bddfbe748b
Another step towards cleaning up shuffles that have been split, often across bitcasts between SSE intrinsic. Strip shuffles entirely if we fold to an identity shuffle.
30 lines
1.7 KiB
LLVM
30 lines
1.7 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt < %s -passes=vector-combine -S -mtriple=x86_64-- -mattr=avx | FileCheck %s
|
|
; RUN: opt < %s -passes=vector-combine -S -mtriple=x86_64-- -mattr=avx2 | FileCheck %s
|
|
|
|
define <4 x i64> @PR67803(<8 x i32> %x, <8 x i32> %y, <8 x float> %a, <8 x float> %b) {
|
|
; CHECK-LABEL: @PR67803(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <8 x i32> [[X:%.*]], [[Y:%.*]]
|
|
; CHECK-NEXT: [[TMP1:%.*]] = sext <8 x i1> [[CMP]] to <8 x i32>
|
|
; CHECK-NEXT: [[CONCAT:%.*]] = bitcast <8 x i32> [[TMP1]] to <4 x i64>
|
|
; CHECK-NEXT: [[MASK:%.*]] = bitcast <4 x i64> [[CONCAT]] to <8 x float>
|
|
; CHECK-NEXT: [[SEL:%.*]] = tail call noundef <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float> [[A:%.*]], <8 x float> [[B:%.*]], <8 x float> [[MASK]])
|
|
; CHECK-NEXT: [[RES:%.*]] = bitcast <8 x float> [[SEL]] to <4 x i64>
|
|
; CHECK-NEXT: ret <4 x i64> [[RES]]
|
|
;
|
|
entry:
|
|
%cmp = icmp sgt <8 x i32> %x, %y
|
|
%cmp.lo = shufflevector <8 x i1> %cmp, <8 x i1> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
|
|
%cmp.hi = shufflevector <8 x i1> %cmp, <8 x i1> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
|
|
%sext.lo = sext <4 x i1> %cmp.lo to <4 x i32>
|
|
%sext.hi = sext <4 x i1> %cmp.hi to <4 x i32>
|
|
%bitcast.lo = bitcast <4 x i32> %sext.lo to <2 x i64>
|
|
%bitcast.hi = bitcast <4 x i32> %sext.hi to <2 x i64>
|
|
%concat = shufflevector <2 x i64> %bitcast.lo, <2 x i64> %bitcast.hi, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
|
|
%mask = bitcast <4 x i64> %concat to <8 x float>
|
|
%sel = tail call noundef <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %mask)
|
|
%res = bitcast <8 x float> %sel to <4 x i64>
|
|
ret <4 x i64> %res
|
|
}
|