xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/RISCV/scatter-vectorize-reversed.ll (revision 38fffa630ee80163dc65e759392ad29798905679)
130d484faSAlexey Bataev; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
230d484faSAlexey Bataev; RUN: opt -S --passes=slp-vectorizer -mtriple=riscv64-unknown-linux-gnu -mattr=+v -slp-threshold=-11 < %s | FileCheck %s
330d484faSAlexey Bataev
430d484faSAlexey Bataevdefine <4 x i32> @test(<2 x i64> %v, ptr %p) {
530d484faSAlexey Bataev; CHECK-LABEL: define <4 x i32> @test(
630d484faSAlexey Bataev; CHECK-SAME: <2 x i64> [[V:%.*]], ptr [[P:%.*]]) #[[ATTR0:[0-9]+]] {
730d484faSAlexey Bataev; CHECK-NEXT:  entry:
8a65a5febSAlexey Bataev; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i64> [[V]], <2 x i64> poison, <2 x i32> <i32 1, i32 0>
930d484faSAlexey Bataev; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x ptr> poison, ptr [[P]], i32 0
1030d484faSAlexey Bataev; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <2 x ptr> [[TMP0]], <2 x ptr> poison, <2 x i32> zeroinitializer
11a65a5febSAlexey Bataev; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i16, <2 x ptr> [[TMP1]], <2 x i64> [[TMP4]]
12*38fffa63SPaul Walker; CHECK-NEXT:    [[TMP3:%.*]] = call <2 x i16> @llvm.masked.gather.v2i16.v2p0(<2 x ptr> [[TMP2]], i32 2, <2 x i1> splat (i1 true), <2 x i16> poison)
13a65a5febSAlexey Bataev; CHECK-NEXT:    [[TMP7:%.*]] = zext <2 x i16> [[TMP3]] to <2 x i32>
1430d484faSAlexey Bataev; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x i32> [[TMP7]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
1530d484faSAlexey Bataev; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <4 x i32> zeroinitializer, <4 x i32> [[TMP6]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
1630d484faSAlexey Bataev; CHECK-NEXT:    ret <4 x i32> [[TMP5]]
1730d484faSAlexey Bataev;
1830d484faSAlexey Bataeventry:
1930d484faSAlexey Bataev  %0 = extractelement <2 x i64> %v, i32 1
2030d484faSAlexey Bataev  %arrayidx127.2 = getelementptr i16, ptr %p, i64 %0
2130d484faSAlexey Bataev  %1 = load i16, ptr %arrayidx127.2, align 2
2230d484faSAlexey Bataev  %conv128.2 = zext i16 %1 to i32
2330d484faSAlexey Bataev  %2 = extractelement <2 x i64> %v, i32 0
2430d484faSAlexey Bataev  %arrayidx127.3 = getelementptr i16, ptr %p, i64 %2
2530d484faSAlexey Bataev  %3 = load i16, ptr %arrayidx127.3, align 2
2630d484faSAlexey Bataev  %conv128.3 = zext i16 %3 to i32
2730d484faSAlexey Bataev  %4 = insertelement <4 x i32> zeroinitializer, i32 %conv128.2, i32 0
2830d484faSAlexey Bataev  %5 = insertelement <4 x i32> %4, i32 %conv128.3, i32 1
2930d484faSAlexey Bataev  ret <4 x i32> %5
3030d484faSAlexey Bataev}
31