Skip to content

[flang][cuda] Convert src and dst to llvm.ptr in tma_bulk_load #165618

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

Jump to bottom
Merged
merged 1 commit into from
Oct 29, 2025
Merged

[flang][cuda] Convert src and dst to llvm.ptr in tma_bulk_load #165618

merged 1 commit into from
Oct 29, 2025

Conversation

@clementval
Copy link
Contributor

@clementval clementval commented Oct 29, 2025

No description provided.

@clementval clementval requested a review from wangzpgi October 29, 2025 20:02
@llvmbot llvmbot added flang Flang issues not falling into any other category flang:fir-hlfir labels Oct 29, 2025
@llvmbot
Copy link
Member

llvmbot commented Oct 29, 2025

@llvm/pr-subscribers-flang-fir-hlfir

Author: Valentin Clement (バレンタイン クレメン) (clementval)

Changes

Full diff: https://github.com/llvm/llvm-project/pull/165618.diff

2 Files Affected:

  • (modified) flang/lib/Optimizer/Builder/IntrinsicCall.cpp (+2)
  • (modified) flang/test/Lower/CUDA/cuda-device-proc.cuf (+7-7)
diff --git a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
index 53fe9c0d2f6f0..ca3e1cd46db7d 100644
--- a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
+++ b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
@@ -9362,6 +9362,8 @@ static void genTMABulkLoad(fir::FirOpBuilder &builder, mlir::Location loc,
   mlir::Value size = mlir::arith::MulIOp::create(builder, loc, nelem, eleSize);
   auto llvmPtrTy = mlir::LLVM::LLVMPointerType::get(builder.getContext());
   barrier = builder.createConvert(loc, llvmPtrTy, barrier);
+  dst = builder.createConvert(loc, llvmPtrTy, dst);
+  src = builder.createConvert(loc, llvmPtrTy, src);
   mlir::NVVM::InlinePtxOp::create(
       builder, loc, mlir::TypeRange{}, {dst, src, size, barrier}, {},
       "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], "
diff --git a/flang/test/Lower/CUDA/cuda-device-proc.cuf b/flang/test/Lower/CUDA/cuda-device-proc.cuf
index d8c78887ff924..e5d3c437d7152 100644
--- a/flang/test/Lower/CUDA/cuda-device-proc.cuf
+++ b/flang/test/Lower/CUDA/cuda-device-proc.cuf
@@ -544,7 +544,7 @@ end subroutine
 ! CHECK: %[[ELEM_SIZE:.*]] = arith.constant 8 : i32
 ! CHECK: %[[SIZE:.*]] = arith.muli %[[COUNT]], %[[ELEM_SIZE]] : i32
 ! CHECK: %[[BARRIER_PTR:.*]] = fir.convert %[[BARRIER]]#0 : (!fir.ref<i64>) -> !llvm.ptr
-! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !fir.ref<!fir.array<1024xcomplex<f32>>>, !fir.ref<complex<f32>>, i32, !llvm.ptr)
+! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !llvm.ptr, !llvm.ptr, i32, !llvm.ptr)
 ! CHECK: nvvm.inline_ptx "mbarrier.expect_tx.relaxed.cta.shared::cta.b64 [%0], %1;" ro(%[[BARRIER_PTR]], %[[SIZE]] : !llvm.ptr, i32)
 
 attributes(global) subroutine test_tma_bulk_load_c8(a, n)
@@ -563,7 +563,7 @@ end subroutine
 ! CHECK: %[[ELEM_SIZE:.*]] = arith.constant 16 : i32
 ! CHECK: %[[SIZE:.*]] = arith.muli %[[COUNT]], %[[ELEM_SIZE]] : i32
 ! CHECK: %[[BARRIER_PTR:.*]] = fir.convert %[[BARRIER]]#0 : (!fir.ref<i64>) -> !llvm.ptr
-! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !fir.ref<!fir.array<1024xcomplex<f64>>>, !fir.ref<complex<f64>>, i32, !llvm.ptr)
+! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !llvm.ptr, !llvm.ptr, i32, !llvm.ptr)
 ! CHECK: nvvm.inline_ptx "mbarrier.expect_tx.relaxed.cta.shared::cta.b64 [%0], %1;" ro(%[[BARRIER_PTR]], %[[SIZE]] : !llvm.ptr, i32)
 
 attributes(global) subroutine test_tma_bulk_load_i4(a, n)
@@ -582,7 +582,7 @@ end subroutine
 ! CHECK: %[[ELEM_SIZE:.*]] = arith.constant 4 : i32
 ! CHECK: %[[SIZE:.*]] = arith.muli %[[COUNT]], %[[ELEM_SIZE]] : i32
 ! CHECK: %[[BARRIER_PTR:.*]] = fir.convert %[[BARRIER]]#0 : (!fir.ref<i64>) -> !llvm.ptr
-! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !fir.ref<!fir.array<1024xi32>>, !fir.ref<i32>, i32, !llvm.ptr)
+! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !llvm.ptr, !llvm.ptr, i32, !llvm.ptr)
 ! CHECK: nvvm.inline_ptx "mbarrier.expect_tx.relaxed.cta.shared::cta.b64 [%0], %1;" ro(%[[BARRIER_PTR]], %[[SIZE]] : !llvm.ptr, i32)
 
 attributes(global) subroutine test_tma_bulk_load_i8(a, n)
@@ -601,7 +601,7 @@ end subroutine
 ! CHECK: %[[ELEM_SIZE:.*]] = arith.constant 8 : i32
 ! CHECK: %[[SIZE:.*]] = arith.muli %[[COUNT]], %[[ELEM_SIZE]] : i32
 ! CHECK: %[[BARRIER_PTR:.*]] = fir.convert %[[BARRIER]]#0 : (!fir.ref<i64>) -> !llvm.ptr
-! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !fir.ref<!fir.array<1024xi64>>, !fir.ref<i64>, i32, !llvm.ptr)
+! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !llvm.ptr, !llvm.ptr, i32, !llvm.ptr)
 ! CHECK: nvvm.inline_ptx "mbarrier.expect_tx.relaxed.cta.shared::cta.b64 [%0], %1;" ro(%[[BARRIER_PTR]], %[[SIZE]] : !llvm.ptr, i32)
 
 attributes(global) subroutine test_tma_bulk_load_r2(a, n)
@@ -620,7 +620,7 @@ end subroutine
 ! CHECK: %[[ELEM_SIZE:.*]] = arith.constant 2 : i32
 ! CHECK: %[[SIZE:.*]] = arith.muli %[[COUNT]], %[[ELEM_SIZE]] : i32
 ! CHECK: %[[BARRIER_PTR:.*]] = fir.convert %[[BARRIER]]#0 : (!fir.ref<i64>) -> !llvm.ptr
-! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !fir.ref<!fir.array<1024xf16>>, !fir.ref<f16>, i32, !llvm.ptr)
+! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !llvm.ptr, !llvm.ptr, i32, !llvm.ptr)
 ! CHECK: nvvm.inline_ptx "mbarrier.expect_tx.relaxed.cta.shared::cta.b64 [%0], %1;" ro(%[[BARRIER_PTR]], %[[SIZE]] : !llvm.ptr, i32)
 
 attributes(global) subroutine test_tma_bulk_load_r4(a, n)
@@ -639,7 +639,7 @@ end subroutine
 ! CHECK: %[[ELEM_SIZE:.*]] = arith.constant 4 : i32
 ! CHECK: %[[SIZE:.*]] = arith.muli %[[COUNT]], %[[ELEM_SIZE]] : i32
 ! CHECK: %[[BARRIER_PTR:.*]] = fir.convert %[[BARRIER]]#0 : (!fir.ref<i64>) -> !llvm.ptr
-! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !fir.ref<!fir.array<1024xf32>>, !fir.ref<f32>, i32, !llvm.ptr)
+! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !llvm.ptr, !llvm.ptr, i32, !llvm.ptr)
 ! CHECK: nvvm.inline_ptx "mbarrier.expect_tx.relaxed.cta.shared::cta.b64 [%0], %1;" ro(%[[BARRIER_PTR]], %[[SIZE]] : !llvm.ptr, i32)
 
 attributes(global) subroutine test_tma_bulk_load_r8(a, n)
@@ -658,7 +658,7 @@ end subroutine
 ! CHECK: %[[ELEM_SIZE:.*]] = arith.constant 8 : i32
 ! CHECK: %[[SIZE:.*]] = arith.muli %[[COUNT]], %[[ELEM_SIZE]] : i32
 ! CHECK: %[[BARRIER_PTR:.*]] = fir.convert %[[BARRIER]]#0 : (!fir.ref<i64>) -> !llvm.ptr
-! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !fir.ref<!fir.array<1024xf64>>, !fir.ref<f64>, i32, !llvm.ptr)
+! CHECK: nvvm.inline_ptx "cp.async.bulk.shared::cluster.global.mbarrier::complete_tx::bytes [%0], [%1], %2, [%3];" ro(%{{.*}}, %{{.*}}, %[[SIZE]], %[[BARRIER_PTR]] : !llvm.ptr, !llvm.ptr, i32, !llvm.ptr)
 ! CHECK: nvvm.inline_ptx "mbarrier.expect_tx.relaxed.cta.shared::cta.b64 [%0], %1;" ro(%[[BARRIER_PTR]], %[[SIZE]] : !llvm.ptr, i32)
 
 attributes(global) subroutine test_tma_bulk_store_c4(c, n)

@clementval clementval enabled auto-merge (squash) October 29, 2025 20:11
@clementval clementval merged commit 5c8492a into llvm:main Oct 29, 2025
13 checks passed
@clementval clementval deleted the cuf_tma_bulk_load_convert branch October 29, 2025 20:56
aokblast pushed a commit to aokblast/llvm-project that referenced this pull request Oct 30, 2025
@clementval @aokblast
[flang][cuda] Convert src and dst to llvm.ptr in tma_bulk_load (llvm#…
24eea91 
…165618)
DEBADRIBASAK pushed a commit to DEBADRIBASAK/llvm-project that referenced this pull request Nov 3, 2025
@clementval
[flang][cuda] Convert src and dst to llvm.ptr in tma_bulk_load (llvm#…
c59db61 
…165618)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@wangzpgi wangzpgi wangzpgi approved these changes

Assignees

No one assigned

Labels

flang:fir-hlfir flang Flang issues not falling into any other category

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@clementval @llvmbot @wangzpgi