60 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: %if x86-registered-target %{ opt -S -passes=slp-vectorizer -mtriple x86_64-unknown-linux-gnu < %s | FileCheck %s --check-prefix=X86 %}3; RUN: %if aarch64-registered-target %{ opt -S -passes=slp-vectorizer -mtriple aarch64-unknown-linux-gnu < %s | FileCheck %s --check-prefix=AARCH86 %}4 5define <4 x double> @test(ptr %p2, double %i1754, double %i1781, double %i1778) {6; X86-LABEL: @test(7; X86-NEXT: entry:8; X86-NEXT: [[I1771:%.*]] = getelementptr inbounds double, ptr [[P2:%.*]], i64 549; X86-NEXT: [[I1772:%.*]] = load double, ptr [[I1771]], align 810; X86-NEXT: [[I1795:%.*]] = getelementptr inbounds double, ptr [[P2]], i64 5511; X86-NEXT: [[I1796:%.*]] = load double, ptr [[I1795]], align 812; X86-NEXT: [[I1797:%.*]] = fmul fast double [[I1796]], [[I1781:%.*]]13; X86-NEXT: [[TMP0:%.*]] = insertelement <4 x double> poison, double [[I1754:%.*]], i32 014; X86-NEXT: [[TMP1:%.*]] = insertelement <4 x double> [[TMP0]], double [[I1778:%.*]], i32 115; X86-NEXT: [[TMP2:%.*]] = insertelement <4 x double> [[TMP1]], double [[I1781]], i32 216; X86-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP2]], double [[I1772]], i32 317; X86-NEXT: [[TMP4:%.*]] = shufflevector <4 x double> [[TMP3]], <4 x double> poison, <4 x i32> zeroinitializer18; X86-NEXT: [[TMP5:%.*]] = fmul fast <4 x double> [[TMP3]], [[TMP4]]19; X86-NEXT: [[TMP6:%.*]] = insertelement <4 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double poison>, double [[I1797]], i32 320; X86-NEXT: [[TMP7:%.*]] = fadd fast <4 x double> [[TMP5]], [[TMP6]]21; X86-NEXT: ret <4 x double> [[TMP7]]22;23; AARCH86-LABEL: @test(24; AARCH86-NEXT: entry:25; AARCH86-NEXT: [[I1771:%.*]] = getelementptr inbounds double, ptr [[P2:%.*]], i64 5426; AARCH86-NEXT: [[I1772:%.*]] = load double, ptr [[I1771]], align 827; AARCH86-NEXT: [[I1795:%.*]] = getelementptr inbounds double, ptr [[P2]], i64 5528; AARCH86-NEXT: [[I1796:%.*]] = load double, ptr [[I1795]], align 829; AARCH86-NEXT: [[I1797:%.*]] = fmul fast double [[I1796]], [[I1781:%.*]]30; AARCH86-NEXT: [[TMP0:%.*]] = insertelement <4 x double> poison, double [[I1754:%.*]], i32 031; AARCH86-NEXT: [[TMP1:%.*]] = insertelement <4 x double> [[TMP0]], double [[I1778:%.*]], i32 132; AARCH86-NEXT: [[TMP2:%.*]] = insertelement <4 x double> [[TMP1]], double [[I1781]], i32 233; AARCH86-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP2]], double [[I1772]], i32 334; AARCH86-NEXT: [[TMP4:%.*]] = shufflevector <4 x double> [[TMP3]], <4 x double> poison, <4 x i32> zeroinitializer35; AARCH86-NEXT: [[TMP5:%.*]] = fmul fast <4 x double> [[TMP3]], [[TMP4]]36; AARCH86-NEXT: [[TMP6:%.*]] = insertelement <4 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double poison>, double [[I1797]], i32 337; AARCH86-NEXT: [[I1994:%.*]] = fadd fast <4 x double> [[TMP5]], [[TMP6]]38; AARCH86-NEXT: ret <4 x double> [[I1994]]39;40entry:41 %i1771 = getelementptr inbounds double, ptr %p2, i64 5442 %i1772 = load double, ptr %i1771, align 843 %i1773 = fmul fast double %i1772, %i175444 %i1782 = fmul fast double %i1754, %i175445 %i1783 = fadd fast double %i1782, 1.000000e+0046 %i1787 = fmul fast double %i1778, %i175447 %i1788 = fadd fast double %i1787, 1.000000e+0048 %i1792 = fmul fast double %i1754, %i178149 %i1793 = fadd fast double %i1792, 1.000000e+0050 %i1795 = getelementptr inbounds double, ptr %p2, i64 5551 %i1796 = load double, ptr %i1795, align 852 %i1797 = fmul fast double %i1796, %i178153 %i1798 = fadd fast double %i1773, %i179754 %i1976 = insertelement <4 x double> zeroinitializer, double %i1783, i64 055 %i1982 = insertelement <4 x double> %i1976, double %i1788, i64 156 %i1988 = insertelement <4 x double> %i1982, double %i1793, i64 257 %i1994 = insertelement <4 x double> %i1988, double %i1798, i64 358 ret <4 x double> %i199459}60