1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -instcombine %s | FileCheck %s 3 4; Eliminate the insertelement. 5 6define <4 x float> @PR29126(<4 x float> %x) { 7; CHECK-LABEL: @PR29126( 8; CHECK-NEXT: [[INS:%.*]] = shufflevector <4 x float> %x, <4 x float> <float undef, float 1.000000e+00, float 2.000000e+00, float 4.200000e+01>, <4 x i32> <i32 0, i32 5, i32 6, i32 7> 9; CHECK-NEXT: ret <4 x float> [[INS]] 10; 11 %shuf = shufflevector <4 x float> %x, <4 x float> <float undef, float 1.0, float 2.0, float undef>, <4 x i32> <i32 0, i32 5, i32 6, i32 3> 12 %ins = insertelement <4 x float> %shuf, float 42.0, i32 3 13 ret <4 x float> %ins 14} 15 16; A chain of inserts should collapse. 17 18define <4 x float> @twoInserts(<4 x float> %x) { 19; CHECK-LABEL: @twoInserts( 20; CHECK-NEXT: [[INS2:%.*]] = shufflevector <4 x float> %x, <4 x float> <float undef, float 0.000000e+00, float 4.200000e+01, float 1.100000e+01>, <4 x i32> <i32 0, i32 5, i32 6, i32 7> 21; CHECK-NEXT: ret <4 x float> [[INS2]] 22; 23 %shuf = shufflevector <4 x float> %x, <4 x float> zeroinitializer, <4 x i32> <i32 0, i32 5, i32 6, i32 3> 24 %ins1 = insertelement <4 x float> %shuf, float 42.0, i32 2 25 %ins2 = insertelement <4 x float> %ins1, float 11.0, i32 3 26 ret <4 x float> %ins2 27} 28 29define <4 x i32> @shuffleRetain(<4 x i32> %base) { 30; CHECK-LABEL: @shuffleRetain( 31; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <4 x i32> %base, <4 x i32> <i32 undef, i32 undef, i32 undef, i32 1>, <4 x i32> <i32 1, i32 2, i32 undef, i32 7> 32; CHECK-NEXT: ret <4 x i32> [[SHUF]] 33; 34 %shuf = shufflevector <4 x i32> %base, <4 x i32> <i32 4, i32 3, i32 2, i32 1>, <4 x i32> <i32 1, i32 2, i32 undef, i32 7> 35 ret <4 x i32> %shuf 36} 37 38; TODO: Transform an arbitrary shuffle with constant into a shuffle that is equivalant to a vector select. 39 40define <4 x float> @disguisedSelect(<4 x float> %x) { 41; CHECK-LABEL: @disguisedSelect( 42; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <4 x float> %x, <4 x float> <float undef, float 1.000000e+00, float 2.000000e+00, float undef>, <4 x i32> <i32 undef, i32 6, i32 5, i32 3> 43; CHECK-NEXT: [[INS:%.*]] = insertelement <4 x float> [[SHUF]], float 4.000000e+00, i32 0 44; CHECK-NEXT: ret <4 x float> [[INS]] 45; 46 %shuf = shufflevector <4 x float> %x, <4 x float> <float undef, float 1.0, float 2.0, float 3.0>, <4 x i32> <i32 7, i32 6, i32 5, i32 3> 47 %ins = insertelement <4 x float> %shuf, float 4.0, i32 0 48 ret <4 x float> %ins 49} 50 51; TODO: Fold arbitrary (non-select-equivalent) shuffles if the new shuffle would have the same shuffle mask. 52 53define <4 x float> @notSelectButNoMaskDifference(<4 x float> %x) { 54; CHECK-LABEL: @notSelectButNoMaskDifference( 55; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <4 x float> %x, <4 x float> <float undef, float 1.000000e+00, float 2.000000e+00, float undef>, <4 x i32> <i32 1, i32 5, i32 6, i32 undef> 56; CHECK-NEXT: [[INS:%.*]] = insertelement <4 x float> [[SHUF]], float 4.000000e+00, i32 3 57; CHECK-NEXT: ret <4 x float> [[INS]] 58; 59 %shuf = shufflevector <4 x float> %x, <4 x float> <float undef, float 1.0, float 2.0, float 3.0>, <4 x i32> <i32 1, i32 5, i32 6, i32 3> 60 %ins = insertelement <4 x float> %shuf, float 4.0, i32 3 61 ret <4 x float> %ins 62} 63 64; We purposely do not touch arbitrary (non-select-equivalent) shuffles because folding the insert may create a more expensive shuffle. 65 66define <4 x float> @tooRisky(<4 x float> %x) { 67; CHECK-LABEL: @tooRisky( 68; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <4 x float> %x, <4 x float> <float 1.000000e+00, float undef, float undef, float undef>, <4 x i32> <i32 1, i32 4, i32 4, i32 undef> 69; CHECK-NEXT: [[INS:%.*]] = insertelement <4 x float> [[SHUF]], float 4.000000e+00, i32 3 70; CHECK-NEXT: ret <4 x float> [[INS]] 71; 72 %shuf = shufflevector <4 x float> %x, <4 x float> <float 1.0, float undef, float undef, float undef>, <4 x i32> <i32 1, i32 4, i32 4, i32 4> 73 %ins = insertelement <4 x float> %shuf, float 4.0, i32 3 74 ret <4 x float> %ins 75} 76 77; Don't transform insert to shuffle if the original shuffle is not removed. 78; TODO: Ease the one-use restriction if the insert scalar would simplify the shuffle to a full vector constant? 79 80define <3 x float> @twoShufUses(<3 x float> %x) { 81; CHECK-LABEL: @twoShufUses( 82; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <3 x float> %x, <3 x float> <float undef, float 1.000000e+00, float 2.000000e+00>, <3 x i32> <i32 0, i32 4, i32 5> 83; CHECK-NEXT: [[INS:%.*]] = insertelement <3 x float> [[SHUF]], float 4.200000e+01, i2 1 84; CHECK-NEXT: [[ADD:%.*]] = fadd <3 x float> [[SHUF]], [[INS]] 85; CHECK-NEXT: ret <3 x float> [[ADD]] 86; 87 %shuf = shufflevector <3 x float> %x, <3 x float> <float undef, float 1.0, float 2.0>, <3 x i32> <i32 0, i32 4, i32 5> 88 %ins = insertelement <3 x float> %shuf, float 42.0, i2 1 89 %add = fadd <3 x float> %shuf, %ins 90 ret <3 x float> %add 91} 92 93; The inserted scalar constant index is out-of-bounds for the shuffle vector constant. 94 95define <5 x i8> @longerMask(<3 x i8> %x) { 96; CHECK-LABEL: @longerMask( 97; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <3 x i8> %x, <3 x i8> <i8 undef, i8 1, i8 undef>, <5 x i32> <i32 2, i32 1, i32 4, i32 undef, i32 undef> 98; CHECK-NEXT: [[INS:%.*]] = insertelement <5 x i8> [[SHUF]], i8 42, i17 4 99; CHECK-NEXT: ret <5 x i8> [[INS]] 100; 101 %shuf = shufflevector <3 x i8> %x, <3 x i8> <i8 undef, i8 1, i8 2>, <5 x i32> <i32 2, i32 1, i32 4, i32 3, i32 0> 102 %ins = insertelement <5 x i8> %shuf, i8 42, i17 4 103 ret <5 x i8> %ins 104} 105 106; TODO: The inserted constant could get folded into the shuffle vector constant. 107 108define <3 x i8> @shorterMask(<5 x i8> %x) { 109; CHECK-LABEL: @shorterMask( 110; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <5 x i8> %x, <5 x i8> undef, <3 x i32> <i32 undef, i32 1, i32 4> 111; CHECK-NEXT: [[INS:%.*]] = insertelement <3 x i8> [[SHUF]], i8 42, i21 0 112; CHECK-NEXT: ret <3 x i8> [[INS]] 113; 114 %shuf = shufflevector <5 x i8> %x, <5 x i8> <i8 undef, i8 1, i8 2, i8 3, i8 4>, <3 x i32> <i32 2, i32 1, i32 4> 115 %ins = insertelement <3 x i8> %shuf, i8 42, i21 0 116 ret <3 x i8> %ins 117} 118 119