7d6e8f2a96
If we vectorize a e.g. store, we leave around a bunch of getelementptrs for the individual scalar stores which we removed. We can go ahead and delete them as well. This is purely for test output quality and readability. It should have no effect in any sane pipeline. Differential Revision: https://reviews.llvm.org/D122493
44 lines
2.3 KiB
LLVM
44 lines
2.3 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt -S -slp-vectorizer -mtriple=aarch64--linux-gnu -mcpu=generic < %s | FileCheck %s
|
|
; RUN: opt -S -slp-vectorizer -mtriple=aarch64-apple-ios -mcpu=cyclone < %s | FileCheck %s
|
|
; Currently disabled for a few subtargets (e.g. Kryo):
|
|
; RUN: opt -S -slp-vectorizer -mtriple=aarch64--linux-gnu -mcpu=kryo < %s | FileCheck --check-prefix=NO_SLP %s
|
|
; RUN: opt -S -slp-vectorizer -mtriple=aarch64--linux-gnu -mcpu=generic -slp-min-reg-size=128 < %s | FileCheck --check-prefix=NO_SLP %s
|
|
|
|
define void @f(float* %r, float* %w) {
|
|
; CHECK-LABEL: @f(
|
|
; CHECK-NEXT: [[R0:%.*]] = getelementptr inbounds float, float* [[R:%.*]], i64 0
|
|
; CHECK-NEXT: [[W0:%.*]] = getelementptr inbounds float, float* [[W:%.*]], i64 0
|
|
; CHECK-NEXT: [[TMP1:%.*]] = bitcast float* [[R0]] to <2 x float>*
|
|
; CHECK-NEXT: [[TMP2:%.*]] = load <2 x float>, <2 x float>* [[TMP1]], align 4
|
|
; CHECK-NEXT: [[TMP3:%.*]] = fadd <2 x float> [[TMP2]], [[TMP2]]
|
|
; CHECK-NEXT: [[TMP4:%.*]] = bitcast float* [[W0]] to <2 x float>*
|
|
; CHECK-NEXT: store <2 x float> [[TMP3]], <2 x float>* [[TMP4]], align 4
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
; NO_SLP-LABEL: @f(
|
|
; NO_SLP-NEXT: [[R0:%.*]] = getelementptr inbounds float, float* [[R:%.*]], i64 0
|
|
; NO_SLP-NEXT: [[R1:%.*]] = getelementptr inbounds float, float* [[R]], i64 1
|
|
; NO_SLP-NEXT: [[F0:%.*]] = load float, float* [[R0]], align 4
|
|
; NO_SLP-NEXT: [[F1:%.*]] = load float, float* [[R1]], align 4
|
|
; NO_SLP-NEXT: [[ADD0:%.*]] = fadd float [[F0]], [[F0]]
|
|
; NO_SLP-NEXT: [[ADD1:%.*]] = fadd float [[F1]], [[F1]]
|
|
; NO_SLP-NEXT: [[W0:%.*]] = getelementptr inbounds float, float* [[W:%.*]], i64 0
|
|
; NO_SLP-NEXT: [[W1:%.*]] = getelementptr inbounds float, float* [[W]], i64 1
|
|
; NO_SLP-NEXT: store float [[ADD0]], float* [[W0]], align 4
|
|
; NO_SLP-NEXT: store float [[ADD1]], float* [[W1]], align 4
|
|
; NO_SLP-NEXT: ret void
|
|
;
|
|
%r0 = getelementptr inbounds float, float* %r, i64 0
|
|
%r1 = getelementptr inbounds float, float* %r, i64 1
|
|
%f0 = load float, float* %r0
|
|
%f1 = load float, float* %r1
|
|
%add0 = fadd float %f0, %f0
|
|
%add1 = fadd float %f1, %f1
|
|
%w0 = getelementptr inbounds float, float* %w, i64 0
|
|
%w1 = getelementptr inbounds float, float* %w, i64 1
|
|
store float %add0, float* %w0
|
|
store float %add1, float* %w1
|
|
ret void
|
|
}
|