diff --git a/llvm/test/Transforms/SLPVectorizer/slp-max-reg-size.ll b/llvm/test/Transforms/SLPVectorizer/slp-max-reg-size.ll new file mode 100644 --- /dev/null +++ b/llvm/test/Transforms/SLPVectorizer/slp-max-reg-size.ll @@ -0,0 +1,62 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; RUN: opt < %s -slp-vectorizer -slp-max-vf=8 -slp-min-reg-size=32 -slp-max-reg-size=160 -S | FileCheck -check-prefix CHECK-VF8-160 %s +; RUN: opt < %s -slp-vectorizer -slp-max-vf=4 -slp-min-reg-size=32 -slp-max-reg-size=160 -S | FileCheck -check-prefix CHECK-VF4-160 %s +; RUN: opt < %s -slp-vectorizer -slp-max-vf=2 -slp-min-reg-size=32 -slp-max-reg-size=160 -S | FileCheck -check-prefix CHECK-VF2-160 %s +; RUN: opt < %s -slp-vectorizer -slp-max-vf=8 -slp-min-reg-size=32 -slp-max-reg-size=128 -S | FileCheck -check-prefix CHECK-VF8-128 %s +; RUN: opt < %s -slp-vectorizer -slp-max-vf=4 -slp-min-reg-size=32 -slp-max-reg-size=128 -S | FileCheck -check-prefix CHECK-VF4-128 %s +; RUN: opt < %s -slp-vectorizer -slp-max-vf=2 -slp-min-reg-size=32 -slp-max-reg-size=128 -S | FileCheck -check-prefix CHECK-VF2-128 %s +; RUN: opt < %s -slp-vectorizer -slp-max-vf=8 -slp-min-reg-size=32 -slp-max-reg-size=256 -S | FileCheck -check-prefix CHECK-VF8-256 %s +; RUN: opt < %s -slp-vectorizer -slp-max-vf=2 -slp-min-reg-size=128 -slp-max-reg-size=128 -S | FileCheck -check-prefix CHECK-VF2-128-128 %s + +; This test case is checking if max register size (and max VF) is honored when +; vecorizing store chains in SLPVectorizer. + +@X = external global [8 x i32], align 1 + +define void @foo() { +; CHECK-VF8-160-LABEL: @foo( +; CHECK-VF8-160-NEXT: store <8 x i32> , <8 x i32>* bitcast ([8 x i32]* @X to <8 x i32>*), align 1 +; CHECK-VF8-160-NEXT: ret void +; +; CHECK-VF4-160-LABEL: @foo( +; CHECK-VF4-160-NEXT: store <8 x i32> , <8 x i32>* bitcast ([8 x i32]* @X to <8 x i32>*), align 1 +; CHECK-VF4-160-NEXT: ret void +; +; CHECK-VF2-160-LABEL: @foo( +; CHECK-VF2-160-NEXT: store <8 x i32> , <8 x i32>* bitcast ([8 x i32]* @X to <8 x i32>*), align 1 +; CHECK-VF2-160-NEXT: ret void +; +; CHECK-VF8-128-LABEL: @foo( +; CHECK-VF8-128-NEXT: store <4 x i32> , <4 x i32>* bitcast ([8 x i32]* @X to <4 x i32>*), align 1 +; CHECK-VF8-128-NEXT: store <4 x i32> , <4 x i32>* bitcast (i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 4) to <4 x i32>*), align 1 +; CHECK-VF8-128-NEXT: ret void +; +; CHECK-VF4-128-LABEL: @foo( +; CHECK-VF4-128-NEXT: store <4 x i32> , <4 x i32>* bitcast ([8 x i32]* @X to <4 x i32>*), align 1 +; CHECK-VF4-128-NEXT: store <4 x i32> , <4 x i32>* bitcast (i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 4) to <4 x i32>*), align 1 +; CHECK-VF4-128-NEXT: ret void +; +; CHECK-VF2-128-LABEL: @foo( +; CHECK-VF2-128-NEXT: store <4 x i32> , <4 x i32>* bitcast ([8 x i32]* @X to <4 x i32>*), align 1 +; CHECK-VF2-128-NEXT: store <4 x i32> , <4 x i32>* bitcast (i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 4) to <4 x i32>*), align 1 +; CHECK-VF2-128-NEXT: ret void +; +; CHECK-VF8-256-LABEL: @foo( +; CHECK-VF8-256-NEXT: store <8 x i32> , <8 x i32>* bitcast ([8 x i32]* @X to <8 x i32>*), align 1 +; CHECK-VF8-256-NEXT: ret void +; +; CHECK-VF2-128-128-LABEL: @foo( +; CHECK-VF2-128-128-NEXT: store <4 x i32> , <4 x i32>* bitcast ([8 x i32]* @X to <4 x i32>*), align 1 +; CHECK-VF2-128-128-NEXT: store <4 x i32> , <4 x i32>* bitcast (i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 4) to <4 x i32>*), align 1 +; CHECK-VF2-128-128-NEXT: ret void +; + store i32 1, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 0), align 1 + store i32 2, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 1), align 1 + store i32 3, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 2), align 1 + store i32 4, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 3), align 1 + store i32 5, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 4), align 1 + store i32 6, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 5), align 1 + store i32 7, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 6), align 1 + store i32 8, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @X, i16 0, i16 7), align 1 + ret void +}