Charitha Saumya
2998c74a1e
[mlir][xegpu] Add SIMT distribution support for GEMM transpose B case. ( #155517 )
...
This PR adds the features needed for supporting the GEMM with transpose
B case.
Summary of changes.
1). Add distribution logic for `vector.bitcast`, `vector.transpose` and
`memref.extract_aligned_pointer_as_index` cases.
2). Add layout propagation support for `vector.shape_cast`,
`vector.broadcast` and `vector.bitcast`
3). Incorporate slice attribute and `DistributeLayoutAttr` interface
with the core logic in layout prop.
2025-09-19 10:33:27 -07:00
Artem Kroviakov
6c6afdd8c2
[MLIR][XeGPU] Reapply attempt for "Scattered ops sg-to-wi distribution #154949 " ( #156924 )
...
This PR is a reapply of
https://github.com/llvm/llvm-project/pull/154949 , which failed one of
sanitizer checks.
The issue was querying the `warpOp` results in `LoadDistribution` after
calling `moveRegionToNewWarpOpAndAppendReturns()`, which resulted in use
after free. This PR solves the issue by moving the op query before the
call and is otherwise identical to the one linked above.
---------
Co-authored-by: Charitha Saumya <136391709+charithaintc@users.noreply.github.com>
2025-09-04 12:04:30 -07:00
Thurston Dang
c1cc9d2c8a
Revert "[MLIR][XeGPU] Scattered ops sg-to-wi distribution" ( #156761 )
...
Reverts llvm/llvm-project#154949 due to suspected buildbot breakage
(https://lab.llvm.org/buildbot/#/builders/55/builds/16630/steps/11/logs/stdio ).
Previously commented on the original pull request:
https://github.com/llvm/llvm-project/pull/154949#issuecomment-3250709417
```
******************** TEST 'MLIR :: Dialect/XeGPU/subgroup-distribute.mlir' FAILED ********************
...
# | PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
# | Stack dump:
# | 0. Program arguments: /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm_build_hwasan/bin/mlir-opt -xegpu-subgroup-distribute -allow-unregistered-dialect -canonicalize -cse -split-input-file /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/mlir/test/Dialect/XeGPU/subgroup-distribute.mlir
# | #0 0x0000c0af4b066df0 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/llvm/lib/Support/Unix/Signals.inc:834:13
# | #1 0x0000c0af4b060e20 llvm::sys::RunSignalHandlers() /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/llvm/lib/Support/Signals.cpp:105:18
# | #2 0x0000c0af4b0691b4 SignalHandler(int, siginfo_t*, void*) /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/llvm/lib/Support/Unix/Signals.inc:426:38
# | #3 0x0000ee25a3dcb8f8 (linux-vdso.so.1+0x8f8)
# | #4 0x0000ee25a36c7608 (/lib/aarch64-linux-gnu/libc.so.6+0x87608)
# | #5 0x0000ee25a367cb3c raise (/lib/aarch64-linux-gnu/libc.so.6+0x3cb3c)
# | #6 0x0000ee25a3667e00 abort (/lib/aarch64-linux-gnu/libc.so.6+0x27e00)
# | #7 0x0000c0af4ae7e4b0 __sanitizer::Atexit(void (*)()) /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/sanitizer_common/sanitizer_posix_libcdep.cpp:168:10
# | #8 0x0000c0af4ae7c354 __sanitizer::Die() /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/sanitizer_common/sanitizer_termination.cpp:52:5
# | #9 0x0000c0af4ae66a30 Unlock /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/../sanitizer_common/sanitizer_mutex.h:250:16
# | #10 0x0000c0af4ae66a30 ~GenericScopedLock /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/../sanitizer_common/sanitizer_mutex.h:386:51
# | #11 0x0000c0af4ae66a30 __hwasan::ScopedReport::~ScopedReport() /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/hwasan_report.cpp:54:5
# | #12 0x0000c0af4ae661b8 __hwasan::(anonymous namespace)::BaseReport::~BaseReport() /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/hwasan_report.cpp:477:7
# | #13 0x0000c0af4ae63f5c __hwasan::ReportTagMismatch(__sanitizer::StackTrace*, unsigned long, unsigned long, bool, bool, unsigned long*) /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/hwasan_report.cpp:1094:1
# | #14 0x0000c0af4ae4f8e0 Destroy /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/../sanitizer_common/sanitizer_common.h:532:31
# | #15 0x0000c0af4ae4f8e0 ~InternalMmapVector /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/../sanitizer_common/sanitizer_common.h:642:56
# | #16 0x0000c0af4ae4f8e0 __hwasan::HandleTagMismatch(__hwasan::AccessInfo, unsigned long, unsigned long, void*, unsigned long*) /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/hwasan.cpp:245:1
# | #17 0x0000c0af4ae51e8c __hwasan_tag_mismatch4 /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/hwasan/hwasan.cpp:764:1
# | #18 0x0000c0af4ae67b30 __interception::InterceptFunction(char const*, unsigned long*, unsigned long, unsigned long) /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/compiler-rt/lib/interception/interception_linux.cpp:60:0
# | #19 0x0000c0af5641cd24 getNumResults /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/mlir/include/mlir/IR/Operation.h:404:37
# | #20 0x0000c0af5641cd24 getOpResultImpl /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/mlir/include/mlir/IR/Operation.h:1010:5
# | #21 0x0000c0af5641cd24 getResult /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/mlir/include/mlir/IR/Operation.h:407:54
# | #22 0x0000c0af5641cd24 mlir::OpTrait::detail::MultiResultTraitBase<mlir::gpu::WarpExecuteOnLane0Op, mlir::OpTrait::VariadicResults>::getResult(unsigned int) /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/mlir/include/mlir/IR/OpDefinition.h:638:62
# | #23 0x0000c0af56426b60 getType /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/mlir/include/mlir/IR/Value.h:63:33
# | #24 0x0000c0af56426b60 getType /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/mlir/include/mlir/IR/Value.h:105:39
# | #25 0x0000c0af56426b60 (anonymous namespace)::LoadDistribution::matchAndRewrite(mlir::gpu::WarpExecuteOnLane0Op, mlir::PatternRewriter&) const /home/b/sanitizer-aarch64-linux-bootstrap-hwasan/build/llvm-project/mlir/lib/Dialect/XeGPU/Transforms/XeGPUSubgroupDistribute.cpp:991:55
...
```
2025-09-03 14:40:18 -07:00
Artem Kroviakov
5777f71bce
[MLIR][XeGPU] Scattered ops sg-to-wi distribution ( #154949 )
...
This PR adds distribution patterns for scattered load and store ops,
chunk size included.
XeGPU moves toward offsets being part of the load/store ops, so the pass
only supports this case. Manipulating a vector of offsets indirectly
through create_tdesc is complex and soon to become obsolete anyway.
This PR assumes the SIMT-adapted scatter ops verification introduced in
https://github.com/llvm/llvm-project/pull/154653 . The distribution
itself can be reviewed in the meantime.
2025-09-03 11:48:55 -07:00
Chao Chen
c96e2cdd13
[mlir][XeGPU] Update utils for LayoutAttr and SliceAttr support ( #154819 )
2025-08-27 12:37:15 -05:00
Kazu Hirata
c06d3a7b72
[mlir] Remove unused includes (NFC) ( #148769 )
...
These are identified by misc-include-cleaner. I've filtered out those
that break builds. Also, I'm staying away from llvm-config.h,
config.h, and Compiler.h, which likely cause platform- or
compiler-specific build failures.
2025-07-14 22:19:23 -07:00
Benjamin Kramer
3287c1c176
[XeGPU] Move targetinfo constants to their own header file
...
This breaks the dependency from Dialect to Utils, which would be cyclic.
2025-06-26 13:53:00 +02:00
Chao Chen
36fbc6a8d2
[MLIR][XeGPU] Remove the transpose attribute from Gather/Scatter ops and Cleanup the documents ( #145389 )
2025-06-25 19:43:53 -05:00
Charitha Saumya
adc6228ea0
[mlir][xegpu] Refine layout assignment in XeGPU SIMT distribution. ( #142687 )
...
Changes:
* Decouple layout propagation from subgroup distribution and move it to
an independent pass.
* Refine layout assignment to handle control-flow ops correctly (scf.for, scf.while).
* Refine test cases.
2025-06-20 10:43:19 -07:00
