Skip to content
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.

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

DSE: pre-commit tests for scalable vectors #110669

Merged
merged 3 commits into from
Nov 28, 2024

Conversation

artagnon
Copy link
Contributor

@artagnon artagnon commented Oct 1, 2024

As AliasAnalysis now has support for scalable sizes, add tests to DeadStoreElimination covering the scalable vectors case, in preparation to extend it.

@llvmbot
Copy link
Member

llvmbot commented Oct 1, 2024

@llvm/pr-subscribers-llvm-transforms

Author: Ramkumar Ramachandra (artagnon)

Changes

As AliasAnalysis now has support for scalable sizes, add tests to DeadStoreElimination covering the scalable vectors case, in preparation to extend it.


Full diff: https://github.com/llvm/llvm-project/pull/110669.diff

2 Files Affected:

  • (modified) llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll (+53-3)
  • (modified) llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll (+52)
diff --git a/llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll b/llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll
index fbab350008f4ee..c2f7adde7a22ee 100644
--- a/llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll
+++ b/llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll
@@ -47,6 +47,29 @@ bb:
   ret void
 }
 
+define void @ScalableVectorTestFullyOverlapping(ptr %arg, i32 %i) {
+; CHECK-LABEL: @ScalableVectorTestFullyOverlapping(
+; CHECK-NEXT:  bb:
+; CHECK-NEXT:    [[I7:%.*]] = add nuw nsw i32 [[I:%.*]], 1
+; CHECK-NEXT:    [[I8:%.*]] = zext i32 [[I7]] to i64
+; CHECK-NEXT:    [[I9:%.*]] = getelementptr inbounds float, ptr [[ARG:%.*]], i64 [[I8]]
+; CHECK-NEXT:    store float 0.000000e+00, ptr [[I9]], align 4
+; CHECK-NEXT:    [[I2:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT:    [[I3:%.*]] = getelementptr inbounds float, ptr [[ARG]], i64 [[I2]]
+; CHECK-NEXT:    store <vscale x 2 x float> zeroinitializer, ptr [[I3]], align 16
+; CHECK-NEXT:    ret void
+;
+bb:
+  %i7 = add nuw nsw i32 %i, 1
+  %i8 = zext i32 %i7 to i64
+  %i9 = getelementptr inbounds float, ptr %arg, i64 %i8
+  store float 0.0, ptr %i9, align 4
+  %i2 = zext i32 %i to i64
+  %i3 = getelementptr inbounds float, ptr %arg, i64 %i2
+  store <vscale x 2 x float> zeroinitializer, ptr %i3, align 16
+  ret void
+}
+
 define void @ArrayTestPartiallyOverlapping(i64 %0) {
 ;
 ; The DSE pass will not kill the store because the overlap is partial
@@ -55,9 +78,9 @@ define void @ArrayTestPartiallyOverlapping(i64 %0) {
 ; CHECK-LABEL: @ArrayTestPartiallyOverlapping(
 ; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[TMP0:%.*]], 10
 ; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [0 x i8], ptr @BUFFER, i64 0, i64 [[TMP2]]
-; CHECK-NEXT:    [[TMP5:%.*]] = add i64 [[TMP0]], 15
-; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [0 x i8], ptr @BUFFER, i64 0, i64 [[TMP5]]
-; CHECK-NEXT:    store i32 1, ptr [[TMP6]], align 4
+; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[TMP0]], 15
+; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [0 x i8], ptr @BUFFER, i64 0, i64 [[TMP4]]
+; CHECK-NEXT:    store i32 1, ptr [[TMP5]], align 4
 ; CHECK-NEXT:    store i64 0, ptr [[TMP3]], align 4
 ; CHECK-NEXT:    ret void
 ;
@@ -97,3 +120,30 @@ bb:
   ret void
 }
 
+define void @ScalableVectorTestPartiallyOverlapping(ptr %arg, i32 %i) {
+;
+; The DSE pass will not kill the store because the overlap is partial
+; and won't fully clobber the original store.
+;
+; CHECK-LABEL: @ScalableVectorTestPartiallyOverlapping(
+; CHECK-NEXT:  bb:
+; CHECK-NEXT:    [[I2:%.*]] = zext i32 [[I:%.*]] to i64
+; CHECK-NEXT:    [[I3:%.*]] = getelementptr inbounds float, ptr [[ARG:%.*]], i64 [[I2]]
+; CHECK-NEXT:    store <vscale x 2 x float> zeroinitializer, ptr [[I3]], align 16
+; CHECK-NEXT:    [[I5:%.*]] = add nuw nsw i32 [[I]], 1
+; CHECK-NEXT:    [[I6:%.*]] = zext i32 [[I5]] to i64
+; CHECK-NEXT:    [[I7:%.*]] = getelementptr inbounds float, ptr [[ARG]], i64 [[I6]]
+; CHECK-NEXT:    store <vscale x 2 x float> zeroinitializer, ptr [[I7]], align 16
+; CHECK-NEXT:    ret void
+;
+bb:
+  %i2 = zext i32 %i to i64
+  %i3 = getelementptr inbounds float, ptr %arg, i64 %i2
+  store <vscale x 2 x float> zeroinitializer, ptr %i3, align 16
+  %i5 = add nuw nsw i32 %i, 1
+  %i6 = zext i32 %i5 to i64
+  %i7 = getelementptr inbounds float, ptr %arg, i64 %i6
+  store <vscale x 2 x float> zeroinitializer, ptr %i7, align 16
+  ret void
+}
+
diff --git a/llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll b/llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll
index c9a0943de8cd98..ae203dfba7ddc1 100644
--- a/llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll
+++ b/llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll
@@ -655,3 +655,55 @@ exit:
   call void @use(ptr %p) argmemonly
   ret void
 }
+
+define void @scalable_scalable_redundant_store(ptr %ptr) {
+; CHECK-LABEL: @scalable_scalable_redundant_store(
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 2
+; CHECK-NEXT:    store <vscale x 2 x i64> zeroinitializer, ptr [[GEP]], align 16
+; CHECK-NEXT:    store <vscale x 4 x i64> zeroinitializer, ptr [[PTR]], align 32
+; CHECK-NEXT:    ret void
+;
+  %gep = getelementptr i64, ptr %ptr, i64 2
+  store <vscale x 2 x i64> zeroinitializer, ptr %gep
+  store <vscale x 4 x i64> zeroinitializer, ptr %ptr
+  ret void
+}
+
+define void @scalable_fixed_redundant_store(ptr %ptr) {
+; CHECK-LABEL: @scalable_fixed_redundant_store(
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 2
+; CHECK-NEXT:    store <2 x i64> zeroinitializer, ptr [[GEP]], align 16
+; CHECK-NEXT:    store <vscale x 4 x i64> zeroinitializer, ptr [[PTR]], align 32
+; CHECK-NEXT:    ret void
+;
+  %gep = getelementptr i64, ptr %ptr, i64 2
+  store <2 x i64> zeroinitializer, ptr %gep
+  store <vscale x 4 x i64> zeroinitializer, ptr %ptr
+  ret void
+}
+
+define void @fixed_scalable_redundant_store(ptr %ptr) {
+; CHECK-LABEL: @fixed_scalable_redundant_store(
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 2
+; CHECK-NEXT:    store <vscale x 2 x i64> zeroinitializer, ptr [[GEP]], align 16
+; CHECK-NEXT:    store <128 x i64> zeroinitializer, ptr [[PTR]], align 1024
+; CHECK-NEXT:    ret void
+;
+  %gep = getelementptr i64, ptr %ptr, i64 2
+  store <vscale x 2 x i64> zeroinitializer, ptr %gep
+  store <128 x i64> zeroinitializer, ptr %ptr
+  ret void
+}
+
+define void @scalable_scalable_neg(ptr %ptr) {
+; CHECK-LABEL: @scalable_scalable_neg(
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 8
+; CHECK-NEXT:    store <vscale x 4 x i64> zeroinitializer, ptr [[GEP]], align 32
+; CHECK-NEXT:    store <vscale x 2 x i64> zeroinitializer, ptr [[PTR]], align 16
+; CHECK-NEXT:    ret void
+;
+  %gep = getelementptr i64, ptr %ptr, i64 8
+  store <vscale x 4 x i64> zeroinitializer, ptr %gep
+  store <vscale x 2 x i64> zeroinitializer, ptr %ptr
+  ret void
+}

Copy link
Contributor

@fhahn fhahn left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks!

Left a few optional suggestions on improving the naming to make the tests a bit easier to read

As AliasAnalysis now has support for scalable sizes, add tests to
DeadStoreElimination covering the scalable vectors case, in preparation
to extend it.
@artagnon artagnon force-pushed the dse-aa-scalable-test branch from 870743f to 2a7de12 Compare November 28, 2024 12:28
@artagnon artagnon merged commit 4e8eabd into llvm:main Nov 28, 2024
8 checks passed
@artagnon artagnon deleted the dse-aa-scalable-test branch November 28, 2024 16:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants