance - maybe
a percent or so in isosurf, nothing elsewhere. May be of use
later on.
Vinson Lee [Fri, 27 Aug 2010 07:40:42 +0000 (00:40 -0700)]
gallium: Remove unnecessary header from p_shader_tokens.h.
Remove p_compiler.h.
Vinson Lee [Fri, 27 Aug 2010 07:34:32 +0000 (00:34 -0700)]
tgsi: Include missing header in tgsi_parse.h.
Include p_compiler.h for boolean and INLINE symbols.
Vinson Lee [Fri, 27 Aug 2010 07:31:27 +0000 (00:31 -0700)]
tgsi: Include missing header in tgsi_info.h.
Include p_compiler.h for uint symbol.
Vinson Lee [Fri, 27 Aug 2010 07:26:59 +0000 (00:26 -0700)]
tgsi: Include missing header in tgsi_dump.h.
Include p_compiler.h for uint symbol.
Vinson Lee [Fri, 27 Aug 2010 07:15:04 +0000 (00:15 -0700)]
gallium: Remove unnecessary header from p_format.h.
Remove p_compiler.h.
Vinson Lee [Fri, 27 Aug 2010 07:07:38 +0000 (00:07 -0700)]
graw: Include missing header in graw.h.
Include p_compiler.h for PUBLIC symbol.
Vinson Lee [Fri, 27 Aug 2010 07:04:50 +0000 (00:04 -0700)]
r300g: Include missing header in r300_texture.h.
Include p_compiler.h for uint32_t and boolean symbols.
Kenneth Graunke [Fri, 27 Aug 2010 06:07:45 +0000 (23:07 -0700)]
Fix typo in function name "shading_laguage_version".
Dave Airlie [Fri, 27 Aug 2010 05:45:58 +0000 (15:45 +1000)]
Revert "r600g: simplify states"
This reverts commit
bd25e23bf3740f59ce8859848c715daeb9e9821f.
Apart from introducing a lot of hex magic numbers and being highly impenetable code,
it causes lots of lockups on an average piglit run that always runs without lockups.
Always run piglit before/after doing big things like this.
Dave Airlie [Wed, 25 Aug 2010 05:57:41 +0000 (15:57 +1000)]
r600g: add initial if/else/endif support
this adds handling for some more CF instructions and conditions
also adds parameter for stack size emission
These seem to pass on VS with the stack size hack but not on FS,
TODO: fix FS + stack size calcs
Dave Airlie [Wed, 25 Aug 2010 06:02:38 +0000 (16:02 +1000)]
r600g: optimise op2 and swapped op2 emission.
this makes op2 emission smaller, since it skips instructions
that don't write to the dst. not sure if this could have unwanted
side effects but try it and see.
Dave Airlie [Tue, 24 Aug 2010 01:47:56 +0000 (11:47 +1000)]
r600g: add exp support in theory.
though it isn't passing the test, and this instruction is pure bonghits.
Dave Airlie [Tue, 24 Aug 2010 00:29:33 +0000 (10:29 +1000)]
r600g: add DPH support.
Eric Anholt [Thu, 26 Aug 2010 23:39:41 +0000 (16:39 -0700)]
i965: Add preliminary support for uniforms to the new FS backend.
+269 piglits
Eric Anholt [Fri, 27 Aug 2010 00:04:30 +0000 (17:04 -0700)]
i965: Abort on gl_FragDepth in the new FS backend for now.
It hangs the GPU due to FB_WRITE handling being incomplete. There are
bigger issues to handle first.
Eric Anholt [Thu, 26 Aug 2010 23:59:55 +0000 (16:59 -0700)]
i965: Fix up and actually enable the NewShader and NewShaderProgram hooks.
Eric Anholt [Thu, 26 Aug 2010 22:43:00 +0000 (15:43 -0700)]
i965: Hack in avoidance of c++ reserved keyword in libdrm.
I'm also fixing this upstream in libdrm, but this avoids new libdrm
dependency for the moment.
Eric Anholt [Thu, 26 Aug 2010 21:42:06 +0000 (14:42 -0700)]
i965: Add GLSL IR-level source annotation and comments to new FS debug.
This should make debugging way easier, as now we have context for
reading large programs.
Eric Anholt [Thu, 26 Aug 2010 21:09:54 +0000 (14:09 -0700)]
i965: Use the implied move in brw_math() in the new FS.
Eric Anholt [Tue, 17 Aug 2010 04:53:02 +0000 (21:53 -0700)]
i965: Add support for in varyings to the new FS codegen.
At least some tests, like glsl-vs-sign, now work.
Eric Anholt [Mon, 16 Aug 2010 01:58:58 +0000 (18:58 -0700)]
i965: Start building the codegen visitor.
This can successfully emit a real program that generates magenta now.
Eric Anholt [Thu, 26 Aug 2010 19:12:00 +0000 (12:12 -0700)]
i965: Start building direct GLSL2 IR to 965 assembly codegen.
Our channel-expressions and vector-splitting changes now happen into a
private copy of the IR that we maintain for ourselves. Uniform
assignment still happens by the core, so we continue using Mesa IR
generation not just for swrast fallbacks but also for uniform values
(since there's no storage for their contents other than
shader_program->FragmentProgram->Parameters->ParameterValues). And
most importantly, at the moment no actual codegen is hooked up other
than emitting our favorite color to the framebuffer.
Eric Anholt [Thu, 26 Aug 2010 19:02:26 +0000 (12:02 -0700)]
i965: Add new pass to split vectors into scalar variables
Combined with the previous pass, this lets other optimization passes
do their work thanks to ir_tree_grafting. Still have regression in
instruction count with INTEL_NEW_FS, but register count is even
better.
Eric Anholt [Fri, 13 Aug 2010 09:20:40 +0000 (02:20 -0700)]
i965: Add a pass for the FS to reduce vector expressions down to scalar.
This is a step towards implementing a GLSL IR backend for the 965
fragment shader. Because it has downsides with the current codegen,
it is hidden under the environment variable INTEL_NEW_FS.
This results in an increase in instruction count at the moment (1444
-> 1752 for glsl-fs-raytrace, 345 -> 359 on my demo), because dot
products are turned into a series of multiplies and adds instead of a
custom expansion of MULs and MACs, and by not splitting the variable
types up we don't get tree grafting and thus there are extra moves of
temporary storage. However, register count drops for the non-GLSL
path (64 -> 56 on my demo shader) because the register allocator sees
all the sub-operations.