[llvm] 4e8eabd - DSE: pre-commit tests for scalable vectors (#110669)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Nov 28 08:16:19 PST 2024
Author: Ramkumar Ramachandra
Date: 2024-11-28T16:16:16Z
New Revision: 4e8eabd93e2c60a9dc2f6afde506012fbca77f81
URL: https://github.com/llvm/llvm-project/commit/4e8eabd93e2c60a9dc2f6afde506012fbca77f81
DIFF: https://github.com/llvm/llvm-project/commit/4e8eabd93e2c60a9dc2f6afde506012fbca77f81.diff
LOG: DSE: pre-commit tests for scalable vectors (#110669)
As AliasAnalysis now has support for scalable sizes, add tests to
DeadStoreElimination covering the scalable vectors case, in preparation
to extend it.
Added:
Modified:
llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll
llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll
Removed:
################################################################################
diff --git a/llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll b/llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll
index bdc023128c8520..48f0c58ceb0534 100644
--- a/llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll
+++ b/llvm/test/Transforms/DeadStoreElimination/offsetted-overlapping-stores.ll
@@ -47,6 +47,90 @@ bb:
ret void
}
+define void @ScalableVectorTestFullyOverlapping(ptr %arg, i32 %i) vscale_range(1, 2) {
+; CHECK-LABEL: @ScalableVectorTestFullyOverlapping(
+; CHECK-NEXT: [[I_1:%.*]] = add nuw nsw i32 [[I:%.*]], 1
+; CHECK-NEXT: [[EXT_I_1:%.*]] = zext i32 [[I_1]] to i64
+; CHECK-NEXT: [[GEP_ARG_I_1:%.*]] = getelementptr inbounds float, ptr [[ARG:%.*]], i64 [[EXT_I_1]]
+; CHECK-NEXT: store float 0.000000e+00, ptr [[GEP_ARG_I_1]], align 4
+; CHECK-NEXT: [[EXT_I:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT: [[GEP_ARG_I:%.*]] = getelementptr inbounds float, ptr [[ARG]], i64 [[EXT_I]]
+; CHECK-NEXT: store <vscale x 2 x float> zeroinitializer, ptr [[GEP_ARG_I]], align 8
+; CHECK-NEXT: ret void
+;
+ %i.1 = add nuw nsw i32 %i, 1
+ %ext.i.1 = zext i32 %i.1 to i64
+ %gep.arg.i.1 = getelementptr inbounds float, ptr %arg, i64 %ext.i.1
+ store float 0.0, ptr %gep.arg.i.1
+ %ext.i = zext i32 %i to i64
+ %gep.arg.i = getelementptr inbounds float, ptr %arg, i64 %ext.i
+ store <vscale x 2 x float> zeroinitializer, ptr %gep.arg.i
+ ret void
+}
+
+define void @ScalableVectorTestFullyOverlapping2(ptr %arg, i32 %i) {
+; CHECK-LABEL: @ScalableVectorTestFullyOverlapping2(
+; CHECK-NEXT: [[I_1:%.*]] = add nuw nsw i32 [[I:%.*]], 1
+; CHECK-NEXT: [[EXT_I_1:%.*]] = zext i32 [[I_1]] to i64
+; CHECK-NEXT: [[GEP_ARG_I_1:%.*]] = getelementptr inbounds float, ptr [[ARG:%.*]], i64 [[EXT_I_1]]
+; CHECK-NEXT: store <vscale x 2 x float> zeroinitializer, ptr [[GEP_ARG_I_1]], align 8
+; CHECK-NEXT: [[EXT_I:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT: [[GEP_ARG_I:%.*]] = getelementptr inbounds float, ptr [[ARG]], i64 [[EXT_I]]
+; CHECK-NEXT: store <vscale x 4 x float> zeroinitializer, ptr [[GEP_ARG_I]], align 16
+; CHECK-NEXT: ret void
+;
+ %i.1 = add nuw nsw i32 %i, 1
+ %ext.i.1 = zext i32 %i.1 to i64
+ %gep.arg.i.1 = getelementptr inbounds float, ptr %arg, i64 %ext.i.1
+ store <vscale x 2 x float> zeroinitializer, ptr %gep.arg.i.1
+ %ext.i = zext i32 %i to i64
+ %gep.arg.i = getelementptr inbounds float, ptr %arg, i64 %ext.i
+ store <vscale x 4 x float> zeroinitializer, ptr %gep.arg.i
+ ret void
+}
+
+define void @ScalableVectorTestNonOverlapping(ptr %arg, i32 %i) vscale_range(1, 2) {
+; CHECK-LABEL: @ScalableVectorTestNonOverlapping(
+; CHECK-NEXT: [[I_10:%.*]] = add nuw nsw i32 [[I:%.*]], 10
+; CHECK-NEXT: [[EXT_I_10:%.*]] = zext i32 [[I_10]] to i64
+; CHECK-NEXT: [[GEP_ARG_I_10:%.*]] = getelementptr inbounds float, ptr [[ARG:%.*]], i64 [[EXT_I_10]]
+; CHECK-NEXT: store float 0.000000e+00, ptr [[GEP_ARG_I_10]], align 4
+; CHECK-NEXT: [[EXT_I:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT: [[GEP_ARG_I:%.*]] = getelementptr inbounds float, ptr [[ARG]], i64 [[EXT_I]]
+; CHECK-NEXT: store <vscale x 2 x float> zeroinitializer, ptr [[GEP_ARG_I]], align 8
+; CHECK-NEXT: ret void
+;
+ %i.10 = add nuw nsw i32 %i, 10
+ %ext.i.10 = zext i32 %i.10 to i64
+ %gep.arg.i.10 = getelementptr inbounds float, ptr %arg, i64 %ext.i.10
+ store float 0.0, ptr %gep.arg.i.10
+ %ext.i = zext i32 %i to i64
+ %gep.arg.i = getelementptr inbounds float, ptr %arg, i64 %ext.i
+ store <vscale x 2 x float> zeroinitializer, ptr %gep.arg.i
+ ret void
+}
+
+define void @ScalableVectorTestNonOverlapping2(ptr %arg, i32 %i) vscale_range(1, 2) {
+; CHECK-LABEL: @ScalableVectorTestNonOverlapping2(
+; CHECK-NEXT: [[I_10:%.*]] = add nuw nsw i32 [[I:%.*]], 10
+; CHECK-NEXT: [[EXT_I_10:%.*]] = zext i32 [[I_10]] to i64
+; CHECK-NEXT: [[GEP_ARG_I_10:%.*]] = getelementptr inbounds float, ptr [[ARG:%.*]], i64 [[EXT_I_10]]
+; CHECK-NEXT: store <vscale x 2 x float> zeroinitializer, ptr [[GEP_ARG_I_10]], align 8
+; CHECK-NEXT: [[EXT_I:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT: [[GEP_ARG_I:%.*]] = getelementptr inbounds float, ptr [[ARG]], i64 [[EXT_I]]
+; CHECK-NEXT: store <vscale x 4 x float> zeroinitializer, ptr [[GEP_ARG_I]], align 16
+; CHECK-NEXT: ret void
+;
+ %i.10 = add nuw nsw i32 %i, 10
+ %ext.i.10 = zext i32 %i.10 to i64
+ %gep.arg.i.10 = getelementptr inbounds float, ptr %arg, i64 %ext.i.10
+ store <vscale x 2 x float> zeroinitializer, ptr %gep.arg.i.10
+ %ext.i = zext i32 %i to i64
+ %gep.arg.i = getelementptr inbounds float, ptr %arg, i64 %ext.i
+ store <vscale x 4 x float> zeroinitializer, ptr %gep.arg.i
+ ret void
+}
+
define void @ArrayTestPartiallyOverlapping(i64 %0) {
;
; The DSE pass will not kill the store because the overlap is partial
@@ -55,9 +139,9 @@ 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 +181,28 @@ 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: [[EXT_I:%.*]] = zext i32 [[I:%.*]] to i64
+; CHECK-NEXT: [[GEP_ARG_I:%.*]] = getelementptr inbounds float, ptr [[ARG:%.*]], i64 [[EXT_I]]
+; CHECK-NEXT: store <vscale x 2 x float> zeroinitializer, ptr [[GEP_ARG_I]], align 8
+; CHECK-NEXT: [[I_1:%.*]] = add nuw nsw i32 [[I]], 1
+; CHECK-NEXT: [[EXT_I_1:%.*]] = zext i32 [[I_1]] to i64
+; CHECK-NEXT: [[GEP_ARG_I_1:%.*]] = getelementptr inbounds float, ptr [[ARG]], i64 [[EXT_I_1]]
+; CHECK-NEXT: store <vscale x 2 x float> zeroinitializer, ptr [[GEP_ARG_I_1]], align 8
+; CHECK-NEXT: ret void
+;
+ %ext.i = zext i32 %i to i64
+ %gep.arg.i = getelementptr inbounds float, ptr %arg, i64 %ext.i
+ store <vscale x 2 x float> zeroinitializer, ptr %gep.arg.i
+ %i.1 = add nuw nsw i32 %i, 1
+ %ext.i.1 = zext i32 %i.1 to i64
+ %gep.arg.i.1 = getelementptr inbounds float, ptr %arg, i64 %ext.i.1
+ store <vscale x 2 x float> zeroinitializer, ptr %gep.arg.i.1
+ 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..7193bb501c89de 100644
--- a/llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll
+++ b/llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll
@@ -641,7 +641,7 @@ define void @later_non_removable_store(i1 %c, ptr %p) {
; CHECK: if:
; CHECK-NEXT: br label [[EXIT]]
; CHECK: exit:
-; CHECK-NEXT: call void @use(ptr [[P]]) #[[ATTR6:[0-9]+]]
+; CHECK-NEXT: call void @use(ptr [[P]]) #[[ATTR7:[0-9]+]]
; CHECK-NEXT: ret void
;
store i8 1, ptr %p
@@ -655,3 +655,94 @@ 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_PTR_2:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 2
+; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[GEP_PTR_2]], align 16
+; CHECK-NEXT: store <vscale x 4 x i64> zeroinitializer, ptr [[PTR]], align 32
+; CHECK-NEXT: ret void
+;
+ %gep.ptr.2 = getelementptr i64, ptr %ptr, i64 2
+ store <vscale x 2 x i64> zeroinitializer, ptr %gep.ptr.2
+ store <vscale x 4 x i64> zeroinitializer, ptr %ptr
+ ret void
+}
+
+define void @scalable_scalable_neg(ptr %ptr) {
+; CHECK-LABEL: @scalable_scalable_neg(
+; CHECK-NEXT: [[GEP_PTR_8:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 8
+; CHECK-NEXT: store <vscale x 4 x i64> zeroinitializer, ptr [[GEP_PTR_8]], align 32
+; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[PTR]], align 16
+; CHECK-NEXT: ret void
+;
+ %gep.ptr.8 = getelementptr i64, ptr %ptr, i64 8
+ store <vscale x 4 x i64> zeroinitializer, ptr %gep.ptr.8
+ store <vscale x 2 x i64> zeroinitializer, ptr %ptr
+ ret void
+}
+
+define void @scalable_scalable_nonconst_offset_neg(ptr %ptr, i64 %i) {
+; CHECK-LABEL: @scalable_scalable_nonconst_offset_neg(
+; CHECK-NEXT: [[GEP_PTR_I:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 [[I:%.*]]
+; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[GEP_PTR_I]], align 16
+; CHECK-NEXT: store <vscale x 4 x i64> zeroinitializer, ptr [[PTR]], align 32
+; CHECK-NEXT: ret void
+;
+ %gep.ptr.i = getelementptr i64, ptr %ptr, i64 %i
+ store <vscale x 2 x i64> zeroinitializer, ptr %gep.ptr.i
+ store <vscale x 4 x i64> zeroinitializer, ptr %ptr
+ ret void
+}
+
+define void @scalable_fixed_redundant_store(ptr %ptr) vscale_range(1, 2) {
+; CHECK-LABEL: @scalable_fixed_redundant_store(
+; CHECK-NEXT: [[GEP_PTR_2:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 2
+; CHECK-NEXT: store <2 x i64> zeroinitializer, ptr [[GEP_PTR_2]], align 16
+; CHECK-NEXT: store <vscale x 4 x i64> zeroinitializer, ptr [[PTR]], align 32
+; CHECK-NEXT: ret void
+;
+ %gep.ptr.2 = getelementptr i64, ptr %ptr, i64 2
+ store <2 x i64> zeroinitializer, ptr %gep.ptr.2
+ store <vscale x 4 x i64> zeroinitializer, ptr %ptr
+ ret void
+}
+
+define void @scalable_fixed_neg(ptr %ptr) vscale_range(1, 2) {
+; CHECK-LABEL: @scalable_fixed_neg(
+; CHECK-NEXT: [[GEP_PTR_16:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 16
+; CHECK-NEXT: store <2 x i64> zeroinitializer, ptr [[GEP_PTR_16]], align 16
+; CHECK-NEXT: store <vscale x 4 x i64> zeroinitializer, ptr [[PTR]], align 32
+; CHECK-NEXT: ret void
+;
+ %gep.ptr.16 = getelementptr i64, ptr %ptr, i64 16
+ store <2 x i64> zeroinitializer, ptr %gep.ptr.16
+ store <vscale x 4 x i64> zeroinitializer, ptr %ptr
+ ret void
+}
+
+define void @fixed_scalable_redundant_store(ptr %ptr) vscale_range(1, 2) {
+; CHECK-LABEL: @fixed_scalable_redundant_store(
+; CHECK-NEXT: [[GEP_PTR_2:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 2
+; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[GEP_PTR_2]], align 16
+; CHECK-NEXT: store <8 x i64> zeroinitializer, ptr [[PTR]], align 64
+; CHECK-NEXT: ret void
+;
+ %gep.ptr.2 = getelementptr i64, ptr %ptr, i64 2
+ store <vscale x 2 x i64> zeroinitializer, ptr %gep.ptr.2
+ store <8 x i64> zeroinitializer, ptr %ptr
+ ret void
+}
+
+define void @fixed_scalable_neg(ptr %ptr) vscale_range(1, 2) {
+; CHECK-LABEL: @fixed_scalable_neg(
+; CHECK-NEXT: [[GEP_PTR_2:%.*]] = getelementptr i64, ptr [[PTR:%.*]], i64 2
+; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[GEP_PTR_2]], align 16
+; CHECK-NEXT: store <4 x i64> zeroinitializer, ptr [[PTR]], align 32
+; CHECK-NEXT: ret void
+;
+ %gep.ptr.2 = getelementptr i64, ptr %ptr, i64 2
+ store <vscale x 2 x i64> zeroinitializer, ptr %gep.ptr.2
+ store <4 x i64> zeroinitializer, ptr %ptr
+ ret void
+}
More information about the llvm-commits
mailing list