OSDN Git Service
Christian König [Mon, 13 Dec 2010 23:43:53 +0000 (00:43 +0100)]
r600g: optimize away CF_INST_POP
If last instruction is an CF_INST_ALU we don't need to emit an
additional CF_INST_POP for stack clean up after an IF ELSE ENDIF.
Christian König [Sun, 12 Dec 2010 23:04:58 +0000 (00:04 +0100)]
[g3dvl] add reg_fixup_label to IF ELSE ENDIF
Christian König [Sun, 12 Dec 2010 21:55:23 +0000 (22:55 +0100)]
[g3dvl] move idct texture addr generation into vertex shader
Christian König [Sun, 12 Dec 2010 20:40:41 +0000 (21:40 +0100)]
[g3dvl] seperate texture addr generation from fetching
Christian König [Sun, 12 Dec 2010 14:37:54 +0000 (15:37 +0100)]
r600g: texture instructions also work fine with TGSI_FILE_INPUT
Christian König [Sun, 12 Dec 2010 14:13:39 +0000 (15:13 +0100)]
r600g: Why all this fiddling with tgsi_helper_copy?
tgsi_helper_copy is used on several occasions to copy a temporary result
into the real destination register to emulate writemasks for OP3 and
reduction operations. According to R600 ISA that's unnecessary.
This patch fixes this use for MAD, CMP and DP4.
Christian König [Sat, 11 Dec 2010 13:34:01 +0000 (14:34 +0100)]
[g3dvl] remove "Nouveau can't writemask tex dst regs" workaround
This now works with r600g, but will probably break Nouveau.
It's just way faster on r600 hardware, so let's fix Nouveau.
Christian König [Sat, 11 Dec 2010 13:00:59 +0000 (14:00 +0100)]
[g3dvl] move mv into vertex stream
Christian König [Sat, 11 Dec 2010 12:43:44 +0000 (13:43 +0100)]
Merge remote branch 'origin/master' into pipe-video
Conflicts:
src/gallium/drivers/r600/r600_pipe.c
src/gallium/drivers/r600/r600_texture.c
Mathias Fröhlich [Sun, 17 Oct 2010 16:16:30 +0000 (18:16 +0200)]
vbo: Avoid the copy to current in dlists if not required.
The current state is allowed to be undefined past DrawElements et al.
Consequently omit that copying at least in the display list code.
This pays us some percents performance.
Signed-off-by: Brian Paul <brianp@vmware.com>
Brian Paul [Fri, 10 Dec 2010 17:02:33 +0000 (10:02 -0700)]
mesa/meta: fix broken assertion, rename stack depth var
assert(current_save_state < MAX_META_OPS_DEPTH) did not compile.
Rename current_save_state to SaveStackDepth to be more consistent with
the style of the other fields.
Jerome Glisse [Fri, 10 Dec 2010 16:17:27 +0000 (11:17 -0500)]
r600g: fix bo size when creating bo from handle
Spoted by Alex Diomin
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Christian König [Fri, 10 Dec 2010 11:05:30 +0000 (12:05 +0100)]
[g3dvl] move z-coord generation for multiple render targets into vertex shader
Christian König [Fri, 10 Dec 2010 10:31:17 +0000 (11:31 +0100)]
[g3dvl] move idct stage 1 scaling into matrix
Christian König [Fri, 10 Dec 2010 10:05:11 +0000 (11:05 +0100)]
[g3dvl] double buffering seems to be enough
Vinson Lee [Fri, 10 Dec 2010 07:52:28 +0000 (23:52 -0800)]
mesa: Clean up header file inclusion in arbprogparse.h.
Xiang, Haihao [Fri, 10 Dec 2010 05:01:44 +0000 (13:01 +0800)]
i965: support for two-sided lighting on Sandybridge
VS places color attributes together so that SF unit can fetch the right
attribute according to object orientation. This fixes light issue in
mesa demo geartrain, projtex.
Xiang, Haihao [Fri, 10 Dec 2010 01:31:19 +0000 (09:31 +0800)]
meta: allow nested meta operations
_mesa_meta_CopyPixels results in nested meta operations on Sandybridge.
Previoulsy the second meta operation overrides all states saved by the
first meta function.
Eric Anholt [Thu, 9 Dec 2010 17:42:05 +0000 (09:42 -0800)]
i965: Add support for gen6 reladdr VS constant loading.
Eric Anholt [Thu, 9 Dec 2010 17:36:17 +0000 (09:36 -0800)]
i965: Add support for gen6 constant-index constant loading.
Chia-I Wu [Fri, 10 Dec 2010 02:44:03 +0000 (10:44 +0800)]
targets/egl: Improve st_GL.so loading.
When the application is not linked to any libGL*.so, loading st_GL.so
would give
/usr/local/lib/egl/st_GL.so: undefined symbol: _glapi_tls_Context
In that case, load libGL.so and try again. This works because
util_dl_open loads with RTLD_GLOBAL.
Fix "clear" OpenGL ES 1.1 demo.
Chia-I Wu [Fri, 10 Dec 2010 02:28:57 +0000 (10:28 +0800)]
target/egl: Fix misleading debug message.
When the name of the module is NULL, the process itself is dlopen()ed.
Do not print
libEGL debug: searching for st module (null)
Christian König [Fri, 10 Dec 2010 01:56:15 +0000 (02:56 +0100)]
[g3dvl] make render targets full configureable
Brian Paul [Fri, 10 Dec 2010 01:37:59 +0000 (18:37 -0700)]
draw/llvm: don't flush in vs_llvm_delete()
Fixes piglit glx-shader-sharing crash.
When shaders are shared by multiple contexts, the shader's draw context
pointer may point to a previously destroyed context. Dereferencing the
context pointer will lead to a crash.
In this case, simply removing the flushing code avoids the crash (the
exec and sse shader paths don't flush here either).
There's a deeper issue here, however, that needs examination. Shaders
should not keep pointers to contexts since contexts might get destroyed
at any time.
NOTE: This is a candidate for the 7.10 branch (after this has been
tested for a while).
Brian Paul [Thu, 9 Dec 2010 23:02:28 +0000 (16:02 -0700)]
draw/llvm: remove redundant comment
Brian Paul [Thu, 9 Dec 2010 23:02:12 +0000 (16:02 -0700)]
draw/llvm: remove extraneous conditional
Luca Barbieri [Wed, 1 Dec 2010 23:12:07 +0000 (15:12 -0800)]
glsl: Unroll loops with conditional breaks anywhere (not just the end)
Currently we only unroll loops with conditional breaks at the end, which is
the form that lower_jumps generates.
However, if breaks are not lowered, they tend to appear at the beginning, so
add support for a conditional break anywhere.
Signed-off-by: Luca Barbieri <luca@luca-barbieri.com>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Kenneth Graunke [Wed, 1 Dec 2010 23:33:56 +0000 (15:33 -0800)]
glsl: Consider the "else" branch when looking for loop breaks.
Found this bug by code inspection. Based off the comments just before
this code, the intent is to find whether the break exists in the "then"
branch or the "else" branch. However, the code actually looked at the
last instruction in the "then" branch twice.
Kenneth Graunke [Wed, 1 Dec 2010 23:06:47 +0000 (15:06 -0800)]
glsl: Clean up code by adding a new is_break() function.
Christian König [Fri, 10 Dec 2010 00:27:21 +0000 (01:27 +0100)]
[g3dvl] move interlaced dct handling into vertex shader
Christian König [Thu, 9 Dec 2010 23:29:59 +0000 (00:29 +0100)]
[g3dvl] move frame predition handling vom fragment into vertex shader
Christian König [Thu, 9 Dec 2010 23:13:32 +0000 (00:13 +0100)]
[g3dvl] avoid some ELSE blocks
Eric Anholt [Thu, 9 Dec 2010 22:31:43 +0000 (14:31 -0800)]
glsl: Correct the marking of InputsRead/OutputsWritten on in/out matrices.
If you used a constant array index to access the matrix, we'd flag a
bunch of wrong inputs/outputs as being used because the index was
multiplied by matrix columns and the actual used index was left out.
Fixes glsl-mat-attribute.
Eric Anholt [Thu, 9 Dec 2010 21:51:21 +0000 (13:51 -0800)]
intel: Set the swizzling for depth textures using the GL_RED depth mode.
Fixes depth-tex-modes-rg.
Eric Anholt [Thu, 9 Dec 2010 21:45:47 +0000 (13:45 -0800)]
intel: Use plain R8 and RG8 for COMPRESSED_RED and COMPRESSED_RG.
Fixes texture-rg.
Vinson Lee [Thu, 9 Dec 2010 22:17:17 +0000 (14:17 -0800)]
i965: Silence uninitialized variable warning.
Fixes this GCC warning.
brw_fs.cpp: In function 'brw_reg brw_reg_from_fs_reg(fs_reg*)':
brw_fs.cpp:3255: warning: 'brw_reg' may be used uninitialized in this function
Vinson Lee [Thu, 9 Dec 2010 22:03:58 +0000 (14:03 -0800)]
r600g: Fix SCons build.
Christian König [Thu, 9 Dec 2010 21:34:43 +0000 (22:34 +0100)]
[g3dvl] try a different buffer aproach
Christian König [Thu, 9 Dec 2010 21:28:31 +0000 (22:28 +0100)]
[g3dvl] fix buffer handling in mc code
Christian König [Thu, 9 Dec 2010 21:27:03 +0000 (22:27 +0100)]
[g3dvl] fix idct matrix upload
Jerome Glisse [Thu, 9 Dec 2010 21:16:22 +0000 (16:16 -0500)]
r600g: indentation cleanup
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Jerome Glisse [Wed, 8 Dec 2010 18:41:25 +0000 (13:41 -0500)]
r600g: specialized upload manager
Allow important performance increase by doing hw specific implementation
of the upload manager helper. Drop the range flushing that is not hit with
this code (and wasn't with previous neither). Performance improvement are
mostly visible on slow CPU.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Jerome Glisse [Thu, 9 Dec 2010 18:07:10 +0000 (13:07 -0500)]
r600g: avoid using pb* helper we are loosing previous cpu cycle with it
r600g is up to a point where all small CPU cycle matter and pb* turn
high on profile. It's mostly because pb try to be generic and thus
trigger unecessary check for r600g driver. To avoid having too much
abstraction & too much depth in the call embedded everythings into
r600_bo. Make code simpler & faster. The performance win highly depend
on the CPU & application considered being more important on slower CPU
and marginal/unoticeable on faster one.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Fabian Bieler [Wed, 1 Dec 2010 13:44:58 +0000 (14:44 +0100)]
glsl: fix lowering conditional returns in subroutines
this fix applies to the lower_sub_return 'branch' of the lower_jumps pass
Fixes piglit tests glsl-functions-5 and glsl-functions-6.
Eric Anholt [Thu, 9 Dec 2010 19:10:31 +0000 (11:10 -0800)]
i965: remove unused variable since brw_wm_glsl.c removal.
Eric Anholt [Thu, 9 Dec 2010 18:03:49 +0000 (10:03 -0800)]
i965: Set render_cache_read_write surface state bit on gen6 constant surfs.
This is said to be required in the spec, even when you aren't doing writes.
Eric Anholt [Wed, 16 Jun 2010 00:58:14 +0000 (17:58 -0700)]
i965: Set up the correct texture border color state struct for Ironlake.
This doesn't actually fix border color on Ironlake, but it appears to
be a requirement, and gen6 needs it too.
Eric Anholt [Thu, 9 Dec 2010 17:26:41 +0000 (09:26 -0800)]
i965: Clean up VS constant buffer location setup.
Eric Anholt [Thu, 9 Dec 2010 17:17:59 +0000 (09:17 -0800)]
i965: Fix VS constants regression pre-gen6.
Last minute change for gen6 with 0 used params dropped the multiply.
José Fonseca [Tue, 7 Dec 2010 16:59:25 +0000 (16:59 +0000)]
llvmpipe: Plug fence leaks.
Shuang He [Thu, 9 Dec 2010 00:49:51 +0000 (08:49 +0800)]
mesa: allow GLfixed arrays for OpenGL ES 2.0
Signed-off-by: Brian Paul <brianp@vmware.com>
Chia-I Wu [Thu, 9 Dec 2010 11:37:15 +0000 (19:37 +0800)]
mesa: Fix glTexCoordPointer with type GL_FIXED.
GL_FIXED is also a legal type for glTexCoordPointer.
Chia-I Wu [Thu, 9 Dec 2010 11:05:50 +0000 (19:05 +0800)]
mesa: Fix GL_FIXED arrays.
It is broken since
433e5e6defc85d8b1d6262aff990e3f5a8b37027.
Eric Anholt [Thu, 9 Dec 2010 01:00:54 +0000 (17:00 -0800)]
i965: Drop push-mode reladdr constant loading and always use constant_map.
This eases the gen6 implementation, which can only handle up to 32
registers of constants, while likely not penalizing real apps using
reladdr since all of those I've seen also end up hitting the pull
constant buffer. On gen6, the constant map means that simple NV VPs
fit under the 32-reg limit and now succeed. Fixes around 10 testcases.
Alex Deucher [Thu, 9 Dec 2010 05:13:21 +0000 (00:13 -0500)]
radeon: bump mip tree levels to 15
I forgot to bump this when I bumped the tex levels.
Brian Paul [Thu, 9 Dec 2010 04:38:35 +0000 (21:38 -0700)]
mesa: simplify target checking for TexImage functions
Brian Paul [Thu, 9 Dec 2010 04:38:35 +0000 (21:38 -0700)]
mesa: revamp error checking for compressed texture images
Simplify some code, remove unneeded checks, etc.
Chad Versace [Tue, 7 Dec 2010 18:35:36 +0000 (10:35 -0800)]
glsl: In ast_to_hir, check sampler array indexing
Raise error if a sampler array is indexed with a non-constant expression.
From section 4.1.7 of the GLSL 1.30 spec:
"Samplers aggregated into arrays within a shader (using square
brackets [ ]) can only be indexed with integral constant
expressions [...]."
Christian König [Wed, 8 Dec 2010 22:37:57 +0000 (23:37 +0100)]
[g3dvl] split mc code into state and buffer
Christian König [Wed, 8 Dec 2010 21:35:30 +0000 (22:35 +0100)]
[g3dvl] move buffer mapping/unmapping out of flush
Christian König [Wed, 8 Dec 2010 20:28:38 +0000 (21:28 +0100)]
[g3dvl] replace IF THEN ELSE with CMP statement
Christian König [Wed, 8 Dec 2010 19:48:34 +0000 (20:48 +0100)]
[g3dvl] remove shadow buffering of vertex buffers
Eric Anholt [Wed, 8 Dec 2010 18:59:58 +0000 (10:59 -0800)]
i965: Drop KIL_NV from the ff/ARB_fp path since it was only used for GLSL.
Eric Anholt [Wed, 8 Dec 2010 18:57:22 +0000 (10:57 -0800)]
i965: Use the new pixel mask location for gen6 ARB_fp KIL instructions.
Fixes:
fp-kil
fp-generic/kil-swizzle.
Eric Anholt [Wed, 8 Dec 2010 18:17:24 +0000 (10:17 -0800)]
i965: Set the render target index in gen6 fixed-function/ARB_fp path.
Fixes:
fbo-drawbuffers2-blend
fbo-drawbuffers2-colormask
Eric Anholt [Wed, 8 Dec 2010 18:12:20 +0000 (10:12 -0800)]
i965: Set up the per-render-target blend state on gen6.
This will let us get EXT_draw_buffers2 blending and colormasking working.
Christian König [Wed, 8 Dec 2010 18:41:01 +0000 (19:41 +0100)]
[g3dvl] remove mb type handler structure
Christian König [Wed, 8 Dec 2010 18:18:55 +0000 (19:18 +0100)]
[g3dvl] move macro block type handling into shaders part 3
Eric Anholt [Wed, 8 Dec 2010 17:52:56 +0000 (09:52 -0800)]
i965: Set up the color masking for the first drawbuffer on gen6.
Fixes glean/maskedClear
Christian König [Wed, 8 Dec 2010 17:42:48 +0000 (18:42 +0100)]
[g3dvl] moving macroblock type handling into shaders part 2
Christian König [Wed, 8 Dec 2010 17:26:58 +0000 (18:26 +0100)]
[g3dvl] start moving macroblock type handling into shaders
Christian König [Wed, 8 Dec 2010 15:11:29 +0000 (16:11 +0100)]
[g3dvl] allways use all motion vectors
reprogramming the vertex buffers takes more time than pumping
all motion vectors through the vertex shader
Chia-I Wu [Wed, 8 Dec 2010 14:33:07 +0000 (22:33 +0800)]
mesa: Do not advertise GL_OES_texture_3D.
GL_OES_texture_3D has a GLSL counterpart. Since it is not implemented,
GL_OES_texture_3D should not be advertised.
Chia-I Wu [Tue, 7 Dec 2010 13:33:53 +0000 (21:33 +0800)]
vbo: Fix GLES2 glVertexAttrib.
Attribute 0 has no special meaning in GLES2. Add VertexAttrib4f_nopos
for that purpose and make _es_VertexAttrib* call the new function.
Rename _vbo_* to _es_* to avoid confusion. These functions are only
used by GLES, and now some of them (_es_VertexAttrib*) even behave
differently than vbo_VertexAttrib*.
Chia-I Wu [Tue, 7 Dec 2010 13:22:28 +0000 (21:22 +0800)]
vbo: Drop second ATTR macro.
There is no need to have a special version of ATTR for
!FEATURE_beginend, since
81ccb3e2ce708619f4c23537a237d61bdffdd35f.
Christian König [Wed, 8 Dec 2010 13:50:28 +0000 (14:50 +0100)]
[g3dvl] split empty block handling
Empty block handling is split between vertex shader (x-axis)
and fragment shader (y-axis).
Brian Paul [Wed, 8 Dec 2010 13:44:42 +0000 (06:44 -0700)]
configure: use llvm-config --cppflags instead of --cflags
Brian Paul [Wed, 8 Dec 2010 04:37:20 +0000 (21:37 -0700)]
mesa: make _mesa_test_proxy_teximage() easier to read
Brian Paul [Wed, 8 Dec 2010 04:37:20 +0000 (21:37 -0700)]
mesa: consolidate glCompressedTexImage1/2/3D() functions
Brian Paul [Wed, 8 Dec 2010 04:37:20 +0000 (21:37 -0700)]
mesa: consolidate glCopyTexSubImage1/2/3D() functions
Brian Paul [Wed, 8 Dec 2010 04:37:19 +0000 (21:37 -0700)]
mesa: consolidate glCopyTexImage1/2D() code
Brian Paul [Wed, 8 Dec 2010 04:37:19 +0000 (21:37 -0700)]
mesa: consolidate the glTexSubImage1/2/3D() functions
Brian Paul [Wed, 8 Dec 2010 04:37:19 +0000 (21:37 -0700)]
mesa: simplify proxy texture code in texture_error_check()
Marek Olšák [Wed, 8 Dec 2010 03:27:58 +0000 (04:27 +0100)]
r300/compiler: remove at least unused immediates if externals cannot be removed
Marek Olšák [Tue, 7 Dec 2010 20:57:18 +0000 (21:57 +0100)]
r300/compiler: make lowering passes possibly use up to two less temps
CMP may now use two less temps, other non-native instructions may end up
using one less temp, except for SIN/COS/SCS, which I am leaving unchanged
for now.
This may reduce register pressure inside loops, because the register
allocator doesn't do a very good job there.
Marek Olšák [Wed, 8 Dec 2010 00:59:33 +0000 (01:59 +0100)]
r300/compiler: handle DPH and XPD in rc_compute_sources_for_writemask
This bug can only be triggered if you put deadcode before native rewrite.
Marek Olšák [Wed, 8 Dec 2010 00:58:00 +0000 (01:58 +0100)]
r300/compiler: do not print pair/tex/presub program stats for vertex shaders
Marek Olšák [Tue, 7 Dec 2010 23:18:05 +0000 (00:18 +0100)]
r300/compiler: cleanup rc_run_compiler
Marek Olšák [Tue, 7 Dec 2010 22:34:21 +0000 (23:34 +0100)]
r300/compiler: add a function to query program stats (alu, tex, temps..)
Marek Olšák [Tue, 7 Dec 2010 20:45:34 +0000 (21:45 +0100)]
r300/compiler: don't terminate regalloc if we surpass max temps limit
The same check is already in a later pass (translate_vertex_program).
Eric Anholt [Wed, 8 Dec 2010 03:29:26 +0000 (19:29 -0800)]
i965: Don't try to store gen6 (float) blend constant color in bytes.
Fixes glean/blendFunc
Ian Romanick [Wed, 8 Dec 2010 03:00:44 +0000 (19:00 -0800)]
linker: Fix regressions caused by previous commit
That's what I get for not running piglit before pushing.
Don't try to patch types of unsized arrays when linking fails.
Don't try to patch types of unsized arrays that are shared between
shader stages.
Ian Romanick [Wed, 8 Dec 2010 02:30:33 +0000 (18:30 -0800)]
linker: Ensure that unsized arrays have a size after linking
Fixes piglit test case glsl-vec-array (bugzilla #31908).
NOTE: This bug does not affect 7.9, but I think this patch is a
candiate for the 7.9 branch anyway.
Eric Anholt [Wed, 8 Dec 2010 01:07:05 +0000 (17:07 -0800)]
i965: Fix flipped value of the not-embedded-in-if on gen6.
Fixes:
glean/glsl1-! (not) operator (1, fail)
glean/glsl1-! (not) operator (1, pass)
Christian König [Wed, 8 Dec 2010 01:12:24 +0000 (02:12 +0100)]
[g3dvl] split idct code into state and buffers
Ian Romanick [Wed, 8 Dec 2010 00:27:22 +0000 (16:27 -0800)]
glsl: Inherrit type of declared variable from initializer
Types of declared variables and their initializer must match excatly
except for unsized arrays. Previously the type inherritance for
unsized arrays happened implicitly in the emitted assignment.
However, this assignment is never emitted for uniforms. Now that type
is explicitly copied unconditionally.
Fixes piglit test array-compare-04.vert (bugzilla #32035) and
glsl-array-uniform-length (bugzilla #31985).
NOTE: This is a candidate for the 7.9 branch.
Eric Anholt [Tue, 7 Dec 2010 22:50:50 +0000 (14:50 -0800)]
i965: Work around gen6 ignoring source modifiers on math instructions.
With the change of extended math from having the arguments moved into
mrfs and handed off through message passing to being directly hooked
up to the EU, it looks like the piece for doing source modifiers
(negate and abs) was left out.
Fixes:
fog-modes
glean/fp1-ARB_fog_exp test
glean/fp1-ARB_fog_exp2 test
glean/fp1-Computed fog exp test
glean/fp1-Computed fog exp2 test
ext_fog_coord-modes
Eric Anholt [Tue, 7 Dec 2010 22:46:27 +0000 (14:46 -0800)]
i965: Add disabled debug code for dumping out the WM constant payload.
This can significantly ease thinking about the asm.
Ian Romanick [Tue, 7 Dec 2010 22:50:34 +0000 (14:50 -0800)]
i965: Correctly emit constants for aggregate types (array, matrix, struct)
Previously the code only handled scalars and vectors. This new code
is modeled somewhat after similar code in ir_to_mesa.
Reviewed-by: Eric Anholt <eric@anholt.net>
Jerome Glisse [Tue, 7 Dec 2010 20:15:58 +0000 (15:15 -0500)]
r600g: fix userspace fence against lastest kernel
R6XX GPU doesn't like to have two partial flush writting
back to memory in row without a prior flush of the pipeline.
Add PS_PARTIAL_FLUSH to flush all work between the CP and
the ES, GS, VS, PS shaders.
Thanks a lot to Alban Browaeys (prahal on irc) for investigating
this issue.
Signed-off-by: Alban Browaeys <prahal@yahoo.com>
Signed-off-by: Jerome Glisse <jglisse@redhat.com>