[llvm] [VPlan] Implement mayHaveSideEffects/mayWriteToMemory for VPInterleav… (PR #71360)
Shih-Po Hung via llvm-commits
llvm-commits at lists.llvm.org
Sun Nov 5 23:28:17 PST 2023
https://github.com/arcbbb created https://github.com/llvm/llvm-project/pull/71360
…eRecipe
This helps VPlanTransforms::removeDeadRecipes to work on VPInterleaveRecipe
>From e3ad6f7c5714203cf580bd92489a08f438ca6e63 Mon Sep 17 00:00:00 2001
From: ShihPo Hung <shihpo.hung at sifive.com>
Date: Sun, 5 Nov 2023 23:16:00 -0800
Subject: [PATCH] [VPlan] Implement mayHaveSideEffects/mayWriteToMemory for
VPInterleaveRecipe
This helps VPlanTransforms::removeDeadRecipes to work on VPInterleaveRecipe
---
llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp | 4 ++++
llvm/test/CodeGen/ARM/loopvectorize_pr33804.ll | 12 ++++++++++--
.../AArch64/loopvectorize_pr33804_double.ll | 12 ++++++++++--
3 files changed, 24 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 6b3218dca1b18b0..f8325c645540d56 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -44,6 +44,8 @@ extern cl::opt<bool> EnableVPlanNativePath;
bool VPRecipeBase::mayWriteToMemory() const {
switch (getVPDefID()) {
+ case VPInterleaveSC:
+ return cast<VPInterleaveRecipe>(this)->getNumStoreOperands() > 0;
case VPWidenMemoryInstructionSC: {
return cast<VPWidenMemoryInstructionRecipe>(this)->isStore();
}
@@ -147,6 +149,8 @@ bool VPRecipeBase::mayHaveSideEffects() const {
"underlying instruction has side-effects");
return false;
}
+ case VPInterleaveSC:
+ return mayWriteToMemory();
case VPWidenMemoryInstructionSC:
assert(cast<VPWidenMemoryInstructionRecipe>(this)
->getIngredient()
diff --git a/llvm/test/CodeGen/ARM/loopvectorize_pr33804.ll b/llvm/test/CodeGen/ARM/loopvectorize_pr33804.ll
index 540cbbfe96e8123..547cb4fe0c91825 100644
--- a/llvm/test/CodeGen/ARM/loopvectorize_pr33804.ll
+++ b/llvm/test/CodeGen/ARM/loopvectorize_pr33804.ll
@@ -64,7 +64,7 @@ for.end22.i.i: ; preds = %for.body14.i.i
; CHECK-LABEL: @cvCalcEMD3
; CHECK: vector.body
; CHECK: inttoptr <{{[0-9]+}} x i32>
-define void @cvCalcEMD3() local_unnamed_addr #0 personality ptr @__gxx_personality_v0 {
+define void @cvCalcEMD3(ptr %dst) local_unnamed_addr #0 personality ptr @__gxx_personality_v0 {
entry:
br label %for.body14.i.i
@@ -74,6 +74,10 @@ for.body14.i.i: ; preds = %for.body14.i.i, %en
%loadf = load float, ptr %arrayidx15.i.i1427, align 4
%next19.i.i = getelementptr inbounds %struct.CvNode1D, ptr undef, i32 %i.1424.i.i, i32 1
%loadp = load ptr, ptr %next19.i.i, align 4
+ %dst.ptr = getelementptr inbounds %struct.CvNode1D, ptr %dst, i32 %i.1424.i.i
+ %dst.ptr.1 = getelementptr inbounds %struct.CvNode1D, ptr %dst, i32 %i.1424.i.i, i32 1
+ store float %loadf, ptr %dst.ptr, align 4
+ store ptr %loadp, ptr %dst.ptr.1, align 4
%inc21.i.i = add nuw nsw i32 %i.1424.i.i, 1
%exitcond438.i.i = icmp eq i32 %inc21.i.i, 0
br i1 %exitcond438.i.i, label %for.end22.i.i, label %for.body14.i.i
@@ -87,7 +91,7 @@ for.end22.i.i: ; preds = %for.body14.i.i
; CHECK-LABEL: @cvCalcEMD3_2
; CHECK: vector.body
; CHECK: ptrtoint <{{[0-9]+}} x ptr>
-define void @cvCalcEMD3_2() local_unnamed_addr #0 personality ptr @__gxx_personality_v0 {
+define void @cvCalcEMD3_2(ptr %dst) local_unnamed_addr #0 personality ptr @__gxx_personality_v0 {
entry:
br label %for.body14.i.i
@@ -98,6 +102,10 @@ for.body14.i.i: ; preds = %for.body14.i.i, %en
%arrayidx15.i.i1427 = getelementptr inbounds %struct.CvNode1D2, ptr undef, i32 %i.1424.i.i
%val.i.i = getelementptr inbounds %struct.CvNode1D2, ptr %arrayidx15.i.i1427, i32 0, i32 1
%loadf = load float, ptr %val.i.i, align 4
+ %dst.ptr = getelementptr inbounds %struct.CvNode1D, ptr %dst, i32 %i.1424.i.i
+ %dst.ptr.1 = getelementptr inbounds %struct.CvNode1D, ptr %dst, i32 %i.1424.i.i, i32 1
+ store float %loadf, ptr %dst.ptr, align 4
+ store ptr %loadp, ptr %dst.ptr.1, align 4
%inc21.i.i = add nuw nsw i32 %i.1424.i.i, 1
%exitcond438.i.i = icmp eq i32 %inc21.i.i, 0
br i1 %exitcond438.i.i, label %for.end22.i.i, label %for.body14.i.i
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/loopvectorize_pr33804_double.ll b/llvm/test/Transforms/LoopVectorize/AArch64/loopvectorize_pr33804_double.ll
index 263da076fbb4e13..b7f4eaef9240894 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/loopvectorize_pr33804_double.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/loopvectorize_pr33804_double.ll
@@ -64,7 +64,7 @@ for.end22.i.i: ; preds = %for.body14.i.i
; CHECK-LABEL: @cvCalcEMD3
; CHECK: vector.body
; CHECK: inttoptr <{{[0-9]+}} x i64>
-define void @cvCalcEMD3() local_unnamed_addr #0 personality ptr @__gxx_personality_v0 {
+define void @cvCalcEMD3(ptr %dst) local_unnamed_addr #0 personality ptr @__gxx_personality_v0 {
entry:
br label %for.body14.i.i
@@ -74,6 +74,10 @@ for.body14.i.i: ; preds = %for.body14.i.i, %en
%load_d = load double, ptr %arrayidx15.i.i1427, align 4
%next19.i.i = getelementptr inbounds %struct.CvNode1D, ptr undef, i32 %i.1424.i.i, i32 1
%load_p = load ptr, ptr %next19.i.i, align 4
+ %dst.ptr = getelementptr inbounds %struct.CvNode1D, ptr %dst, i32 %i.1424.i.i
+ %dst.ptr.1 = getelementptr inbounds %struct.CvNode1D, ptr %dst, i32 %i.1424.i.i, i32 1
+ store double %load_d, ptr %dst.ptr, align 4
+ store ptr %load_p, ptr %dst.ptr.1, align 4
%inc21.i.i = add nuw nsw i32 %i.1424.i.i, 1
%exitcond438.i.i = icmp eq i32 %inc21.i.i, 0
br i1 %exitcond438.i.i, label %for.end22.i.i, label %for.body14.i.i
@@ -87,7 +91,7 @@ for.end22.i.i: ; preds = %for.body14.i.i
; CHECK-LABEL: @cvCalcEMD3_2
; CHECK: vector.body
; CHECK: ptrtoint <{{[0-9]+}} x ptr>
-define void @cvCalcEMD3_2() local_unnamed_addr #0 personality ptr @__gxx_personality_v0 {
+define void @cvCalcEMD3_2(ptr %dst) local_unnamed_addr #0 personality ptr @__gxx_personality_v0 {
entry:
br label %for.body14.i.i
@@ -98,6 +102,10 @@ for.body14.i.i: ; preds = %for.body14.i.i, %en
%arrayidx15.i.i1427 = getelementptr inbounds %struct.CvNode1D2, ptr undef, i32 %i.1424.i.i
%val.i.i = getelementptr inbounds %struct.CvNode1D2, ptr %arrayidx15.i.i1427, i32 0, i32 1
%load_d = load double, ptr %val.i.i, align 4
+ %dst.ptr = getelementptr inbounds %struct.CvNode1D, ptr %dst, i32 %i.1424.i.i
+ %dst.ptr.1 = getelementptr inbounds %struct.CvNode1D, ptr %dst, i32 %i.1424.i.i, i32 1
+ store double %load_d, ptr %dst.ptr, align 4
+ store ptr %load_p, ptr %dst.ptr.1, align 4
%inc21.i.i = add nuw nsw i32 %i.1424.i.i, 1
%exitcond438.i.i = icmp eq i32 %inc21.i.i, 0
br i1 %exitcond438.i.i, label %for.end22.i.i, label %for.body14.i.i
More information about the llvm-commits
mailing list