110 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 22; RUN: opt -S -passes=slp-vectorizer < %s -mtriple=riscv64-unknown-linux -mattr=+v | FileCheck %s3 4define i32 @sum_of_abs_stride_2(ptr noalias %a, ptr noalias %b) {5; CHECK-LABEL: define i32 @sum_of_abs_stride_26; CHECK-SAME: (ptr noalias [[A:%.*]], ptr noalias [[B:%.*]]) #[[ATTR0:[0-9]+]] {7; CHECK-NEXT: entry:8; CHECK-NEXT: [[TMP0:%.*]] = call <15 x i8> @llvm.masked.load.v15i8.p0(ptr align 1 [[A]], <15 x i1> <i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true>, <15 x i8> poison)9; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <15 x i8> [[TMP0]], <15 x i8> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>10; CHECK-NEXT: [[TMP2:%.*]] = call <8 x i8> @llvm.abs.v8i8(<8 x i8> [[TMP1]], i1 false)11; CHECK-NEXT: [[TMP3:%.*]] = sext <8 x i8> [[TMP2]] to <8 x i32>12; CHECK-NEXT: [[TMP4:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP3]])13; CHECK-NEXT: ret i32 [[TMP4]]14;15entry:16 %0 = load i8, ptr %a, align 117 %spec.select.i = tail call i8 @llvm.abs.i8(i8 %0, i1 false)18 %conv = sext i8 %spec.select.i to i3219 %arrayidx.1 = getelementptr inbounds i8, ptr %a, i64 220 %1 = load i8, ptr %arrayidx.1, align 121 %spec.select.i.1 = tail call i8 @llvm.abs.i8(i8 %1, i1 false)22 %conv.1 = sext i8 %spec.select.i.1 to i3223 %add.1 = add nsw i32 %conv, %conv.124 %arrayidx.2 = getelementptr inbounds i8, ptr %a, i64 425 %2 = load i8, ptr %arrayidx.2, align 126 %spec.select.i.2 = tail call i8 @llvm.abs.i8(i8 %2, i1 false)27 %conv.2 = sext i8 %spec.select.i.2 to i3228 %add.2 = add nsw i32 %add.1, %conv.229 %arrayidx.3 = getelementptr inbounds i8, ptr %a, i64 630 %3 = load i8, ptr %arrayidx.3, align 131 %spec.select.i.3 = tail call i8 @llvm.abs.i8(i8 %3, i1 false)32 %conv.3 = sext i8 %spec.select.i.3 to i3233 %add.3 = add nsw i32 %add.2, %conv.334 %arrayidx.4 = getelementptr inbounds i8, ptr %a, i64 835 %4 = load i8, ptr %arrayidx.4, align 136 %spec.select.i.4 = tail call i8 @llvm.abs.i8(i8 %4, i1 false)37 %conv.4 = sext i8 %spec.select.i.4 to i3238 %add.4 = add nsw i32 %add.3, %conv.439 %arrayidx.5 = getelementptr inbounds i8, ptr %a, i64 1040 %5 = load i8, ptr %arrayidx.5, align 141 %spec.select.i.5 = tail call i8 @llvm.abs.i8(i8 %5, i1 false)42 %conv.5 = sext i8 %spec.select.i.5 to i3243 %add.5 = add nsw i32 %add.4, %conv.544 %arrayidx.6 = getelementptr inbounds i8, ptr %a, i64 1245 %6 = load i8, ptr %arrayidx.6, align 146 %spec.select.i.6 = tail call i8 @llvm.abs.i8(i8 %6, i1 false)47 %conv.6 = sext i8 %spec.select.i.6 to i3248 %add.6 = add nsw i32 %add.5, %conv.649 %arrayidx.7 = getelementptr inbounds i8, ptr %a, i64 1450 %7 = load i8, ptr %arrayidx.7, align 151 %spec.select.i.7 = tail call i8 @llvm.abs.i8(i8 %7, i1 false)52 %conv.7 = sext i8 %spec.select.i.7 to i3253 %add.7 = add nsw i32 %add.6, %conv.754 ret i32 %add.755}56 57define i32 @sum_of_abs_stride_3(ptr noalias %a, ptr noalias %b) {58; CHECK-LABEL: define i32 @sum_of_abs_stride_359; CHECK-SAME: (ptr noalias [[A:%.*]], ptr noalias [[B:%.*]]) #[[ATTR0]] {60; CHECK-NEXT: entry:61; CHECK-NEXT: [[TMP0:%.*]] = call <22 x i8> @llvm.masked.load.v22i8.p0(ptr align 1 [[A]], <22 x i1> <i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true>, <22 x i8> poison)62; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <22 x i8> [[TMP0]], <22 x i8> poison, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>63; CHECK-NEXT: [[TMP2:%.*]] = call <8 x i8> @llvm.abs.v8i8(<8 x i8> [[TMP1]], i1 false)64; CHECK-NEXT: [[TMP3:%.*]] = sext <8 x i8> [[TMP2]] to <8 x i32>65; CHECK-NEXT: [[TMP4:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP3]])66; CHECK-NEXT: ret i32 [[TMP4]]67;68entry:69 %0 = load i8, ptr %a, align 170 %spec.select.i = tail call i8 @llvm.abs.i8(i8 %0, i1 false)71 %conv = sext i8 %spec.select.i to i3272 %arrayidx.1 = getelementptr inbounds i8, ptr %a, i64 373 %1 = load i8, ptr %arrayidx.1, align 174 %spec.select.i.1 = tail call i8 @llvm.abs.i8(i8 %1, i1 false)75 %conv.1 = sext i8 %spec.select.i.1 to i3276 %add.1 = add nsw i32 %conv, %conv.177 %arrayidx.2 = getelementptr inbounds i8, ptr %a, i64 678 %2 = load i8, ptr %arrayidx.2, align 179 %spec.select.i.2 = tail call i8 @llvm.abs.i8(i8 %2, i1 false)80 %conv.2 = sext i8 %spec.select.i.2 to i3281 %add.2 = add nsw i32 %add.1, %conv.282 %arrayidx.3 = getelementptr inbounds i8, ptr %a, i64 983 %3 = load i8, ptr %arrayidx.3, align 184 %spec.select.i.3 = tail call i8 @llvm.abs.i8(i8 %3, i1 false)85 %conv.3 = sext i8 %spec.select.i.3 to i3286 %add.3 = add nsw i32 %add.2, %conv.387 %arrayidx.4 = getelementptr inbounds i8, ptr %a, i64 1288 %4 = load i8, ptr %arrayidx.4, align 189 %spec.select.i.4 = tail call i8 @llvm.abs.i8(i8 %4, i1 false)90 %conv.4 = sext i8 %spec.select.i.4 to i3291 %add.4 = add nsw i32 %add.3, %conv.492 %arrayidx.5 = getelementptr inbounds i8, ptr %a, i64 1593 %5 = load i8, ptr %arrayidx.5, align 194 %spec.select.i.5 = tail call i8 @llvm.abs.i8(i8 %5, i1 false)95 %conv.5 = sext i8 %spec.select.i.5 to i3296 %add.5 = add nsw i32 %add.4, %conv.597 %arrayidx.6 = getelementptr inbounds i8, ptr %a, i64 1898 %6 = load i8, ptr %arrayidx.6, align 199 %spec.select.i.6 = tail call i8 @llvm.abs.i8(i8 %6, i1 false)100 %conv.6 = sext i8 %spec.select.i.6 to i32101 %add.6 = add nsw i32 %add.5, %conv.6102 %arrayidx.7 = getelementptr inbounds i8, ptr %a, i64 21103 %7 = load i8, ptr %arrayidx.7, align 1104 %spec.select.i.7 = tail call i8 @llvm.abs.i8(i8 %7, i1 false)105 %conv.7 = sext i8 %spec.select.i.7 to i32106 %add.7 = add nsw i32 %add.6, %conv.7107 ret i32 %add.7108}109declare i8 @llvm.abs.i8(i8, i1 immarg)110