Skip to content

[BACKEND] bump to llvm/llvm-project@e12cbd8 #6880

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ThomasRaoux merged 2 commits into from
May 23, 2025
Merged

[BACKEND] bump to llvm/llvm-project@e12cbd8 #6880

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9d2ce41
[BACKEND] bump llvm to e33e623
makslevental May 20, 2025
4437a09
Update llvm-hash.txt
makslevental May 21, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion cmake/llvm-hash.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1 +1 @@
3c709802d31b5bc5ed3af8284b40593ff39b9eec
e33e623cdf188faf56da62677910c707a7e94bf7
9 changes: 3 additions & 6 deletions test/Conversion/amd/async_ops_to_llvm.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -259,16 +259,13 @@ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 16 : i32, ttg.sha
// Each thread needs to load 1 element and we load 1 (sizePerThread) per global.load.lds

// CHECK: llvm.getelementptr
// CHECK: %[[aux_ca:.*]] = llvm.mlir.constant(0 : i32) : i32
// CHECK: rocdl.global.load.lds {{.*}}, {{.*}}, {{.*}}, {{.*}}, %[[aux_ca]]
// CHECK: rocdl.global.load.lds {{.*}}, {{.*}}, 4, 0, 0
%2 = ttg.async_copy_global_to_local %1, %arg2 cacheModifier = ca: tensor<32x32x!tt.ptr<f32>, #blocked> -> <32x32xf32, #shared, #smem, mutable>
// CHECK: llvm.getelementptr
// CHECK: %[[aux_cg:.*]] = llvm.mlir.constant(3 : i32) : i32
// CHECK: rocdl.global.load.lds {{.*}}, {{.*}}, {{.*}}, {{.*}}, %[[aux_cg]]
// CHECK: rocdl.global.load.lds {{.*}}, {{.*}}, 4, 0, 3
%3 = ttg.async_copy_global_to_local %1, %arg2 cacheModifier = cg: tensor<32x32x!tt.ptr<f32>, #blocked> -> <32x32xf32, #shared, #smem, mutable>
// CHECK: llvm.getelementptr
// CHECK: %[[aux_cv:.*]] = llvm.mlir.constant(17 : i32) : i32
// CHECK: rocdl.global.load.lds {{.*}}, {{.*}}, {{.*}}, {{.*}}, %[[aux_cv]]
// CHECK: rocdl.global.load.lds {{.*}}, {{.*}}, 4, 0, 17
%4 = ttg.async_copy_global_to_local %1, %arg2 cacheModifier = cv: tensor<32x32x!tt.ptr<f32>, #blocked> -> <32x32xf32, #shared, #smem, mutable>
tt.return
}
Expand Down
13 changes: 6 additions & 7 deletions third_party/amd/lib/TritonAMDGPUToLLVM/LoadStoreOpToLLVM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -639,10 +639,9 @@ struct AsyncCopyGlobalToLocalOpConversion
(vecTy.getNumElements() * vecTy.getElementTypeBitWidth()) / 8;
assert(llvm::isPowerOf2_32(vecBytes));
Value vecBytesVal = b.i32_val(vecBytes);

Value cacheModifiers =
b.i32_val(mlir::LLVM::AMD::getCtrlBitsForCacheModifierOnTarget(
op.getCache(), /*isLoad=*/true, targetInfo));
int32_t cacheModifiers =
mlir::LLVM::AMD::getCtrlBitsForCacheModifierOnTarget(
op.getCache(), /*isLoad=*/true, targetInfo);

Value llMask = adaptor.getMask();
SmallVector<Value> maskElems;
Expand Down Expand Up @@ -680,7 +679,7 @@ struct AsyncCopyGlobalToLocalOpConversion
auto globalLoadLdsOp = rewriter.create<ROCDL::GlobalLoadLDSOp>(
loc,
/*globalPtr=*/srcPtr, /*ldsPtr=*/coalescedShmemAddr[i],
/*size=*/vecBytesVal, /*offset=*/b.i32_val(0),
/*size=*/vecBytes, /*offset=*/0,
/*aux=*/cacheModifiers, /*alias_scopes=*/nullptr,
/*noalias_scopes=*/nullptr, /*tbaa=*/nullptr);
LLVM::AMD::addAsyncCopyAliasScope(globalLoadLdsOp);
Expand All @@ -695,8 +694,8 @@ struct AsyncCopyGlobalToLocalOpConversion
rewriter.create<LLVM::CondBrOp>(loc, pred, loadBlock, afterLoad);
rewriter.setInsertionPointToStart(loadBlock);
auto globalLoadLdsOp = rewriter.create<ROCDL::GlobalLoadLDSOp>(
loc, srcPtr, coalescedShmemAddr[i], vecBytesVal,
/*offset=*/b.i32_val(0), cacheModifiers, nullptr, nullptr, nullptr);
loc, srcPtr, coalescedShmemAddr[i], vecBytes,
/*offset=*/0, cacheModifiers, nullptr, nullptr, nullptr);
LLVM::AMD::addAsyncCopyAliasScope(globalLoadLdsOp);

rewriter.create<LLVM::BrOp>(loc, afterLoad);
Expand Down
Loading