54 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -mtriple=x86_64 -passes=slp-vectorizer -S -mcpu=skylake-avx512 | FileCheck %s3 4declare void @llvm.masked.scatter.v2f64.v2p0(<2 x double>, <2 x ptr>, i32 immarg, <2 x i1>)5 6define void @rdx_feeds_single_insert(<2 x double> %v, ptr nocapture readonly %arg, ptr nocapture readonly %arg1, ptr nocapture %arg2) {7; CHECK-LABEL: @rdx_feeds_single_insert(8; CHECK-NEXT: entry:9; CHECK-NEXT: [[TMP0:%.*]] = load <8 x double>, ptr [[ARG1:%.*]], align 810; CHECK-NEXT: [[TMP1:%.*]] = fmul fast <8 x double> [[TMP0]], <double 1.000000e+01, double 1.100000e+01, double 1.200000e+01, double 1.300000e+01, double 1.400000e+01, double 1.500000e+01, double 1.600000e+01, double 1.700000e+01>11; CHECK-NEXT: [[TMP2:%.*]] = call fast double @llvm.vector.reduce.fadd.v8f64(double 0.000000e+00, <8 x double> [[TMP1]])12; CHECK-NEXT: [[I:%.*]] = insertelement <2 x double> [[V:%.*]], double [[TMP2]], i64 113; CHECK-NEXT: [[P:%.*]] = getelementptr inbounds double, ptr [[ARG2:%.*]], <2 x i64> <i64 0, i64 16>14; CHECK-NEXT: call void @llvm.masked.scatter.v2f64.v2p0(<2 x double> [[I]], <2 x ptr> align 8 [[P]], <2 x i1> splat (i1 true))15; CHECK-NEXT: ret void16;17entry:18 %ld0.0 = load double, ptr %arg1, align 819 %mul1.0 = fmul fast double %ld0.0, 10.020 %gep0.1 = getelementptr inbounds double, ptr %arg1, i64 121 %ld0.1 = load double, ptr %gep0.1, align 822 %mul1.1 = fmul fast double %ld0.1, 11.023 %rdx1.0 = fadd fast double %mul1.0, %mul1.124 %gep0.2 = getelementptr inbounds double, ptr %arg1, i64 225 %ld0.2 = load double, ptr %gep0.2, align 826 %mul1.2 = fmul fast double %ld0.2, 12.027 %rdx1.1 = fadd fast double %rdx1.0, %mul1.228 %gep0.3 = getelementptr inbounds double, ptr %arg1, i64 329 %ld0.3 = load double, ptr %gep0.3, align 830 %mul1.3 = fmul fast double %ld0.3, 13.031 %rdx1.2 = fadd fast double %rdx1.1, %mul1.332 %gep0.4 = getelementptr inbounds double, ptr %arg1, i64 433 %ld0.4 = load double, ptr %gep0.4, align 834 %mul1.4 = fmul fast double %ld0.4, 14.035 %rdx1.3 = fadd fast double %rdx1.2, %mul1.436 %gep0.5 = getelementptr inbounds double, ptr %arg1, i64 537 %ld0.5 = load double, ptr %gep0.5, align 838 %mul1.5 = fmul fast double %ld0.5, 15.039 %rdx1.4 = fadd fast double %rdx1.3, %mul1.540 %gep0.6 = getelementptr inbounds double, ptr %arg1, i64 641 %ld0.6 = load double, ptr %gep0.6, align 842 %mul1.6 = fmul fast double %ld0.6, 16.043 %rdx1.5 = fadd fast double %rdx1.4, %mul1.644 %gep0.7 = getelementptr inbounds double, ptr %arg1, i64 745 %ld0.7 = load double, ptr %gep0.7, align 846 %mul1.7 = fmul fast double %ld0.7, 17.047 %rdx1 = fadd fast double %rdx1.5, %mul1.748 %i = insertelement <2 x double> %v, double %rdx1, i64 149 %p = getelementptr inbounds double, ptr %arg2, <2 x i64> <i64 0, i64 16>50 call void @llvm.masked.scatter.v2f64.v2p0(<2 x double> %i, <2 x ptr> %p, i32 8, <2 x i1> <i1 true, i1 true>)51 ret void52}53 54