Eric Anholt [Fri, 15 Oct 2010 19:04:52 +0000 (12:04 -0700)]
i965: Set the type of the null register to fix gen6 FS comparisons.
We often use reg_null as the destination when setting up the flag
regs. However, on gen6 there aren't general implicit conversions to
destination types from src types, so the comparison to produce the
flag regs would be done on the integer result interpreted as a float.
Hilarity ensued.
Fixes 20 piglit cases.
Ian Romanick [Fri, 15 Oct 2010 17:32:02 +0000 (10:32 -0700)]
i965: Fix indentation after commit
3322fbaf
Ian Romanick [Fri, 15 Oct 2010 00:55:17 +0000 (17:55 -0700)]
linker: Trivial indention fix
Jakob Bornecrantz [Fri, 15 Oct 2010 14:46:10 +0000 (15:46 +0100)]
target-helpers: Remove per target software wrapper check
Instead of having a NAME_SOFTWARE check just use the GALLIUM_DRIVER
instead but set the default to native which is the same as not wrapped.
Jakob Bornecrantz [Fri, 15 Oct 2010 14:58:54 +0000 (15:58 +0100)]
egl: Remove unnecessary headers
Jakob Bornecrantz [Fri, 15 Oct 2010 14:57:55 +0000 (15:57 +0100)]
wrapper: Add a way to dewrap a pipe screen without destroying it
Jakob Bornecrantz [Fri, 15 Oct 2010 14:38:29 +0000 (15:38 +0100)]
wrapper: Fix spelling
Jakob Bornecrantz [Tue, 12 Oct 2010 17:41:24 +0000 (18:41 +0100)]
llvmpipe: Move makefile include to before targets
Or plain make inside of the directory wont build libllvmpipe.a
Xavier Chantry [Fri, 15 Oct 2010 13:53:13 +0000 (15:53 +0200)]
nv50: apply layout_mask to tile_flags
The tile_flags now store more than just nv50 page table entry bits.
Keith Whitwell [Fri, 15 Oct 2010 12:23:30 +0000 (13:23 +0100)]
llvmpipe: validate color outputs against key->nr_cbufs
Keith Whitwell [Fri, 15 Oct 2010 12:23:05 +0000 (13:23 +0100)]
llvmpipe: check shader outputs are non-null before using
Vinson Lee [Fri, 15 Oct 2010 13:12:26 +0000 (06:12 -0700)]
mesa: Add missing header to shaderobj.h.
Include compiler.h for ASSERT symbol.
Keith Whitwell [Fri, 15 Oct 2010 13:11:22 +0000 (14:11 +0100)]
llvmpipe: fix non-sse build after recent changes
Keith Whitwell [Fri, 15 Oct 2010 12:52:00 +0000 (13:52 +0100)]
llvmpipe: use aligned loads/stores for plane values
Keith Whitwell [Tue, 12 Oct 2010 17:59:15 +0000 (18:59 +0100)]
llvmpipe: do plane calculations with intrinsics
This is a step towards moving this code into the rasterizer.
Keith Whitwell [Tue, 12 Oct 2010 17:58:05 +0000 (18:58 +0100)]
gallium: move some intrinsics helpers to u_sse.h
Keith Whitwell [Fri, 15 Oct 2010 12:04:19 +0000 (13:04 +0100)]
llvmpipe: don't store plane.ei value in binned data
Further reduce the size of a binned triangle.
Keith Whitwell [Fri, 15 Oct 2010 11:23:22 +0000 (12:23 +0100)]
llvmpipe: slightly shrink the size of a binned triangle
Keith Whitwell [Thu, 14 Oct 2010 23:12:19 +0000 (00:12 +0100)]
llvmpipe: don't pass frontfacing as a float
Keith Whitwell [Thu, 14 Oct 2010 22:28:10 +0000 (23:28 +0100)]
llvmpipe: reintroduce SET_STATE binner command
But bin lazily only into bins which are receiving geometry.
Chad Versace [Wed, 13 Oct 2010 15:29:58 +0000 (08:29 -0700)]
glsl: Fix ir validation for bit logic ops
In ir_validate::visit_leave(), the cases for
- ir_binop_bit_and
- ir_binop_bit_xor
- ir_binop_bit_or
were incorrect. It was incorrectly asserted that both operands must be the
same type, when in fact one may be scalar and the other a vector. It was also
incorrectly asserted that the resultant type was the type of the left operand,
which in fact does not hold when the left operand is a scalar and the right
operand is a vector.
Chad Versace [Sun, 10 Oct 2010 04:27:41 +0000 (21:27 -0700)]
glsl: Implement constant expr evaluation for bitwise logic ops
Implement by adding the following cases to
ir_exporession::constant_expression_value():
- ir_binop_bit_and
- ir_binop_bit_or
- ir_binop_bit_xor
Chad Versace [Sun, 10 Oct 2010 03:56:14 +0000 (20:56 -0700)]
glsl: Implement constant expr evaluation for bit-shift ops
Implement by adding the following cases to
ir_expression::constant_expression_value():
- ir_binop_lshfit
- ir_binop_rshfit
Chad Versace [Sat, 9 Oct 2010 00:28:34 +0000 (17:28 -0700)]
glsl: Implement constant expr evaluation for bitwise-not
Implement by adding a case to ir_expression::constant_expression_value()
for ir_unop_bit_not.
Chad Versace [Fri, 8 Oct 2010 23:22:28 +0000 (16:22 -0700)]
glsl: Implement ast-to-hir for binary shifts in GLSL 1.30
Implement by adding the following cases to ast_expression::hir():
- ast_lshift
- ast_rshift
Also, implement ir validation for the new operators by adding the following
cases to ir_validate::visit_leave():
- ir_binop_lshift
- ir_binop_rshift
Chad Versace [Fri, 8 Oct 2010 23:12:56 +0000 (16:12 -0700)]
glsl: Change generated file glsl_lexer.cpp
Chad Versace [Fri, 8 Oct 2010 23:12:24 +0000 (16:12 -0700)]
glsl: Add lexer rules for << and >> in GLSL 1.30
Commit for generated file glsl_lexer.cpp follows this commit.
Dave Airlie [Thu, 14 Oct 2010 04:40:51 +0000 (14:40 +1000)]
r600g: evergreen interpolation support.
On evergreen, interpolation has moved into the fragment shader,
with the interpolation parmaters being passed via GPRs and LDS entries.
This works out the number of interps required and reserves GPR/LDS
storage for them, it also correctly routes face/position values which
aren't interpolated from the vertex shader.
Also if we noticed nothing is to be interpolated we always setup perspective
interpolation for one value otherwise the GPU appears to lockup.
This fixes about 15 piglit tests on evergreen.
Dave Airlie [Thu, 14 Oct 2010 04:40:24 +0000 (14:40 +1000)]
tgsi: add scanner support for centroid inputs
Ian Romanick [Thu, 14 Oct 2010 20:28:42 +0000 (13:28 -0700)]
glsl: Slightly change the semantic of _LinkedShaders
Previously _LinkedShaders was a compact array of the linked shaders
for each shader stage. Now it is arranged such that each slot,
indexed by the MESA_SHADER_* defines, refers to a specific shader
stage. As a result, some slots will be NULL. This makes things a
little more complex in the linker, but it simplifies things in other
places.
As a side effect _NumLinkedShaders is removed.
NOTE: This may be a candidate for the 7.9 branch. If there are other
patches that get backported to 7.9 that use _LinkedShader, this patch
should be cherry picked also.
Eric Anholt [Fri, 15 Oct 2010 00:01:06 +0000 (17:01 -0700)]
i965: Fix texturing on pre-gen5.
I broke it in
06fd639c519214b6ebcbf29127b6d9ed429f8641 by only testing
2 generations of hardware :(
Brian Paul [Thu, 14 Oct 2010 22:31:54 +0000 (16:31 -0600)]
llvmpipe: code to dump bytecode to file (disabled)
Brian Paul [Thu, 14 Oct 2010 22:21:32 +0000 (16:21 -0600)]
gallivm: add compile-time option to emit inst addrs and/or line numbers
Disabling address printing is helpful for diffing.
Brian Paul [Thu, 14 Oct 2010 20:03:46 +0000 (14:03 -0600)]
mesa: remove post-convolution width/height vars
These were left-over bits from when convolution was removed.
Kenneth Graunke [Thu, 14 Oct 2010 20:41:29 +0000 (13:41 -0700)]
glsl: Refresh autogenerated file builtin_function.cpp.
Kenneth Graunke [Thu, 14 Oct 2010 21:33:30 +0000 (14:33 -0700)]
glsl: Add support for the 1.30 round() built-in.
This implements round() via the ir_unop_round_even opcode, rather than
adding a new opcode. We may wish to add one in the future, since it
might enable a small performance increase on some hardware, but for now,
this should suffice.
Kenneth Graunke [Thu, 14 Oct 2010 21:31:54 +0000 (14:31 -0700)]
i965: Add support for ir_unop_round_even via the RNDE instruction.
Kenneth Graunke [Thu, 14 Oct 2010 20:40:26 +0000 (13:40 -0700)]
glsl: Add front-end support for GLSL 1.30's roundEven built-in.
Implemented using the op-code introduced in the previous commit.
Kenneth Graunke [Thu, 14 Oct 2010 20:37:03 +0000 (13:37 -0700)]
glsl: Add a new ir_unop_round_even opcode for GLSL 1.30's roundEven.
Also, update ir_to_mesa's "1.30 is unsupported" case to "handle" it.
Dave Airlie [Thu, 14 Oct 2010 22:46:16 +0000 (08:46 +1000)]
r300g: clean up warning due to unknown cap.
Keith Whitwell [Thu, 14 Oct 2010 16:19:00 +0000 (17:19 +0100)]
r600g: handle absolute modifier in shader translator
This was being classed as unsupported in one place but used in others.
Enabling it seems to work fine.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Keith Whitwell [Thu, 14 Oct 2010 15:42:39 +0000 (16:42 +0100)]
r600g: emit hardware linewidth
Tested with demos/pixeltest - line rasterization doesn't seem to be
set up for GL conventions yet, but at least width is respected now.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Keith Whitwell [Thu, 14 Oct 2010 15:42:38 +0000 (16:42 +0100)]
r600/drm: fix segfaults in winsys create failure path
Would try to destroy radeon->cman, radeon->kman both which were still
NULL.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Kenneth Graunke [Thu, 14 Oct 2010 20:19:23 +0000 (13:19 -0700)]
i965: Clean up a warning in the old fragment backend.
Hopefully this code can just go away soon.
Eric Anholt [Thu, 14 Oct 2010 19:23:29 +0000 (12:23 -0700)]
i965: Enable the new FS backend on pre-gen6 as well.
It is now to the point where we have no regressing piglit tests. It
also fixes Yo Frankie! and Humus DynamicBranching, probably due to the
piglit bias tests that work that didn't on the Mesa IR backend.
As a downside, performance takes about a 5-10% performance hit at the
moment (e.g. nexuiz 19.8fps -> 18.8fps), which I plan to resolve by
reintroducing 16-wide fragment shaders where possible. It is a win,
though, for fragment shaders using flow control.
Kenneth Graunke [Thu, 14 Oct 2010 18:40:19 +0000 (11:40 -0700)]
i965: Correctly emit the RNDZ instruction.
Simply using RNDU, RNDZ, or RNDE does not produce the desired result.
Rather, the RND* instructions place a value in the destination register
that may be 1 less than the correct answer. They can also set per-channel
"increment bits" in a flag register, which, if set, mean dest needs to
be incremented by 1. A second instruction - a predicated add -
completes the job.
Notably, RNDD always produces the correct answer in a single
instruction.
Fixes piglit test glsl-fs-trunc.
Kenneth Graunke [Thu, 14 Oct 2010 18:39:09 +0000 (11:39 -0700)]
i965: Use RNDZ for ir_unop_trunc in the new FS.
The existing code used RNDD, which rounds down, rather than toward zero.
Kenneth Graunke [Thu, 14 Oct 2010 17:51:09 +0000 (10:51 -0700)]
glsl: Refresh autogenerated file builtin_function.cpp.
Kenneth Graunke [Thu, 14 Oct 2010 17:49:13 +0000 (10:49 -0700)]
glsl: Add front-end support for the "trunc" built-in.
Kenneth Graunke [Wed, 13 Oct 2010 23:35:34 +0000 (16:35 -0700)]
i965: Use logical-not when emitting ir_unop_ceil.
Fixes piglit test glsl-fs-ceil.
Eric Anholt [Thu, 14 Oct 2010 18:27:17 +0000 (11:27 -0700)]
i965: Add peepholing of conditional mod generation from expressions.
This cuts usually 2 out of 3 instructions for flag reg generation (if
statements, conditional assignment) by producing the conditional mod
in the expression representing the boolean value.
Fixes glsl-fs-vec4-indexing-temp-dst-in-nested-loop-combined (register
allocation no longer fails for the conditional generation
proliferation)
Eric Anholt [Thu, 14 Oct 2010 18:11:29 +0000 (11:11 -0700)]
i965: Add a function for handling the move of boolean values to flag regs.
This will be a place to peephole comparisions directly to the flag
regs, and for now avoids using MOV with conditional mod on gen6, which
is now illegal.
Kristian Høgsberg [Thu, 14 Oct 2010 18:57:47 +0000 (14:57 -0400)]
Only install vtxfmt tables for OpenGL
GLES1 and GLES2 install their own exec pointers and don't need the
Save table. Also, the SET_* macros use different indices for the different
APIs so the offsets used in vtxfmt.c are actually wrong for the ES APIs.
Eric Anholt [Thu, 14 Oct 2010 03:17:15 +0000 (20:17 -0700)]
i965: Add a pass to the FS to split virtual GRFs to float channels.
Improves nexuiz performance 0.91% (+/- 0.54%, n=8)
Eric Anholt [Thu, 14 Oct 2010 05:39:06 +0000 (22:39 -0700)]
i965: Update the live interval when coalescing regs.
Eric Anholt [Thu, 14 Oct 2010 02:21:32 +0000 (19:21 -0700)]
i965: Set class_sizes[] for the aligned reg pair class.
So far, I've only seen this be a valgrind warning and not a real failure.
Keith Whitwell [Thu, 14 Oct 2010 12:15:28 +0000 (13:15 +0100)]
llvmpipe: don't try to emit non-existent color outputs
Kristian Høgsberg [Wed, 13 Oct 2010 15:11:47 +0000 (11:11 -0400)]
Drop the "neutral" tnl module
Just always check for FLUSH_UPDATE_CURRENT and call Driver.BeginVertices
when necessary. By using the unlikely() macros, this ends up as
a 10% performance improvement (for isosurf, anyway) over the old,
complicated function pointer swapping.
Chia-I Wu [Thu, 14 Oct 2010 09:13:36 +0000 (17:13 +0800)]
st/egl: Do not finish a fence that is NULL.
i915g would dereference the NULL pointer.
Chia-I Wu [Thu, 14 Oct 2010 09:06:21 +0000 (17:06 +0800)]
st/egl: Access _EGLConfig directly.
Drop the use of SET_CONFIG_ATTRIB. Fix the value of EGL_SAMPLE_BUFFERS
along the way.
Chia-I Wu [Thu, 14 Oct 2010 08:38:58 +0000 (16:38 +0800)]
egl: Access config attributes directly.
Replace SET_CONFIG_ATTRIB/GET_CONFIG_ATTRIB by direct dereferences.
Chia-I Wu [Thu, 14 Oct 2010 08:25:30 +0000 (16:25 +0800)]
egl: Use attribute names as the _EGLConfig member names.
This makes _EGLConfig more accessible and scales better when new
attributes are added.
Dave Airlie [Thu, 14 Oct 2010 04:27:34 +0000 (14:27 +1000)]
r600g: select linear interpolate if tgsi input requests it
Dave Airlie [Thu, 14 Oct 2010 04:19:37 +0000 (14:19 +1000)]
r600g: fixup typo in macro name
Dave Airlie [Thu, 14 Oct 2010 04:14:22 +0000 (14:14 +1000)]
r600g: fixup pos/face ena/address properly
Dave Airlie [Thu, 14 Oct 2010 03:29:11 +0000 (13:29 +1000)]
r600g: only pick centroid coordinate when asked.
TGSI tells us when to use this, its not hooked up from GLSL to MESA to TGSI yet though.
Zhenyu Wang [Thu, 14 Oct 2010 03:24:49 +0000 (11:24 +0800)]
Revert "i965: fallback lineloop on sandybridge for now"
This reverts commit
73dab75b4165f7d2214a68d4ba8e3cb7aab9b4ac.
Zhenyu Wang [Thu, 14 Oct 2010 02:54:53 +0000 (10:54 +0800)]
i965: Fix GS hang on Sandybridge
Don't use r0 for FF_SYNC dest reg on Sandybridge, which would
smash FFID field in GS payload, that cause later URB write fail.
Also not use r0 in any URB write requiring allocate.
Eric Anholt [Wed, 13 Oct 2010 23:13:45 +0000 (16:13 -0700)]
i965: Add support for rescaling GL_TEXTURE_RECTANGLE coords to new FS.
Fredrik Höglund [Wed, 13 Oct 2010 15:49:15 +0000 (17:49 +0200)]
r600g: Fix texture sampling with swizzled coords
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Wed, 13 Oct 2010 23:10:16 +0000 (09:10 +1000)]
r600g: drop unused context members
Ian Romanick [Mon, 11 Oct 2010 23:19:19 +0000 (16:19 -0700)]
mesa: Clean up various 'unused parameter' warnings in shaderapi
Ian Romanick [Mon, 11 Oct 2010 23:13:11 +0000 (16:13 -0700)]
mesa: Clean up two 'comparison between signed and unsigned' warnings
Ian Romanick [Mon, 11 Oct 2010 22:21:17 +0000 (15:21 -0700)]
mesa: Refactor validation of shader targets
Actually validate that the implementation supports the particular
shader target as well. Previously if a driver only supported vertex
shaders, for example, glCreateShaderObjectARB would gladly create a
fragment shader.
NOTE: this is a candidate for the 7.9 branch.
Ian Romanick [Tue, 12 Oct 2010 22:29:43 +0000 (15:29 -0700)]
mesa: Silence unused variable warning
Ian Romanick [Wed, 13 Oct 2010 22:13:02 +0000 (15:13 -0700)]
linker: Reject shaders that have unresolved function calls
This really amounts to just using the return value from
link_function_calls. All the work was being done, but the result was
being ignored.
Fixes piglit test link-unresolved-funciton.
NOTE: this is a candidate for the 7.9 branch.
Vinson Lee [Wed, 13 Oct 2010 21:21:08 +0000 (14:21 -0700)]
glsl: Initialize variable in ir_derefence_array::constant_expression_value
Completely initialize data passed to ir_constant constructor.
Fixes piglit glsl-mat-from-int-ctor-03 valgrind uninitialized value
error on softpipe.
José Fonseca [Wed, 13 Oct 2010 19:25:17 +0000 (20:25 +0100)]
llvmpipe: Generalize the x8z24 fast path to all depth formats.
Together with the previous commit, this generalize the benefits of
d2cf757f44f4ee5554243f3279483a25886d9927 to all depth formats, in
particular:
- simpler float -> 24unorm conversion
- avoid unsigned comparisons (not directly supported on SSE) by aligning
to the least significant bit
- avoid unecessary/repeated mask ANDing
Verified with trivial/tri-z that the exact same assembly is produced for
X8Z24.
José Fonseca [Wed, 13 Oct 2010 14:45:24 +0000 (15:45 +0100)]
gallivm: More accurate float -> 24bit & 32bit unorm conversion.
Brian Paul [Wed, 13 Oct 2010 18:35:38 +0000 (12:35 -0600)]
gallivm: work-around trilinear mipmap filtering regression with LLVM 2.8
The bug only happens on the AOS / fixed-pt path.
Vinson Lee [Wed, 13 Oct 2010 18:18:40 +0000 (11:18 -0700)]
gallivm: Remove unnecessary header.
Brian Paul [Wed, 13 Oct 2010 14:32:08 +0000 (08:32 -0600)]
x11: fix breakage from gl_config::visualType removal
José Fonseca [Wed, 13 Oct 2010 13:28:51 +0000 (14:28 +0100)]
llvmpipe: Unbreak Z32_FLOAT.
Z32_FLOAT uses <4 x float> as intermediate/destination type,
instead of <4 x i32>.
The necessary bitcasts got removed with commit
5b7eb868fde98388d80601d8dea39e679828f42f
Also use depth/stencil type and build contexts consistently, and
make the depth pointer argument a ordinary <i8 *>, to catch this
sort of issues in the future (and also to pave way for Z16 and
Z32_FLOAT_S8_X24 support).
Kristian Høgsberg [Tue, 12 Oct 2010 16:26:10 +0000 (12:26 -0400)]
Drop GLcontext typedef and use struct gl_context instead
Kristian Høgsberg [Tue, 12 Oct 2010 16:02:01 +0000 (12:02 -0400)]
Drop GLframebuffer typedef and just use struct gl_framebuffer
Kristian Høgsberg [Tue, 12 Oct 2010 15:58:47 +0000 (11:58 -0400)]
Rename GLvisual and __GLcontextModes to struct gl_config
Kristian Høgsberg [Tue, 12 Oct 2010 15:40:05 +0000 (11:40 -0400)]
gl: Remove unused GLcontextModes fields
Kristian Høgsberg [Tue, 12 Oct 2010 14:34:03 +0000 (10:34 -0400)]
Get rid of GL/internal/glcore.h
__GLcontextModes is always only used as an implementation internal struct
at this point and we shouldn't install glcore.h anymore. Anything that
needs __GLcontextModes should just include the struct in its headers files
directly.
Roland Scheidegger [Wed, 13 Oct 2010 13:26:37 +0000 (15:26 +0200)]
gallivm: only use lp_build_conv 4x4f -> 1x16 ub fastpath with sse2
This is relying on lp_build_pack2 using the sse2 pack intrinsics which
handle clamping.
(Alternatively could have make it use lp_build_packs2 but it might
not even produce more efficient code than not using the fastpath
in the first place.)
Dave Airlie [Wed, 13 Oct 2010 08:50:37 +0000 (18:50 +1000)]
r600g: fix stencil export for evergreen harder
Stephan Schmid [Mon, 11 Oct 2010 13:49:56 +0000 (15:49 +0200)]
r600g: fix relative addressing when splitting constant accesses
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Wed, 13 Oct 2010 07:45:10 +0000 (17:45 +1000)]
r600g: add missing eg reg definition
Dave Airlie [Wed, 13 Oct 2010 07:40:32 +0000 (17:40 +1000)]
r600g: evergreen add stencil export bit
Dave Airlie [Wed, 13 Oct 2010 01:17:44 +0000 (11:17 +1000)]
r600g: use blitter for hw copy region
at the moment depth copies are failing (piglit depth-level-clamp)
so use the fallback for now until get some time to investigate.
Dave Airlie [Wed, 13 Oct 2010 01:08:44 +0000 (11:08 +1000)]
r600g: drop all use of unsigned long
this changes size on 32/64 bit so is definitely no what you want to use here.
Dave Airlie [Wed, 13 Oct 2010 00:14:55 +0000 (10:14 +1000)]
r600g: fix transfer stride.
fixes segfaults
Dave Airlie [Wed, 13 Oct 2010 01:03:18 +0000 (11:03 +1000)]
r600g: remove bpt and start using pitch_in_bytes/pixels.
this mirror changes in r300g, bpt is kinda useless when it comes to some
of the non-simple texture formats.
Dave Airlie [Wed, 13 Oct 2010 01:02:52 +0000 (11:02 +1000)]
r600g: rename pitch in texture to pitch_in_bytes
Dave Airlie [Tue, 12 Oct 2010 00:51:03 +0000 (10:51 +1000)]
r600g: use common texture object create function
Dave Airlie [Mon, 11 Oct 2010 23:53:17 +0000 (09:53 +1000)]
r600g: split out miptree setup like r300g
just a cleanup step towards tiling