b10ecfa914
Currently SLP vectorizer tries to keep only GEPs as scalar, if they are vectorized but used externally. Same approach can be used for all scalar values. This patch tries to keep original scalars if all its operands remain scalar or externally used, the cost of the original scalar is lower than the cost of the extractelement instruction, or if the number of externally used scalars in the same entry is power of 2. Last criterion allows better revectorization for multiply used scalars. Reviewers: RKSimon Reviewed By: RKSimon Pull Request: https://github.com/llvm/llvm-project/pull/100904
50 lines
2.8 KiB
LLVM
50 lines
2.8 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
|
|
; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-linux-gnu < %s | FileCheck %s
|
|
|
|
define void @test(ptr noalias %arg, ptr noalias %arg1, ptr %arg2) {
|
|
; CHECK-LABEL: define void @test(
|
|
; CHECK-SAME: ptr noalias [[ARG:%.*]], ptr noalias [[ARG1:%.*]], ptr [[ARG2:%.*]]) {
|
|
; CHECK-NEXT: bb:
|
|
; CHECK-NEXT: [[TMP_I_I:%.*]] = getelementptr i8, ptr [[ARG1]], i64 24
|
|
; CHECK-NEXT: [[TMP_I_I4:%.*]] = getelementptr i8, ptr [[ARG]], i64 24
|
|
; CHECK-NEXT: [[TMP_I_I13:%.*]] = getelementptr i8, ptr [[ARG1]], i64 28
|
|
; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr [[TMP_I_I13]], align 4
|
|
; CHECK-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[TMP_I_I]], align 8
|
|
; CHECK-NEXT: store float [[TMP1]], ptr [[ARG2]], align 4
|
|
; CHECK-NEXT: [[TMP2:%.*]] = fcmp olt <4 x float> [[TMP0]], zeroinitializer
|
|
; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
|
|
; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
|
|
; CHECK-NEXT: [[TMP5:%.*]] = select <4 x i1> [[TMP2]], <4 x float> [[TMP3]], <4 x float> [[TMP4]]
|
|
; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <4 x float> [[TMP5]], <4 x float> poison, <4 x i32> <i32 2, i32 0, i32 3, i32 1>
|
|
; CHECK-NEXT: store <4 x float> [[TMP6]], ptr [[TMP_I_I4]], align 8
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
bb:
|
|
%tmp.i.i = getelementptr i8, ptr %arg1, i64 24
|
|
%tmp1.i.i = load float, ptr %tmp.i.i, align 8
|
|
%tmp.i.i2 = getelementptr i8, ptr %arg1, i64 32
|
|
%tmp1.i.i3 = load float, ptr %tmp.i.i2, align 8
|
|
%tmp1.i.i.i = fcmp olt float %tmp1.i.i3, 0.000000e+00
|
|
%tmp9 = select i1 %tmp1.i.i.i, float %tmp1.i.i3, float %tmp1.i.i
|
|
%tmp.i.i4 = getelementptr i8, ptr %arg, i64 24
|
|
store float %tmp9, ptr %tmp.i.i4, align 8
|
|
%tmp1.i.i.i10 = fcmp olt float %tmp1.i.i, 0.000000e+00
|
|
%tmp13 = select i1 %tmp1.i.i.i10, float %tmp1.i.i3, float %tmp1.i.i
|
|
%tmp.i.i12 = getelementptr i8, ptr %arg, i64 28
|
|
store float %tmp13, ptr %tmp.i.i12, align 4
|
|
%tmp.i.i13 = getelementptr i8, ptr %arg1, i64 28
|
|
%tmp1.i.i14 = load float, ptr %tmp.i.i13, align 4
|
|
%tmp.i.i15 = getelementptr i8, ptr %arg1, i64 36
|
|
%tmp1.i.i16 = load float, ptr %tmp.i.i15, align 4
|
|
%tmp1.i.i.i18 = fcmp olt float %tmp1.i.i16, 0.000000e+00
|
|
%tmp17 = select i1 %tmp1.i.i.i18, float %tmp1.i.i16, float %tmp1.i.i14
|
|
%tmp.i.i20 = getelementptr i8, ptr %arg, i64 32
|
|
store float %tmp17, ptr %tmp.i.i20, align 8
|
|
store float %tmp1.i.i14, ptr %arg2, align 4
|
|
%tmp1.i.i.i24 = fcmp olt float %tmp1.i.i14, 0.000000e+00
|
|
%tmp20 = select i1 %tmp1.i.i.i24, float %tmp1.i.i16, float %tmp1.i.i14
|
|
%tmp.i.i26 = getelementptr i8, ptr %arg, i64 36
|
|
store float %tmp20, ptr %tmp.i.i26, align 4
|
|
ret void
|
|
}
|