Eric Anholt [Thu, 19 Aug 2010 00:39:57 +0000 (17:39 -0700)]
glsl: Also strdup the names of uniform list entries for >vec4 types.
Fixes double-free since the fix to free all of the uniform list.
Kenneth Graunke [Thu, 19 Aug 2010 00:38:24 +0000 (17:38 -0700)]
glcpp: Refresh autogenerated lexer.
Kenneth Graunke [Thu, 19 Aug 2010 00:38:05 +0000 (17:38 -0700)]
glcpp: Add basic #line support (adapted from the main compiler).
Eric Anholt [Thu, 19 Aug 2010 00:12:18 +0000 (17:12 -0700)]
ir_to_mesa: Fix leak by improper freeing of a uniform list.
Eric Anholt [Wed, 18 Aug 2010 23:56:39 +0000 (16:56 -0700)]
glsl2: Free the shader compiler at dri screen destruction.
Hooray, we can valgrind again without adding suppressions. This also
adds an interface for use by an implementation of
glReleaseShaderCompiler().
Eric Anholt [Wed, 18 Aug 2010 23:29:02 +0000 (16:29 -0700)]
ir_to_mesa: Don't leak the whole linked assembly program.
Eric Anholt [Wed, 18 Aug 2010 23:17:10 +0000 (16:17 -0700)]
dri2: Clean up the common dri2 options at screen destroy.
Eric Anholt [Wed, 18 Aug 2010 22:59:31 +0000 (15:59 -0700)]
glsl: Fix leak-causing typo in destructor that made it another constructor.
Vinson Lee [Wed, 18 Aug 2010 23:10:15 +0000 (16:10 -0700)]
glsl: Fix uninitialized member in ir_hierarchical_vistor constructor.
Class member base_ir was not initialized by the default constructor.
Vinson Lee [Wed, 18 Aug 2010 21:36:08 +0000 (14:36 -0700)]
mesa: fpclassify is available on OpenSolaris.
There is no explicit predefined macro to distinguish between OpenSolaris
and Solaris. This patch assumes that the difference is in the compilers.
OpenSolaris uses GCC and not the Sun Studio compiler. Assume that the
availability of fpclassify is due to GCC.
This patch was not tested on Solaris. It would break the build on
Solaris with GCC if GCC on Solaris does not have fpclassify.
Eric Anholt [Wed, 18 Aug 2010 21:13:30 +0000 (14:13 -0700)]
ir_to_mesa: Fix leak of set_branchtargets temp data.
Eric Anholt [Wed, 18 Aug 2010 19:04:43 +0000 (12:04 -0700)]
glsl: Garbage collect old prototype for ir_to_mesa.
Eric Anholt [Wed, 18 Aug 2010 20:48:32 +0000 (13:48 -0700)]
ir_to_mesa: Stop leaking the ir_instruction * annotation of our compile.
Eric Anholt [Wed, 18 Aug 2010 21:12:52 +0000 (14:12 -0700)]
mesa: Free linked shaders when deleting the shader program.
Eric Anholt [Wed, 18 Aug 2010 19:02:35 +0000 (12:02 -0700)]
mesa: Free old linked shaders when relinking new shaders.
Eric Anholt [Wed, 18 Aug 2010 18:42:42 +0000 (11:42 -0700)]
glsl2: Fix copy'n'paste hilarity leading to leaking in the refcount visitor.
Eric Anholt [Wed, 18 Aug 2010 18:37:18 +0000 (11:37 -0700)]
ir_to_mesa: Free the ir_to_mesa temporary storage when we're done.
Eric Anholt [Wed, 18 Aug 2010 18:36:09 +0000 (11:36 -0700)]
mesa: Hang the compiled shader off the shader->ir, not the shader.
Otherwise, with repeated program recompile, we never free the results
of the previous compile.
Eric Anholt [Wed, 18 Aug 2010 18:35:15 +0000 (11:35 -0700)]
mesa: Make MESA_GLSL=dump include when compile/link fails, and the info log.
Kenneth Graunke [Wed, 18 Aug 2010 20:54:50 +0000 (13:54 -0700)]
glsl: Fix scoping bug in if statements.
Fixes glslparsertest/glsl2/scoping-01.frag (successfully compiled but
should've failed) and scoping-02.frag (assertion triggered).
Kenneth Graunke [Wed, 18 Aug 2010 20:20:58 +0000 (13:20 -0700)]
glsl: Refresh autogenerated file builtin_function.cpp.
Kenneth Graunke [Wed, 18 Aug 2010 20:17:27 +0000 (13:17 -0700)]
generate_builtins.py: Clean up generated output a bit.
This should make it easier to diff the output, clean up some of the
insane whitespace, and make the strings a bit smaller.
We'll probably need to split up the prototype strings eventually, but
for now, this gets it under the 65K mark.
Kenneth Graunke [Wed, 18 Aug 2010 20:16:50 +0000 (13:16 -0700)]
glsl/builtins: Add forgotten hyperbolic trig builtins in 1.30 profiles.
Kenneth Graunke [Wed, 18 Aug 2010 19:06:25 +0000 (12:06 -0700)]
ir_constant_expression: Implement equal/notEqual for booleans.
Calls to equal(bvec, bvec) or notEqual(bvec, bvec) previously caused an
assertion. Fixes piglit tests glsl-const-builtin-equal-bool and
glsl-const-builtin-notEqual-bool.
Andre Maasikas [Wed, 18 Aug 2010 11:14:38 +0000 (14:14 +0300)]
r600: implement SSG instruction
michal [Wed, 18 Aug 2010 11:16:42 +0000 (13:16 +0200)]
gallivm: Use proper index to lookup predicate register array.
Doesn't fix anything, as those indices were both always 0.
Andre Maasikas [Wed, 18 Aug 2010 08:57:28 +0000 (11:57 +0300)]
r600: implement DP2 opcode
Kenneth Graunke [Wed, 18 Aug 2010 06:47:51 +0000 (23:47 -0700)]
glsl2: Regenerate builtin_function.cpp.
Kenneth Graunke [Wed, 18 Aug 2010 06:47:11 +0000 (23:47 -0700)]
glsl2: Fix cut and paste error in EXT_texture_array builtins.
Fixes fd.o bug #29629.
Carl Worth [Wed, 18 Aug 2010 06:22:42 +0000 (23:22 -0700)]
glcpp: Refresh generated files.
After a recent change to glcpp-parse.y (adding "redefined macro" error).
Carl Worth [Wed, 18 Aug 2010 06:20:58 +0000 (23:20 -0700)]
glcpp: Add support for "redefined macro" error.
Carefully avoiding printing any error when the new definition matches
the existing definition.
This fixes the recently-added 088-redefine-macro-legitimate.c and
089-redefine-macro-error.c tests as well as glsparsertest/preprocess1
in piglit.
Carl Worth [Wed, 18 Aug 2010 06:19:01 +0000 (23:19 -0700)]
glcpp: Add two new tests for testing redefined macros.
The specification says that redefining a macro is an error, unless the
new definitions is identical to the old one, (identical replacement
lists but ignoring differing amounts of whitespace).
Carl Worth [Wed, 18 Aug 2010 06:13:56 +0000 (23:13 -0700)]
glcpp: Allow standalone glcpp to accept a filename as first argument.
This is useful for debugging the preprocessor.
Dave Airlie [Wed, 18 Aug 2010 06:10:16 +0000 (16:10 +1000)]
r600g: add FRC, FLR, DDX and DDY
the first two are straight op2's and the DDX/DDY are taken from r600c.
Dave Airlie [Wed, 18 Aug 2010 05:53:29 +0000 (15:53 +1000)]
r600g: add SGE and SLE opcodes
fixes fp-set-01 and glsl-fs-step
Dave Airlie [Wed, 18 Aug 2010 05:26:28 +0000 (15:26 +1000)]
r600g: add TXB support
fixes biased texturing tests
Dave Airlie [Wed, 18 Aug 2010 05:05:34 +0000 (15:05 +1000)]
r600g: fix TXP vs TEX in shader.
Don't do perspective for TEX, and also copy input to a temporary for TEX
also add tex opcode names
Carl Worth [Wed, 18 Aug 2010 05:23:43 +0000 (22:23 -0700)]
glcpp: Fix 064-version.c expected result to track recent change.
In commit
6be3a8b70af4ba4fa4d037d54ecf6d5f055edbc9, the #version directive
was fixed to stop generating a spurious newline. Here we simply update
the expected result for the single test which includes a #version directive.
Carl Worth [Wed, 18 Aug 2010 05:22:13 +0000 (22:22 -0700)]
glcpp: Regenerated glcpp-lex.c from previous commit.
The previous commit changed glcpp-lex.l so we commit the resulting
generated file here.
Carl Worth [Wed, 18 Aug 2010 05:17:09 +0000 (22:17 -0700)]
glcpp: Don't include the newline when discarding single-line comments
Matching the newline here meant having to do some redundant work here,
(incrementing line number, resetting column number, and returning a
NEWLINE token), that could otherwise simply be left to the existing rule
which matches a newline.
Worse, when the comment rule matches the newline as well, the parser
can lookahead and see a token for something that should actually be skipped.
For example, in a case like this:
#if 0 // comment here
fail
#else
win
#endif
Both fail and win appear in the output, (not that the condition is being
evaluated incorrectly---merely that one token after the comment's newline
was being lexed/parse regardless of the condition).
This commit fixes the above test case, (which is also remarkably similar
to 087-if-comments which now passes).
Dave Airlie [Wed, 18 Aug 2010 03:48:04 +0000 (13:48 +1000)]
r600g: add two simple tgsi opcodes.
makes glsl-fs-log2 and glsl1-integer division with uniform var pass
Dave Airlie [Wed, 18 Aug 2010 03:29:41 +0000 (13:29 +1000)]
r600g: fix point size
fixes piglit pointAtten and point-sprite tests
Eric Anholt [Wed, 18 Aug 2010 02:41:27 +0000 (19:41 -0700)]
ir_to_mesa: Allow ir_return in main().
I didn't expect that this would really work, but it turns out there
are shaders in the wild that do it.
Fixes: (with swrast)
glsl-fs-main-return
glsl-vs-main-return
Eric Anholt [Wed, 18 Aug 2010 02:51:00 +0000 (19:51 -0700)]
i965: Throw a link error when we see a "return" in main().
We'll need to use the HALT instruction to do this right, like returns
from other functions.
Dave Airlie [Wed, 18 Aug 2010 01:49:51 +0000 (11:49 +1000)]
r600g: fixup pitch alignment like r600c.
This still needs work, passes tex3d, fbo-scissor-bitmap, scissor-bitmap
Luca Barbieri [Wed, 18 Aug 2010 02:14:43 +0000 (04:14 +0200)]
translate_sse: don't overwrite source buffer pointer
We were putting the source pointer in a register used as a temporary,
breaking all paths that don't read the data in a single instruction.
Eric Anholt [Wed, 18 Aug 2010 00:50:05 +0000 (17:50 -0700)]
glcpp: Add testcase for #if handling bug that breaks Savage2.
Dave Airlie [Wed, 18 Aug 2010 00:50:19 +0000 (10:50 +1000)]
r600g: fix height calcs for miptree
h needs to be rounded up, this probably needs revisiting when we get
to tiling etc.
fixes fbo-generatemipmap-npot
Dave Airlie [Tue, 17 Aug 2010 23:56:21 +0000 (09:56 +1000)]
r600g: emit texture level offset in CB/DB setup.
8 more piglit tests pass,
fbo-clearmipmap, fbo-copyteximage, fbo-generatemipmap,
fbo-generatemipmap-nonsquare, fbo-generatemipmap-scissor,
fbo-generatemipmap-viewport, gen-teximage, gen-texsubimage
Eric Anholt [Wed, 18 Aug 2010 00:24:42 +0000 (17:24 -0700)]
ir_to_mesa: Fix implementation of ir_binop_equal, ir_binop_notequal.
These binops are the vector-to-bool comparisons, not vec-to-bvec. We
likely want both operations avilable as expression, since 915 and 965
FS naturally does the vector version, while 965 VS can also naturally
do the scalar version. However, we can save that until later.
Fixes:
glsl-fs-vec4-operator-equal.shader_test
glsl-fs-vec4-operator-notequal.shader_test
glsl-vs-vec4-operator-equal.shader_test
glsl-vs-vec4-operator-notequal.shader_test
Eric Anholt [Tue, 17 Aug 2010 23:12:15 +0000 (16:12 -0700)]
i915: Add support for SSG opcode.
Fixes glsl-fs-sign and many of the tests of trig builtins.
Eric Anholt [Tue, 17 Aug 2010 22:51:34 +0000 (15:51 -0700)]
i915: Add support for reading output regs in the FS.
Fixes glsl-unused-varying and many others, since we produce an output
reg read any time gl_FragColor is written inside an if statement.
Eric Anholt [Tue, 17 Aug 2010 22:42:40 +0000 (15:42 -0700)]
i915: Add support for OPCODE_DP2.
Fixes glsl-fs-dot-vec2.
Eric Anholt [Tue, 17 Aug 2010 22:07:22 +0000 (15:07 -0700)]
i915: Enable ARB_fragment_shader by default.
Now that we have glsl2 with if flattening in place, most shaders will
just work. Remaining failing shaders will mostly be due to loop
unrolling (in progress), some possible if flattening failures in
inlining functions (planning on fixing), and the register/instruction
count limits.
While the GLSL and GLSL-ES specs say that shaders shouldn't fail to
compile/link due to register/instruction limits, in practice we're not
the first vendor to expose GLSL on hardware with these limitations.
The benefit to application developers of providing a better language
for GPU programming is greater than the pain of having to handle
instruction limits (which they had to for ARB_fp on this hardware
anyway)
Ian Romanick [Tue, 17 Aug 2010 22:57:48 +0000 (15:57 -0700)]
glsl2: Fix transpose of rows and cols
This error led to an assertion failure for some constructors of
non-square matrices. It only occured in matrices where the number of
columns was greater than the number of rows. It didn't even always
occur on those.
Fixes piglit glslparsertest case constructor-16.vert.
Henri Verbeet [Tue, 17 Aug 2010 22:25:32 +0000 (18:25 -0400)]
r600c: Handle reads from PROGRAM_OUTPUT
with glsl2, reads from outputs are legal
Ian Romanick [Tue, 17 Aug 2010 21:55:50 +0000 (14:55 -0700)]
linker: Demote user-defined varyings in the VS-only case
Fixes piglit test case glsl-vs-ff-frag and bugzilla #29623.
Benjamin Segovia [Tue, 17 Aug 2010 03:40:09 +0000 (20:40 -0700)]
prog_optimize: Only merge writes to temporary registers
In one optimization pass, register files may have been messed therefore
merging instructions which use the same index in two different register
files.
Jerome Glisse [Tue, 17 Aug 2010 21:23:20 +0000 (17:23 -0400)]
r600g: fix fake pixel output
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Eric Anholt [Tue, 17 Aug 2010 20:59:08 +0000 (13:59 -0700)]
i965: Add support for DP2 in the VS.
Fixes glsl-vs-dot-vec2.
Eric Anholt [Tue, 17 Aug 2010 20:24:50 +0000 (13:24 -0700)]
glsl: When doing algebraic simplification, make sure the type still matches.
When simplifying (vec4(1.0) / (float(x))) to rcp(float(x)), we forgot
to produce a vec4, angering ir_validate when starting alien-arena.
Fixes:
glsl-algebraic-add-zero-2
glsl-algebraic-div-one-2
glsl-algebraic-mul-one-2
glsl-algebraic-sub-zero-3
glsl-algebraic-rcp-sqrt-2
Eric Anholt [Tue, 17 Aug 2010 20:27:44 +0000 (13:27 -0700)]
glsl: Make ir_algebraic new expressions allocate out of the parent.
This could reduce the amount of memory used by a shader tree after
optimization, and increases consistency with other passes.
Ian Romanick [Tue, 17 Aug 2010 19:57:28 +0000 (12:57 -0700)]
ir_constant: Don't assert on out-of-bounds array accesses
Several optimization paths, including constant folding, can lead to
accessing an ir_constant array with an out of bounds index. The GLSL
spec lets us produce "undefined" results, but it does not let us
crash.
Fixes piglit test case glsl-array-bounds-01 and glsl-array-bounds-03.
Eric Anholt [Tue, 17 Aug 2010 19:37:49 +0000 (12:37 -0700)]
mesa: Dump shader source before validating the shader.
This will make extracting source to produce minimal testcases for
shader compile issues easier.
Alex Deucher [Tue, 17 Aug 2010 14:42:06 +0000 (10:42 -0400)]
r600c: fix dword miscount in blit emit code
Chia-I Wu [Tue, 17 Aug 2010 11:24:18 +0000 (19:24 +0800)]
targets/egl: Link with DRI_LIB_DEPS.
Use DRI_LIB_DEPS when linking GL/GLES state trackers. This fixes
missing talloc symbol errors, and is hopefully more future proof.
nobled [Mon, 16 Aug 2010 16:46:14 +0000 (16:46 +0000)]
gallivm: Fix and re-enable MMX-disabling code
Signed-off-by: José Fonseca <jfonseca@vmware.com>
Dave Airlie [Tue, 17 Aug 2010 10:54:45 +0000 (20:54 +1000)]
mesa: fix es1/2 build hopefully
needed to add cpp rules and includes properly for es1/es2
Dave Airlie [Tue, 17 Aug 2010 09:01:18 +0000 (19:01 +1000)]
r300g: fix context destroy under hyperz
we were destroying the mm before unrefing all the objects, so segfault.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Vinson Lee [Tue, 17 Aug 2010 07:20:58 +0000 (00:20 -0700)]
scons: Add -fno-common for 64-bit builds on Mac OS X.
This option is also needed for 64-bit builds if llvm is enabled.
Other the build fails during linking.
Dave Airlie [Tue, 17 Aug 2010 05:16:53 +0000 (15:16 +1000)]
r600g: kill event type magic number in winsys
these events have names, use them.
Vinson Lee [Tue, 17 Aug 2010 05:41:44 +0000 (22:41 -0700)]
configs: Add -ltalloc to linux-dri.
Fixes build after glsl2 branch merge, which added talloc dependency.
Dave Airlie [Tue, 17 Aug 2010 03:40:15 +0000 (13:40 +1000)]
r600g: add user clip plane support.
Apart from the fact that the radeon.h/r600_states.h editing is a nightmare, this
wasn't so bad.
passes piglit user-clip test now also trivial tests.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Eric Anholt [Tue, 17 Aug 2010 02:18:11 +0000 (19:18 -0700)]
i965: Use the implied move available in most brw_wm_emit brw_math() calls.
This saves an extra message reg move in the program, though I'm not
clear on whether it will have any performance impact other than cache
footprint. It will also fix those math calls on Sandybridge, where
the brw_eu_emit.c brw_math() support relies on the implied move being
used.
Eric Anholt [Fri, 9 Jul 2010 00:05:42 +0000 (17:05 -0700)]
i965: Add disasm for Compr4 instruction compression.
Ian Romanick [Tue, 17 Aug 2010 02:08:53 +0000 (19:08 -0700)]
Merge branch 'glsl2'
Conflicts:
src/mesa/program/prog_optimize.c
Vinson Lee [Tue, 17 Aug 2010 01:52:37 +0000 (18:52 -0700)]
util: Remove check_os_katmai_support.
check_os_katmai_support checks that the operating system running on a
SSE-capable processor supports SSE. This is necessary for unpatched
2.2.x and earlier kernels. 2.4.x and later kernels support SSE.
check_os_katmai_support will disable SSE capabilities for 32-bit x86
operating systems for which there is no code path. Currently, this
function handles Linux, Windows, and several BSDs. Mac OS, Cygwin, and
Solaris are several operating systems with no code paths.
Rather than add code for the unhandled operating systems, remove this
function altogether. This will fix SSE detection on all recent 32-bit
x86 operating systems. This completely breaks functionality on unpatched
2.2.x and earlier kernels, although there are likely no Gallium3D users
on such operating systems.
Dave Airlie [Fri, 13 Aug 2010 03:54:25 +0000 (13:54 +1000)]
r600g: drop libdrm_radeon link
Kenneth Graunke [Mon, 16 Aug 2010 22:18:23 +0000 (15:18 -0700)]
glsl2: Refresh autogenerated file builtin_function.cpp.
Kenneth Graunke [Mon, 16 Aug 2010 22:14:04 +0000 (15:14 -0700)]
glsl2: Add builtins profile for GLSL 1.30.
Many functions are currently wrapped with #if 0 since we haven't
implemented them yet.
Ian Romanick [Mon, 16 Aug 2010 20:59:34 +0000 (13:59 -0700)]
linker: Include compiler.h to avoid spurious warnings about INLINE
Ian Romanick [Mon, 16 Aug 2010 20:59:01 +0000 (13:59 -0700)]
glsl2: Silence unused variable warning
Vinson Lee [Mon, 16 Aug 2010 20:52:57 +0000 (13:52 -0700)]
translate: Move loop variable declaration outside for loop.
Fixes MSVC build.
Kenneth Graunke [Mon, 16 Aug 2010 20:43:10 +0000 (13:43 -0700)]
glcpp: Refresh autogenerated lexer and parser.
Kenneth Graunke [Mon, 16 Aug 2010 20:42:04 +0000 (13:42 -0700)]
glcpp: Remove spurious newline generated by #version handling.
This was causing line numbering to be off by one. The newline comes
from the NEWLINE token at the end of the line; there's no need to
insert one.
Marek Olšák [Mon, 16 Aug 2010 17:17:02 +0000 (19:17 +0200)]
r300g: fix assert in the rasterizer block for r3xx-r4xx
Reported-by: Niels Ole Salscheider <niels_ole@salscheider-online.de>
Ian Romanick [Mon, 16 Aug 2010 16:45:01 +0000 (09:45 -0700)]
ir_to_mesa: Silence unused variable warnings
Ian Romanick [Mon, 16 Aug 2010 16:43:00 +0000 (09:43 -0700)]
ir_to_mesa: Clean up assertions in ir_to_mesa_visitor::visit(ir_texture *)
Ian Romanick [Mon, 16 Aug 2010 16:39:58 +0000 (09:39 -0700)]
ir_to_mesa: Support texture rectangle targets
José Fonseca [Mon, 16 Aug 2010 16:20:54 +0000 (17:20 +0100)]
translate: Remove unused temporary register.
Assuming the side-effect of x86_make_reg is also unnecessary.
José Fonseca [Mon, 16 Aug 2010 16:18:14 +0000 (17:18 +0100)]
translate: Eliminate void pointer arithmetic.
Non-portable.
Chia-I Wu [Sun, 15 Aug 2010 09:24:14 +0000 (17:24 +0800)]
st/egl: Add support for EGL_KHR_fence_sync.
The extension is implemented by pipe_fence_handle.
Chia-I Wu [Sun, 15 Aug 2010 09:09:48 +0000 (17:09 +0800)]
egl: Add support for EGL_KHR_fence_sync.
Individual drivers still need to support and enable the extension.
Chia-I Wu [Sat, 14 Aug 2010 15:56:46 +0000 (23:56 +0800)]
st/egl: Add support for EGL_KHR_reusable_sync.
The extension is implemented by pipe_condvar.
Chia-I Wu [Sat, 14 Aug 2010 15:09:12 +0000 (23:09 +0800)]
egl: Add support for EGL_KHR_reusable_sync.
Individual drivers still need to support and enable the extension.
Chia-I Wu [Sun, 15 Aug 2010 16:18:30 +0000 (00:18 +0800)]
egl: Update eglext.h.
Update to version 7 for EGL_KHR_fence_sync.
Alex Deucher [Mon, 16 Aug 2010 15:34:17 +0000 (11:34 -0400)]
r600c: blit emit updates
- set VGT_MAX_VTX_INDX to a larger value
- emit PA_SC_AA_CONFIG. The command checker in 2.6.36+
requires this reg.
Luca Barbieri [Sun, 15 Aug 2010 05:37:31 +0000 (07:37 +0200)]
draw_llvm: fix segfaults on non-SSE2 CPUs where it is disabled (v2)
Changes in v2:
- Change function name
Currently draw_llvm refuses to create itself on non-SSE2 CPUs due to
an alleged LLVM bug.
However, this is implemented improperly, because other parts of draw
still attempt to access draw->llvm, resulting in segfaults.
Instead, put the check in debug_get_option_draw_use_llvm, check that
before calling draw_llvm_create, and then check whether draw->llvm is
non-null everywhere else.
Luca Barbieri [Tue, 10 Aug 2010 00:14:04 +0000 (02:14 +0200)]
translate_sse: major rewrite (v5)
NOTE: Win64 is untested, and is thus currently disabled.
If you have such a system, please enable it and report whether it works.
To enable it, change src/gallium/auxiliary/translate/translate.c
Changes in v5:
- On Win64, preserve %xmm6 and %xmm7 as required by the ABI
- Use _WIN64 instead of WIN64
Changes in v4:
- Use x86_target() and x86_target_caps()
- Enable translate_sse in x86-64, but not in Win64
Changes in v3:
- Win64 support (untested)
- Use u_cpu_detect.h constants instead of #ifs
Changes in v2:
- Minimize #ifs
- Give a name to magic number CHANNELS_0001
- Add support for CPUs without SSE (only memcpy and swizzles, like non SSE2)
- Fixed comments
translate_sse is currently very limited to the point of
being useless in essentially all cases.
In particular, it only support some float32 and unorm8
formats and doesn't work on x86-64.
This commit rewrites it to support:
1. Dumb memory copy for any pair of identical formats
2. All formats that are swizzles of each other
3. Converting 32/64-bit floats and all 8/16/32-bit integers to 32-bit float
4. Converting unorm8/snorm8 to snorm16 and uscaled8/sscaled8 to sscaled16
5. Support for x86-64 (doesn't take advantage of it in any way though)
This new translate can even be useful to translate index buffers for
cards that lack 8-bit index support.
It passes the testsuite I wrote, but note that this is a major change, and more
testing would be great.
Luca Barbieri [Thu, 12 Aug 2010 16:27:02 +0000 (18:27 +0200)]
rtasm: add minimal x86-64 support and new instructions (v5)
Changes in v5:
- Add sse2_movdqa
Changes in v4:
- Use _WIN64 instead of WIN64
Changes in v3:
- Add target and target caps functions, so that they could be different in
principle from the current CPU and they don't need #ifs to check
Changes in v2:
- Win64 support (untested)
- Use u_cpu_detect.h constants instead of #ifs
This commit adds minimal x86-64 support: only movs between registers
are supported for r8-r15, and x64_rexw() must be used to ask for 64-bit
operations.
It also adds several new instructions for the new translate_sse code.
movdqa