1; RUN: opt < %s -instcombine -S | FileCheck %s 2 3define <4 x float> @test1(<4 x float> %v1) { 4; CHECK: @test1 5; CHECK: ret <4 x float> %v1 6 %v2 = shufflevector <4 x float> %v1, <4 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 7 ret <4 x float> %v2 8} 9 10define <4 x float> @test2(<4 x float> %v1) { 11; CHECK: @test2 12; CHECK: ret <4 x float> %v1 13 %v2 = shufflevector <4 x float> %v1, <4 x float> %v1, <4 x i32> <i32 0, i32 5, i32 2, i32 7> 14 ret <4 x float> %v2 15} 16 17define float @test3(<4 x float> %A, <4 x float> %B, float %f) { 18; CHECK: @test3 19; CHECK: ret float %f 20 %C = insertelement <4 x float> %A, float %f, i32 0 21 %D = shufflevector <4 x float> %C, <4 x float> %B, <4 x i32> <i32 5, i32 0, i32 2, i32 7> 22 %E = extractelement <4 x float> %D, i32 1 23 ret float %E 24} 25 26define i32 @test4(<4 x i32> %X) { 27; CHECK: @test4 28; CHECK-NEXT: extractelement 29; CHECK-NEXT: ret 30 %tmp152.i53899.i = shufflevector <4 x i32> %X, <4 x i32> undef, <4 x i32> zeroinitializer 31 %tmp34 = extractelement <4 x i32> %tmp152.i53899.i, i32 0 32 ret i32 %tmp34 33} 34 35define i32 @test5(<4 x i32> %X) { 36; CHECK: @test5 37; CHECK-NEXT: extractelement 38; CHECK-NEXT: ret 39 %tmp152.i53899.i = shufflevector <4 x i32> %X, <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 undef, i32 undef> 40 %tmp34 = extractelement <4 x i32> %tmp152.i53899.i, i32 0 41 ret i32 %tmp34 42} 43 44define float @test6(<4 x float> %X) { 45; CHECK: @test6 46; CHECK-NEXT: extractelement 47; CHECK-NEXT: ret 48 %X1 = bitcast <4 x float> %X to <4 x i32> 49 %tmp152.i53899.i = shufflevector <4 x i32> %X1, <4 x i32> undef, <4 x i32> zeroinitializer 50 %tmp152.i53900.i = bitcast <4 x i32> %tmp152.i53899.i to <4 x float> 51 %tmp34 = extractelement <4 x float> %tmp152.i53900.i, i32 0 52 ret float %tmp34 53} 54 55define <4 x float> @test7(<4 x float> %tmp45.i) { 56; CHECK: @test7 57; CHECK-NEXT: ret <4 x float> %tmp45.i 58 %tmp1642.i = shufflevector <4 x float> %tmp45.i, <4 x float> undef, <4 x i32> < i32 0, i32 1, i32 6, i32 7 > 59 ret <4 x float> %tmp1642.i 60} 61 62; This should turn into a single shuffle. 63define <4 x float> @test8(<4 x float> %tmp, <4 x float> %tmp1) { 64; CHECK: @test8 65; CHECK-NEXT: shufflevector 66; CHECK-NEXT: ret 67 %tmp4 = extractelement <4 x float> %tmp, i32 1 68 %tmp2 = extractelement <4 x float> %tmp, i32 3 69 %tmp1.upgrd.1 = extractelement <4 x float> %tmp1, i32 0 70 %tmp128 = insertelement <4 x float> undef, float %tmp4, i32 0 71 %tmp130 = insertelement <4 x float> %tmp128, float undef, i32 1 72 %tmp132 = insertelement <4 x float> %tmp130, float %tmp2, i32 2 73 %tmp134 = insertelement <4 x float> %tmp132, float %tmp1.upgrd.1, i32 3 74 ret <4 x float> %tmp134 75} 76 77; Test fold of two shuffles where the first shuffle vectors inputs are a 78; different length then the second. 79define <4 x i8> @test9(<16 x i8> %tmp6) nounwind { 80; CHECK: @test9 81; CHECK-NEXT: shufflevector 82; CHECK-NEXT: ret 83 %tmp7 = shufflevector <16 x i8> %tmp6, <16 x i8> undef, <4 x i32> < i32 13, i32 9, i32 4, i32 13 > ; <<4 x i8>> [#uses=1] 84 %tmp9 = shufflevector <4 x i8> %tmp7, <4 x i8> undef, <4 x i32> < i32 3, i32 1, i32 2, i32 0 > ; <<4 x i8>> [#uses=1] 85 ret <4 x i8> %tmp9 86} 87 88; Same as test9, but make sure that "undef" mask values are not confused with 89; mask values of 2*N, where N is the mask length. These shuffles should not 90; be folded (because [8,9,4,8] may not be a mask supported by the target). 91define <4 x i8> @test9a(<16 x i8> %tmp6) nounwind { 92; CHECK: @test9a 93; CHECK-NEXT: shufflevector 94; CHECK-NEXT: shufflevector 95; CHECK-NEXT: ret 96 %tmp7 = shufflevector <16 x i8> %tmp6, <16 x i8> undef, <4 x i32> < i32 undef, i32 9, i32 4, i32 8 > ; <<4 x i8>> [#uses=1] 97 %tmp9 = shufflevector <4 x i8> %tmp7, <4 x i8> undef, <4 x i32> < i32 3, i32 1, i32 2, i32 0 > ; <<4 x i8>> [#uses=1] 98 ret <4 x i8> %tmp9 99} 100 101; Test fold of two shuffles where the first shuffle vectors inputs are a 102; different length then the second. 103define <4 x i8> @test9b(<4 x i8> %tmp6, <4 x i8> %tmp7) nounwind { 104; CHECK: @test9 105; CHECK-NEXT: shufflevector 106; CHECK-NEXT: ret 107 %tmp1 = shufflevector <4 x i8> %tmp6, <4 x i8> %tmp7, <8 x i32> <i32 0, i32 1, i32 4, i32 5, i32 4, i32 5, i32 2, i32 3> ; <<4 x i8>> [#uses=1] 108 %tmp9 = shufflevector <8 x i8> %tmp1, <8 x i8> undef, <4 x i32> <i32 0, i32 1, i32 4, i32 5> ; <<4 x i8>> [#uses=1] 109 ret <4 x i8> %tmp9 110} 111 112; Redundant vector splats should be removed. Radar 8597790. 113define <4 x i32> @test10(<4 x i32> %tmp5) nounwind { 114; CHECK: @test10 115; CHECK-NEXT: shufflevector 116; CHECK-NEXT: ret 117 %tmp6 = shufflevector <4 x i32> %tmp5, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef> 118 %tmp7 = shufflevector <4 x i32> %tmp6, <4 x i32> undef, <4 x i32> zeroinitializer 119 ret <4 x i32> %tmp7 120} 121 122; Test fold of two shuffles where the two shufflevector inputs's op1 are 123; the same 124define <8 x i8> @test11(<16 x i8> %tmp6) nounwind { 125; CHECK: @test11 126; CHECK-NEXT: shufflevector <16 x i8> %tmp6, <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 127; CHECK-NEXT: ret 128 %tmp1 = shufflevector <16 x i8> %tmp6, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x i8>> [#uses=1] 129 %tmp2 = shufflevector <16 x i8> %tmp6, <16 x i8> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7> ; <<4 x i8>> [#uses=1] 130 %tmp3 = shufflevector <4 x i8> %tmp1, <4 x i8> %tmp2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> ; <<8 x i8>> [#uses=1] 131 ret <8 x i8> %tmp3 132} 133 134; Test fold of two shuffles where the first shufflevector's inputs are 135; the same as the second 136define <8 x i8> @test12(<8 x i8> %tmp6, <8 x i8> %tmp2) nounwind { 137; CHECK: @test12 138; CHECK-NEXT: shufflevector <8 x i8> %tmp6, <8 x i8> %tmp2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 9, i32 8, i32 11, i32 12> 139; CHECK-NEXT: ret 140 %tmp1 = shufflevector <8 x i8> %tmp6, <8 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 undef, i32 7> ; <<8 x i8>> [#uses=1] 141 %tmp3 = shufflevector <8 x i8> %tmp1, <8 x i8> %tmp2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 9, i32 8, i32 11, i32 12> ; <<8 x i8>> [#uses=1] 142 ret <8 x i8> %tmp3 143} 144 145; Test fold of two shuffles where the first shufflevector's inputs are 146; the same as the second 147define <8 x i8> @test12a(<8 x i8> %tmp6, <8 x i8> %tmp2) nounwind { 148; CHECK: @test12a 149; CHECK-NEXT: shufflevector <8 x i8> %tmp2, <8 x i8> %tmp6, <8 x i32> <i32 0, i32 3, i32 1, i32 4, i32 8, i32 9, i32 10, i32 11> 150; CHECK-NEXT: ret 151 %tmp1 = shufflevector <8 x i8> %tmp6, <8 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 undef, i32 7> ; <<8 x i8>> [#uses=1] 152 %tmp3 = shufflevector <8 x i8> %tmp2, <8 x i8> %tmp1, <8 x i32> <i32 0, i32 3, i32 1, i32 4, i32 8, i32 9, i32 10, i32 11> ; <<8 x i8>> [#uses=1] 153 ret <8 x i8> %tmp3 154} 155 156; We should form a shuffle out of a select with constant condition. 157define <4 x i16> @test13a(<4 x i16> %lhs, <4 x i16> %rhs) { 158; CHECK: @test13a 159; CHECK-NEXT: shufflevector <4 x i16> %lhs, <4 x i16> %rhs, <4 x i32> <i32 0, i32 5, i32 2, i32 7> 160; CHECK-NEXT: ret 161 %A = select <4 x i1> <i1 true, i1 false, i1 true, i1 false>, 162 <4 x i16> %lhs, <4 x i16> %rhs 163 ret <4 x i16> %A 164} 165 166define <4 x i16> @test13b(<4 x i16> %lhs, <4 x i16> %rhs) { 167; CHECK: @test13b 168; CHECK-NEXT: ret <4 x i16> %lhs 169 %A = select <4 x i1> <i1 true, i1 undef, i1 true, i1 true>, 170 <4 x i16> %lhs, <4 x i16> %rhs 171 ret <4 x i16> %A 172} 173 174define <4 x i16> @test13c(<4 x i16> %lhs, <4 x i16> %rhs) { 175; CHECK: @test13c 176; CHECK-NEXT: shufflevector <4 x i16> %lhs, <4 x i16> %rhs, <4 x i32> <i32 0, i32 undef, i32 2, i32 7> 177; CHECK-NEXT: ret 178 %A = select <4 x i1> <i1 true, i1 undef, i1 true, i1 false>, 179 <4 x i16> %lhs, <4 x i16> %rhs 180 ret <4 x i16> %A 181} 182 183define <4 x i16> @test13d(<4 x i16> %lhs, <4 x i16> %rhs) { 184; CHECK: @test13d 185; CHECK: select 186; CHECK-NEXT: ret 187 %A = select <4 x i1> <i1 true, i1 icmp ugt (<4 x i16>(<4 x i16>, <4 x i16>)* @test13a, <4 x i16>(<4 x i16>, <4 x i16>)* @test13b), i1 true, i1 false>, 188 <4 x i16> %lhs, <4 x i16> %rhs 189 ret <4 x i16> %A 190} 191 192define <4 x i16> @test13e(<4 x i16> %lhs, <4 x i16> %rhs) { 193; CHECK: @test13e 194; CHECK-NEXT: ret <4 x i16> %rhs 195 %A = select <4 x i1> <i1 false, i1 false, i1 false, i1 false>, 196 <4 x i16> %lhs, <4 x i16> %rhs 197 ret <4 x i16> %A 198} 199