74 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 52; RUN: opt -S --passes=slp-vectorizer -mtriple=riscv64-unknown-linux-gnu -mattr=+v -slp-threshold=-1000 %s | FileCheck %s3 4define void @test(ptr %mdct_forward_x) {5; CHECK-LABEL: define void @test(6; CHECK-SAME: ptr [[MDCT_FORWARD_X:%.*]]) #[[ATTR0:[0-9]+]] {7; CHECK-NEXT: [[ENTRY:.*:]]8; CHECK-NEXT: br label %[[FOR_COND:.*]]9; CHECK: [[FOR_COND]]:10; CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[MDCT_FORWARD_X]], align 811; CHECK-NEXT: [[ARRAYIDX2_I_I:%.*]] = getelementptr i8, ptr [[TMP0]], i64 3212; CHECK-NEXT: [[ARRAYIDX5_I_I:%.*]] = getelementptr i8, ptr [[TMP0]], i64 4013; CHECK-NEXT: [[ADD_PTR_I:%.*]] = getelementptr i8, ptr [[TMP0]], i64 2414; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x ptr> poison, ptr [[TMP0]], i32 015; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x ptr> [[TMP1]], <4 x ptr> poison, <4 x i32> zeroinitializer16; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, <4 x ptr> [[TMP2]], <4 x i64> <i64 28, i64 36, i64 24, i64 28>17; CHECK-NEXT: [[TMP5:%.*]] = call <3 x float> @llvm.masked.load.v3f32.p0(ptr align 4 [[ADD_PTR_I]], <3 x i1> <i1 true, i1 false, i1 true>, <3 x float> poison)18; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <3 x float> [[TMP5]], <3 x float> poison, <2 x i32> <i32 2, i32 0>19; CHECK-NEXT: [[TMP6:%.*]] = call <3 x float> @llvm.masked.load.v3f32.p0(ptr align 4 [[ARRAYIDX5_I_I]], <3 x i1> <i1 true, i1 false, i1 true>, <3 x float> poison)20; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <3 x float> [[TMP6]], <3 x float> poison, <2 x i32> <i32 2, i32 0>21; CHECK-NEXT: [[TMP8:%.*]] = call <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr> align 4 [[TMP3]], <4 x i1> splat (i1 true), <4 x float> poison)22; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <3 x float> [[TMP6]], <3 x float> poison, <4 x i32> <i32 2, i32 0, i32 2, i32 2>23; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>24; CHECK-NEXT: [[TMP22:%.*]] = shufflevector <3 x float> [[TMP5]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>25; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <4 x float> <float poison, float poison, float 0.000000e+00, float poison>, <4 x float> [[TMP22]], <4 x i32> <i32 poison, i32 poison, i32 2, i32 6>26; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <4 x float> [[TMP11]], <4 x float> [[TMP10]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>27; CHECK-NEXT: [[TMP13:%.*]] = fsub <4 x float> [[TMP9]], [[TMP12]]28; CHECK-NEXT: [[TMP14:%.*]] = fadd <4 x float> [[TMP9]], [[TMP12]]29; CHECK-NEXT: [[TMP15:%.*]] = shufflevector <4 x float> [[TMP13]], <4 x float> [[TMP14]], <4 x i32> <i32 0, i32 1, i32 6, i32 3>30; CHECK-NEXT: [[TMP16:%.*]] = fsub <4 x float> zeroinitializer, [[TMP8]]31; CHECK-NEXT: [[TMP17:%.*]] = fadd <4 x float> zeroinitializer, [[TMP8]]32; CHECK-NEXT: [[TMP18:%.*]] = shufflevector <4 x float> [[TMP16]], <4 x float> [[TMP17]], <4 x i32> <i32 0, i32 1, i32 6, i32 3>33; CHECK-NEXT: store float 0.000000e+00, ptr [[ADD_PTR_I]], align 434; CHECK-NEXT: [[TMP19:%.*]] = fsub <4 x float> [[TMP15]], [[TMP18]]35; CHECK-NEXT: [[TMP20:%.*]] = fadd <4 x float> [[TMP15]], [[TMP18]]36; CHECK-NEXT: [[TMP21:%.*]] = shufflevector <4 x float> [[TMP19]], <4 x float> [[TMP20]], <4 x i32> <i32 0, i32 5, i32 2, i32 3>37; CHECK-NEXT: store <4 x float> [[TMP21]], ptr [[ARRAYIDX2_I_I]], align 438; CHECK-NEXT: br label %[[FOR_COND]]39;40entry:41 br label %for.cond42 43for.cond:44 %0 = load ptr, ptr %mdct_forward_x, align 845 %add.ptr.i = getelementptr i8, ptr %0, i64 2446 %arrayidx.i.i = getelementptr i8, ptr %0, i64 4847 %1 = load float, ptr %arrayidx.i.i, align 448 %add.i.i = fadd float %1, 0.000000e+0049 %arrayidx2.i.i = getelementptr i8, ptr %0, i64 3250 %2 = load float, ptr %arrayidx2.i.i, align 451 %sub.i.i = fsub float %1, %252 %3 = load float, ptr %add.ptr.i, align 453 %add4.i.i = fadd float %3, 0.000000e+0054 %arrayidx5.i.i = getelementptr i8, ptr %0, i64 4055 %4 = load float, ptr %arrayidx5.i.i, align 456 %sub7.i.i = fsub float %4, %357 %sub8.i.i = fsub float %add.i.i, %add4.i.i58 store float %sub8.i.i, ptr %arrayidx5.i.i, align 459 %arrayidx10.i.i = getelementptr i8, ptr %0, i64 2860 %5 = load float, ptr %arrayidx10.i.i, align 461 %sub11.i.i = fsub float 0.000000e+00, %562 %arrayidx12.i.i = getelementptr i8, ptr %0, i64 3663 %6 = load float, ptr %arrayidx12.i.i, align 464 %sub13.i.i = fsub float 0.000000e+00, %665 store float 0.000000e+00, ptr %add.ptr.i, align 466 %sub15.i.i = fsub float %sub.i.i, %sub11.i.i67 store float %sub15.i.i, ptr %arrayidx2.i.i, align 468 %add17.i.i = fadd float %sub7.i.i, %sub13.i.i69 store float %add17.i.i, ptr %arrayidx12.i.i, align 470 %arrayidx20.i.i = getelementptr i8, ptr %0, i64 4471 store float %sub15.i.i, ptr %arrayidx20.i.i, align 472 br label %for.cond73}74