summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/gallivm
AgeCommit message (Collapse)Author
2010-08-21gallivm: Emit DIVPS instead of RCPPS.José Fonseca
See comments for detailed rationale. Thanks to Michal Krol and Zack Rusin for detecting and investigating this in detail.
2010-08-20gallium: make all checks for PIPE_TEXTURE_2D check for PIPE_TEXTURE_RECT tooLuca Barbieri
Searched for them with: git grep -E '[!=]=.*PIPE_TEXTURE_2D|PIPE_TEXTURE_2D.*[!=]=|case.*PIPE_TEXTURE_2D' Behavior hasn't been changed.
2010-08-18gallivm: Use proper index to lookup predicate register array.michal
Doesn't fix anything, as those indices were both always 0.
2010-08-17gallivm: Fix and re-enable MMX-disabling codenobled
Signed-off-by: José Fonseca <jfonseca@vmware.com>
2010-08-14gallivm: Remove unnecessary header.Vinson Lee
2010-08-14u_cpu_detect: remove arch and little_endianLuca Barbieri
This logic duplicates the one in p_config.h, so remove it and adjust the only two places that were using it.
2010-08-14gallivm: Refactor the Newton-Rapshon steps, and disable once again.José Fonseca
It causes a very ugly corruption on the Earth's halo on Google Earth.
2010-08-11gallivm: Fix and enable the extra Newton/Raphson step in lp_build_rcp().José Fonseca
Thanks to Michal for spotting this.
2010-08-11gallivm: Use lp_build_div instead of lp_build_mul + lp_build_rcp.José Fonseca
Single divide, so let lp_build_div decide how to implement this. This will save a multiplication in architectures which don't have a RCP intrinsic.
2010-08-11gallivm: Use unsigned shift in lp_build_minify.José Fonseca
Texture dimensions are unsigned.
2010-08-10gallivm: fix non-SSE4.1 case in lp_build_pack2()Brian Paul
Since there's no SSE instruction for this case, fall through to the generic shuffle code. Fixes bug fd.o 29468.
2010-08-10gallivm: Fix bitwise operations for floats, division for integersnobled
http://bugs.freedesktop.org/29407 Signed-off-by: José Fonseca <jfonseca@vmware.com>
2010-08-10gallivm: Even more type checkingnobled
http://bugs.freedesktop.org/29407 Signed-off-by: José Fonseca <jfonseca@vmware.com>
2010-08-09gallivm: More type checks.José Fonseca
2010-08-09gallivm: Don't call LLVMBuildFNeg on llvm-2.6.José Fonseca
It didn't exist yet.
2010-08-09gallivm: Always use floating-point operators for floating-point typesnobled
This fixes the assert added in LLVM 2.8: assert(getType()->isIntOrIntVectorTy() && "Tried to create an integer operation on a non-integer type!") But it also fixes some subtle bugs, since we should've been doing this since LLVM 2.6 anyway. Includes a modified patch from steckdenis@yahoo.fr for the FNeg instructions in emit_fetch(); thanks for pointing those out. http://bugs.freedesktop.org/29404 http://bugs.freedesktop.org/29407 Signed-off-by: José Fonseca <jfonseca@vmware.com>
2010-08-08gallivm: Fix more integer operations.José Fonseca
2010-08-08gallivm: Use the correct context for integersnobled
See: http://bugs.freedesktop.org/29407
2010-08-08gallivm: Add type checks for the basic operations.José Fonseca
2010-08-06gallium: remove stray semicolonsBrian Paul
2010-08-05gallivm: Only get debug option onceJakob Bornecrantz
2010-07-30llvmpipe: delete function bodies after generating machine codeZack Rusin
2010-07-29gallivm: added lp_build_assert() function to make assertions in LLVM codeBrian Paul
2010-07-29Revert "gallivm: fix lp_build_sample_offset() crash when indexing a 1-D texture"José Fonseca
This reverts commit 5f90e76c54bbf4456c977b3cbca450d7a570179e. Bad cherry-pick.
2010-07-29gallivm: fix lp_build_sample_offset() crash when indexing a 1-D textureBrian Paul
If y==NULL and y_stride==NULL it means the texture is 1D. Return zero for out_i and the offset instead of garbage.
2010-07-29llvmpipe: Avoid corrupting the FPU stack with MMX instructions on 32bit OSes.José Fonseca
Unfortunately LLVM doesn't emit EMMS itself, and there is no easy/effective way to disable MMX. http://llvm.org/bugs/show_bug.cgi?id=3287
2010-07-27gallivm: Add lp_build_select_bitwise() alternative to lp_build_select_bitwiseJosé Fonseca
When (mask & a) | (~mask & b) is meant instead of mask ? a : b
2010-07-21gallivm: replace has_indirect_addressing field with indirect_files fieldBrian Paul
Instead of one big boolean indicating indirect addressing, use a bitfield indicating which register files are accessed with indirect addressing. Most shaders that use indirect addressing only use it to access the constant buffer. So no need to use an array for temporary registers in this case.
2010-07-21gallivm: refactor code into get_indirect_offsets() functionBrian Paul
2010-07-21gallivm: added commentBrian Paul
2010-07-21gallivm: remove extraneous bracesBrian Paul
2010-07-21gallivm: no longer do indirect addressing in get_temp_ptr()Brian Paul
2010-07-21gallivm: implement correct indirect addressing of temp registersBrian Paul
As with indexing the const buffer, the ADDR reg may have totally different values for each element. Need to use a gather operation.
2010-07-21gallivm: re-org, comments for get_temp_ptr()Brian Paul
2010-07-21gallivm: rename a var to avoid compiler warningsBrian Paul
2010-07-20gallivm: fix indirect addressing of constant bufferBrian Paul
The previous code assumed that all elements of the address register were the same. But it can vary from pixel to pixel or vertex to vertex so we must use a gather operation when dynamically indexing the constant buffer. Still need to fix this for the temporary register file...
2010-07-20gallivm: added lp_build_const_int32() helperBrian Paul
2010-07-07gallivm: restore const qualifierBrian Paul
2010-07-07gallivm: fix cube map LOD computationBrian Paul
First, this undoes commit e503af4baa2c709ae5743bb278b277d3faaba076 so we use iround() in lp_build_nearest_mip_level(). Second, in lp_build_sample_general() we need to check if we're sampling a cube map before anything else. Choose the cube face and then recompute the partial derivatives of (S,T) with respect to the chosen cube face. Before, we were using the directional (S,T,R) derivatives to compute the LOD. Third, work around an apparent bug in LLVM 2.7 where setting the lod variable to a const(0) value results in bad x86 code. See comments in the code.
2010-07-06gallivm: Remove unnecessary header.Vinson Lee
2010-07-06gallivm: use trunc, not round in lp_build_nearest_mip_level()Brian Paul
Fixes fd.o bug 28036 (piglit fbo-cubemap.c regression)
2010-07-06gallivm: finish implementation of lp_build_iceil()Brian Paul
Plus fix minor error in lp_build_iceil() by tweaking the offset value. And add a bunch of comments for the round(), trunc(), floor(), ceil() functions.
2010-07-06gallivm: Use SSE4.1's BLENDV instructions for lp_build_select().José Fonseca
2010-07-06gallivm: Fix 8bit comparisons.José Fonseca
2010-07-03gallivm: Remove unnecessary headers.Vinson Lee
2010-07-02gallivm: Do 4ubyte AoS texture filtering for any format that can be expressed.José Fonseca
Except if it has only one channel, as it would take the same number of instructions.
2010-07-02gallivm: Use util_format_description::fetch_rgba_8unorm() when available.José Fonseca
2010-07-02gallivm: Support multiple pixels in lp_build_fetch_rgba_aos().José Fonseca
This allows to do the unpacking of formats that fit in 4 x unorm8 in parallel, 4 pixels at a time.
2010-07-02gallivm: Move lp_build_rgba8_to_f32_soa() to lp_bld_format_soa.cJosé Fonseca
It will be more useful here.
2010-07-02gallivm: Move gather functions to its own module.José Fonseca
They need to grow, and they provide basic functionality which is not specific to sampling.
generated by cgit v1.2.3 (git 2.39.1) at 2024-06-27 18:21:46 +0000