1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -passes=instcombine -S | FileCheck %s 3 4define <4 x float> @test1(<4 x float> %v1) { 5; CHECK-LABEL: @test1( 6; CHECK-NEXT: ret <4 x float> [[V1:%.*]] 7; 8 %v2 = shufflevector <4 x float> %v1, <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 9 ret <4 x float> %v2 10} 11 12define <4 x float> @test2(<4 x float> %v1) { 13; CHECK-LABEL: @test2( 14; CHECK-NEXT: ret <4 x float> [[V1:%.*]] 15; 16 %v2 = shufflevector <4 x float> %v1, <4 x float> %v1, <4 x i32> <i32 0, i32 5, i32 2, i32 7> 17 ret <4 x float> %v2 18} 19 20define float @test3(<4 x float> %A, <4 x float> %B, float %f) { 21; CHECK-LABEL: @test3( 22; CHECK-NEXT: ret float [[F:%.*]] 23; 24 %C = insertelement <4 x float> %A, float %f, i32 0 25 %D = shufflevector <4 x float> %C, <4 x float> %B, <4 x i32> <i32 5, i32 0, i32 2, i32 7> 26 %E = extractelement <4 x float> %D, i32 1 27 ret float %E 28} 29 30define i32 @test4(<4 x i32> %X) { 31; CHECK-LABEL: @test4( 32; CHECK-NEXT: [[R:%.*]] = extractelement <4 x i32> [[X:%.*]], i64 0 33; CHECK-NEXT: ret i32 [[R]] 34; 35 %t = shufflevector <4 x i32> %X, <4 x i32> poison, <4 x i32> zeroinitializer 36 %r = extractelement <4 x i32> %t, i32 0 37 ret i32 %r 38} 39 40define i32 @test5(<4 x i32> %X) { 41; CHECK-LABEL: @test5( 42; CHECK-NEXT: [[R:%.*]] = extractelement <4 x i32> [[X:%.*]], i64 3 43; CHECK-NEXT: ret i32 [[R]] 44; 45 %t = shufflevector <4 x i32> %X, <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 undef, i32 undef> 46 %r = extractelement <4 x i32> %t, i32 0 47 ret i32 %r 48} 49 50define float @test6(<4 x float> %X) { 51; CHECK-LABEL: @test6( 52; CHECK-NEXT: [[R:%.*]] = extractelement <4 x float> [[X:%.*]], i64 0 53; CHECK-NEXT: ret float [[R]] 54; 55 %X1 = bitcast <4 x float> %X to <4 x i32> 56 %t = shufflevector <4 x i32> %X1, <4 x i32> poison, <4 x i32> zeroinitializer 57 %t2 = bitcast <4 x i32> %t to <4 x float> 58 %r = extractelement <4 x float> %t2, i32 0 59 ret float %r 60} 61 62define float @testvscale6(<vscale x 4 x float> %X) { 63; CHECK-LABEL: @testvscale6( 64; CHECK-NEXT: [[T2:%.*]] = shufflevector <vscale x 4 x float> [[X:%.*]], <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer 65; CHECK-NEXT: [[R:%.*]] = extractelement <vscale x 4 x float> [[T2]], i64 0 66; CHECK-NEXT: ret float [[R]] 67; 68 %X1 = bitcast <vscale x 4 x float> %X to <vscale x 4 x i32> 69 %t = shufflevector <vscale x 4 x i32> %X1, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer 70 %t2 = bitcast <vscale x 4 x i32> %t to <vscale x 4 x float> 71 %r = extractelement <vscale x 4 x float> %t2, i32 0 72 ret float %r 73} 74 75 76define <4 x float> @test7(<4 x float> %x) { 77; CHECK-LABEL: @test7( 78; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 79; CHECK-NEXT: ret <4 x float> [[R]] 80; 81 %r = shufflevector <4 x float> %x, <4 x float> poison, <4 x i32> < i32 0, i32 1, i32 6, i32 7 > 82 ret <4 x float> %r 83} 84 85; This should turn into a single shuffle. 86define <4 x float> @test8(<4 x float> %x, <4 x float> %y) { 87; CHECK-LABEL: @test8( 88; CHECK-NEXT: [[T134:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> [[Y:%.*]], <4 x i32> <i32 1, i32 poison, i32 3, i32 4> 89; CHECK-NEXT: ret <4 x float> [[T134]] 90; 91 %t4 = extractelement <4 x float> %x, i32 1 92 %t2 = extractelement <4 x float> %x, i32 3 93 %t1 = extractelement <4 x float> %y, i32 0 94 %t128 = insertelement <4 x float> poison, float %t4, i32 0 95 %t130 = insertelement <4 x float> %t128, float poison, i32 1 96 %t132 = insertelement <4 x float> %t130, float %t2, i32 2 97 %t134 = insertelement <4 x float> %t132, float %t1, i32 3 98 ret <4 x float> %t134 99} 100 101; This shouldn't turn into a single shuffle 102define <4 x float> @test8_undef(<4 x float> %x, <4 x float> %y) { 103; CHECK-LABEL: @test8_undef( 104; CHECK-NEXT: [[T132:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> <float poison, float undef, float poison, float poison>, <4 x i32> <i32 1, i32 5, i32 3, i32 poison> 105; CHECK-NEXT: [[T134:%.*]] = shufflevector <4 x float> [[T132]], <4 x float> [[Y:%.*]], <4 x i32> <i32 0, i32 1, i32 2, i32 4> 106; CHECK-NEXT: ret <4 x float> [[T134]] 107; 108 %t4 = extractelement <4 x float> %x, i32 1 109 %t2 = extractelement <4 x float> %x, i32 3 110 %t1 = extractelement <4 x float> %y, i32 0 111 %t128 = insertelement <4 x float> poison, float %t4, i32 0 112 %t130 = insertelement <4 x float> %t128, float undef, i32 1 113 %t132 = insertelement <4 x float> %t130, float %t2, i32 2 114 %t134 = insertelement <4 x float> %t132, float %t1, i32 3 115 ret <4 x float> %t134 116} 117 118; Test fold of two shuffles where the first shuffle vectors inputs are a 119; different length then the second. 120define <4 x i8> @test9(<16 x i8> %t6) { 121; CHECK-LABEL: @test9( 122; CHECK-NEXT: [[T9:%.*]] = shufflevector <16 x i8> [[T6:%.*]], <16 x i8> poison, <4 x i32> <i32 13, i32 9, i32 4, i32 13> 123; CHECK-NEXT: ret <4 x i8> [[T9]] 124; 125 %t7 = shufflevector <16 x i8> %t6, <16 x i8> poison, <4 x i32> < i32 13, i32 9, i32 4, i32 13 > 126 %t9 = shufflevector <4 x i8> %t7, <4 x i8> poison, <4 x i32> < i32 3, i32 1, i32 2, i32 0 > 127 ret <4 x i8> %t9 128} 129 130; Same as test9, but make sure that "undef" mask values are not confused with 131; mask values of 2*N, where N is the mask length. These shuffles should not 132; be folded (because [8,9,4,8] may not be a mask supported by the target). 133 134define <4 x i8> @test9a(<16 x i8> %t6) { 135; CHECK-LABEL: @test9a( 136; CHECK-NEXT: [[T7:%.*]] = shufflevector <16 x i8> [[T6:%.*]], <16 x i8> poison, <4 x i32> <i32 poison, i32 9, i32 4, i32 8> 137; CHECK-NEXT: [[T9:%.*]] = shufflevector <4 x i8> [[T7]], <4 x i8> poison, <4 x i32> <i32 3, i32 1, i32 2, i32 poison> 138; CHECK-NEXT: ret <4 x i8> [[T9]] 139; 140 %t7 = shufflevector <16 x i8> %t6, <16 x i8> poison, <4 x i32> < i32 undef, i32 9, i32 4, i32 8 > 141 %t9 = shufflevector <4 x i8> %t7, <4 x i8> poison, <4 x i32> < i32 3, i32 1, i32 2, i32 0 > 142 ret <4 x i8> %t9 143} 144 145; Test fold of two shuffles where the first shuffle vectors inputs are a 146; different length then the second. 147define <4 x i8> @test9b(<4 x i8> %t6, <4 x i8> %t7) { 148; CHECK-LABEL: @test9b( 149; CHECK-NEXT: [[T9:%.*]] = shufflevector <4 x i8> [[T6:%.*]], <4 x i8> [[T7:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 5> 150; CHECK-NEXT: ret <4 x i8> [[T9]] 151; 152 %t1 = shufflevector <4 x i8> %t6, <4 x i8> %t7, <8 x i32> <i32 0, i32 1, i32 4, i32 5, i32 4, i32 5, i32 2, i32 3> 153 %t9 = shufflevector <8 x i8> %t1, <8 x i8> poison, <4 x i32> <i32 0, i32 1, i32 4, i32 5> 154 ret <4 x i8> %t9 155} 156 157; Redundant vector splats should be removed. Radar 8597790. 158define <4 x i32> @test10(<4 x i32> %t5) { 159; CHECK-LABEL: @test10( 160; CHECK-NEXT: [[T7:%.*]] = shufflevector <4 x i32> [[T5:%.*]], <4 x i32> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 161; CHECK-NEXT: ret <4 x i32> [[T7]] 162; 163 %t6 = shufflevector <4 x i32> %t5, <4 x i32> poison, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef> 164 %t7 = shufflevector <4 x i32> %t6, <4 x i32> poison, <4 x i32> zeroinitializer 165 ret <4 x i32> %t7 166} 167 168; Test fold of two shuffles where the two shufflevector inputs's op1 are the same. 169 170define <8 x i8> @test11(<16 x i8> %t6) { 171; CHECK-LABEL: @test11( 172; CHECK-NEXT: [[T3:%.*]] = shufflevector <16 x i8> [[T6:%.*]], <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 173; CHECK-NEXT: ret <8 x i8> [[T3]] 174; 175 %t1 = shufflevector <16 x i8> %t6, <16 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 176 %t2 = shufflevector <16 x i8> %t6, <16 x i8> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7> 177 %t3 = shufflevector <4 x i8> %t1, <4 x i8> %t2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 178 ret <8 x i8> %t3 179} 180 181; Test fold of two shuffles where the first shufflevector's inputs are the same as the second. 182 183define <8 x i8> @test12(<8 x i8> %t6, <8 x i8> %t2) { 184; CHECK-LABEL: @test12( 185; CHECK-NEXT: [[T3:%.*]] = shufflevector <8 x i8> [[T6:%.*]], <8 x i8> [[T2:%.*]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 9, i32 8, i32 11, i32 12> 186; CHECK-NEXT: ret <8 x i8> [[T3]] 187; 188 %t1 = shufflevector <8 x i8> %t6, <8 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 undef, i32 7> 189 %t3 = shufflevector <8 x i8> %t1, <8 x i8> %t2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 9, i32 8, i32 11, i32 12> 190 ret <8 x i8> %t3 191} 192 193; Test fold of two shuffles where the first shufflevector's inputs are the same as the second. 194 195define <8 x i8> @test12a(<8 x i8> %t6, <8 x i8> %t2) { 196; CHECK-LABEL: @test12a( 197; CHECK-NEXT: [[T3:%.*]] = shufflevector <8 x i8> [[T2:%.*]], <8 x i8> [[T6:%.*]], <8 x i32> <i32 0, i32 3, i32 1, i32 4, i32 8, i32 9, i32 10, i32 11> 198; CHECK-NEXT: ret <8 x i8> [[T3]] 199; 200 %t1 = shufflevector <8 x i8> %t6, <8 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 undef, i32 7> 201 %t3 = shufflevector <8 x i8> %t2, <8 x i8> %t1, <8 x i32> <i32 0, i32 3, i32 1, i32 4, i32 8, i32 9, i32 10, i32 11> 202 ret <8 x i8> %t3 203} 204 205; The mask length of the 1st shuffle can be reduced to eliminate the 2nd shuffle. 206 207define <2 x i8> @extract_subvector_of_shuffle(<2 x i8> %x, <2 x i8> %y) { 208; CHECK-LABEL: @extract_subvector_of_shuffle( 209; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <2 x i32> <i32 0, i32 2> 210; CHECK-NEXT: ret <2 x i8> [[EXTRACT_SUBV]] 211; 212 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <3 x i32> <i32 0, i32 2, i32 0> 213 %extract_subv = shufflevector <3 x i8> %shuf, <3 x i8> poison, <2 x i32> <i32 0, i32 1> 214 ret <2 x i8> %extract_subv 215} 216 217; Undef elements in either mask are ok. Undefs from the 2nd shuffle mask should propagate to the new shuffle. 218; The type of the inputs does not have to match the output type. 219 220define <4 x i8> @extract_subvector_of_shuffle_undefs_types(<2 x i8> %x, <2 x i8> %y) { 221; CHECK-LABEL: @extract_subvector_of_shuffle_undefs_types( 222; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <4 x i32> <i32 poison, i32 2, i32 0, i32 poison> 223; CHECK-NEXT: ret <4 x i8> [[EXTRACT_SUBV]] 224; 225 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <5 x i32> <i32 undef, i32 2, i32 0, i32 1, i32 0> 226 %extract_subv = shufflevector <5 x i8> %shuf, <5 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 227 ret <4 x i8> %extract_subv 228} 229 230; Extra uses are not ok - we only do the transform when we can eliminate an instruction. 231 232declare void @use_v5i8(<5 x i8>) 233 234define <4 x i8> @extract_subvector_of_shuffle_extra_use(<2 x i8> %x, <2 x i8> %y) { 235; CHECK-LABEL: @extract_subvector_of_shuffle_extra_use( 236; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <5 x i32> <i32 poison, i32 2, i32 0, i32 1, i32 0> 237; CHECK-NEXT: call void @use_v5i8(<5 x i8> [[SHUF]]) 238; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <5 x i8> [[SHUF]], <5 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> 239; CHECK-NEXT: ret <4 x i8> [[EXTRACT_SUBV]] 240; 241 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <5 x i32> <i32 undef, i32 2, i32 0, i32 1, i32 0> 242 call void @use_v5i8(<5 x i8> %shuf) 243 %extract_subv = shufflevector <5 x i8> %shuf, <5 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 244 ret <4 x i8> %extract_subv 245} 246 247define <2 x i8> @test13a(i8 %x1, i8 %x2) { 248; CHECK-LABEL: @test13a( 249; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i8> poison, i8 [[X2:%.*]], i64 0 250; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i8> [[TMP1]], i8 [[X1:%.*]], i64 1 251; CHECK-NEXT: [[D:%.*]] = add <2 x i8> [[TMP2]], <i8 7, i8 5> 252; CHECK-NEXT: ret <2 x i8> [[D]] 253; 254 %A = insertelement <2 x i8> poison, i8 %x1, i32 0 255 %B = insertelement <2 x i8> %A, i8 %x2, i32 1 256 %C = add <2 x i8> %B, <i8 5, i8 7> 257 %D = shufflevector <2 x i8> %C, <2 x i8> poison, <2 x i32> <i32 1, i32 0> 258 ret <2 x i8> %D 259} 260 261; Increasing length of vector ops is not a good canonicalization. 262 263define <3 x i32> @add_wider(i32 %y, i32 %z) { 264; CHECK-LABEL: @add_wider( 265; CHECK-NEXT: [[I0:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0 266; CHECK-NEXT: [[I1:%.*]] = insertelement <2 x i32> [[I0]], i32 [[Z:%.*]], i64 1 267; CHECK-NEXT: [[A:%.*]] = add <2 x i32> [[I1]], splat (i32 255) 268; CHECK-NEXT: [[EXT:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> poison, <3 x i32> <i32 0, i32 1, i32 poison> 269; CHECK-NEXT: ret <3 x i32> [[EXT]] 270; 271 %i0 = insertelement <2 x i32> poison, i32 %y, i32 0 272 %i1 = insertelement <2 x i32> %i0, i32 %z, i32 1 273 %a = add <2 x i32> %i1, <i32 255, i32 255> 274 %ext = shufflevector <2 x i32> %a, <2 x i32> poison, <3 x i32> <i32 0, i32 1, i32 undef> 275 ret <3 x i32> %ext 276} 277 278; Increasing length of vector ops must be safe from illegal undef propagation. 279 280define <3 x i32> @div_wider(i32 %y, i32 %z) { 281; CHECK-LABEL: @div_wider( 282; CHECK-NEXT: [[I0:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0 283; CHECK-NEXT: [[I1:%.*]] = insertelement <2 x i32> [[I0]], i32 [[Z:%.*]], i64 1 284; CHECK-NEXT: [[A:%.*]] = sdiv <2 x i32> [[I1]], splat (i32 255) 285; CHECK-NEXT: [[EXT:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> poison, <3 x i32> <i32 0, i32 1, i32 poison> 286; CHECK-NEXT: ret <3 x i32> [[EXT]] 287; 288 %i0 = insertelement <2 x i32> poison, i32 %y, i32 0 289 %i1 = insertelement <2 x i32> %i0, i32 %z, i32 1 290 %a = sdiv <2 x i32> %i1, <i32 255, i32 255> 291 %ext = shufflevector <2 x i32> %a, <2 x i32> poison, <3 x i32> <i32 0, i32 1, i32 undef> 292 ret <3 x i32> %ext 293} 294 295; Increasing length of insertelements (no math ops) is a good canonicalization. 296 297define <3 x i8> @fold_inselts_with_widening_shuffle(i8 %x, i8 %y) { 298; CHECK-LABEL: @fold_inselts_with_widening_shuffle( 299; CHECK-NEXT: [[TMP1:%.*]] = insertelement <3 x i8> poison, i8 [[X:%.*]], i64 0 300; CHECK-NEXT: [[WIDEN:%.*]] = insertelement <3 x i8> [[TMP1]], i8 [[Y:%.*]], i64 1 301; CHECK-NEXT: ret <3 x i8> [[WIDEN]] 302; 303 %ins0 = insertelement <2 x i8> poison, i8 %x, i32 0 304 %ins1 = insertelement <2 x i8> %ins0, i8 %y, i32 1 305 %widen = shufflevector <2 x i8> %ins1, <2 x i8> poison, <3 x i32> <i32 0, i32 1, i32 undef> 306 ret <3 x i8> %widen 307} 308 309define <2 x i8> @test13b(i8 %x) { 310; CHECK-LABEL: @test13b( 311; CHECK-NEXT: [[B:%.*]] = insertelement <2 x i8> poison, i8 [[X:%.*]], i64 1 312; CHECK-NEXT: ret <2 x i8> [[B]] 313; 314 %A = insertelement <2 x i8> poison, i8 %x, i32 0 315 %B = shufflevector <2 x i8> %A, <2 x i8> poison, <2 x i32> <i32 undef, i32 0> 316 ret <2 x i8> %B 317} 318 319define <2 x i8> @test13c(i8 %x1, i8 %x2) { 320; CHECK-LABEL: @test13c( 321; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i8> poison, i8 [[X1:%.*]], i64 0 322; CHECK-NEXT: [[C:%.*]] = insertelement <2 x i8> [[TMP1]], i8 [[X2:%.*]], i64 1 323; CHECK-NEXT: ret <2 x i8> [[C]] 324; 325 %A = insertelement <4 x i8> poison, i8 %x1, i32 0 326 %B = insertelement <4 x i8> %A, i8 %x2, i32 2 327 %C = shufflevector <4 x i8> %B, <4 x i8> poison, <2 x i32> <i32 0, i32 2> 328 ret <2 x i8> %C 329} 330 331define void @test14(i16 %conv10) { 332; CHECK-LABEL: @test14( 333; CHECK-NEXT: store <4 x i16> <i16 poison, i16 poison, i16 poison, i16 23>, ptr undef, align 8 334; CHECK-NEXT: ret void 335; 336 %t = alloca <4 x i16>, align 8 337 %vecinit6 = insertelement <4 x i16> poison, i16 23, i32 3 338 store <4 x i16> %vecinit6, ptr undef 339 %t1 = load <4 x i16>, ptr undef 340 %vecinit11 = insertelement <4 x i16> poison, i16 %conv10, i32 3 341 %div = udiv <4 x i16> %t1, %vecinit11 342 store <4 x i16> %div, ptr %t 343 %t4 = load <4 x i16>, ptr %t 344 %t5 = shufflevector <4 x i16> %t4, <4 x i16> poison, <2 x i32> <i32 2, i32 0> 345 %cmp = icmp ule <2 x i16> %t5, undef 346 %sext = sext <2 x i1> %cmp to <2 x i16> 347 ret void 348} 349 350; Check that sequences of insert/extract element are 351; collapsed into valid shuffle instruction with correct shuffle indexes. 352 353define <4 x float> @test15a(<4 x float> %LHS, <4 x float> %RHS) { 354; CHECK-LABEL: @test15a( 355; CHECK-NEXT: [[T4:%.*]] = shufflevector <4 x float> [[LHS:%.*]], <4 x float> [[RHS:%.*]], <4 x i32> <i32 4, i32 0, i32 6, i32 6> 356; CHECK-NEXT: ret <4 x float> [[T4]] 357; 358 %t1 = extractelement <4 x float> %LHS, i32 0 359 %t2 = insertelement <4 x float> %RHS, float %t1, i32 1 360 %t3 = extractelement <4 x float> %RHS, i32 2 361 %t4 = insertelement <4 x float> %t2, float %t3, i32 3 362 ret <4 x float> %t4 363} 364 365define <4 x float> @test15b(<4 x float> %LHS, <4 x float> %RHS) { 366; CHECK-LABEL: @test15b( 367; CHECK-NEXT: [[T5:%.*]] = shufflevector <4 x float> [[LHS:%.*]], <4 x float> [[RHS:%.*]], <4 x i32> <i32 4, i32 3, i32 6, i32 6> 368; CHECK-NEXT: ret <4 x float> [[T5]] 369; 370 %t0 = extractelement <4 x float> %LHS, i32 3 371 %t1 = insertelement <4 x float> %RHS, float %t0, i32 0 372 %t2 = extractelement <4 x float> %t1, i32 0 373 %t3 = insertelement <4 x float> %RHS, float %t2, i32 1 374 %t4 = extractelement <4 x float> %RHS, i32 2 375 %t5 = insertelement <4 x float> %t3, float %t4, i32 3 376 ret <4 x float> %t5 377} 378 379define <1 x i32> @test16a(i32 %ele) { 380; CHECK-LABEL: @test16a( 381; CHECK-NEXT: ret <1 x i32> splat (i32 2) 382; 383 %t0 = insertelement <2 x i32> <i32 1, i32 undef>, i32 %ele, i32 1 384 %t1 = shl <2 x i32> %t0, <i32 1, i32 1> 385 %t2 = shufflevector <2 x i32> %t1, <2 x i32> poison, <1 x i32> <i32 0> 386 ret <1 x i32> %t2 387} 388 389define <4 x i8> @test16b(i8 %ele) { 390; CHECK-LABEL: @test16b( 391; CHECK-NEXT: ret <4 x i8> splat (i8 2) 392; 393 %t0 = insertelement <8 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 undef, i8 1>, i8 %ele, i32 6 394 %t1 = shl <8 x i8> %t0, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1> 395 %t2 = shufflevector <8 x i8> %t1, <8 x i8> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 4> 396 ret <4 x i8> %t2 397} 398 399; If composition of two shuffles is identity, shuffles can be removed. 400define <4 x i32> @shuffle_17ident(<4 x i32> %v) { 401; CHECK-LABEL: @shuffle_17ident( 402; CHECK-NEXT: ret <4 x i32> [[V:%.*]] 403; 404 %shuffle = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 405 %shuffle2 = shufflevector <4 x i32> %shuffle, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 0, i32 1, i32 2> 406 ret <4 x i32> %shuffle2 407} 408 409; swizzle can be put after operation 410define <4 x i32> @shuffle_17and(<4 x i32> %v1, <4 x i32> %v2) { 411; CHECK-LABEL: @shuffle_17and( 412; CHECK-NEXT: [[TMP1:%.*]] = and <4 x i32> [[V1:%.*]], [[V2:%.*]] 413; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 414; CHECK-NEXT: ret <4 x i32> [[R]] 415; 416 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 417 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 418 %r = and <4 x i32> %t1, %t2 419 ret <4 x i32> %r 420} 421 422declare void @use(<2 x float>) 423 424; One extra use is ok to transform. 425 426define <2 x float> @shuffle_fadd_multiuse(<2 x float> %v1, <2 x float> %v2) { 427; CHECK-LABEL: @shuffle_fadd_multiuse( 428; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V1:%.*]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 429; CHECK-NEXT: [[TMP1:%.*]] = fadd <2 x float> [[V1]], [[V2:%.*]] 430; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 431; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 432; CHECK-NEXT: ret <2 x float> [[R]] 433; 434 %t1 = shufflevector <2 x float> %v1, <2 x float> poison, <2 x i32> <i32 1, i32 0> 435 %t2 = shufflevector <2 x float> %v2, <2 x float> poison, <2 x i32> <i32 1, i32 0> 436 %r = fadd <2 x float> %t1, %t2 437 call void @use(<2 x float> %t1) 438 ret <2 x float> %r 439} 440 441define <2 x float> @shuffle_fdiv_multiuse(<2 x float> %v1, <2 x float> %v2) { 442; CHECK-LABEL: @shuffle_fdiv_multiuse( 443; CHECK-NEXT: [[T2:%.*]] = shufflevector <2 x float> [[V2:%.*]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 444; CHECK-NEXT: [[TMP1:%.*]] = fdiv <2 x float> [[V1:%.*]], [[V2]] 445; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 446; CHECK-NEXT: call void @use(<2 x float> [[T2]]) 447; CHECK-NEXT: ret <2 x float> [[R]] 448; 449 %t1 = shufflevector <2 x float> %v1, <2 x float> poison, <2 x i32> <i32 1, i32 0> 450 %t2 = shufflevector <2 x float> %v2, <2 x float> poison, <2 x i32> <i32 1, i32 0> 451 %r = fdiv <2 x float> %t1, %t2 452 call void @use(<2 x float> %t2) 453 ret <2 x float> %r 454} 455 456; But 2 extra uses would require an extra instruction. 457 458define <2 x float> @shuffle_fsub_multiuse(<2 x float> %v1, <2 x float> %v2) { 459; CHECK-LABEL: @shuffle_fsub_multiuse( 460; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V1:%.*]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 461; CHECK-NEXT: [[T2:%.*]] = shufflevector <2 x float> [[V2:%.*]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 462; CHECK-NEXT: [[R:%.*]] = fsub <2 x float> [[T1]], [[T2]] 463; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 464; CHECK-NEXT: call void @use(<2 x float> [[T2]]) 465; CHECK-NEXT: ret <2 x float> [[R]] 466; 467 %t1 = shufflevector <2 x float> %v1, <2 x float> poison, <2 x i32> <i32 1, i32 0> 468 %t2 = shufflevector <2 x float> %v2, <2 x float> poison, <2 x i32> <i32 1, i32 0> 469 %r = fsub <2 x float> %t1, %t2 470 call void @use(<2 x float> %t1) 471 call void @use(<2 x float> %t2) 472 ret <2 x float> %r 473} 474 475define <4 x i32> @shuffle_17add(<4 x i32> %v1, <4 x i32> %v2) { 476; CHECK-LABEL: @shuffle_17add( 477; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[V1:%.*]], [[V2:%.*]] 478; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 479; CHECK-NEXT: ret <4 x i32> [[R]] 480; 481 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 482 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 483 %r = add <4 x i32> %t1, %t2 484 ret <4 x i32> %r 485} 486 487define <4 x i32> @shuffle_17addnsw(<4 x i32> %v1, <4 x i32> %v2) { 488; CHECK-LABEL: @shuffle_17addnsw( 489; CHECK-NEXT: [[TMP1:%.*]] = add nsw <4 x i32> [[V1:%.*]], [[V2:%.*]] 490; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 491; CHECK-NEXT: ret <4 x i32> [[R]] 492; 493 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 494 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 495 %r = add nsw <4 x i32> %t1, %t2 496 ret <4 x i32> %r 497} 498 499define <4 x i32> @shuffle_17addnuw(<4 x i32> %v1, <4 x i32> %v2) { 500; CHECK-LABEL: @shuffle_17addnuw( 501; CHECK-NEXT: [[TMP1:%.*]] = add nuw <4 x i32> [[V1:%.*]], [[V2:%.*]] 502; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 503; CHECK-NEXT: ret <4 x i32> [[R]] 504; 505 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 506 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 507 %r = add nuw <4 x i32> %t1, %t2 508 ret <4 x i32> %r 509} 510 511define <4 x float> @shuffle_17fsub_fast(<4 x float> %v1, <4 x float> %v2) { 512; CHECK-LABEL: @shuffle_17fsub_fast( 513; CHECK-NEXT: [[TMP1:%.*]] = fsub fast <4 x float> [[V1:%.*]], [[V2:%.*]] 514; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 515; CHECK-NEXT: ret <4 x float> [[R]] 516; 517 %t1 = shufflevector <4 x float> %v1, <4 x float> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 518 %t2 = shufflevector <4 x float> %v2, <4 x float> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 519 %r = fsub fast <4 x float> %t1, %t2 520 ret <4 x float> %r 521} 522 523define <4 x i32> @add_const(<4 x i32> %v) { 524; CHECK-LABEL: @add_const( 525; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[V:%.*]], <i32 44, i32 41, i32 42, i32 43> 526; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 527; CHECK-NEXT: ret <4 x i32> [[R]] 528; 529 %t1 = shufflevector <4 x i32> %v, <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 530 %r = add <4 x i32> %t1, <i32 41, i32 42, i32 43, i32 44> 531 ret <4 x i32> %r 532} 533 534define <4 x i32> @sub_const(<4 x i32> %v) { 535; CHECK-LABEL: @sub_const( 536; CHECK-NEXT: [[TMP1:%.*]] = sub <4 x i32> <i32 44, i32 43, i32 42, i32 41>, [[V:%.*]] 537; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 538; CHECK-NEXT: ret <4 x i32> [[R]] 539; 540 %t1 = shufflevector <4 x i32> %v, <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 541 %r = sub <4 x i32> <i32 41, i32 42, i32 43, i32 44>, %t1 542 ret <4 x i32> %r 543} 544 545; Math before shuffle requires an extra shuffle. 546 547define <2 x float> @fadd_const_multiuse(<2 x float> %v) { 548; CHECK-LABEL: @fadd_const_multiuse( 549; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 550; CHECK-NEXT: [[R:%.*]] = fadd <2 x float> [[T1]], <float 4.100000e+01, float 4.200000e+01> 551; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 552; CHECK-NEXT: ret <2 x float> [[R]] 553; 554 %t1 = shufflevector <2 x float> %v, <2 x float> poison, <2 x i32> <i32 1, i32 0> 555 %r = fadd <2 x float> %t1, <float 41.0, float 42.0> 556 call void @use(<2 x float> %t1) 557 ret <2 x float> %r 558} 559 560; Math before splat allows replacing constant elements with undef lanes. 561 562define <4 x i32> @mul_const_splat(<4 x i32> %v) { 563; CHECK-LABEL: @mul_const_splat( 564; CHECK-NEXT: [[TMP1:%.*]] = mul <4 x i32> [[V:%.*]], <i32 poison, i32 42, i32 poison, i32 poison> 565; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 566; CHECK-NEXT: ret <4 x i32> [[R]] 567; 568 %t1 = shufflevector <4 x i32> %v, <4 x i32> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 569 %r = mul <4 x i32> <i32 42, i32 42, i32 42, i32 42>, %t1 570 ret <4 x i32> %r 571} 572 573; Take 2 elements of a vector and shift each of those by a different amount 574 575define <4 x i32> @lshr_const_half_splat(<4 x i32> %v) { 576; CHECK-LABEL: @lshr_const_half_splat( 577; CHECK-NEXT: [[TMP1:%.*]] = lshr <4 x i32> <i32 poison, i32 8, i32 9, i32 poison>, [[V:%.*]] 578; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 1, i32 2, i32 2> 579; CHECK-NEXT: ret <4 x i32> [[R]] 580; 581 %t1 = shufflevector <4 x i32> %v, <4 x i32> poison, <4 x i32> <i32 1, i32 1, i32 2, i32 2> 582 %r = lshr <4 x i32> <i32 8, i32 8, i32 9, i32 9>, %t1 583 ret <4 x i32> %r 584} 585 586; We can't change this because there's no pre-shuffle version of the fmul constant. 587 588define <2 x float> @fmul_const_invalid_constant(<2 x float> %v) { 589; CHECK-LABEL: @fmul_const_invalid_constant( 590; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> poison, <2 x i32> zeroinitializer 591; CHECK-NEXT: [[R:%.*]] = fmul <2 x float> [[T1]], <float 4.100000e+01, float 4.200000e+01> 592; CHECK-NEXT: ret <2 x float> [[R]] 593; 594 %t1 = shufflevector <2 x float> %v, <2 x float> poison, <2 x i32> <i32 0, i32 0> 595 %r = fmul <2 x float> %t1, <float 41.0, float 42.0> 596 ret <2 x float> %r 597} 598 599; Reduce the width of the binop by moving it ahead of a shuffle. 600 601define <4 x i8> @widening_shuffle_add_1(<2 x i8> %x) { 602; CHECK-LABEL: @widening_shuffle_add_1( 603; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i8> [[X:%.*]], <i8 42, i8 43> 604; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i8> [[TMP1]], <2 x i8> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 605; CHECK-NEXT: ret <4 x i8> [[R]] 606; 607 %widex = shufflevector <2 x i8> %x, <2 x i8> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 608 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 609 ret <4 x i8> %r 610} 611 612; Reduce the width of the binop by moving it ahead of a shuffle. 613 614define <4 x i8> @widening_shuffle_add_2(<2 x i8> %x) { 615; CHECK-LABEL: @widening_shuffle_add_2( 616; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i8> [[X:%.*]], <i8 43, i8 42> 617; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i8> [[TMP1]], <2 x i8> poison, <4 x i32> <i32 1, i32 0, i32 poison, i32 poison> 618; CHECK-NEXT: ret <4 x i8> [[R]] 619; 620 %widex = shufflevector <2 x i8> %x, <2 x i8> poison, <4 x i32> <i32 1, i32 0, i32 undef, i32 undef> 621 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 622 ret <4 x i8> %r 623} 624 625; Negative test - widening shuffles have the same mask/constant constraint as non-size-changing shuffles. 626 627define <4 x i8> @widening_shuffle_add_invalid_constant(<2 x i8> %x) { 628; CHECK-LABEL: @widening_shuffle_add_invalid_constant( 629; CHECK-NEXT: [[WIDEX:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> poison, <4 x i32> <i32 1, i32 1, i32 poison, i32 poison> 630; CHECK-NEXT: [[R:%.*]] = add <4 x i8> [[WIDEX]], <i8 42, i8 43, i8 44, i8 45> 631; CHECK-NEXT: ret <4 x i8> [[R]] 632; 633 %widex = shufflevector <2 x i8> %x, <2 x i8> poison, <4 x i32> <i32 1, i32 1, i32 undef, i32 undef> 634 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 635 ret <4 x i8> %r 636} 637 638; Negative test - widening shuffles have an additional constraint: they must not extend with anything but undefs. 639 640define <4 x i8> @widening_shuffle_add_invalid_mask(<2 x i8> %x) { 641; CHECK-LABEL: @widening_shuffle_add_invalid_mask( 642; CHECK-NEXT: [[WIDEX:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 0> 643; CHECK-NEXT: [[R:%.*]] = add <4 x i8> [[WIDEX]], <i8 42, i8 43, i8 44, i8 45> 644; CHECK-NEXT: ret <4 x i8> [[R]] 645; 646 %widex = shufflevector <2 x i8> %x, <2 x i8> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 0> 647 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 648 ret <4 x i8> %r 649} 650 651; A binop that produces undef in the high lanes can be moved before the shuffle. 652; This is ok because 'shl C, undef --> undef'. 653 654define <4 x i16> @widening_shuffle_shl_constant_op0(<2 x i16> %v) { 655; CHECK-LABEL: @widening_shuffle_shl_constant_op0( 656; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i16> <i16 42, i16 -42>, [[V:%.*]] 657; CHECK-NEXT: [[BO:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 658; CHECK-NEXT: ret <4 x i16> [[BO]] 659; 660 %shuf = shufflevector <2 x i16> %v, <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 661 %bo = shl <4 x i16> <i16 42, i16 -42, i16 -1, i16 -1>, %shuf 662 ret <4 x i16> %bo 663} 664 665; A binop that produces undef in the high lanes can be moved before the shuffle. 666; This is ok because 'shl undef, 0 --> undef'. 667 668define <4 x i16> @widening_shuffle_shl_constant_op1(<2 x i16> %v) { 669; CHECK-LABEL: @widening_shuffle_shl_constant_op1( 670; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i16> [[V:%.*]], <i16 2, i16 4> 671; CHECK-NEXT: [[BO:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 672; CHECK-NEXT: ret <4 x i16> [[BO]] 673; 674 %shuf = shufflevector <2 x i16> %v, <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 675 %bo = shl <4 x i16> %shuf, <i16 2, i16 4, i16 0, i16 0> 676 ret <4 x i16> %bo 677} 678 679; This is valid for poison, but would not be valid for undef. 680 681define <4 x i16> @widening_shuffle_shl_constant_op1_non0(<2 x i16> %v) { 682; CHECK-LABEL: @widening_shuffle_shl_constant_op1_non0( 683; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i16> [[V:%.*]], <i16 2, i16 4> 684; CHECK-NEXT: [[BO:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 685; CHECK-NEXT: ret <4 x i16> [[BO]] 686; 687 %shuf = shufflevector <2 x i16> %v, <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 688 %bo = shl <4 x i16> %shuf, <i16 2, i16 4, i16 1, i16 2> 689 ret <4 x i16> %bo 690} 691 692define <4 x i16> @widening_shuffle_or(<2 x i16> %v) { 693; CHECK-LABEL: @widening_shuffle_or( 694; CHECK-NEXT: [[TMP1:%.*]] = or <2 x i16> [[V:%.*]], <i16 42, i16 -42> 695; CHECK-NEXT: [[BO:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 696; CHECK-NEXT: ret <4 x i16> [[BO]] 697; 698 %shuf = shufflevector <2 x i16> %v, <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 699 %bo = or <4 x i16> %shuf, <i16 42, i16 -42, i16 -1, i16 -1> 700 ret <4 x i16> %bo 701} 702 703define <4 x i32> @shuffle_17add2(<4 x i32> %v) { 704; CHECK-LABEL: @shuffle_17add2( 705; CHECK-NEXT: [[TMP1:%.*]] = shl <4 x i32> [[V:%.*]], splat (i32 1) 706; CHECK-NEXT: ret <4 x i32> [[TMP1]] 707; 708 %t1 = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 709 %t2 = add <4 x i32> %t1, %t1 710 %r = shufflevector <4 x i32> %t2, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 711 ret <4 x i32> %r 712} 713 714define <4 x i32> @shuffle_17mulsplat(<4 x i32> %v) { 715; CHECK-LABEL: @shuffle_17mulsplat( 716; CHECK-NEXT: [[TMP1:%.*]] = mul <4 x i32> [[V:%.*]], [[V]] 717; CHECK-NEXT: [[M1:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer 718; CHECK-NEXT: ret <4 x i32> [[M1]] 719; 720 %s1 = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer 721 %m1 = mul <4 x i32> %s1, %s1 722 %s2 = shufflevector <4 x i32> %m1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 723 ret <4 x i32> %s2 724} 725 726; Do not reorder shuffle and binop if LHS of shuffles are of different size 727define <2 x i32> @pr19717(<4 x i32> %in0, <2 x i32> %in1) { 728; CHECK-LABEL: @pr19717( 729; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[IN0:%.*]], <4 x i32> poison, <2 x i32> zeroinitializer 730; CHECK-NEXT: [[SHUFFLE4:%.*]] = shufflevector <2 x i32> [[IN1:%.*]], <2 x i32> poison, <2 x i32> zeroinitializer 731; CHECK-NEXT: [[MUL:%.*]] = mul <2 x i32> [[SHUFFLE]], [[SHUFFLE4]] 732; CHECK-NEXT: ret <2 x i32> [[MUL]] 733; 734 %shuffle = shufflevector <4 x i32> %in0, <4 x i32> %in0, <2 x i32> zeroinitializer 735 %shuffle4 = shufflevector <2 x i32> %in1, <2 x i32> %in1, <2 x i32> zeroinitializer 736 %mul = mul <2 x i32> %shuffle, %shuffle4 737 ret <2 x i32> %mul 738} 739 740define <4 x i16> @pr19717a(<8 x i16> %in0, <8 x i16> %in1) { 741; CHECK-LABEL: @pr19717a( 742; CHECK-NEXT: [[TMP1:%.*]] = mul <8 x i16> [[IN0:%.*]], [[IN1:%.*]] 743; CHECK-NEXT: [[MUL:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> poison, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 744; CHECK-NEXT: ret <4 x i16> [[MUL]] 745; 746 %shuffle = shufflevector <8 x i16> %in0, <8 x i16> %in0, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 747 %shuffle1 = shufflevector <8 x i16> %in1, <8 x i16> %in1, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 748 %mul = mul <4 x i16> %shuffle, %shuffle1 749 ret <4 x i16> %mul 750} 751 752define <8 x i8> @pr19730(<16 x i8> %in0) { 753; CHECK-LABEL: @pr19730( 754; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <16 x i8> [[IN0:%.*]], <16 x i8> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> 755; CHECK-NEXT: [[SHUFFLE1:%.*]] = shufflevector <8 x i8> [[SHUFFLE]], <8 x i8> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> 756; CHECK-NEXT: ret <8 x i8> [[SHUFFLE1]] 757; 758 %shuffle = shufflevector <16 x i8> %in0, <16 x i8> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> 759 %shuffle1 = shufflevector <8 x i8> %shuffle, <8 x i8> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> 760 ret <8 x i8> %shuffle1 761} 762 763define i32 @pr19737(<4 x i32> %in0) { 764; CHECK-LABEL: @pr19737( 765; CHECK-NEXT: [[TMP1:%.*]] = extractelement <4 x i32> [[IN0:%.*]], i64 0 766; CHECK-NEXT: ret i32 [[TMP1]] 767; 768 %shuffle.i = shufflevector <4 x i32> zeroinitializer, <4 x i32> %in0, <4 x i32> <i32 0, i32 4, i32 2, i32 6> 769 %neg.i = xor <4 x i32> %shuffle.i, <i32 -1, i32 -1, i32 -1, i32 -1> 770 %and.i = and <4 x i32> %in0, %neg.i 771 %rv = extractelement <4 x i32> %and.i, i32 0 772 ret i32 %rv 773} 774 775; In PR20059 ( http://llvm.org/pr20059 ), shufflevector operations are reordered/removed 776; for an srem operation. This is not a valid optimization because it may cause a trap 777; on div-by-zero. 778 779define <4 x i32> @pr20059(<4 x i32> %p1, <4 x i32> %p2) { 780; CHECK-LABEL: @pr20059( 781; CHECK-NEXT: [[SPLAT1:%.*]] = shufflevector <4 x i32> [[P1:%.*]], <4 x i32> poison, <4 x i32> zeroinitializer 782; CHECK-NEXT: [[SPLAT2:%.*]] = shufflevector <4 x i32> [[P2:%.*]], <4 x i32> poison, <4 x i32> zeroinitializer 783; CHECK-NEXT: [[RETVAL:%.*]] = srem <4 x i32> [[SPLAT1]], [[SPLAT2]] 784; CHECK-NEXT: ret <4 x i32> [[RETVAL]] 785; 786 %splat1 = shufflevector <4 x i32> %p1, <4 x i32> poison, <4 x i32> zeroinitializer 787 %splat2 = shufflevector <4 x i32> %p2, <4 x i32> poison, <4 x i32> zeroinitializer 788 %retval = srem <4 x i32> %splat1, %splat2 789 ret <4 x i32> %retval 790} 791 792define <4 x i32> @pr20114(<4 x i32> %__mask) { 793; CHECK-LABEL: @pr20114( 794; CHECK-NEXT: [[MASK01_I:%.*]] = shufflevector <4 x i32> [[__MASK:%.*]], <4 x i32> poison, <4 x i32> <i32 0, i32 0, i32 1, i32 1> 795; CHECK-NEXT: [[MASKED_NEW_I_I_I:%.*]] = and <4 x i32> [[MASK01_I]], bitcast (<2 x i64> <i64 ptrtoint (ptr @pr20114 to i64), i64 ptrtoint (ptr @pr20114 to i64)> to <4 x i32>) 796; CHECK-NEXT: ret <4 x i32> [[MASKED_NEW_I_I_I]] 797; 798 %mask01.i = shufflevector <4 x i32> %__mask, <4 x i32> poison, <4 x i32> <i32 0, i32 0, i32 1, i32 1> 799 %masked_new.i.i.i = and <4 x i32> bitcast (<2 x i64> <i64 ptrtoint (ptr @pr20114 to i64), i64 ptrtoint (ptr @pr20114 to i64)> to <4 x i32>), %mask01.i 800 ret <4 x i32> %masked_new.i.i.i 801} 802 803define <2 x ptr> @pr23113(<4 x ptr> %A) { 804; CHECK-LABEL: @pr23113( 805; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x ptr> [[A:%.*]], <4 x ptr> poison, <2 x i32> <i32 0, i32 1> 806; CHECK-NEXT: ret <2 x ptr> [[TMP1]] 807; 808 %1 = shufflevector <4 x ptr> %A, <4 x ptr> poison, <2 x i32> <i32 0, i32 1> 809 ret <2 x ptr> %1 810} 811 812; Unused lanes in the new binop should not kill the entire op (although it may simplify anyway as shown here). 813 814define <2 x i32> @PR37648(<2 x i32> %x) { 815; CHECK-LABEL: @PR37648( 816; CHECK-NEXT: ret <2 x i32> zeroinitializer 817; 818 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 819 %r = urem <2 x i32> %splat, <i32 1, i32 1> 820 ret <2 x i32> %r 821} 822 823; Test shuffle followed by binop with splat constant for all 18 binop opcodes. 824; Test with constant as operand 0 and operand 1 for non-commutative opcodes. 825 826define <2 x i32> @add_splat_constant(<2 x i32> %x) { 827; CHECK-LABEL: @add_splat_constant( 828; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 42, i32 poison> 829; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 830; CHECK-NEXT: ret <2 x i32> [[R]] 831; 832 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 833 %r = add <2 x i32> %splat, <i32 42, i32 42> 834 ret <2 x i32> %r 835} 836 837define <2 x i32> @sub_splat_constant0(<2 x i32> %x) { 838; CHECK-LABEL: @sub_splat_constant0( 839; CHECK-NEXT: [[TMP1:%.*]] = sub <2 x i32> <i32 42, i32 poison>, [[X:%.*]] 840; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 841; CHECK-NEXT: ret <2 x i32> [[R]] 842; 843 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 844 %r = sub <2 x i32> <i32 42, i32 42>, %splat 845 ret <2 x i32> %r 846} 847 848define <2 x i32> @sub_splat_constant1(<2 x i32> %x) { 849; CHECK-LABEL: @sub_splat_constant1( 850; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 -42, i32 poison> 851; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 852; CHECK-NEXT: ret <2 x i32> [[R]] 853; 854 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 855 %r = sub <2 x i32> %splat, <i32 42, i32 42> 856 ret <2 x i32> %r 857} 858 859define <2 x i32> @mul_splat_constant(<2 x i32> %x) { 860; CHECK-LABEL: @mul_splat_constant( 861; CHECK-NEXT: [[TMP1:%.*]] = mul <2 x i32> [[X:%.*]], <i32 42, i32 poison> 862; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 863; CHECK-NEXT: ret <2 x i32> [[R]] 864; 865 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 866 %r = mul <2 x i32> %splat, <i32 42, i32 42> 867 ret <2 x i32> %r 868} 869 870define <2 x i32> @shl_splat_constant0(<2 x i32> %x) { 871; CHECK-LABEL: @shl_splat_constant0( 872; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> <i32 5, i32 poison>, [[X:%.*]] 873; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 874; CHECK-NEXT: ret <2 x i32> [[R]] 875; 876 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 877 %r = shl <2 x i32> <i32 5, i32 5>, %splat 878 ret <2 x i32> %r 879} 880 881define <2 x i32> @shl_splat_constant1(<2 x i32> %x) { 882; CHECK-LABEL: @shl_splat_constant1( 883; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> [[X:%.*]], <i32 5, i32 0> 884; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 885; CHECK-NEXT: ret <2 x i32> [[R]] 886; 887 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 888 %r = shl <2 x i32> %splat, <i32 5, i32 5> 889 ret <2 x i32> %r 890} 891 892define <2 x i32> @ashr_splat_constant0(<2 x i32> %x) { 893; CHECK-LABEL: @ashr_splat_constant0( 894; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i32> <i32 5, i32 poison>, [[X:%.*]] 895; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 896; CHECK-NEXT: ret <2 x i32> [[R]] 897; 898 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 899 %r = ashr <2 x i32> <i32 5, i32 5>, %splat 900 ret <2 x i32> %r 901} 902 903define <2 x i32> @ashr_splat_constant1(<2 x i32> %x) { 904; CHECK-LABEL: @ashr_splat_constant1( 905; CHECK-NEXT: [[TMP1:%.*]] = ashr <2 x i32> [[X:%.*]], <i32 5, i32 0> 906; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 907; CHECK-NEXT: ret <2 x i32> [[R]] 908; 909 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 910 %r = ashr <2 x i32> %splat, <i32 5, i32 5> 911 ret <2 x i32> %r 912} 913 914define <2 x i32> @lshr_splat_constant0(<2 x i32> %x) { 915; CHECK-LABEL: @lshr_splat_constant0( 916; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i32> <i32 5, i32 poison>, [[X:%.*]] 917; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 918; CHECK-NEXT: ret <2 x i32> [[R]] 919; 920 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 921 %r = lshr <2 x i32> <i32 5, i32 5>, %splat 922 ret <2 x i32> %r 923} 924 925define <2 x i32> @lshr_splat_constant1(<2 x i32> %x) { 926; CHECK-LABEL: @lshr_splat_constant1( 927; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i32> [[X:%.*]], <i32 5, i32 0> 928; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 929; CHECK-NEXT: ret <2 x i32> [[R]] 930; 931 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 932 %r = lshr <2 x i32> %splat, <i32 5, i32 5> 933 ret <2 x i32> %r 934} 935 936define <2 x i32> @urem_splat_constant0(<2 x i32> %x) { 937; CHECK-LABEL: @urem_splat_constant0( 938; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> poison, <2 x i32> zeroinitializer 939; CHECK-NEXT: [[R:%.*]] = urem <2 x i32> splat (i32 42), [[SPLAT]] 940; CHECK-NEXT: ret <2 x i32> [[R]] 941; 942 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 943 %r = urem <2 x i32> <i32 42, i32 42>, %splat 944 ret <2 x i32> %r 945} 946 947define <2 x i32> @urem_splat_constant1(<2 x i32> %x) { 948; CHECK-LABEL: @urem_splat_constant1( 949; CHECK-NEXT: [[TMP1:%.*]] = urem <2 x i32> [[X:%.*]], <i32 42, i32 1> 950; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 951; CHECK-NEXT: ret <2 x i32> [[R]] 952; 953 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 954 %r = urem <2 x i32> %splat, <i32 42, i32 42> 955 ret <2 x i32> %r 956} 957 958define <2 x i32> @srem_splat_constant0(<2 x i32> %x) { 959; CHECK-LABEL: @srem_splat_constant0( 960; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> poison, <2 x i32> zeroinitializer 961; CHECK-NEXT: [[R:%.*]] = srem <2 x i32> splat (i32 42), [[SPLAT]] 962; CHECK-NEXT: ret <2 x i32> [[R]] 963; 964 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 965 %r = srem <2 x i32> <i32 42, i32 42>, %splat 966 ret <2 x i32> %r 967} 968 969define <2 x i32> @srem_splat_constant1(<2 x i32> %x) { 970; CHECK-LABEL: @srem_splat_constant1( 971; CHECK-NEXT: [[TMP1:%.*]] = srem <2 x i32> [[X:%.*]], <i32 42, i32 1> 972; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 973; CHECK-NEXT: ret <2 x i32> [[R]] 974; 975 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 976 %r = srem <2 x i32> %splat, <i32 42, i32 42> 977 ret <2 x i32> %r 978} 979 980define <2 x i32> @udiv_splat_constant0(<2 x i32> %x) { 981; CHECK-LABEL: @udiv_splat_constant0( 982; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> poison, <2 x i32> zeroinitializer 983; CHECK-NEXT: [[R:%.*]] = udiv <2 x i32> splat (i32 42), [[SPLAT]] 984; CHECK-NEXT: ret <2 x i32> [[R]] 985; 986 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 987 %r = udiv <2 x i32> <i32 42, i32 42>, %splat 988 ret <2 x i32> %r 989} 990 991define <2 x i32> @udiv_splat_constant1(<2 x i32> %x) { 992; CHECK-LABEL: @udiv_splat_constant1( 993; CHECK-NEXT: [[TMP1:%.*]] = udiv <2 x i32> [[X:%.*]], <i32 42, i32 1> 994; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 995; CHECK-NEXT: ret <2 x i32> [[R]] 996; 997 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 998 %r = udiv <2 x i32> %splat, <i32 42, i32 42> 999 ret <2 x i32> %r 1000} 1001 1002define <2 x i32> @sdiv_splat_constant0(<2 x i32> %x) { 1003; CHECK-LABEL: @sdiv_splat_constant0( 1004; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> poison, <2 x i32> zeroinitializer 1005; CHECK-NEXT: [[R:%.*]] = sdiv <2 x i32> splat (i32 42), [[SPLAT]] 1006; CHECK-NEXT: ret <2 x i32> [[R]] 1007; 1008 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 1009 %r = sdiv <2 x i32> <i32 42, i32 42>, %splat 1010 ret <2 x i32> %r 1011} 1012 1013define <2 x i32> @sdiv_splat_constant1(<2 x i32> %x) { 1014; CHECK-LABEL: @sdiv_splat_constant1( 1015; CHECK-NEXT: [[TMP1:%.*]] = sdiv <2 x i32> [[X:%.*]], <i32 42, i32 1> 1016; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 1017; CHECK-NEXT: ret <2 x i32> [[R]] 1018; 1019 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 1020 %r = sdiv <2 x i32> %splat, <i32 42, i32 42> 1021 ret <2 x i32> %r 1022} 1023 1024define <2 x i32> @and_splat_constant(<2 x i32> %x) { 1025; CHECK-LABEL: @and_splat_constant( 1026; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i32> [[X:%.*]], <i32 42, i32 poison> 1027; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 1028; CHECK-NEXT: ret <2 x i32> [[R]] 1029; 1030 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 1031 %r = and <2 x i32> %splat, <i32 42, i32 42> 1032 ret <2 x i32> %r 1033} 1034 1035define <4 x i16> @and_constant_mask_poison(<4 x i16> %add) { 1036; CHECK-LABEL: @and_constant_mask_poison( 1037; CHECK-NEXT: entry: 1038; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[ADD:%.*]], <4 x i16> poison, <4 x i32> <i32 poison, i32 poison, i32 1, i32 1> 1039; CHECK-NEXT: ret <4 x i16> [[AND]] 1040; 1041entry: 1042 %shuffle = shufflevector <4 x i16> %add, <4 x i16> poison, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1043 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 -1, i16 -1> 1044 ret <4 x i16> %and 1045} 1046 1047define <4 x i16> @and_constant_mask_poison_2(<4 x i16> %add) { 1048; CHECK-LABEL: @and_constant_mask_poison_2( 1049; CHECK-NEXT: entry: 1050; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[ADD:%.*]], <4 x i16> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 poison> 1051; CHECK-NEXT: ret <4 x i16> [[AND]] 1052; 1053entry: 1054 %shuffle = shufflevector <4 x i16> %add, <4 x i16> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 undef> 1055 %and = and <4 x i16> %shuffle, <i16 -1, i16 -1, i16 -1, i16 -0> 1056 ret <4 x i16> %and 1057} 1058 1059; We can move the AND across the shuffle, as -1 (AND identity value) is used for undef lanes. 1060define <4 x i16> @and_constant_mask_poison_3(<4 x i16> %add) { 1061; CHECK-LABEL: @and_constant_mask_poison_3( 1062; CHECK-NEXT: entry: 1063; CHECK-NEXT: ret <4 x i16> <i16 0, i16 0, i16 0, i16 undef> 1064; 1065entry: 1066 %shuffle = shufflevector <4 x i16> %add, <4 x i16> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 undef> 1067 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 0, i16 -1> 1068 ret <4 x i16> %and 1069} 1070 1071; We can move the AND across the shuffle, as -1 (AND identity value) is used for undef lanes. 1072define <4 x i16> @and_constant_mask_poison_4(<4 x i16> %add) { 1073; CHECK-LABEL: @and_constant_mask_poison_4( 1074; CHECK-NEXT: entry: 1075; CHECK-NEXT: [[TMP0:%.*]] = and <4 x i16> [[ADD:%.*]], <i16 9, i16 20, i16 poison, i16 poison> 1076; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 poison> 1077; CHECK-NEXT: ret <4 x i16> [[AND]] 1078; 1079entry: 1080 %shuffle = shufflevector <4 x i16> %add, <4 x i16> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 undef> 1081 %and = and <4 x i16> %shuffle, <i16 9, i16 20, i16 20, i16 -1> 1082 ret <4 x i16> %and 1083} 1084 1085define <4 x i16> @and_constant_mask_not_undef(<4 x i16> %add) { 1086; CHECK-LABEL: @and_constant_mask_not_undef( 1087; CHECK-NEXT: entry: 1088; CHECK-NEXT: [[TMP0:%.*]] = and <4 x i16> [[ADD:%.*]], <i16 poison, i16 -1, i16 0, i16 0> 1089; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1090; CHECK-NEXT: ret <4 x i16> [[AND]] 1091; 1092entry: 1093 %shuffle = shufflevector <4 x i16> %add, <4 x i16> poison, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1094 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 -1, i16 -1> 1095 ret <4 x i16> %and 1096} 1097 1098define <4 x i16> @or_constant_mask_poison(<4 x i16> %in) { 1099; CHECK-LABEL: @or_constant_mask_poison( 1100; CHECK-NEXT: entry: 1101; CHECK-NEXT: [[OR:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> poison, <4 x i32> <i32 poison, i32 poison, i32 1, i32 1> 1102; CHECK-NEXT: ret <4 x i16> [[OR]] 1103; 1104entry: 1105 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1106 %or = or <4 x i16> %shuffle, <i16 -1, i16 -1, i16 0, i16 0> 1107 ret <4 x i16> %or 1108} 1109 1110define <4 x i16> @or_constant_mask_poison_2(<4 x i16> %in) { 1111; CHECK-LABEL: @or_constant_mask_poison_2( 1112; CHECK-NEXT: entry: 1113; CHECK-NEXT: [[OR:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> poison, <4 x i32> <i32 poison, i32 1, i32 1, i32 poison> 1114; CHECK-NEXT: ret <4 x i16> [[OR]] 1115; 1116entry: 1117 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef> 1118 %or = or <4 x i16> %shuffle, <i16 -1, i16 0, i16 0, i16 -1> 1119 ret <4 x i16> %or 1120} 1121 1122; We can move the OR across the shuffle, as 0 (OR identity value) is used for undef lanes. 1123define <4 x i16> @or_constant_mask_poison_3(<4 x i16> %in) { 1124; CHECK-LABEL: @or_constant_mask_poison_3( 1125; CHECK-NEXT: entry: 1126; CHECK-NEXT: ret <4 x i16> <i16 undef, i16 -1, i16 -1, i16 undef> 1127; 1128entry: 1129 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef> 1130 %or = or <4 x i16> %shuffle, <i16 0, i16 -1, i16 -1, i16 0> 1131 ret <4 x i16> %or 1132} 1133 1134; We can move the OR across the shuffle, as 0 (OR identity value) is used for undef lanes. 1135define <4 x i16> @or_constant_mask_poison_4(<4 x i16> %in) { 1136; CHECK-LABEL: @or_constant_mask_poison_4( 1137; CHECK-NEXT: entry: 1138; CHECK-NEXT: [[TMP0:%.*]] = or <4 x i16> [[IN:%.*]], <i16 poison, i16 99, i16 poison, i16 poison> 1139; CHECK-NEXT: [[OR:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 poison, i32 1, i32 1, i32 poison> 1140; CHECK-NEXT: ret <4 x i16> [[OR]] 1141; 1142entry: 1143 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef> 1144 %or = or <4 x i16> %shuffle, <i16 0, i16 99, i16 99, i16 0> 1145 ret <4 x i16> %or 1146} 1147 1148define <4 x i16> @or_constant_mask_not_undef(<4 x i16> %in) { 1149; CHECK-LABEL: @or_constant_mask_not_undef( 1150; CHECK-NEXT: entry: 1151; CHECK-NEXT: [[TMP0:%.*]] = or <4 x i16> [[IN:%.*]], <i16 poison, i16 -1, i16 0, i16 0> 1152; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1153; CHECK-NEXT: ret <4 x i16> [[AND]] 1154; 1155entry: 1156 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1157 %and = or <4 x i16> %shuffle, <i16 0, i16 0, i16 -1, i16 -1> 1158 ret <4 x i16> %and 1159} 1160 1161define <4 x i16> @shl_constant_mask_undef(<4 x i16> %in) { 1162; CHECK-LABEL: @shl_constant_mask_undef( 1163; CHECK-NEXT: entry: 1164; CHECK-NEXT: [[TMP0:%.*]] = shl <4 x i16> [[IN:%.*]], <i16 10, i16 0, i16 0, i16 0> 1165; CHECK-NEXT: [[SHL:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 0, i32 poison, i32 1, i32 1> 1166; CHECK-NEXT: ret <4 x i16> [[SHL]] 1167; 1168entry: 1169 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 0, i32 undef, i32 1, i32 1> 1170 %shl = shl <4 x i16> %shuffle, <i16 10, i16 3, i16 0, i16 0> 1171 ret <4 x i16> %shl 1172} 1173 1174define <4 x i16> @add_constant_mask_undef(<4 x i16> %in) { 1175; CHECK-LABEL: @add_constant_mask_undef( 1176; CHECK-NEXT: entry: 1177; CHECK-NEXT: [[ADD:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> poison, <4 x i32> <i32 poison, i32 poison, i32 1, i32 1> 1178; CHECK-NEXT: ret <4 x i16> [[ADD]] 1179; 1180entry: 1181 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1182 %add = add <4 x i16> %shuffle, <i16 10, i16 3, i16 0, i16 0> 1183 ret <4 x i16> %add 1184} 1185 1186define <4 x i16> @add_constant_mask_undef_2(<4 x i16> %in) { 1187; CHECK-LABEL: @add_constant_mask_undef_2( 1188; CHECK-NEXT: entry: 1189; CHECK-NEXT: [[TMP0:%.*]] = add <4 x i16> [[IN:%.*]], <i16 poison, i16 0, i16 3, i16 poison> 1190; CHECK-NEXT: [[ADD:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 poison, i32 2, i32 1, i32 1> 1191; CHECK-NEXT: ret <4 x i16> [[ADD]] 1192; 1193entry: 1194 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 undef, i32 2, i32 1, i32 1> 1195 %add = add <4 x i16> %shuffle, <i16 10, i16 3, i16 0, i16 0> 1196 ret <4 x i16> %add 1197} 1198 1199define <4 x i16> @sub_constant_mask_undef(<4 x i16> %in) { 1200; CHECK-LABEL: @sub_constant_mask_undef( 1201; CHECK-NEXT: entry: 1202; CHECK-NEXT: [[SUB:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> poison, <4 x i32> <i32 poison, i32 poison, i32 1, i32 1> 1203; CHECK-NEXT: ret <4 x i16> [[SUB]] 1204; 1205entry: 1206 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1207 %sub = sub <4 x i16> %shuffle, <i16 10, i16 3, i16 0, i16 0> 1208 ret <4 x i16> %sub 1209} 1210 1211define <4 x i16> @sub_constant_mask_undef_2(<4 x i16> %in) { 1212; CHECK-LABEL: @sub_constant_mask_undef_2( 1213; CHECK-NEXT: entry: 1214; CHECK-NEXT: [[TMP0:%.*]] = add <4 x i16> [[IN:%.*]], <i16 poison, i16 0, i16 -10, i16 poison> 1215; CHECK-NEXT: [[SUB:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 1, i32 1, i32 2, i32 poison> 1216; CHECK-NEXT: ret <4 x i16> [[SUB]] 1217; 1218entry: 1219 %shuffle = shufflevector <4 x i16> %in, <4 x i16> poison, <4 x i32> <i32 1, i32 1, i32 2, i32 undef> 1220 %sub = sub <4 x i16> %shuffle, <i16 0, i16 0, i16 10, i16 99> 1221 ret <4 x i16> %sub 1222} 1223 1224define <2 x i32> @or_splat_constant(<2 x i32> %x) { 1225; CHECK-LABEL: @or_splat_constant( 1226; CHECK-NEXT: [[TMP1:%.*]] = or <2 x i32> [[X:%.*]], <i32 42, i32 poison> 1227; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 1228; CHECK-NEXT: ret <2 x i32> [[R]] 1229; 1230 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 1231 %r = or <2 x i32> %splat, <i32 42, i32 42> 1232 ret <2 x i32> %r 1233} 1234 1235define <2 x i32> @xor_splat_constant(<2 x i32> %x) { 1236; CHECK-LABEL: @xor_splat_constant( 1237; CHECK-NEXT: [[TMP1:%.*]] = xor <2 x i32> [[X:%.*]], <i32 42, i32 poison> 1238; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 1239; CHECK-NEXT: ret <2 x i32> [[R]] 1240; 1241 %splat = shufflevector <2 x i32> %x, <2 x i32> poison, <2 x i32> zeroinitializer 1242 %r = xor <2 x i32> %splat, <i32 42, i32 42> 1243 ret <2 x i32> %r 1244} 1245 1246define <2 x float> @fadd_splat_constant(<2 x float> %x) { 1247; CHECK-LABEL: @fadd_splat_constant( 1248; CHECK-NEXT: [[TMP1:%.*]] = fadd <2 x float> [[X:%.*]], <float 4.200000e+01, float poison> 1249; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer 1250; CHECK-NEXT: ret <2 x float> [[R]] 1251; 1252 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1253 %r = fadd <2 x float> %splat, <float 42.0, float 42.0> 1254 ret <2 x float> %r 1255} 1256 1257define <2 x float> @fsub_splat_constant0(<2 x float> %x) { 1258; CHECK-LABEL: @fsub_splat_constant0( 1259; CHECK-NEXT: [[TMP1:%.*]] = fsub <2 x float> <float 4.200000e+01, float poison>, [[X:%.*]] 1260; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer 1261; CHECK-NEXT: ret <2 x float> [[R]] 1262; 1263 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1264 %r = fsub <2 x float> <float 42.0, float 42.0>, %splat 1265 ret <2 x float> %r 1266} 1267 1268define <2 x float> @fsub_splat_constant1(<2 x float> %x) { 1269; CHECK-LABEL: @fsub_splat_constant1( 1270; CHECK-NEXT: [[TMP1:%.*]] = fadd <2 x float> [[X:%.*]], <float -4.200000e+01, float poison> 1271; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer 1272; CHECK-NEXT: ret <2 x float> [[R]] 1273; 1274 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1275 %r = fsub <2 x float> %splat, <float 42.0, float 42.0> 1276 ret <2 x float> %r 1277} 1278 1279define <2 x float> @fneg(<2 x float> %x) { 1280; CHECK-LABEL: @fneg( 1281; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> poison, <2 x i32> zeroinitializer 1282; CHECK-NEXT: [[R:%.*]] = fneg <2 x float> [[TMP1]] 1283; CHECK-NEXT: ret <2 x float> [[R]] 1284; 1285 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1286 %r = fsub <2 x float> <float -0.0, float -0.0>, %splat 1287 ret <2 x float> %r 1288} 1289 1290define <2 x float> @fmul_splat_constant(<2 x float> %x) { 1291; CHECK-LABEL: @fmul_splat_constant( 1292; CHECK-NEXT: [[TMP1:%.*]] = fmul <2 x float> [[X:%.*]], <float 4.200000e+01, float poison> 1293; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer 1294; CHECK-NEXT: ret <2 x float> [[R]] 1295; 1296 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1297 %r = fmul <2 x float> %splat, <float 42.0, float 42.0> 1298 ret <2 x float> %r 1299} 1300 1301define <2 x float> @fdiv_splat_constant0(<2 x float> %x) { 1302; CHECK-LABEL: @fdiv_splat_constant0( 1303; CHECK-NEXT: [[TMP1:%.*]] = fdiv <2 x float> <float 4.200000e+01, float poison>, [[X:%.*]] 1304; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer 1305; CHECK-NEXT: ret <2 x float> [[R]] 1306; 1307 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1308 %r = fdiv <2 x float> <float 42.0, float 42.0>, %splat 1309 ret <2 x float> %r 1310} 1311 1312define <2 x float> @fdiv_splat_constant1(<2 x float> %x) { 1313; CHECK-LABEL: @fdiv_splat_constant1( 1314; CHECK-NEXT: [[TMP1:%.*]] = fdiv <2 x float> [[X:%.*]], <float 4.200000e+01, float poison> 1315; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer 1316; CHECK-NEXT: ret <2 x float> [[R]] 1317; 1318 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1319 %r = fdiv <2 x float> %splat, <float 42.0, float 42.0> 1320 ret <2 x float> %r 1321} 1322 1323define <2 x float> @frem_splat_constant0(<2 x float> %x) { 1324; CHECK-LABEL: @frem_splat_constant0( 1325; CHECK-NEXT: [[TMP1:%.*]] = frem <2 x float> <float 4.200000e+01, float poison>, [[X:%.*]] 1326; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer 1327; CHECK-NEXT: ret <2 x float> [[R]] 1328; 1329 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1330 %r = frem <2 x float> <float 42.0, float 42.0>, %splat 1331 ret <2 x float> %r 1332} 1333 1334define <2 x float> @frem_splat_constant1(<2 x float> %x) { 1335; CHECK-LABEL: @frem_splat_constant1( 1336; CHECK-NEXT: [[TMP1:%.*]] = frem <2 x float> [[X:%.*]], <float 4.200000e+01, float poison> 1337; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer 1338; CHECK-NEXT: ret <2 x float> [[R]] 1339; 1340 %splat = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1341 %r = frem <2 x float> %splat, <float 42.0, float 42.0> 1342 ret <2 x float> %r 1343} 1344 1345; Equivalent shuffle masks, but only one is a narrowing op. 1346 1347define <2 x i1> @PR40734(<1 x i1> %x, <4 x i1> %y) { 1348; CHECK-LABEL: @PR40734( 1349; CHECK-NEXT: [[WIDEN:%.*]] = shufflevector <1 x i1> zeroinitializer, <1 x i1> [[X:%.*]], <2 x i32> <i32 0, i32 1> 1350; CHECK-NEXT: [[NARROW:%.*]] = shufflevector <4 x i1> [[Y:%.*]], <4 x i1> poison, <2 x i32> <i32 0, i32 1> 1351; CHECK-NEXT: [[R:%.*]] = and <2 x i1> [[WIDEN]], [[NARROW]] 1352; CHECK-NEXT: ret <2 x i1> [[R]] 1353; 1354 %widen = shufflevector <1 x i1> zeroinitializer, <1 x i1> %x, <2 x i32> <i32 0, i32 1> 1355 %narrow = shufflevector <4 x i1> %y, <4 x i1> poison, <2 x i32> <i32 0, i32 1> 1356 %r = and <2 x i1> %widen, %narrow 1357 ret <2 x i1> %r 1358} 1359 1360; Negative test - do not transform non-power-of-2 unless we know the backend handles these sequences identically. 1361 1362define <7 x i8> @insert_subvector_shuffles(<3 x i8> %x, <3 x i8> %y) { 1363; CHECK-LABEL: @insert_subvector_shuffles( 1364; CHECK-NEXT: [[S1:%.*]] = shufflevector <3 x i8> [[X:%.*]], <3 x i8> poison, <7 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison> 1365; CHECK-NEXT: [[S2:%.*]] = shufflevector <3 x i8> [[Y:%.*]], <3 x i8> poison, <7 x i32> <i32 poison, i32 1, i32 2, i32 poison, i32 poison, i32 poison, i32 poison> 1366; CHECK-NEXT: [[S3:%.*]] = shufflevector <7 x i8> [[S1]], <7 x i8> [[S2]], <7 x i32> <i32 0, i32 8, i32 1, i32 poison, i32 8, i32 1, i32 9> 1367; CHECK-NEXT: ret <7 x i8> [[S3]] 1368; 1369 %s1 = shufflevector <3 x i8> %x, <3 x i8> poison, <7 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 1370 %s2 = shufflevector <3 x i8> %y, <3 x i8> poison, <7 x i32> <i32 undef, i32 1, i32 2, i32 undef, i32 undef, i32 undef, i32 undef> 1371 %s3 = shufflevector <7 x i8> %s1, <7 x i8> %s2, <7 x i32> <i32 0, i32 8, i32 1, i32 undef, i32 8, i32 1, i32 9> 1372 ret <7 x i8> %s3 1373} 1374 1375define <8 x i8> @insert_subvector_shuffles_pow2elts(<2 x i8> %x, <2 x i8> %y) { 1376; CHECK-LABEL: @insert_subvector_shuffles_pow2elts( 1377; CHECK-NEXT: [[S3:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <8 x i32> <i32 0, i32 2, i32 1, i32 poison, i32 2, i32 1, i32 3, i32 0> 1378; CHECK-NEXT: ret <8 x i8> [[S3]] 1379; 1380 %s1 = shufflevector <2 x i8> %x, <2 x i8> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 1381 %s2 = shufflevector <2 x i8> %y, <2 x i8> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 1382 %s3 = shufflevector <8 x i8> %s1, <8 x i8> %s2, <8 x i32> <i32 0, i32 8, i32 1, i32 undef, i32 8, i32 1, i32 9, i32 0> 1383 ret <8 x i8> %s3 1384} 1385 1386; The last shuffle may change the vector type. 1387; Negative test - do not transform non-power-of-2 unless we know the backend handles these sequences identically. 1388 1389define <2 x i8> @insert_subvector_shuffles_narrowing(<3 x i8> %x, <3 x i8> %y) { 1390; CHECK-LABEL: @insert_subvector_shuffles_narrowing( 1391; CHECK-NEXT: [[S1:%.*]] = shufflevector <3 x i8> [[X:%.*]], <3 x i8> poison, <7 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison> 1392; CHECK-NEXT: [[S2:%.*]] = shufflevector <3 x i8> [[Y:%.*]], <3 x i8> poison, <7 x i32> <i32 poison, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison> 1393; CHECK-NEXT: [[S3:%.*]] = shufflevector <7 x i8> [[S1]], <7 x i8> [[S2]], <2 x i32> <i32 0, i32 8> 1394; CHECK-NEXT: ret <2 x i8> [[S3]] 1395; 1396 %s1 = shufflevector <3 x i8> %x, <3 x i8> poison, <7 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 1397 %s2 = shufflevector <3 x i8> %y, <3 x i8> poison, <7 x i32> <i32 undef, i32 1, i32 2, i32 undef, i32 undef, i32 undef, i32 undef> 1398 %s3 = shufflevector <7 x i8> %s1, <7 x i8> %s2, <2 x i32> <i32 0, i32 8> 1399 ret <2 x i8> %s3 1400} 1401 1402define <2 x i8> @insert_subvector_shuffles_narrowing_pow2elts(<4 x i8> %x, <4 x i8> %y) { 1403; CHECK-LABEL: @insert_subvector_shuffles_narrowing_pow2elts( 1404; CHECK-NEXT: [[S3:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]], <2 x i32> <i32 0, i32 4> 1405; CHECK-NEXT: ret <2 x i8> [[S3]] 1406; 1407 %s1 = shufflevector <4 x i8> %x, <4 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 1408 %s2 = shufflevector <4 x i8> %y, <4 x i8> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 3, i32 undef, i32 undef, i32 undef, i32 undef> 1409 %s3 = shufflevector <8 x i8> %s1, <8 x i8> %s2, <2 x i32> <i32 0, i32 8> 1410 ret <2 x i8> %s3 1411} 1412 1413; Similar to above, but this reduces to a widen with undefs of 'x'. 1414 1415define <4 x double> @insert_subvector_shuffles_identity(<2 x double> %x) { 1416; CHECK-LABEL: @insert_subvector_shuffles_identity( 1417; CHECK-NEXT: [[S3:%.*]] = shufflevector <2 x double> [[X:%.*]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 1418; CHECK-NEXT: ret <4 x double> [[S3]] 1419; 1420 %s1 = shufflevector <2 x double> %x, <2 x double> poison, <4 x i32> <i32 undef, i32 1, i32 undef, i32 undef> 1421 %s2 = shufflevector <2 x double> %x, <2 x double> poison, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef> 1422 %s3 = shufflevector <4 x double> %s2, <4 x double> %s1, <4 x i32> <i32 0, i32 5, i32 undef, i32 undef> 1423 ret <4 x double> %s3 1424} 1425 1426; Negative test - not identity with padding (although this could be folded with better analysis). 1427 1428define <4 x double> @not_insert_subvector_shuffle(<2 x double> %x) { 1429; CHECK-LABEL: @not_insert_subvector_shuffle( 1430; CHECK-NEXT: [[S1:%.*]] = shufflevector <2 x double> [[X:%.*]], <2 x double> poison, <4 x i32> <i32 poison, i32 1, i32 poison, i32 1> 1431; CHECK-NEXT: [[S2:%.*]] = shufflevector <2 x double> [[X]], <2 x double> poison, <4 x i32> <i32 0, i32 poison, i32 poison, i32 poison> 1432; CHECK-NEXT: [[S3:%.*]] = shufflevector <4 x double> [[S2]], <4 x double> [[S1]], <4 x i32> <i32 0, i32 5, i32 7, i32 poison> 1433; CHECK-NEXT: ret <4 x double> [[S3]] 1434; 1435 %s1 = shufflevector <2 x double> %x, <2 x double> poison, <4 x i32> <i32 undef, i32 1, i32 undef, i32 1> 1436 %s2 = shufflevector <2 x double> %x, <2 x double> poison, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef> 1437 %s3 = shufflevector <4 x double> %s2, <4 x double> %s1, <4 x i32> <i32 0, i32 5, i32 7, i32 undef> 1438 ret <4 x double> %s3 1439} 1440 1441; Negative test - operands are not the same size (although this could be partly folded with better analysis). 1442 1443define <4 x double> @not_insert_subvector_shuffles_with_same_size(<2 x double> %x, <3 x double> %y) { 1444; CHECK-LABEL: @not_insert_subvector_shuffles_with_same_size( 1445; CHECK-NEXT: [[S1:%.*]] = shufflevector <2 x double> [[X:%.*]], <2 x double> poison, <4 x i32> <i32 poison, i32 1, i32 poison, i32 poison> 1446; CHECK-NEXT: [[S2:%.*]] = shufflevector <3 x double> [[Y:%.*]], <3 x double> poison, <4 x i32> <i32 0, i32 poison, i32 poison, i32 poison> 1447; CHECK-NEXT: [[S3:%.*]] = shufflevector <4 x double> [[S2]], <4 x double> [[S1]], <4 x i32> <i32 0, i32 5, i32 poison, i32 poison> 1448; CHECK-NEXT: ret <4 x double> [[S3]] 1449; 1450 %s1 = shufflevector <2 x double> %x, <2 x double> poison, <4 x i32> <i32 undef, i32 1, i32 undef, i32 undef> 1451 %s2 = shufflevector <3 x double> %y, <3 x double> poison, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef> 1452 %s3 = shufflevector <4 x double> %s2, <4 x double> %s1, <4 x i32> <i32 0, i32 5, i32 undef, i32 undef> 1453 ret <4 x double> %s3 1454} 1455 1456; Demanded vector elements may not be able to simplify a shuffle mask 1457; before we try to narrow it. This used to crash. 1458 1459define <4 x float> @insert_subvector_crash_invalid_mask_elt(<2 x float> %x, ptr %p) { 1460; CHECK-LABEL: @insert_subvector_crash_invalid_mask_elt( 1461; CHECK-NEXT: [[WIDEN:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 1462; CHECK-NEXT: [[I:%.*]] = shufflevector <2 x float> [[X]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 1463; CHECK-NEXT: store <4 x float> [[I]], ptr [[P:%.*]], align 16 1464; CHECK-NEXT: ret <4 x float> [[WIDEN]] 1465; 1466 %widen = shufflevector <2 x float> %x, <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 1467 %ext2 = extractelement <2 x float> %x, i32 0 1468 %I = insertelement <4 x float> %widen, float %ext2, i16 0 1469 store <4 x float> %I, ptr %p 1470 ret <4 x float> %widen 1471} 1472 1473define <4 x i32> @splat_assoc_add(<4 x i32> %x, <4 x i32> %y) { 1474; CHECK-LABEL: @splat_assoc_add( 1475; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[X:%.*]], <i32 317426, i32 poison, i32 poison, i32 poison> 1476; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer 1477; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[TMP2]], [[Y:%.*]] 1478; CHECK-NEXT: ret <4 x i32> [[R]] 1479; 1480 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> zeroinitializer 1481 %a = add <4 x i32> %y, <i32 317426, i32 317426, i32 317426, i32 317426> 1482 %r = add <4 x i32> %splatx, %a 1483 ret <4 x i32> %r 1484} 1485 1486define <vscale x 4 x i32> @vsplat_assoc_add(<vscale x 4 x i32> %x, <vscale x 4 x i32> %y) { 1487; CHECK-LABEL: @vsplat_assoc_add( 1488; CHECK-NEXT: [[TMP1:%.*]] = add <vscale x 4 x i32> [[X:%.*]], splat (i32 317426) 1489; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer 1490; CHECK-NEXT: [[R:%.*]] = add <vscale x 4 x i32> [[TMP2]], [[Y:%.*]] 1491; CHECK-NEXT: ret <vscale x 4 x i32> [[R]] 1492; 1493 1494 %splatx = shufflevector <vscale x 4 x i32> %x, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer 1495 %a = add <vscale x 4 x i32> %y, splat (i32 317426) 1496 %r = add <vscale x 4 x i32> %splatx, %a 1497 ret <vscale x 4 x i32> %r 1498} 1499 1500; Undefs in splat mask are replaced with defined splat index 1501 1502define <4 x i32> @splat_assoc_add_undef_mask_elts(<4 x i32> %x, <4 x i32> %y) { 1503; CHECK-LABEL: @splat_assoc_add_undef_mask_elts( 1504; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[X:%.*]], <i32 42, i32 poison, i32 poison, i32 poison> 1505; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer 1506; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[TMP2]], [[Y:%.*]] 1507; CHECK-NEXT: ret <4 x i32> [[R]] 1508; 1509 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 0, i32 0, i32 undef, i32 undef> 1510 %a = add <4 x i32> %y, <i32 42, i32 42, i32 42, i32 42> 1511 %r = add <4 x i32> %splatx, %a 1512 ret <4 x i32> %r 1513} 1514 1515; Undefs in splat mask are replaced with defined splat index 1516 1517define <4 x i32> @splat_assoc_add_undef_mask_elt_at_splat_index(<4 x i32> %x, <4 x i32> %y) { 1518; CHECK-LABEL: @splat_assoc_add_undef_mask_elt_at_splat_index( 1519; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[X:%.*]], <i32 42, i32 poison, i32 poison, i32 poison> 1520; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer 1521; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[TMP2]], [[Y:%.*]] 1522; CHECK-NEXT: ret <4 x i32> [[R]] 1523; 1524 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 undef, i32 0, i32 0, i32 0> 1525 %a = add <4 x i32> %y, <i32 42, i32 42, i32 42, i32 42> 1526 %r = add <4 x i32> %splatx, %a 1527 ret <4 x i32> %r 1528} 1529 1530define <4 x i32> @splat_assoc_add_undef_constant_elts(<4 x i32> %x, <4 x i32> %y) { 1531; CHECK-LABEL: @splat_assoc_add_undef_constant_elts( 1532; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> zeroinitializer 1533; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 42, i32 undef, i32 undef, i32 42> 1534; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1535; CHECK-NEXT: ret <4 x i32> [[R]] 1536; 1537 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> zeroinitializer 1538 %a = add <4 x i32> %y, <i32 42, i32 undef, i32 undef, i32 42> 1539 %r = add <4 x i32> %splatx, %a 1540 ret <4 x i32> %r 1541} 1542 1543define <4 x i32> @splat_assoc_add_undef_constant_elt_at_splat_index(<4 x i32> %x, <4 x i32> %y) { 1544; CHECK-LABEL: @splat_assoc_add_undef_constant_elt_at_splat_index( 1545; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> zeroinitializer 1546; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 undef, i32 42, i32 undef, i32 42> 1547; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1548; CHECK-NEXT: ret <4 x i32> [[R]] 1549; 1550 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> zeroinitializer 1551 %a = add <4 x i32> %y, <i32 undef, i32 42, i32 undef, i32 42> 1552 %r = add <4 x i32> %splatx, %a 1553 ret <4 x i32> %r 1554} 1555 1556define <4 x i32> @splat_assoc_add_undef_mask_elts_undef_constant_elts(<4 x i32> %x, <4 x i32> %y) { 1557; CHECK-LABEL: @splat_assoc_add_undef_mask_elts_undef_constant_elts( 1558; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 0, i32 poison, i32 0, i32 poison> 1559; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 42, i32 undef, i32 undef, i32 42> 1560; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1561; CHECK-NEXT: ret <4 x i32> [[R]] 1562; 1563 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 0, i32 undef, i32 0, i32 undef> 1564 %a = add <4 x i32> %y, <i32 42, i32 undef, i32 undef, i32 42> 1565 %r = add <4 x i32> %splatx, %a 1566 ret <4 x i32> %r 1567} 1568 1569define <4 x i32> @splat_assoc_add_undef_mask_elt_at_splat_index_undef_constant_elts(<4 x i32> %x, <4 x i32> %y) { 1570; CHECK-LABEL: @splat_assoc_add_undef_mask_elt_at_splat_index_undef_constant_elts( 1571; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 poison, i32 0, i32 0, i32 0> 1572; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 42, i32 undef, i32 undef, i32 42> 1573; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1574; CHECK-NEXT: ret <4 x i32> [[R]] 1575; 1576 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 undef, i32 0, i32 0, i32 0> 1577 %a = add <4 x i32> %y, <i32 42, i32 undef, i32 undef, i32 42> 1578 %r = add <4 x i32> %splatx, %a 1579 ret <4 x i32> %r 1580} 1581 1582define <4 x i32> @splat_assoc_add_undef_mask_elt_at_splat_index_undef_constant_elt_at_splat_index(<4 x i32> %x, <4 x i32> %y) { 1583; CHECK-LABEL: @splat_assoc_add_undef_mask_elt_at_splat_index_undef_constant_elt_at_splat_index( 1584; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 poison, i32 0, i32 0, i32 0> 1585; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 undef, i32 42, i32 undef, i32 42> 1586; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1587; CHECK-NEXT: ret <4 x i32> [[R]] 1588; 1589 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 undef, i32 0, i32 0, i32 0> 1590 %a = add <4 x i32> %y, <i32 undef, i32 42, i32 undef, i32 42> 1591 %r = add <4 x i32> %splatx, %a 1592 ret <4 x i32> %r 1593} 1594 1595; Non-zero splat index; commute operands; FMF intersect 1596 1597define <2 x float> @splat_assoc_fmul(<2 x float> %x, <2 x float> %y) { 1598; CHECK-LABEL: @splat_assoc_fmul( 1599; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc nsz <2 x float> [[X:%.*]], <float poison, float 3.000000e+00> 1600; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> <i32 1, i32 1> 1601; CHECK-NEXT: [[R:%.*]] = fmul reassoc nsz <2 x float> [[TMP2]], [[Y:%.*]] 1602; CHECK-NEXT: ret <2 x float> [[R]] 1603; 1604 %splatx = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> <i32 1, i32 1> 1605 %a = fmul reassoc nsz <2 x float> %y, <float 3.0, float 3.0> 1606 %r = fmul reassoc nsz nnan <2 x float> %a, %splatx 1607 ret <2 x float> %r 1608} 1609 1610; Two splat shuffles; drop poison-generating flags 1611 1612define <3 x i8> @splat_assoc_mul(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1613; CHECK-LABEL: @splat_assoc_mul( 1614; CHECK-NEXT: [[TMP1:%.*]] = mul <3 x i8> [[X:%.*]], [[Z:%.*]] 1615; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <3 x i8> [[TMP1]], <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1616; CHECK-NEXT: [[R:%.*]] = mul <3 x i8> [[TMP2]], [[Y:%.*]] 1617; CHECK-NEXT: ret <3 x i8> [[R]] 1618; 1619 %splatx = shufflevector <3 x i8> %x, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1620 %splatz = shufflevector <3 x i8> %z, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1621 %a = mul nsw <3 x i8> %y, %splatz 1622 %r = mul <3 x i8> %a, %splatx 1623 ret <3 x i8> %r 1624} 1625 1626define <3 x i8> @splat_assoc_mul_undef_elt1(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1627; CHECK-LABEL: @splat_assoc_mul_undef_elt1( 1628; CHECK-NEXT: [[TMP1:%.*]] = mul <3 x i8> [[X:%.*]], [[Z:%.*]] 1629; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <3 x i8> [[TMP1]], <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1630; CHECK-NEXT: [[R:%.*]] = mul <3 x i8> [[TMP2]], [[Y:%.*]] 1631; CHECK-NEXT: ret <3 x i8> [[R]] 1632; 1633 %splatx = shufflevector <3 x i8> %x, <3 x i8> poison, <3 x i32> <i32 undef, i32 2, i32 2> 1634 %splatz = shufflevector <3 x i8> %z, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1635 %a = mul nsw <3 x i8> %y, %splatz 1636 %r = mul nsw nuw <3 x i8> %a, %splatx 1637 ret <3 x i8> %r 1638} 1639 1640define <3 x i8> @splat_assoc_mul_undef_elt2(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1641; CHECK-LABEL: @splat_assoc_mul_undef_elt2( 1642; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <3 x i8> [[X:%.*]], <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1643; CHECK-NEXT: [[SPLATZ:%.*]] = shufflevector <3 x i8> [[Z:%.*]], <3 x i8> poison, <3 x i32> <i32 poison, i32 2, i32 2> 1644; CHECK-NEXT: [[A:%.*]] = mul nsw <3 x i8> [[Y:%.*]], [[SPLATZ]] 1645; CHECK-NEXT: [[R:%.*]] = mul nuw nsw <3 x i8> [[A]], [[SPLATX]] 1646; CHECK-NEXT: ret <3 x i8> [[R]] 1647; 1648 %splatx = shufflevector <3 x i8> %x, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1649 %splatz = shufflevector <3 x i8> %z, <3 x i8> poison, <3 x i32> <i32 undef, i32 2, i32 2> 1650 %a = mul nsw <3 x i8> %y, %splatz 1651 %r = mul nsw nuw <3 x i8> %a, %splatx 1652 ret <3 x i8> %r 1653} 1654 1655define <3 x i8> @splat_assoc_mul_undef_elt_at_splat_index1(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1656; CHECK-LABEL: @splat_assoc_mul_undef_elt_at_splat_index1( 1657; CHECK-NEXT: [[TMP1:%.*]] = mul <3 x i8> [[X:%.*]], [[Z:%.*]] 1658; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <3 x i8> [[TMP1]], <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1659; CHECK-NEXT: [[R:%.*]] = mul <3 x i8> [[TMP2]], [[Y:%.*]] 1660; CHECK-NEXT: ret <3 x i8> [[R]] 1661; 1662 %splatx = shufflevector <3 x i8> %x, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 undef> 1663 %splatz = shufflevector <3 x i8> %z, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1664 %a = mul nsw <3 x i8> %y, %splatz 1665 %r = mul nsw nuw <3 x i8> %a, %splatx 1666 ret <3 x i8> %r 1667} 1668 1669define <3 x i8> @splat_assoc_mul_undef_elt_at_splat_index2(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1670; CHECK-LABEL: @splat_assoc_mul_undef_elt_at_splat_index2( 1671; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <3 x i8> [[X:%.*]], <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1672; CHECK-NEXT: [[SPLATZ:%.*]] = shufflevector <3 x i8> [[Z:%.*]], <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 poison> 1673; CHECK-NEXT: [[A:%.*]] = mul nsw <3 x i8> [[Y:%.*]], [[SPLATZ]] 1674; CHECK-NEXT: [[R:%.*]] = mul nuw nsw <3 x i8> [[A]], [[SPLATX]] 1675; CHECK-NEXT: ret <3 x i8> [[R]] 1676; 1677 %splatx = shufflevector <3 x i8> %x, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1678 %splatz = shufflevector <3 x i8> %z, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 undef> 1679 %a = mul nsw <3 x i8> %y, %splatz 1680 %r = mul nsw nuw <3 x i8> %a, %splatx 1681 ret <3 x i8> %r 1682} 1683 1684; Negative test - mismatched splat elements 1685 1686define <3 x i8> @splat_assoc_or(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1687; CHECK-LABEL: @splat_assoc_or( 1688; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <3 x i8> [[X:%.*]], <3 x i8> poison, <3 x i32> <i32 1, i32 1, i32 1> 1689; CHECK-NEXT: [[SPLATZ:%.*]] = shufflevector <3 x i8> [[Z:%.*]], <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1690; CHECK-NEXT: [[A:%.*]] = or <3 x i8> [[Y:%.*]], [[SPLATZ]] 1691; CHECK-NEXT: [[R:%.*]] = or <3 x i8> [[A]], [[SPLATX]] 1692; CHECK-NEXT: ret <3 x i8> [[R]] 1693; 1694 %splatx = shufflevector <3 x i8> %x, <3 x i8> poison, <3 x i32> <i32 1, i32 1, i32 1> 1695 %splatz = shufflevector <3 x i8> %z, <3 x i8> poison, <3 x i32> <i32 2, i32 2, i32 2> 1696 %a = or <3 x i8> %y, %splatz 1697 %r = or <3 x i8> %a, %splatx 1698 ret <3 x i8> %r 1699} 1700 1701; Negative test - not associative 1702 1703define <2 x float> @splat_assoc_fdiv(<2 x float> %x, <2 x float> %y) { 1704; CHECK-LABEL: @splat_assoc_fdiv( 1705; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> poison, <2 x i32> zeroinitializer 1706; CHECK-NEXT: [[A:%.*]] = fdiv reassoc nsz <2 x float> [[Y:%.*]], splat (float 3.000000e+00) 1707; CHECK-NEXT: [[R:%.*]] = fdiv reassoc nsz <2 x float> [[A]], [[SPLATX]] 1708; CHECK-NEXT: ret <2 x float> [[R]] 1709; 1710 %splatx = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> zeroinitializer 1711 %a = fdiv reassoc nsz <2 x float> %y, <float 3.0, float 3.0> 1712 %r = fdiv reassoc nsz <2 x float> %a, %splatx 1713 ret <2 x float> %r 1714} 1715 1716; Negative test - extra use 1717 1718define <2 x float> @splat_assoc_fadd(<2 x float> %x, <2 x float> %y) { 1719; CHECK-LABEL: @splat_assoc_fadd( 1720; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> poison, <2 x i32> <i32 1, i32 1> 1721; CHECK-NEXT: [[A:%.*]] = fadd fast <2 x float> [[Y:%.*]], splat (float 3.000000e+00) 1722; CHECK-NEXT: call void @use(<2 x float> [[A]]) 1723; CHECK-NEXT: [[R:%.*]] = fadd fast <2 x float> [[A]], [[SPLATX]] 1724; CHECK-NEXT: ret <2 x float> [[R]] 1725; 1726 %splatx = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> <i32 1, i32 1> 1727 %a = fadd fast <2 x float> %y, <float 3.0, float 3.0> 1728 call void @use(<2 x float> %a) 1729 %r = fadd fast <2 x float> %a, %splatx 1730 ret <2 x float> %r 1731} 1732 1733; Negative test - narrowing splat 1734 1735define <3 x i32> @splat_assoc_and(<4 x i32> %x, <3 x i32> %y) { 1736; CHECK-LABEL: @splat_assoc_and( 1737; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <3 x i32> zeroinitializer 1738; CHECK-NEXT: [[A:%.*]] = and <3 x i32> [[Y:%.*]], splat (i32 42) 1739; CHECK-NEXT: [[R:%.*]] = and <3 x i32> [[SPLATX]], [[A]] 1740; CHECK-NEXT: ret <3 x i32> [[R]] 1741; 1742 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <3 x i32> zeroinitializer 1743 %a = and <3 x i32> %y, <i32 42, i32 42, i32 42> 1744 %r = and <3 x i32> %splatx, %a 1745 ret <3 x i32> %r 1746} 1747 1748; Negative test - widening splat 1749 1750define <5 x i32> @splat_assoc_xor(<4 x i32> %x, <5 x i32> %y) { 1751; CHECK-LABEL: @splat_assoc_xor( 1752; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <5 x i32> zeroinitializer 1753; CHECK-NEXT: [[TMP1:%.*]] = xor <5 x i32> [[Y:%.*]], [[SPLATX]] 1754; CHECK-NEXT: [[R:%.*]] = xor <5 x i32> [[TMP1]], splat (i32 42) 1755; CHECK-NEXT: ret <5 x i32> [[R]] 1756; 1757 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <5 x i32> zeroinitializer 1758 %a = xor <5 x i32> %y, <i32 42, i32 42, i32 42, i32 42, i32 42> 1759 %r = xor <5 x i32> %splatx, %a 1760 ret <5 x i32> %r 1761} 1762 1763; Negative test - opcode mismatch 1764 1765define <4 x i32> @splat_assoc_add_mul(<4 x i32> %x, <4 x i32> %y) { 1766; CHECK-LABEL: @splat_assoc_add_mul( 1767; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> zeroinitializer 1768; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], splat (i32 42) 1769; CHECK-NEXT: [[R:%.*]] = mul <4 x i32> [[SPLATX]], [[A]] 1770; CHECK-NEXT: ret <4 x i32> [[R]] 1771; 1772 %splatx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> zeroinitializer 1773 %a = add <4 x i32> %y, <i32 42, i32 42, i32 42, i32 42> 1774 %r = mul <4 x i32> %splatx, %a 1775 ret <4 x i32> %r 1776} 1777 1778 1779; Do not crash on constant expressions. 1780 1781define <4 x i32> @PR46872(<4 x i32> %x) { 1782; CHECK-LABEL: @PR46872( 1783; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 poison, i32 0, i32 1, i32 1> 1784; CHECK-NEXT: [[A:%.*]] = and <4 x i32> [[S]], bitcast (<2 x i64> <i64 ptrtoint (ptr @PR46872 to i64), i64 ptrtoint (ptr @PR46872 to i64)> to <4 x i32>) 1785; CHECK-NEXT: ret <4 x i32> [[A]] 1786; 1787 %s = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 undef, i32 0, i32 1, i32 1> 1788 %a = and <4 x i32> %s, bitcast (<2 x i64> <i64 ptrtoint (ptr @PR46872 to i64), i64 ptrtoint (ptr @PR46872 to i64)> to <4 x i32>) 1789 ret <4 x i32> %a 1790} 1791 1792