InferAddressSpaces: Handle prefetch intrinsic

[arsenm]

No description provided.

[arsenm]

• InferAddressSpaces: Handle llvm.is.constant #102010
• InferAddressSpaces: Handle masked load and store intrinsics #102007
• InferAddressSpaces: Handle prefetch intrinsic #101982 👈
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @arsenm and the rest of your teammates on Graphite

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/101982.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/Scalar/InferAddressSpaces.cpp (+8)
• (added) llvm/test/Transforms/InferAddressSpaces/AMDGPU/prefetch.ll (+59)

diff --git a/llvm/lib/Transforms/Scalar/InferAddressSpaces.cpp b/llvm/lib/Transforms/Scalar/InferAddressSpaces.cpp
index 830c15249582c..87b885447cc02 100644
--- a/llvm/lib/Transforms/Scalar/InferAddressSpaces.cpp
+++ b/llvm/lib/Transforms/Scalar/InferAddressSpaces.cpp
@@ -401,6 +401,13 @@ bool InferAddressSpacesImpl::rewriteIntrinsicOperands(IntrinsicInst *II,
     II->setCalledFunction(NewDecl);
     return true;
   }
+  case Intrinsic::prefetch: {
+    Function *NewDecl =
+        Intrinsic::getDeclaration(M, II->getIntrinsicID(), {NewV->getType()});
+    II->setArgOperand(0, NewV);
+    II->setCalledFunction(NewDecl);
+    return true;
+  }
   default: {
     Value *Rewrite = TTI->rewriteIntrinsicWithAddressSpace(II, OldV, NewV);
     if (!Rewrite)
@@ -423,6 +430,7 @@ void InferAddressSpacesImpl::collectRewritableIntrinsicOperands(
                                                  PostorderStack, Visited);
     break;
   case Intrinsic::masked_gather:
+  case Intrinsic::prefetch:
     appendsFlatAddressExpressionToPostorderStack(II->getArgOperand(0),
                                                  PostorderStack, Visited);
     break;
diff --git a/llvm/test/Transforms/InferAddressSpaces/AMDGPU/prefetch.ll b/llvm/test/Transforms/InferAddressSpaces/AMDGPU/prefetch.ll
new file mode 100644
index 0000000000000..e2c12c4c37141
--- /dev/null
+++ b/llvm/test/Transforms/InferAddressSpaces/AMDGPU/prefetch.ll
@@ -0,0 +1,59 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=infer-address-spaces %s | FileCheck %s
+
+define void @prefetch_shared_to_flat(ptr addrspace(3) %group.ptr) {
+; CHECK-LABEL: define void @prefetch_shared_to_flat(
+; CHECK-SAME: ptr addrspace(3) [[GROUP_PTR:%.*]]) {
+; CHECK-NEXT:    tail call void @llvm.prefetch.p3(ptr addrspace(3) [[GROUP_PTR]], i32 0, i32 0, i32 1)
+; CHECK-NEXT:    ret void
+;
+  %cast = addrspacecast ptr addrspace(3) %group.ptr to ptr
+  tail call void @llvm.prefetch.p0(ptr %cast, i32 0, i32 0, i32 1)
+  ret void
+}
+
+define void @prefetch_global_to_flat(ptr addrspace(1) %global.ptr) {
+; CHECK-LABEL: define void @prefetch_global_to_flat(
+; CHECK-SAME: ptr addrspace(1) [[GLOBAL_PTR:%.*]]) {
+; CHECK-NEXT:    tail call void @llvm.prefetch.p1(ptr addrspace(1) [[GLOBAL_PTR]], i32 0, i32 0, i32 1)
+; CHECK-NEXT:    ret void
+;
+  %cast = addrspacecast ptr addrspace(1) %global.ptr to ptr
+  tail call void @llvm.prefetch.p0(ptr addrspace(0) %cast, i32 0, i32 0, i32 1)
+  ret void
+}
+
+define void @prefetch_constant_to_flat(ptr addrspace(4) %const.ptr) {
+; CHECK-LABEL: define void @prefetch_constant_to_flat(
+; CHECK-SAME: ptr addrspace(4) [[CONST_PTR:%.*]]) {
+; CHECK-NEXT:    tail call void @llvm.prefetch.p4(ptr addrspace(4) [[CONST_PTR]], i32 0, i32 0, i32 1)
+; CHECK-NEXT:    ret void
+;
+  %cast = addrspacecast ptr addrspace(4) %const.ptr to ptr
+  tail call void @llvm.prefetch.p0(ptr %cast, i32 0, i32 0, i32 1)
+  ret void
+}
+
+define void @prefetch_flat_to_shared(ptr %flat.ptr) {
+; CHECK-LABEL: define void @prefetch_flat_to_shared(
+; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) {
+; CHECK-NEXT:    [[CAST:%.*]] = addrspacecast ptr [[FLAT_PTR]] to ptr addrspace(3)
+; CHECK-NEXT:    tail call void @llvm.prefetch.p3(ptr addrspace(3) [[CAST]], i32 0, i32 0, i32 1)
+; CHECK-NEXT:    ret void
+;
+  %cast = addrspacecast ptr %flat.ptr to ptr addrspace(3)
+  tail call void @llvm.prefetch.p3(ptr addrspace(3) %cast, i32 0, i32 0, i32 1)
+  ret void
+}
+
+define void @prefetch_flat_to_global(ptr %flat.ptr) {
+; CHECK-LABEL: define void @prefetch_flat_to_global(
+; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) {
+; CHECK-NEXT:    [[CAST:%.*]] = addrspacecast ptr [[FLAT_PTR]] to ptr addrspace(1)
+; CHECK-NEXT:    tail call void @llvm.prefetch.p1(ptr addrspace(1) [[CAST]], i32 0, i32 0, i32 1)
+; CHECK-NEXT:    ret void
+;
+  %cast = addrspacecast ptr %flat.ptr to ptr addrspace(1)
+  tail call void @llvm.prefetch.p1(ptr addrspace(1) %cast, i32 0, i32 0, i32 1)
+  ret void
+}

[arsenm]

Merge activity

• Aug 5, 4:10 PM EDT: @arsenm started a stack merge that includes this pull request via Graphite.
• Aug 5, 4:12 PM EDT: Graphite rebased this pull request as part of a merge.
• Aug 5, 4:14 PM EDT: @arsenm merged this pull request with Graphite.