The results of this and the other 2 patches look promising. Some initial benchmarks show:
- rL LLVM
- Sun, Aug 20, 2:38 PM
- Subversion
Today
Today
sortMemAccesses() is analogous to the formation of InterleaveGroups in the LoopVectorizer, which also scans a collection of Loads (or Stores) to determine if they are adjacent in some order and can be combined into one Vector Load of a given width; and if so, in what order. This requires a single scan to compute the distances relative to the first access, as done here. But knowing that we're looking for a permutation of a given width, we can more easily sort the accesses as they are entered into a map, holding the minimum and maximum indices. See insertMember() there.
Nice work!
craig.topper updated the diff for D36944: [InstCombine] Fix a weakness in canEvaluateZExtd around 'and' instructions.
Remove TODO from test case
add check for declaration to avoid post-link transformation of pipe functions.
This patch segfaults in the following three tests:
refactor AMDGPULibFunc to handle unmangled lib functions.
[InstCombine] Add a test case for a weakness in canEvaluateZExtd. NFC
Meinersbur committed rL311302: [MatMul] Make MatMul detection independent of internal isl representations..
[MatMul] Make MatMul detection independent of internal isl representations.
This patch adds support for -polly-mse to the new PassManager infrastructure.
Meinersbur added a reviewer for D36926: [Polly][WIP] Clear DependenceInfo after expansion: philip.pfaffe.
bmakam updated the diff for D36900: [WIP] Re-land MachineInstr: Reason locally about some memory objects before going to AA..
Trim the patch and avoid handling stack references for non-value MMOs to prevent LNT test failures.
Ping. @NoQ would you please have a look? Thanks!
Results with the patch.
krobelus committed rL311301: Allow thiscall attribute in test/Tooling/clang-diff-ast.cpp.
Allow thiscall attribute in test/Tooling/clang-diff-ast.cpp
craig.topper retitled D36938: [X86] When selecting sse_load_f32/f64 pattern, make sure there's only one use of every node all the way back to the root of the match from [X86] When selecting sse_load_f32/f64 pattern, make sure there's only one use of every note all the way back to the root of the match to [X86] When selecting sse_load_f32/f64 pattern, make sure there's only one use of every node all the way back to the root of the match.
ctopper committed rL311300: [AVX512] Add a test to check what happens when a load is referenced by two….
[AVX512] Add a test to check what happens when a load is referenced by two…
Updated the patch as per Simon's comments.
Added the FP instruction itineraries which includes SSE4A and SHA instructions.
kuba.brecka committed rL311299: Fix archive-update.test after r311296..
Fix archive-update.test after r311296.
kuba.brecka committed rL311298: Remove "%T" from ASan Darwin tests..
Remove "%T" from ASan Darwin tests.
[AVX-512] Use a scalar load pattern for FPCLASSSS/FPCLASSSD patterns.
kuba.brecka committed rL311296: Remove uses of "%T" from test/Object/archive-* tests..
Remove uses of "%T" from test/Object/archive-* tests.
Updated the patch to use LVILatticeValue (which I moved to a separate header file for now, I'd move it out in a separate patch if we decide to use it) for tracking function parameter values and updated tryToReplaceWithConstant to use range information to removed ICmp instructions.