[NVPTX] Add intrinsics for wgmma.fence PTX instructions

[Wolfram70]

This PR adds NVVM intrinsics and NVPTX codegen for:

• wgmma.fence.sync.aligned
• wgmma.commit_group.sync.aligned
• wgmma.wait_group.sync.aligned

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-llvm-ir

@llvm/pr-subscribers-backend-nvptx

Author: Srinivasa R (Wolfram70)

Changes

This PR adds NVVM intrinsics and NVPTX codegen for:

• wgmma.fence.sync.aligned
• wgmma.commit_group.sync.aligned
• wgmma.wait_group.sync.aligned

---

Full diff: https://github.com/llvm/llvm-project/pull/120523.diff

5 Files Affected:

• (modified) llvm/include/llvm/IR/IntrinsicsNVVM.td (+15)
• (modified) llvm/lib/Target/NVPTX/NVPTXIntrinsics.td (+12)
• (added) llvm/test/CodeGen/NVPTX/wgmma-commit-group.ll (+15)
• (added) llvm/test/CodeGen/NVPTX/wgmma-fence.ll (+15)
• (added) llvm/test/CodeGen/NVPTX/wgmma-wait-group.ll (+15)

diff --git a/llvm/include/llvm/IR/IntrinsicsNVVM.td b/llvm/include/llvm/IR/IntrinsicsNVVM.td
index 9834dbb70d4c1f..6f64311cc06c03 100644
--- a/llvm/include/llvm/IR/IntrinsicsNVVM.td
+++ b/llvm/include/llvm/IR/IntrinsicsNVVM.td
@@ -4805,6 +4805,21 @@ def int_nvvm_redux_sync_or : ClangBuiltin<"__nvvm_redux_sync_or">,
   Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
             [IntrConvergent, IntrInaccessibleMemOnly, IntrNoCallback]>;
 
+//
+// WGMMA instructions
+//
+// wgmma.fence.sync.aligned;
+def int_nvvm_wgmma_fence_sync_aligned 
+  : Intrinsic<[], [], [IntrConvergent], "llvm.nvvm.wgmma.fence.sync.aligned">;
+
+// wgmma.commit_group.sync.aligned;
+def int_nvvm_wgmma_commit_group_sync_aligned
+  : Intrinsic<[], [], [IntrConvergent], "llvm.nvvm.wgmma.commit_group.sync.aligned">;
+
+// wgmma.wait_group.sync.aligned N;
+def int_nvvm_wgmma_wait_group_sync_aligned
+  : Intrinsic<[], [llvm_i32_ty], [IntrConvergent, ImmArg<ArgIndex<0>>], "llvm.nvvm.wgmma.wait_group.sync.aligned">;
+
 //
 // WMMA instructions
 //
diff --git a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
index 256161d5d79c77..d83985b46abe56 100644
--- a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
+++ b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -7484,4 +7484,16 @@ defm INT_SET_MAXNREG_DEC : SET_MAXNREG<"dec", int_nvvm_setmaxnreg_dec_sync_align
 
 } // isConvergent
 
+//
+// WGMMA instructions
+//
+def INT_NVVM_WGMMA_FENCE_SYNC_ALIGNED : NVPTXInst<(outs), (ins), "wgmma.fence.sync.aligned;",
+                             [(int_nvvm_wgmma_fence_sync_aligned)]>, Requires<[hasSM90a, hasPTX<80>]>;
+
+def INT_NVVM_WGMMA_COMMIT_GROUP_SYNC_ALIGNED : NVPTXInst<(outs), (ins), "wgmma.commit_group.sync.aligned;",
+                             [(int_nvvm_wgmma_commit_group_sync_aligned)]>, Requires<[hasSM90a, hasPTX<80>]>;
+
+def INT_NVVM_WGMMA_WAIT_GROUP_SYNC_ALIGNED : NVPTXInst<(outs), (ins i32imm:$n), "wgmma.wait_group.sync.aligned \t$n;",
+                             [(int_nvvm_wgmma_wait_group_sync_aligned timm:$n)]>, Requires<[hasSM90a, hasPTX<80>]>;
+
 def INT_EXIT : NVPTXInst<(outs), (ins), "exit;", [(int_nvvm_exit)]>;
diff --git a/llvm/test/CodeGen/NVPTX/wgmma-commit-group.ll b/llvm/test/CodeGen/NVPTX/wgmma-commit-group.ll
new file mode 100644
index 00000000000000..d2a41f44b28776
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/wgmma-commit-group.ll
@@ -0,0 +1,15 @@
+; RUN: llc < %s -march=nvptx64 -mcpu=sm_90a -mattr=+ptx80 | FileCheck %s
+; RUN: %if ptxas-12.0 %{ llc < %s -march=nvptx64 -mcpu=sm_90a -mattr=+ptx80 | %ptxas-verify -arch=sm_90a %}
+
+target triple = "nvptx64-nvidia-cuda"
+
+declare void @llvm.nvvm.wgmma.commit_group.sync.aligned()
+
+define void @test_wgmma_commit_group_sync_aligned() {
+  ; CHECK-LABEL:  test_wgmma_commit_group_sync_aligned(
+  ; CHECK:        // %bb.0:
+  ; CHECK-NEXT:     wgmma.commit_group.sync.aligned;
+  ; CHECK-NEXT:     ret;
+  call void @llvm.nvvm.wgmma.commit_group.sync.aligned()
+  ret void
+}
\ No newline at end of file
diff --git a/llvm/test/CodeGen/NVPTX/wgmma-fence.ll b/llvm/test/CodeGen/NVPTX/wgmma-fence.ll
new file mode 100644
index 00000000000000..24075ba93786a6
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/wgmma-fence.ll
@@ -0,0 +1,15 @@
+; RUN: llc < %s -march=nvptx64 -mcpu=sm_90a -mattr=+ptx80 | FileCheck %s
+; RUN: %if ptxas-12.0 %{ llc < %s -march=nvptx64 -mcpu=sm_90a -mattr=+ptx80 | %ptxas-verify -arch=sm_90a %}
+
+target triple = "nvptx64-nvidia-cuda"
+
+declare void @llvm.nvvm.wgmma.fence.sync.aligned()
+
+define void @test_wgmma_fence_sync_aligned() {
+  ; CHECK-LABEL:  test_wgmma_fence_sync_aligned(
+  ; CHECK:        // %bb.0:
+  ; CHECK-NEXT:     wgmma.fence.sync.aligned;
+  ; CHECK-NEXT:     ret;
+  call void @llvm.nvvm.wgmma.fence.sync.aligned()
+  ret void
+}
\ No newline at end of file
diff --git a/llvm/test/CodeGen/NVPTX/wgmma-wait-group.ll b/llvm/test/CodeGen/NVPTX/wgmma-wait-group.ll
new file mode 100644
index 00000000000000..ad314d2faa5fd1
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/wgmma-wait-group.ll
@@ -0,0 +1,15 @@
+; RUN: llc < %s -march=nvptx64 -mcpu=sm_90a -mattr=+ptx80 | FileCheck %s
+; RUN: %if ptxas-12.0 %{ llc < %s -march=nvptx64 -mcpu=sm_90a -mattr=+ptx80 | %ptxas-verify -arch=sm_90a %}
+
+target triple = "nvptx64-nvidia-cuda"
+
+declare void @llvm.nvvm.wgmma.wait_group.sync.aligned(i32)
+
+define void @test_wgmma_wait_group_sync_aligned() {
+  ; CHECK-LABEL:  test_wgmma_wait_group_sync_aligned(
+  ; CHECK:        // %bb.0:
+  ; CHECK-NEXT:     wgmma.wait_group.sync.aligned   10;
+  ; CHECK-NEXT:     ret;
+  call void @llvm.nvvm.wgmma.wait_group.sync.aligned(i32 10)
+  ret void
+}
\ No newline at end of file

[gchak]

Can these intrinsics be documented in the NVPTX usage guide in this same change?

[durga4github]

@Wolfram70 , Can you confirm the generated ptx compiles smoothly with 12.0 ptxas ? (using the LLVM_PTXAS_EXECUTABLE option)

[Wolfram70]

@Wolfram70 , Can you confirm the generated ptx compiles smoothly with 12.0 ptxas ? (using the LLVM_PTXAS_EXECUTABLE option)

Yes, I have verified that it passes through ptxas without issues.

[Wolfram70]

Can these intrinsics be documented in the NVPTX usage guide in this same change?

Yes, I have now added the corresponding documentation for these intrinsics, thanks!

[durga4github]

Changes LGTM,

@AlexMaclean , @Artem-B , Could one of you please help with a review?

[justinfargnoli]

Merging at @Wolfram70 offline request.

[github-actions]

@Wolfram70 Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!