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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 79; CHECK-NEXT: ret <4 x float> [[R]] 80; 81 %r = shufflevector <4 x float> %x, <4 x float> undef, <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 undef, 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> undef, float %t4, i32 0 95 %t130 = insertelement <4 x float> %t128, float undef, 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; Test fold of two shuffles where the first shuffle vectors inputs are a 102; different length then the second. 103define <4 x i8> @test9(<16 x i8> %t6) { 104; CHECK-LABEL: @test9( 105; CHECK-NEXT: [[T9:%.*]] = shufflevector <16 x i8> [[T6:%.*]], <16 x i8> undef, <4 x i32> <i32 13, i32 9, i32 4, i32 13> 106; CHECK-NEXT: ret <4 x i8> [[T9]] 107; 108 %t7 = shufflevector <16 x i8> %t6, <16 x i8> undef, <4 x i32> < i32 13, i32 9, i32 4, i32 13 > 109 %t9 = shufflevector <4 x i8> %t7, <4 x i8> undef, <4 x i32> < i32 3, i32 1, i32 2, i32 0 > 110 ret <4 x i8> %t9 111} 112 113; Same as test9, but make sure that "undef" mask values are not confused with 114; mask values of 2*N, where N is the mask length. These shuffles should not 115; be folded (because [8,9,4,8] may not be a mask supported by the target). 116 117define <4 x i8> @test9a(<16 x i8> %t6) { 118; CHECK-LABEL: @test9a( 119; CHECK-NEXT: [[T7:%.*]] = shufflevector <16 x i8> [[T6:%.*]], <16 x i8> undef, <4 x i32> <i32 undef, i32 9, i32 4, i32 8> 120; CHECK-NEXT: [[T9:%.*]] = shufflevector <4 x i8> [[T7]], <4 x i8> undef, <4 x i32> <i32 3, i32 1, i32 2, i32 undef> 121; CHECK-NEXT: ret <4 x i8> [[T9]] 122; 123 %t7 = shufflevector <16 x i8> %t6, <16 x i8> undef, <4 x i32> < i32 undef, i32 9, i32 4, i32 8 > 124 %t9 = shufflevector <4 x i8> %t7, <4 x i8> undef, <4 x i32> < i32 3, i32 1, i32 2, i32 0 > 125 ret <4 x i8> %t9 126} 127 128; Test fold of two shuffles where the first shuffle vectors inputs are a 129; different length then the second. 130define <4 x i8> @test9b(<4 x i8> %t6, <4 x i8> %t7) { 131; CHECK-LABEL: @test9b( 132; CHECK-NEXT: [[T9:%.*]] = shufflevector <4 x i8> [[T6:%.*]], <4 x i8> [[T7:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 5> 133; CHECK-NEXT: ret <4 x i8> [[T9]] 134; 135 %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> 136 %t9 = shufflevector <8 x i8> %t1, <8 x i8> undef, <4 x i32> <i32 0, i32 1, i32 4, i32 5> 137 ret <4 x i8> %t9 138} 139 140; Redundant vector splats should be removed. Radar 8597790. 141define <4 x i32> @test10(<4 x i32> %t5) { 142; CHECK-LABEL: @test10( 143; CHECK-NEXT: [[T7:%.*]] = shufflevector <4 x i32> [[T5:%.*]], <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 144; CHECK-NEXT: ret <4 x i32> [[T7]] 145; 146 %t6 = shufflevector <4 x i32> %t5, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef> 147 %t7 = shufflevector <4 x i32> %t6, <4 x i32> undef, <4 x i32> zeroinitializer 148 ret <4 x i32> %t7 149} 150 151; Test fold of two shuffles where the two shufflevector inputs's op1 are the same. 152 153define <8 x i8> @test11(<16 x i8> %t6) { 154; CHECK-LABEL: @test11( 155; CHECK-NEXT: [[T3:%.*]] = shufflevector <16 x i8> [[T6:%.*]], <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 156; CHECK-NEXT: ret <8 x i8> [[T3]] 157; 158 %t1 = shufflevector <16 x i8> %t6, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 159 %t2 = shufflevector <16 x i8> %t6, <16 x i8> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7> 160 %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> 161 ret <8 x i8> %t3 162} 163 164; Test fold of two shuffles where the first shufflevector's inputs are the same as the second. 165 166define <8 x i8> @test12(<8 x i8> %t6, <8 x i8> %t2) { 167; CHECK-LABEL: @test12( 168; 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> 169; CHECK-NEXT: ret <8 x i8> [[T3]] 170; 171 %t1 = shufflevector <8 x i8> %t6, <8 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 undef, i32 7> 172 %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> 173 ret <8 x i8> %t3 174} 175 176; Test fold of two shuffles where the first shufflevector's inputs are the same as the second. 177 178define <8 x i8> @test12a(<8 x i8> %t6, <8 x i8> %t2) { 179; CHECK-LABEL: @test12a( 180; 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> 181; CHECK-NEXT: ret <8 x i8> [[T3]] 182; 183 %t1 = shufflevector <8 x i8> %t6, <8 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 undef, i32 7> 184 %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> 185 ret <8 x i8> %t3 186} 187 188; The mask length of the 1st shuffle can be reduced to eliminate the 2nd shuffle. 189 190define <2 x i8> @extract_subvector_of_shuffle(<2 x i8> %x, <2 x i8> %y) { 191; CHECK-LABEL: @extract_subvector_of_shuffle( 192; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <2 x i32> <i32 0, i32 2> 193; CHECK-NEXT: ret <2 x i8> [[EXTRACT_SUBV]] 194; 195 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <3 x i32> <i32 0, i32 2, i32 0> 196 %extract_subv = shufflevector <3 x i8> %shuf, <3 x i8> undef, <2 x i32> <i32 0, i32 1> 197 ret <2 x i8> %extract_subv 198} 199 200; Undef elements in either mask are ok. Undefs from the 2nd shuffle mask should propagate to the new shuffle. 201; The type of the inputs does not have to match the output type. 202 203define <4 x i8> @extract_subvector_of_shuffle_undefs_types(<2 x i8> %x, <2 x i8> %y) { 204; CHECK-LABEL: @extract_subvector_of_shuffle_undefs_types( 205; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <4 x i32> <i32 undef, i32 2, i32 0, i32 undef> 206; CHECK-NEXT: ret <4 x i8> [[EXTRACT_SUBV]] 207; 208 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <5 x i32> <i32 undef, i32 2, i32 0, i32 1, i32 0> 209 %extract_subv = shufflevector <5 x i8> %shuf, <5 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 210 ret <4 x i8> %extract_subv 211} 212 213; Extra uses are not ok - we only do the transform when we can eliminate an instruction. 214 215declare void @use_v5i8(<5 x i8>) 216 217define <4 x i8> @extract_subvector_of_shuffle_extra_use(<2 x i8> %x, <2 x i8> %y) { 218; CHECK-LABEL: @extract_subvector_of_shuffle_extra_use( 219; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <5 x i32> <i32 undef, i32 2, i32 0, i32 1, i32 0> 220; CHECK-NEXT: call void @use_v5i8(<5 x i8> [[SHUF]]) 221; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <5 x i8> [[SHUF]], <5 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 222; CHECK-NEXT: ret <4 x i8> [[EXTRACT_SUBV]] 223; 224 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <5 x i32> <i32 undef, i32 2, i32 0, i32 1, i32 0> 225 call void @use_v5i8(<5 x i8> %shuf) 226 %extract_subv = shufflevector <5 x i8> %shuf, <5 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 227 ret <4 x i8> %extract_subv 228} 229 230define <2 x i8> @test13a(i8 %x1, i8 %x2) { 231; CHECK-LABEL: @test13a( 232; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i8> undef, i8 [[X1:%.*]], i64 1 233; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i8> [[TMP1]], i8 [[X2:%.*]], i64 0 234; CHECK-NEXT: [[TMP3:%.*]] = add <2 x i8> [[TMP2]], <i8 7, i8 5> 235; CHECK-NEXT: ret <2 x i8> [[TMP3]] 236; 237 %A = insertelement <2 x i8> undef, i8 %x1, i32 0 238 %B = insertelement <2 x i8> %A, i8 %x2, i32 1 239 %C = add <2 x i8> %B, <i8 5, i8 7> 240 %D = shufflevector <2 x i8> %C, <2 x i8> undef, <2 x i32> <i32 1, i32 0> 241 ret <2 x i8> %D 242} 243 244; Increasing length of vector ops is not a good canonicalization. 245 246define <3 x i32> @add_wider(i32 %y, i32 %z) { 247; CHECK-LABEL: @add_wider( 248; CHECK-NEXT: [[I0:%.*]] = insertelement <2 x i32> undef, i32 [[Y:%.*]], i64 0 249; CHECK-NEXT: [[I1:%.*]] = insertelement <2 x i32> [[I0]], i32 [[Z:%.*]], i64 1 250; CHECK-NEXT: [[A:%.*]] = add <2 x i32> [[I1]], <i32 255, i32 255> 251; CHECK-NEXT: [[EXT:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> undef, <3 x i32> <i32 0, i32 1, i32 undef> 252; CHECK-NEXT: ret <3 x i32> [[EXT]] 253; 254 %i0 = insertelement <2 x i32> undef, i32 %y, i32 0 255 %i1 = insertelement <2 x i32> %i0, i32 %z, i32 1 256 %a = add <2 x i32> %i1, <i32 255, i32 255> 257 %ext = shufflevector <2 x i32> %a, <2 x i32> undef, <3 x i32> <i32 0, i32 1, i32 undef> 258 ret <3 x i32> %ext 259} 260 261; Increasing length of vector ops must be safe from illegal undef propagation. 262 263define <3 x i32> @div_wider(i32 %y, i32 %z) { 264; CHECK-LABEL: @div_wider( 265; CHECK-NEXT: [[I0:%.*]] = insertelement <2 x i32> undef, i32 [[Y:%.*]], i64 0 266; CHECK-NEXT: [[I1:%.*]] = insertelement <2 x i32> [[I0]], i32 [[Z:%.*]], i64 1 267; CHECK-NEXT: [[A:%.*]] = sdiv <2 x i32> [[I1]], <i32 255, i32 255> 268; CHECK-NEXT: [[EXT:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> undef, <3 x i32> <i32 0, i32 1, i32 undef> 269; CHECK-NEXT: ret <3 x i32> [[EXT]] 270; 271 %i0 = insertelement <2 x i32> undef, i32 %y, i32 0 272 %i1 = insertelement <2 x i32> %i0, i32 %z, i32 1 273 %a = sdiv <2 x i32> %i1, <i32 255, i32 255> 274 %ext = shufflevector <2 x i32> %a, <2 x i32> undef, <3 x i32> <i32 0, i32 1, i32 undef> 275 ret <3 x i32> %ext 276} 277 278; Increasing length of insertelements (no math ops) is a good canonicalization. 279 280define <3 x i8> @fold_inselts_with_widening_shuffle(i8 %x, i8 %y) { 281; CHECK-LABEL: @fold_inselts_with_widening_shuffle( 282; CHECK-NEXT: [[TMP1:%.*]] = insertelement <3 x i8> undef, i8 [[X:%.*]], i64 0 283; CHECK-NEXT: [[TMP2:%.*]] = insertelement <3 x i8> [[TMP1]], i8 [[Y:%.*]], i64 1 284; CHECK-NEXT: ret <3 x i8> [[TMP2]] 285; 286 %ins0 = insertelement <2 x i8> undef, i8 %x, i32 0 287 %ins1 = insertelement <2 x i8> %ins0, i8 %y, i32 1 288 %widen = shufflevector <2 x i8> %ins1, <2 x i8> undef, <3 x i32> <i32 0, i32 1, i32 undef> 289 ret <3 x i8> %widen 290} 291 292define <2 x i8> @test13b(i8 %x) { 293; CHECK-LABEL: @test13b( 294; CHECK-NEXT: [[B:%.*]] = insertelement <2 x i8> undef, i8 [[X:%.*]], i64 1 295; CHECK-NEXT: ret <2 x i8> [[B]] 296; 297 %A = insertelement <2 x i8> undef, i8 %x, i32 0 298 %B = shufflevector <2 x i8> %A, <2 x i8> undef, <2 x i32> <i32 undef, i32 0> 299 ret <2 x i8> %B 300} 301 302define <2 x i8> @test13c(i8 %x1, i8 %x2) { 303; CHECK-LABEL: @test13c( 304; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i8> undef, i8 [[X1:%.*]], i64 0 305; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i8> [[TMP1]], i8 [[X2:%.*]], i64 1 306; CHECK-NEXT: ret <2 x i8> [[TMP2]] 307; 308 %A = insertelement <4 x i8> undef, i8 %x1, i32 0 309 %B = insertelement <4 x i8> %A, i8 %x2, i32 2 310 %C = shufflevector <4 x i8> %B, <4 x i8> undef, <2 x i32> <i32 0, i32 2> 311 ret <2 x i8> %C 312} 313 314define void @test14(i16 %conv10) { 315; CHECK-LABEL: @test14( 316; CHECK-NEXT: store <4 x i16> <i16 undef, i16 undef, i16 undef, i16 23>, <4 x i16>* undef, align 8 317; CHECK-NEXT: ret void 318; 319 %t = alloca <4 x i16>, align 8 320 %vecinit6 = insertelement <4 x i16> undef, i16 23, i32 3 321 store <4 x i16> %vecinit6, <4 x i16>* undef 322 %t1 = load <4 x i16>, <4 x i16>* undef 323 %vecinit11 = insertelement <4 x i16> undef, i16 %conv10, i32 3 324 %div = udiv <4 x i16> %t1, %vecinit11 325 store <4 x i16> %div, <4 x i16>* %t 326 %t4 = load <4 x i16>, <4 x i16>* %t 327 %t5 = shufflevector <4 x i16> %t4, <4 x i16> undef, <2 x i32> <i32 2, i32 0> 328 %cmp = icmp ule <2 x i16> %t5, undef 329 %sext = sext <2 x i1> %cmp to <2 x i16> 330 ret void 331} 332 333; Check that sequences of insert/extract element are 334; collapsed into valid shuffle instruction with correct shuffle indexes. 335 336define <4 x float> @test15a(<4 x float> %LHS, <4 x float> %RHS) { 337; CHECK-LABEL: @test15a( 338; CHECK-NEXT: [[T4:%.*]] = shufflevector <4 x float> [[LHS:%.*]], <4 x float> [[RHS:%.*]], <4 x i32> <i32 4, i32 0, i32 6, i32 6> 339; CHECK-NEXT: ret <4 x float> [[T4]] 340; 341 %t1 = extractelement <4 x float> %LHS, i32 0 342 %t2 = insertelement <4 x float> %RHS, float %t1, i32 1 343 %t3 = extractelement <4 x float> %RHS, i32 2 344 %t4 = insertelement <4 x float> %t2, float %t3, i32 3 345 ret <4 x float> %t4 346} 347 348define <4 x float> @test15b(<4 x float> %LHS, <4 x float> %RHS) { 349; CHECK-LABEL: @test15b( 350; CHECK-NEXT: [[T5:%.*]] = shufflevector <4 x float> [[LHS:%.*]], <4 x float> [[RHS:%.*]], <4 x i32> <i32 4, i32 3, i32 6, i32 6> 351; CHECK-NEXT: ret <4 x float> [[T5]] 352; 353 %t0 = extractelement <4 x float> %LHS, i32 3 354 %t1 = insertelement <4 x float> %RHS, float %t0, i32 0 355 %t2 = extractelement <4 x float> %t1, i32 0 356 %t3 = insertelement <4 x float> %RHS, float %t2, i32 1 357 %t4 = extractelement <4 x float> %RHS, i32 2 358 %t5 = insertelement <4 x float> %t3, float %t4, i32 3 359 ret <4 x float> %t5 360} 361 362define <1 x i32> @test16a(i32 %ele) { 363; CHECK-LABEL: @test16a( 364; CHECK-NEXT: ret <1 x i32> <i32 2> 365; 366 %t0 = insertelement <2 x i32> <i32 1, i32 undef>, i32 %ele, i32 1 367 %t1 = shl <2 x i32> %t0, <i32 1, i32 1> 368 %t2 = shufflevector <2 x i32> %t1, <2 x i32> undef, <1 x i32> <i32 0> 369 ret <1 x i32> %t2 370} 371 372define <4 x i8> @test16b(i8 %ele) { 373; CHECK-LABEL: @test16b( 374; CHECK-NEXT: ret <4 x i8> <i8 2, i8 2, i8 2, i8 2> 375; 376 %t0 = insertelement <8 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 undef, i8 1>, i8 %ele, i32 6 377 %t1 = shl <8 x i8> %t0, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1> 378 %t2 = shufflevector <8 x i8> %t1, <8 x i8> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 4> 379 ret <4 x i8> %t2 380} 381 382; If composition of two shuffles is identity, shuffles can be removed. 383define <4 x i32> @shuffle_17ident(<4 x i32> %v) { 384; CHECK-LABEL: @shuffle_17ident( 385; CHECK-NEXT: ret <4 x i32> [[V:%.*]] 386; 387 %shuffle = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 388 %shuffle2 = shufflevector <4 x i32> %shuffle, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 0, i32 1, i32 2> 389 ret <4 x i32> %shuffle2 390} 391 392; swizzle can be put after operation 393define <4 x i32> @shuffle_17and(<4 x i32> %v1, <4 x i32> %v2) { 394; CHECK-LABEL: @shuffle_17and( 395; CHECK-NEXT: [[TMP1:%.*]] = and <4 x i32> [[V1:%.*]], [[V2:%.*]] 396; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 397; CHECK-NEXT: ret <4 x i32> [[R]] 398; 399 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 400 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 401 %r = and <4 x i32> %t1, %t2 402 ret <4 x i32> %r 403} 404 405declare void @use(<2 x float>) 406declare void @use4(<4 x float>) 407 408; One extra use is ok to transform. 409 410define <2 x float> @shuffle_fadd_multiuse(<2 x float> %v1, <2 x float> %v2) { 411; CHECK-LABEL: @shuffle_fadd_multiuse( 412; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V1:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 413; CHECK-NEXT: [[TMP1:%.*]] = fadd <2 x float> [[V1]], [[V2:%.*]] 414; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 415; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 416; CHECK-NEXT: ret <2 x float> [[R]] 417; 418 %t1 = shufflevector <2 x float> %v1, <2 x float> undef, <2 x i32> <i32 1, i32 0> 419 %t2 = shufflevector <2 x float> %v2, <2 x float> undef, <2 x i32> <i32 1, i32 0> 420 %r = fadd <2 x float> %t1, %t2 421 call void @use(<2 x float> %t1) 422 ret <2 x float> %r 423} 424 425define <2 x float> @shuffle_fdiv_multiuse(<2 x float> %v1, <2 x float> %v2) { 426; CHECK-LABEL: @shuffle_fdiv_multiuse( 427; CHECK-NEXT: [[T2:%.*]] = shufflevector <2 x float> [[V2:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 428; CHECK-NEXT: [[TMP1:%.*]] = fdiv <2 x float> [[V1:%.*]], [[V2]] 429; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 430; CHECK-NEXT: call void @use(<2 x float> [[T2]]) 431; CHECK-NEXT: ret <2 x float> [[R]] 432; 433 %t1 = shufflevector <2 x float> %v1, <2 x float> undef, <2 x i32> <i32 1, i32 0> 434 %t2 = shufflevector <2 x float> %v2, <2 x float> undef, <2 x i32> <i32 1, i32 0> 435 %r = fdiv <2 x float> %t1, %t2 436 call void @use(<2 x float> %t2) 437 ret <2 x float> %r 438} 439 440; But 2 extra uses would require an extra instruction. 441 442define <2 x float> @shuffle_fsub_multiuse(<2 x float> %v1, <2 x float> %v2) { 443; CHECK-LABEL: @shuffle_fsub_multiuse( 444; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V1:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 445; CHECK-NEXT: [[T2:%.*]] = shufflevector <2 x float> [[V2:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 446; CHECK-NEXT: [[R:%.*]] = fsub <2 x float> [[T1]], [[T2]] 447; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 448; CHECK-NEXT: call void @use(<2 x float> [[T2]]) 449; CHECK-NEXT: ret <2 x float> [[R]] 450; 451 %t1 = shufflevector <2 x float> %v1, <2 x float> undef, <2 x i32> <i32 1, i32 0> 452 %t2 = shufflevector <2 x float> %v2, <2 x float> undef, <2 x i32> <i32 1, i32 0> 453 %r = fsub <2 x float> %t1, %t2 454 call void @use(<2 x float> %t1) 455 call void @use(<2 x float> %t2) 456 ret <2 x float> %r 457} 458 459define <4 x i32> @shuffle_17add(<4 x i32> %v1, <4 x i32> %v2) { 460; CHECK-LABEL: @shuffle_17add( 461; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[V1:%.*]], [[V2:%.*]] 462; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 463; CHECK-NEXT: ret <4 x i32> [[R]] 464; 465 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 466 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 467 %r = add <4 x i32> %t1, %t2 468 ret <4 x i32> %r 469} 470 471define <4 x i32> @shuffle_17addnsw(<4 x i32> %v1, <4 x i32> %v2) { 472; CHECK-LABEL: @shuffle_17addnsw( 473; CHECK-NEXT: [[TMP1:%.*]] = add nsw <4 x i32> [[V1:%.*]], [[V2:%.*]] 474; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 475; CHECK-NEXT: ret <4 x i32> [[R]] 476; 477 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 478 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 479 %r = add nsw <4 x i32> %t1, %t2 480 ret <4 x i32> %r 481} 482 483define <4 x i32> @shuffle_17addnuw(<4 x i32> %v1, <4 x i32> %v2) { 484; CHECK-LABEL: @shuffle_17addnuw( 485; CHECK-NEXT: [[TMP1:%.*]] = add nuw <4 x i32> [[V1:%.*]], [[V2:%.*]] 486; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 487; CHECK-NEXT: ret <4 x i32> [[R]] 488; 489 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 490 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 491 %r = add nuw <4 x i32> %t1, %t2 492 ret <4 x i32> %r 493} 494 495define <4 x float> @shuffle_17fsub_fast(<4 x float> %v1, <4 x float> %v2) { 496; CHECK-LABEL: @shuffle_17fsub_fast( 497; CHECK-NEXT: [[TMP1:%.*]] = fsub fast <4 x float> [[V1:%.*]], [[V2:%.*]] 498; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 499; CHECK-NEXT: ret <4 x float> [[R]] 500; 501 %t1 = shufflevector <4 x float> %v1, <4 x float> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 502 %t2 = shufflevector <4 x float> %v2, <4 x float> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 503 %r = fsub fast <4 x float> %t1, %t2 504 ret <4 x float> %r 505} 506 507define <4 x i32> @add_const(<4 x i32> %v) { 508; CHECK-LABEL: @add_const( 509; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[V:%.*]], <i32 44, i32 41, i32 42, i32 43> 510; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 511; CHECK-NEXT: ret <4 x i32> [[R]] 512; 513 %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 514 %r = add <4 x i32> %t1, <i32 41, i32 42, i32 43, i32 44> 515 ret <4 x i32> %r 516} 517 518define <4 x i32> @sub_const(<4 x i32> %v) { 519; CHECK-LABEL: @sub_const( 520; CHECK-NEXT: [[TMP1:%.*]] = sub <4 x i32> <i32 44, i32 43, i32 42, i32 41>, [[V:%.*]] 521; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 522; CHECK-NEXT: ret <4 x i32> [[R]] 523; 524 %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 525 %r = sub <4 x i32> <i32 41, i32 42, i32 43, i32 44>, %t1 526 ret <4 x i32> %r 527} 528 529; Math before shuffle requires an extra shuffle. 530 531define <2 x float> @fadd_const_multiuse(<2 x float> %v) { 532; CHECK-LABEL: @fadd_const_multiuse( 533; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 534; CHECK-NEXT: [[R:%.*]] = fadd <2 x float> [[T1]], <float 4.100000e+01, float 4.200000e+01> 535; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 536; CHECK-NEXT: ret <2 x float> [[R]] 537; 538 %t1 = shufflevector <2 x float> %v, <2 x float> undef, <2 x i32> <i32 1, i32 0> 539 %r = fadd <2 x float> %t1, <float 41.0, float 42.0> 540 call void @use(<2 x float> %t1) 541 ret <2 x float> %r 542} 543 544; Math before splat allows replacing constant elements with undef lanes. 545 546define <4 x i32> @mul_const_splat(<4 x i32> %v) { 547; CHECK-LABEL: @mul_const_splat( 548; CHECK-NEXT: [[TMP1:%.*]] = mul <4 x i32> [[V:%.*]], <i32 poison, i32 42, i32 poison, i32 poison> 549; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 550; CHECK-NEXT: ret <4 x i32> [[R]] 551; 552 %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 553 %r = mul <4 x i32> <i32 42, i32 42, i32 42, i32 42>, %t1 554 ret <4 x i32> %r 555} 556 557; Take 2 elements of a vector and shift each of those by a different amount 558 559define <4 x i32> @lshr_const_half_splat(<4 x i32> %v) { 560; CHECK-LABEL: @lshr_const_half_splat( 561; CHECK-NEXT: [[TMP1:%.*]] = lshr <4 x i32> <i32 undef, i32 8, i32 9, i32 undef>, [[V:%.*]] 562; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 1, i32 1, i32 2, i32 2> 563; CHECK-NEXT: ret <4 x i32> [[R]] 564; 565 %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2> 566 %r = lshr <4 x i32> <i32 8, i32 8, i32 9, i32 9>, %t1 567 ret <4 x i32> %r 568} 569 570; We can't change this because there's no pre-shuffle version of the fmul constant. 571 572define <2 x float> @fmul_const_invalid_constant(<2 x float> %v) { 573; CHECK-LABEL: @fmul_const_invalid_constant( 574; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> undef, <2 x i32> zeroinitializer 575; CHECK-NEXT: [[R:%.*]] = fmul <2 x float> [[T1]], <float 4.100000e+01, float 4.200000e+01> 576; CHECK-NEXT: ret <2 x float> [[R]] 577; 578 %t1 = shufflevector <2 x float> %v, <2 x float> undef, <2 x i32> <i32 0, i32 0> 579 %r = fmul <2 x float> %t1, <float 41.0, float 42.0> 580 ret <2 x float> %r 581} 582 583; Reduce the width of the binop by moving it ahead of a shuffle. 584 585define <4 x i8> @widening_shuffle_add_1(<2 x i8> %x) { 586; CHECK-LABEL: @widening_shuffle_add_1( 587; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i8> [[X:%.*]], <i8 42, i8 43> 588; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i8> [[TMP1]], <2 x i8> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 589; CHECK-NEXT: ret <4 x i8> [[R]] 590; 591 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 592 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 593 ret <4 x i8> %r 594} 595 596; Reduce the width of the binop by moving it ahead of a shuffle. 597 598define <4 x i8> @widening_shuffle_add_2(<2 x i8> %x) { 599; CHECK-LABEL: @widening_shuffle_add_2( 600; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i8> [[X:%.*]], <i8 43, i8 42> 601; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i8> [[TMP1]], <2 x i8> poison, <4 x i32> <i32 1, i32 0, i32 undef, i32 undef> 602; CHECK-NEXT: ret <4 x i8> [[R]] 603; 604 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 1, i32 0, i32 undef, i32 undef> 605 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 606 ret <4 x i8> %r 607} 608 609; Negative test - widening shuffles have the same mask/constant constraint as non-size-changing shuffles. 610 611define <4 x i8> @widening_shuffle_add_invalid_constant(<2 x i8> %x) { 612; CHECK-LABEL: @widening_shuffle_add_invalid_constant( 613; CHECK-NEXT: [[WIDEX:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> undef, <4 x i32> <i32 1, i32 1, i32 undef, i32 undef> 614; CHECK-NEXT: [[R:%.*]] = add <4 x i8> [[WIDEX]], <i8 42, i8 43, i8 44, i8 45> 615; CHECK-NEXT: ret <4 x i8> [[R]] 616; 617 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 1, i32 1, i32 undef, i32 undef> 618 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 619 ret <4 x i8> %r 620} 621 622; Negative test - widening shuffles have an additional constraint: they must not extend with anything but undefs. 623 624define <4 x i8> @widening_shuffle_add_invalid_mask(<2 x i8> %x) { 625; CHECK-LABEL: @widening_shuffle_add_invalid_mask( 626; CHECK-NEXT: [[WIDEX:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 0> 627; CHECK-NEXT: [[R:%.*]] = add <4 x i8> [[WIDEX]], <i8 42, i8 43, i8 44, i8 45> 628; CHECK-NEXT: ret <4 x i8> [[R]] 629; 630 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 0> 631 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 632 ret <4 x i8> %r 633} 634 635; A binop that produces undef in the high lanes can be moved before the shuffle. 636; This is ok because 'shl C, undef --> undef'. 637 638define <4 x i16> @widening_shuffle_shl_constant_op0(<2 x i16> %v) { 639; CHECK-LABEL: @widening_shuffle_shl_constant_op0( 640; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i16> <i16 42, i16 -42>, [[V:%.*]] 641; CHECK-NEXT: [[BO:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 642; CHECK-NEXT: ret <4 x i16> [[BO]] 643; 644 %shuf = shufflevector <2 x i16> %v, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 645 %bo = shl <4 x i16> <i16 42, i16 -42, i16 -1, i16 -1>, %shuf 646 ret <4 x i16> %bo 647} 648 649; A binop that produces undef in the high lanes can be moved before the shuffle. 650; This is ok because 'shl undef, 0 --> undef'. 651 652define <4 x i16> @widening_shuffle_shl_constant_op1(<2 x i16> %v) { 653; CHECK-LABEL: @widening_shuffle_shl_constant_op1( 654; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i16> [[V:%.*]], <i16 2, i16 4> 655; CHECK-NEXT: [[BO:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 656; CHECK-NEXT: ret <4 x i16> [[BO]] 657; 658 %shuf = shufflevector <2 x i16> %v, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 659 %bo = shl <4 x i16> %shuf, <i16 2, i16 4, i16 0, i16 0> 660 ret <4 x i16> %bo 661} 662 663; A binop that does not produce undef in the high lanes can not be moved before the shuffle. 664; This is not ok because 'shl undef, 1 (or 2)' --> 0' but moving the shuffle results in undef instead. 665 666define <4 x i16> @widening_shuffle_shl_constant_op1_non0(<2 x i16> %v) { 667; CHECK-LABEL: @widening_shuffle_shl_constant_op1_non0( 668; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <2 x i16> [[V:%.*]], <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 669; CHECK-NEXT: [[BO:%.*]] = shl <4 x i16> [[SHUF]], <i16 2, i16 4, i16 1, i16 2> 670; CHECK-NEXT: ret <4 x i16> [[BO]] 671; 672 %shuf = shufflevector <2 x i16> %v, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 673 %bo = shl <4 x i16> %shuf, <i16 2, i16 4, i16 1, i16 2> 674 ret <4 x i16> %bo 675} 676 677; A binop that does not produce undef in the high lanes can not be moved before the shuffle. 678; This is not ok because 'or -1, undef --> -1' but moving the shuffle results in undef instead. 679 680define <4 x i16> @widening_shuffle_or(<2 x i16> %v) { 681; CHECK-LABEL: @widening_shuffle_or( 682; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <2 x i16> [[V:%.*]], <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 683; CHECK-NEXT: [[BO:%.*]] = or <4 x i16> [[SHUF]], <i16 42, i16 -42, i16 -1, i16 -1> 684; CHECK-NEXT: ret <4 x i16> [[BO]] 685; 686 %shuf = shufflevector <2 x i16> %v, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 687 %bo = or <4 x i16> %shuf, <i16 42, i16 -42, i16 -1, i16 -1> 688 ret <4 x i16> %bo 689} 690 691define <4 x i32> @shuffle_17add2(<4 x i32> %v) { 692; CHECK-LABEL: @shuffle_17add2( 693; CHECK-NEXT: [[TMP1:%.*]] = shl <4 x i32> [[V:%.*]], <i32 1, i32 1, i32 1, i32 1> 694; CHECK-NEXT: ret <4 x i32> [[TMP1]] 695; 696 %t1 = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 697 %t2 = add <4 x i32> %t1, %t1 698 %r = shufflevector <4 x i32> %t2, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 699 ret <4 x i32> %r 700} 701 702define <4 x i32> @shuffle_17mulsplat(<4 x i32> %v) { 703; CHECK-LABEL: @shuffle_17mulsplat( 704; CHECK-NEXT: [[TMP1:%.*]] = mul <4 x i32> [[V:%.*]], [[V]] 705; CHECK-NEXT: [[M1:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer 706; CHECK-NEXT: ret <4 x i32> [[M1]] 707; 708 %s1 = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer 709 %m1 = mul <4 x i32> %s1, %s1 710 %s2 = shufflevector <4 x i32> %m1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 711 ret <4 x i32> %s2 712} 713 714; Do not reorder shuffle and binop if LHS of shuffles are of different size 715define <2 x i32> @pr19717(<4 x i32> %in0, <2 x i32> %in1) { 716; CHECK-LABEL: @pr19717( 717; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[IN0:%.*]], <4 x i32> poison, <2 x i32> zeroinitializer 718; CHECK-NEXT: [[SHUFFLE4:%.*]] = shufflevector <2 x i32> [[IN1:%.*]], <2 x i32> poison, <2 x i32> zeroinitializer 719; CHECK-NEXT: [[MUL:%.*]] = mul <2 x i32> [[SHUFFLE]], [[SHUFFLE4]] 720; CHECK-NEXT: ret <2 x i32> [[MUL]] 721; 722 %shuffle = shufflevector <4 x i32> %in0, <4 x i32> %in0, <2 x i32> zeroinitializer 723 %shuffle4 = shufflevector <2 x i32> %in1, <2 x i32> %in1, <2 x i32> zeroinitializer 724 %mul = mul <2 x i32> %shuffle, %shuffle4 725 ret <2 x i32> %mul 726} 727 728define <4 x i16> @pr19717a(<8 x i16> %in0, <8 x i16> %in1) { 729; CHECK-LABEL: @pr19717a( 730; CHECK-NEXT: [[TMP1:%.*]] = mul <8 x i16> [[IN0:%.*]], [[IN1:%.*]] 731; CHECK-NEXT: [[MUL:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> poison, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 732; CHECK-NEXT: ret <4 x i16> [[MUL]] 733; 734 %shuffle = shufflevector <8 x i16> %in0, <8 x i16> %in0, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 735 %shuffle1 = shufflevector <8 x i16> %in1, <8 x i16> %in1, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 736 %mul = mul <4 x i16> %shuffle, %shuffle1 737 ret <4 x i16> %mul 738} 739 740define <8 x i8> @pr19730(<16 x i8> %in0) { 741; CHECK-LABEL: @pr19730( 742; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <16 x i8> [[IN0:%.*]], <16 x i8> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> 743; CHECK-NEXT: [[SHUFFLE1:%.*]] = shufflevector <8 x i8> [[SHUFFLE]], <8 x i8> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> 744; CHECK-NEXT: ret <8 x i8> [[SHUFFLE1]] 745; 746 %shuffle = shufflevector <16 x i8> %in0, <16 x i8> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> 747 %shuffle1 = shufflevector <8 x i8> %shuffle, <8 x i8> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> 748 ret <8 x i8> %shuffle1 749} 750 751define i32 @pr19737(<4 x i32> %in0) { 752; CHECK-LABEL: @pr19737( 753; CHECK-NEXT: [[TMP1:%.*]] = extractelement <4 x i32> [[IN0:%.*]], i64 0 754; CHECK-NEXT: ret i32 [[TMP1]] 755; 756 %shuffle.i = shufflevector <4 x i32> zeroinitializer, <4 x i32> %in0, <4 x i32> <i32 0, i32 4, i32 2, i32 6> 757 %neg.i = xor <4 x i32> %shuffle.i, <i32 -1, i32 -1, i32 -1, i32 -1> 758 %and.i = and <4 x i32> %in0, %neg.i 759 %rv = extractelement <4 x i32> %and.i, i32 0 760 ret i32 %rv 761} 762 763; In PR20059 ( http://llvm.org/pr20059 ), shufflevector operations are reordered/removed 764; for an srem operation. This is not a valid optimization because it may cause a trap 765; on div-by-zero. 766 767define <4 x i32> @pr20059(<4 x i32> %p1, <4 x i32> %p2) { 768; CHECK-LABEL: @pr20059( 769; CHECK-NEXT: [[SPLAT1:%.*]] = shufflevector <4 x i32> [[P1:%.*]], <4 x i32> undef, <4 x i32> zeroinitializer 770; CHECK-NEXT: [[SPLAT2:%.*]] = shufflevector <4 x i32> [[P2:%.*]], <4 x i32> undef, <4 x i32> zeroinitializer 771; CHECK-NEXT: [[RETVAL:%.*]] = srem <4 x i32> [[SPLAT1]], [[SPLAT2]] 772; CHECK-NEXT: ret <4 x i32> [[RETVAL]] 773; 774 %splat1 = shufflevector <4 x i32> %p1, <4 x i32> undef, <4 x i32> zeroinitializer 775 %splat2 = shufflevector <4 x i32> %p2, <4 x i32> undef, <4 x i32> zeroinitializer 776 %retval = srem <4 x i32> %splat1, %splat2 777 ret <4 x i32> %retval 778} 779 780define <4 x i32> @pr20114(<4 x i32> %__mask) { 781; CHECK-LABEL: @pr20114( 782; CHECK-NEXT: [[MASK01_I:%.*]] = shufflevector <4 x i32> [[__MASK:%.*]], <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 1, i32 1> 783; CHECK-NEXT: [[MASKED_NEW_I_I_I:%.*]] = and <4 x i32> [[MASK01_I]], bitcast (<2 x i64> <i64 ptrtoint (<4 x i32> (<4 x i32>)* @pr20114 to i64), i64 ptrtoint (<4 x i32> (<4 x i32>)* @pr20114 to i64)> to <4 x i32>) 784; CHECK-NEXT: ret <4 x i32> [[MASKED_NEW_I_I_I]] 785; 786 %mask01.i = shufflevector <4 x i32> %__mask, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 1, i32 1> 787 %masked_new.i.i.i = and <4 x i32> bitcast (<2 x i64> <i64 ptrtoint (<4 x i32> (<4 x i32>)* @pr20114 to i64), i64 ptrtoint (<4 x i32> (<4 x i32>)* @pr20114 to i64)> to <4 x i32>), %mask01.i 788 ret <4 x i32> %masked_new.i.i.i 789} 790 791define <2 x i32*> @pr23113(<4 x i32*> %A) { 792; CHECK-LABEL: @pr23113( 793; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32*> [[A:%.*]], <4 x i32*> undef, <2 x i32> <i32 0, i32 1> 794; CHECK-NEXT: ret <2 x i32*> [[TMP1]] 795; 796 %1 = shufflevector <4 x i32*> %A, <4 x i32*> undef, <2 x i32> <i32 0, i32 1> 797 ret <2 x i32*> %1 798} 799 800; Unused lanes in the new binop should not kill the entire op (although it may simplify anyway as shown here). 801 802define <2 x i32> @PR37648(<2 x i32> %x) { 803; CHECK-LABEL: @PR37648( 804; CHECK-NEXT: ret <2 x i32> zeroinitializer 805; 806 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 807 %r = urem <2 x i32> %splat, <i32 1, i32 1> 808 ret <2 x i32> %r 809} 810 811; Test shuffle followed by binop with splat constant for all 18 binop opcodes. 812; Test with constant as operand 0 and operand 1 for non-commutative opcodes. 813 814define <2 x i32> @add_splat_constant(<2 x i32> %x) { 815; CHECK-LABEL: @add_splat_constant( 816; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 42, i32 poison> 817; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 818; CHECK-NEXT: ret <2 x i32> [[R]] 819; 820 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 821 %r = add <2 x i32> %splat, <i32 42, i32 42> 822 ret <2 x i32> %r 823} 824 825define <2 x i32> @sub_splat_constant0(<2 x i32> %x) { 826; CHECK-LABEL: @sub_splat_constant0( 827; CHECK-NEXT: [[TMP1:%.*]] = sub <2 x i32> <i32 42, i32 poison>, [[X:%.*]] 828; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 829; CHECK-NEXT: ret <2 x i32> [[R]] 830; 831 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 832 %r = sub <2 x i32> <i32 42, i32 42>, %splat 833 ret <2 x i32> %r 834} 835 836define <2 x i32> @sub_splat_constant1(<2 x i32> %x) { 837; CHECK-LABEL: @sub_splat_constant1( 838; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 -42, i32 poison> 839; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 840; CHECK-NEXT: ret <2 x i32> [[R]] 841; 842 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 843 %r = sub <2 x i32> %splat, <i32 42, i32 42> 844 ret <2 x i32> %r 845} 846 847define <2 x i32> @mul_splat_constant(<2 x i32> %x) { 848; CHECK-LABEL: @mul_splat_constant( 849; CHECK-NEXT: [[TMP1:%.*]] = mul <2 x i32> [[X:%.*]], <i32 42, i32 poison> 850; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 851; CHECK-NEXT: ret <2 x i32> [[R]] 852; 853 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 854 %r = mul <2 x i32> %splat, <i32 42, i32 42> 855 ret <2 x i32> %r 856} 857 858define <2 x i32> @shl_splat_constant0(<2 x i32> %x) { 859; CHECK-LABEL: @shl_splat_constant0( 860; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> <i32 5, i32 undef>, [[X:%.*]] 861; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 862; CHECK-NEXT: ret <2 x i32> [[R]] 863; 864 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 865 %r = shl <2 x i32> <i32 5, i32 5>, %splat 866 ret <2 x i32> %r 867} 868 869define <2 x i32> @shl_splat_constant1(<2 x i32> %x) { 870; CHECK-LABEL: @shl_splat_constant1( 871; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> [[X:%.*]], <i32 5, i32 0> 872; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 873; CHECK-NEXT: ret <2 x i32> [[R]] 874; 875 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 876 %r = shl <2 x i32> %splat, <i32 5, i32 5> 877 ret <2 x i32> %r 878} 879 880define <2 x i32> @ashr_splat_constant0(<2 x i32> %x) { 881; CHECK-LABEL: @ashr_splat_constant0( 882; CHECK-NEXT: [[TMP1:%.*]] = ashr <2 x i32> <i32 5, i32 undef>, [[X:%.*]] 883; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 884; CHECK-NEXT: ret <2 x i32> [[R]] 885; 886 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 887 %r = ashr <2 x i32> <i32 5, i32 5>, %splat 888 ret <2 x i32> %r 889} 890 891define <2 x i32> @ashr_splat_constant1(<2 x i32> %x) { 892; CHECK-LABEL: @ashr_splat_constant1( 893; CHECK-NEXT: [[TMP1:%.*]] = ashr <2 x i32> [[X:%.*]], <i32 5, i32 0> 894; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 895; CHECK-NEXT: ret <2 x i32> [[R]] 896; 897 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 898 %r = ashr <2 x i32> %splat, <i32 5, i32 5> 899 ret <2 x i32> %r 900} 901 902define <2 x i32> @lshr_splat_constant0(<2 x i32> %x) { 903; CHECK-LABEL: @lshr_splat_constant0( 904; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i32> <i32 5, i32 undef>, [[X:%.*]] 905; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 906; CHECK-NEXT: ret <2 x i32> [[R]] 907; 908 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 909 %r = lshr <2 x i32> <i32 5, i32 5>, %splat 910 ret <2 x i32> %r 911} 912 913define <2 x i32> @lshr_splat_constant1(<2 x i32> %x) { 914; CHECK-LABEL: @lshr_splat_constant1( 915; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i32> [[X:%.*]], <i32 5, i32 0> 916; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 917; CHECK-NEXT: ret <2 x i32> [[R]] 918; 919 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 920 %r = lshr <2 x i32> %splat, <i32 5, i32 5> 921 ret <2 x i32> %r 922} 923 924define <2 x i32> @urem_splat_constant0(<2 x i32> %x) { 925; CHECK-LABEL: @urem_splat_constant0( 926; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 927; CHECK-NEXT: [[R:%.*]] = urem <2 x i32> <i32 42, i32 42>, [[SPLAT]] 928; CHECK-NEXT: ret <2 x i32> [[R]] 929; 930 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 931 %r = urem <2 x i32> <i32 42, i32 42>, %splat 932 ret <2 x i32> %r 933} 934 935define <2 x i32> @urem_splat_constant1(<2 x i32> %x) { 936; CHECK-LABEL: @urem_splat_constant1( 937; CHECK-NEXT: [[TMP1:%.*]] = urem <2 x i32> [[X:%.*]], <i32 42, i32 1> 938; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 939; CHECK-NEXT: ret <2 x i32> [[R]] 940; 941 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 942 %r = urem <2 x i32> %splat, <i32 42, i32 42> 943 ret <2 x i32> %r 944} 945 946define <2 x i32> @srem_splat_constant0(<2 x i32> %x) { 947; CHECK-LABEL: @srem_splat_constant0( 948; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 949; CHECK-NEXT: [[R:%.*]] = srem <2 x i32> <i32 42, i32 42>, [[SPLAT]] 950; CHECK-NEXT: ret <2 x i32> [[R]] 951; 952 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 953 %r = srem <2 x i32> <i32 42, i32 42>, %splat 954 ret <2 x i32> %r 955} 956 957define <2 x i32> @srem_splat_constant1(<2 x i32> %x) { 958; CHECK-LABEL: @srem_splat_constant1( 959; CHECK-NEXT: [[TMP1:%.*]] = srem <2 x i32> [[X:%.*]], <i32 42, i32 1> 960; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 961; CHECK-NEXT: ret <2 x i32> [[R]] 962; 963 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 964 %r = srem <2 x i32> %splat, <i32 42, i32 42> 965 ret <2 x i32> %r 966} 967 968define <2 x i32> @udiv_splat_constant0(<2 x i32> %x) { 969; CHECK-LABEL: @udiv_splat_constant0( 970; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 971; CHECK-NEXT: [[R:%.*]] = udiv <2 x i32> <i32 42, i32 42>, [[SPLAT]] 972; CHECK-NEXT: ret <2 x i32> [[R]] 973; 974 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 975 %r = udiv <2 x i32> <i32 42, i32 42>, %splat 976 ret <2 x i32> %r 977} 978 979define <2 x i32> @udiv_splat_constant1(<2 x i32> %x) { 980; CHECK-LABEL: @udiv_splat_constant1( 981; CHECK-NEXT: [[TMP1:%.*]] = udiv <2 x i32> [[X:%.*]], <i32 42, i32 1> 982; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 983; CHECK-NEXT: ret <2 x i32> [[R]] 984; 985 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 986 %r = udiv <2 x i32> %splat, <i32 42, i32 42> 987 ret <2 x i32> %r 988} 989 990define <2 x i32> @sdiv_splat_constant0(<2 x i32> %x) { 991; CHECK-LABEL: @sdiv_splat_constant0( 992; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 993; CHECK-NEXT: [[R:%.*]] = sdiv <2 x i32> <i32 42, i32 42>, [[SPLAT]] 994; CHECK-NEXT: ret <2 x i32> [[R]] 995; 996 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 997 %r = sdiv <2 x i32> <i32 42, i32 42>, %splat 998 ret <2 x i32> %r 999} 1000 1001define <2 x i32> @sdiv_splat_constant1(<2 x i32> %x) { 1002; CHECK-LABEL: @sdiv_splat_constant1( 1003; CHECK-NEXT: [[TMP1:%.*]] = sdiv <2 x i32> [[X:%.*]], <i32 42, i32 1> 1004; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 1005; CHECK-NEXT: ret <2 x i32> [[R]] 1006; 1007 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 1008 %r = sdiv <2 x i32> %splat, <i32 42, i32 42> 1009 ret <2 x i32> %r 1010} 1011 1012define <2 x i32> @and_splat_constant(<2 x i32> %x) { 1013; CHECK-LABEL: @and_splat_constant( 1014; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i32> [[X:%.*]], <i32 42, i32 poison> 1015; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer 1016; CHECK-NEXT: ret <2 x i32> [[R]] 1017; 1018 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 1019 %r = and <2 x i32> %splat, <i32 42, i32 42> 1020 ret <2 x i32> %r 1021} 1022 1023; AND does not fold to undef for undef operands, we cannot move it 1024; across a shuffle with undef masks. 1025define <4 x i16> @and_constant_mask_undef(<4 x i16> %add) { 1026; CHECK-LABEL: @and_constant_mask_undef( 1027; CHECK-NEXT: entry: 1028; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[ADD:%.*]], <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1029; CHECK-NEXT: [[AND:%.*]] = and <4 x i16> [[SHUFFLE]], <i16 0, i16 0, i16 -1, i16 -1> 1030; CHECK-NEXT: ret <4 x i16> [[AND]] 1031; 1032entry: 1033 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1034 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 -1, i16 -1> 1035 ret <4 x i16> %and 1036} 1037 1038; AND does not fold to undef for undef operands, we cannot move it 1039; across a shuffle with undef masks. 1040define <4 x i16> @and_constant_mask_undef_2(<4 x i16> %add) { 1041; CHECK-LABEL: @and_constant_mask_undef_2( 1042; CHECK-NEXT: entry: 1043; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[ADD:%.*]], <4 x i16> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 undef> 1044; CHECK-NEXT: [[AND:%.*]] = and <4 x i16> [[SHUFFLE]], <i16 -1, i16 -1, i16 -1, i16 0> 1045; CHECK-NEXT: ret <4 x i16> [[AND]] 1046; 1047entry: 1048 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 undef> 1049 %and = and <4 x i16> %shuffle, <i16 -1, i16 -1, i16 -1, i16 -0> 1050 ret <4 x i16> %and 1051} 1052 1053; We can move the AND across the shuffle, as -1 (AND identity value) is used for undef lanes. 1054define <4 x i16> @and_constant_mask_undef_3(<4 x i16> %add) { 1055; CHECK-LABEL: @and_constant_mask_undef_3( 1056; CHECK-NEXT: entry: 1057; CHECK-NEXT: ret <4 x i16> <i16 0, i16 0, i16 0, i16 undef> 1058; 1059entry: 1060 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 0, i32 1, i32 1, i32 undef> 1061 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 0, i16 -1> 1062 ret <4 x i16> %and 1063} 1064 1065; We can move the AND across the shuffle, as -1 (AND identity value) is used for undef lanes. 1066define <4 x i16> @and_constant_mask_undef_4(<4 x i16> %add) { 1067; CHECK-LABEL: @and_constant_mask_undef_4( 1068; CHECK-NEXT: entry: 1069; CHECK-NEXT: [[TMP0:%.*]] = and <4 x i16> [[ADD:%.*]], <i16 9, i16 20, i16 poison, i16 poison> 1070; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 undef> 1071; CHECK-NEXT: ret <4 x i16> [[AND]] 1072; 1073entry: 1074 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 0, i32 1, i32 1, i32 undef> 1075 %and = and <4 x i16> %shuffle, <i16 9, i16 20, i16 20, i16 -1> 1076 ret <4 x i16> %and 1077} 1078 1079define <4 x i16> @and_constant_mask_not_undef(<4 x i16> %add) { 1080; CHECK-LABEL: @and_constant_mask_not_undef( 1081; CHECK-NEXT: entry: 1082; CHECK-NEXT: [[TMP0:%.*]] = and <4 x i16> [[ADD:%.*]], <i16 poison, i16 -1, i16 0, i16 0> 1083; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> poison, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1084; CHECK-NEXT: ret <4 x i16> [[AND]] 1085; 1086entry: 1087 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1088 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 -1, i16 -1> 1089 ret <4 x i16> %and 1090} 1091 1092; OR does not fold to undef for undef operands, we cannot move it 1093; across a shuffle with undef masks. 1094define <4 x i16> @or_constant_mask_undef(<4 x i16> %in) { 1095; CHECK-LABEL: @or_constant_mask_undef( 1096; CHECK-NEXT: entry: 1097; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1098; CHECK-NEXT: [[OR:%.*]] = or <4 x i16> [[SHUFFLE]], <i16 -1, i16 -1, i16 0, i16 0> 1099; CHECK-NEXT: ret <4 x i16> [[OR]] 1100; 1101entry: 1102 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1103 %or = or <4 x i16> %shuffle, <i16 -1, i16 -1, i16 0, i16 0> 1104 ret <4 x i16> %or 1105} 1106 1107; OR does not fold to undef for undef operands, we cannot move it 1108; across a shuffle with undef masks. 1109define <4 x i16> @or_constant_mask_undef_2(<4 x i16> %in) { 1110; CHECK-LABEL: @or_constant_mask_undef_2( 1111; CHECK-NEXT: entry: 1112; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> undef, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef> 1113; CHECK-NEXT: [[OR:%.*]] = or <4 x i16> [[SHUFFLE]], <i16 -1, i16 0, i16 0, i16 -1> 1114; CHECK-NEXT: ret <4 x i16> [[OR]] 1115; 1116entry: 1117 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <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_undef_3(<4 x i16> %in) { 1124; CHECK-LABEL: @or_constant_mask_undef_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> undef, <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_undef_4(<4 x i16> %in) { 1136; CHECK-LABEL: @or_constant_mask_undef_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 undef, i32 1, i32 1, i32 undef> 1140; CHECK-NEXT: ret <4 x i16> [[OR]] 1141; 1142entry: 1143 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <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> undef, <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: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> undef, <4 x i32> <i32 0, i32 undef, i32 1, i32 1> 1165; CHECK-NEXT: [[SHL:%.*]] = shl <4 x i16> [[SHUFFLE]], <i16 10, i16 3, i16 0, i16 0> 1166; CHECK-NEXT: ret <4 x i16> [[SHL]] 1167; 1168entry: 1169 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <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 undef, i32 undef, i32 1, i32 1> 1178; CHECK-NEXT: ret <4 x i16> [[ADD]] 1179; 1180entry: 1181 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <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 undef, 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> undef, <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 undef, i32 undef, i32 1, i32 1> 1203; CHECK-NEXT: ret <4 x i16> [[SUB]] 1204; 1205entry: 1206 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <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 undef> 1216; CHECK-NEXT: ret <4 x i16> [[SUB]] 1217; 1218entry: 1219 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <7 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 1365; CHECK-NEXT: [[S2:%.*]] = shufflevector <3 x i8> [[Y:%.*]], <3 x i8> undef, <7 x i32> <i32 undef, i32 1, i32 2, i32 undef, i32 undef, i32 undef, i32 undef> 1366; CHECK-NEXT: [[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> 1367; CHECK-NEXT: ret <7 x i8> [[S3]] 1368; 1369 %s1 = shufflevector <3 x i8> %x, <3 x i8> undef, <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> undef, <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 undef, 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> undef, <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> undef, <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> undef, <7 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 1392; CHECK-NEXT: [[S2:%.*]] = shufflevector <3 x i8> [[Y:%.*]], <3 x i8> undef, <7 x i32> <i32 undef, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 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> undef, <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> undef, <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> undef, <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> undef, <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 undef, i32 undef> 1418; CHECK-NEXT: ret <4 x double> [[S3]] 1419; 1420 %s1 = shufflevector <2 x double> %x, <2 x double> undef, <4 x i32> <i32 undef, i32 1, i32 undef, i32 undef> 1421 %s2 = shufflevector <2 x double> %x, <2 x double> undef, <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> undef, <4 x i32> <i32 undef, i32 1, i32 undef, i32 1> 1431; CHECK-NEXT: [[S2:%.*]] = shufflevector <2 x double> [[X]], <2 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef> 1432; CHECK-NEXT: [[S3:%.*]] = shufflevector <4 x double> [[S2]], <4 x double> [[S1]], <4 x i32> <i32 0, i32 5, i32 7, i32 undef> 1433; CHECK-NEXT: ret <4 x double> [[S3]] 1434; 1435 %s1 = shufflevector <2 x double> %x, <2 x double> undef, <4 x i32> <i32 undef, i32 1, i32 undef, i32 1> 1436 %s2 = shufflevector <2 x double> %x, <2 x double> undef, <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> undef, <4 x i32> <i32 undef, i32 1, i32 undef, i32 undef> 1446; CHECK-NEXT: [[S2:%.*]] = shufflevector <3 x double> [[Y:%.*]], <3 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef> 1447; CHECK-NEXT: [[S3:%.*]] = shufflevector <4 x double> [[S2]], <4 x double> [[S1]], <4 x i32> <i32 0, i32 5, i32 undef, i32 undef> 1448; CHECK-NEXT: ret <4 x double> [[S3]] 1449; 1450 %s1 = shufflevector <2 x double> %x, <2 x double> undef, <4 x i32> <i32 undef, i32 1, i32 undef, i32 undef> 1451 %s2 = shufflevector <3 x double> %y, <3 x double> undef, <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, <4 x float>* %p) { 1460; CHECK-LABEL: @insert_subvector_crash_invalid_mask_elt( 1461; CHECK-NEXT: [[WIDEN:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 1462; CHECK-NEXT: [[I:%.*]] = shufflevector <2 x float> [[X]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 1463; CHECK-NEXT: store <4 x float> [[I]], <4 x float>* [[P:%.*]], align 16 1464; CHECK-NEXT: ret <4 x float> [[WIDEN]] 1465; 1466 %widen = shufflevector <2 x float> %x, <2 x float> undef, <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, <4 x float>* %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> undef, <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:%.*]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> undef, i32 317426, i32 0), <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer) 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> undef, <vscale x 4 x i32> zeroinitializer 1495 %a = add <vscale x 4 x i32> %y, shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> undef, i32 317426, i32 0), <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer) 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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <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> undef, <4 x i32> <i32 0, i32 undef, i32 0, i32 undef> 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> undef, <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> undef, <4 x i32> <i32 undef, 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> undef, <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> undef, <4 x i32> <i32 undef, 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> undef, <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> undef, <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> [[Z:%.*]], [[X:%.*]] 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> undef, <3 x i32> <i32 2, i32 2, i32 2> 1620 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <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> [[Z:%.*]], [[X:%.*]] 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> undef, <3 x i32> <i32 undef, i32 2, i32 2> 1634 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <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> undef, <3 x i32> <i32 2, i32 2, i32 2> 1643; CHECK-NEXT: [[SPLATZ:%.*]] = shufflevector <3 x i8> [[Z:%.*]], <3 x i8> undef, <3 x i32> <i32 undef, i32 2, i32 2> 1644; CHECK-NEXT: [[A:%.*]] = mul nsw <3 x i8> [[SPLATZ]], [[Y:%.*]] 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> undef, <3 x i32> <i32 2, i32 2, i32 2> 1649 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <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> [[Z:%.*]], [[X:%.*]] 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> undef, <3 x i32> <i32 2, i32 2, i32 undef> 1663 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <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> undef, <3 x i32> <i32 2, i32 2, i32 2> 1672; CHECK-NEXT: [[SPLATZ:%.*]] = shufflevector <3 x i8> [[Z:%.*]], <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 undef> 1673; CHECK-NEXT: [[A:%.*]] = mul nsw <3 x i8> [[SPLATZ]], [[Y:%.*]] 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> undef, <3 x i32> <i32 2, i32 2, i32 2> 1678 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <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> undef, <3 x i32> <i32 1, i32 1, i32 1> 1689; CHECK-NEXT: [[SPLATZ:%.*]] = shufflevector <3 x i8> [[Z:%.*]], <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1690; CHECK-NEXT: [[A:%.*]] = or <3 x i8> [[SPLATZ]], [[Y:%.*]] 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> undef, <3 x i32> <i32 1, i32 1, i32 1> 1695 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <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> undef, <2 x i32> zeroinitializer 1706; CHECK-NEXT: [[A:%.*]] = fdiv reassoc nsz <2 x float> [[Y:%.*]], <float 3.000000e+00, 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> undef, <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> undef, <2 x i32> <i32 1, i32 1> 1721; CHECK-NEXT: [[A:%.*]] = fadd fast <2 x float> [[Y:%.*]], <float 3.000000e+00, 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> undef, <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> undef, <3 x i32> zeroinitializer 1738; CHECK-NEXT: [[A:%.*]] = and <3 x i32> [[Y:%.*]], <i32 42, i32 42, 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> undef, <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> undef, <5 x i32> zeroinitializer 1753; CHECK-NEXT: [[TMP1:%.*]] = xor <5 x i32> [[SPLATX]], [[Y:%.*]] 1754; CHECK-NEXT: [[R:%.*]] = xor <5 x i32> [[TMP1]], <i32 42, i32 42, i32 42, i32 42, i32 42> 1755; CHECK-NEXT: ret <5 x i32> [[R]] 1756; 1757 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <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> undef, <4 x i32> zeroinitializer 1768; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 42, i32 42, i32 42, 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> undef, <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> undef, <4 x i32> <i32 undef, i32 0, i32 1, i32 1> 1784; CHECK-NEXT: [[A:%.*]] = and <4 x i32> [[S]], bitcast (<2 x i64> <i64 ptrtoint (<4 x i32> (<4 x i32>)* @PR46872 to i64), i64 ptrtoint (<4 x i32> (<4 x i32>)* @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> undef, <4 x i32> <i32 undef, i32 0, i32 1, i32 1> 1788 %a = and <4 x i32> %s, bitcast (<2 x i64> <i64 ptrtoint (<4 x i32> (<4 x i32>)* @PR46872 to i64), i64 ptrtoint (<4 x i32> (<4 x i32>)* @PR46872 to i64)> to <4 x i32>) 1789 ret <4 x i32> %a 1790} 1791 1792define <2 x float> @fneg_unary_shuf(<2 x float> %x) { 1793; CHECK-LABEL: @fneg_unary_shuf( 1794; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> poison, <2 x i32> <i32 1, i32 0> 1795; CHECK-NEXT: [[R:%.*]] = fneg nnan nsz <2 x float> [[TMP1]] 1796; CHECK-NEXT: ret <2 x float> [[R]] 1797; 1798 %nx = fneg nsz nnan <2 x float> %x 1799 %r = shufflevector <2 x float> %nx, <2 x float> poison, <2 x i32> <i32 1, i32 0> 1800 ret <2 x float> %r 1801} 1802 1803define <4 x half> @fneg_unary_shuf_widen(<2 x half> %x) { 1804; CHECK-LABEL: @fneg_unary_shuf_widen( 1805; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <2 x half> [[X:%.*]], <2 x half> poison, <4 x i32> <i32 1, i32 0, i32 0, i32 undef> 1806; CHECK-NEXT: [[R:%.*]] = fneg ninf <4 x half> [[TMP1]] 1807; CHECK-NEXT: ret <4 x half> [[R]] 1808; 1809 %nx = fneg ninf <2 x half> %x 1810 %r = shufflevector <2 x half> %nx, <2 x half> poison, <4 x i32> <i32 1, i32 0, i32 0, i32 poison> 1811 ret <4 x half> %r 1812} 1813 1814define <2 x double> @fneg_unary_shuf_narrow(<4 x double> %x) { 1815; CHECK-LABEL: @fneg_unary_shuf_narrow( 1816; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[X:%.*]], <4 x double> poison, <2 x i32> <i32 3, i32 0> 1817; CHECK-NEXT: [[R:%.*]] = fneg nsz <2 x double> [[TMP1]] 1818; CHECK-NEXT: ret <2 x double> [[R]] 1819; 1820 %nx = fneg nsz <4 x double> %x 1821 %r = shufflevector <4 x double> %nx, <4 x double> poison, <2 x i32> <i32 3, i32 0> 1822 ret <2 x double> %r 1823} 1824 1825; negative test - extra use prevents canonicalization 1826 1827define <2 x float> @fneg_unary_shuf_use(<2 x float> %x) { 1828; CHECK-LABEL: @fneg_unary_shuf_use( 1829; CHECK-NEXT: [[NX:%.*]] = fneg nsz <2 x float> [[X:%.*]] 1830; CHECK-NEXT: call void @use(<2 x float> [[NX]]) 1831; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[NX]], <2 x float> poison, <2 x i32> <i32 undef, i32 0> 1832; CHECK-NEXT: ret <2 x float> [[R]] 1833; 1834 %nx = fneg nsz <2 x float> %x 1835 call void @use(<2 x float> %nx) 1836 %r = shufflevector <2 x float> %nx, <2 x float> poison, <2 x i32> <i32 3, i32 0> 1837 ret <2 x float> %r 1838} 1839 1840; intersect FMF 1841 1842define <4 x float> @fneg_shuf(<4 x float> %x, <4 x float> %y) { 1843; CHECK-LABEL: @fneg_shuf( 1844; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> [[Y:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 5> 1845; CHECK-NEXT: [[R:%.*]] = fneg ninf <4 x float> [[TMP1]] 1846; CHECK-NEXT: ret <4 x float> [[R]] 1847; 1848 %nx = fneg nsz ninf <4 x float> %x 1849 %ny = fneg nnan ninf <4 x float> %y 1850 %r = shufflevector <4 x float> %nx, <4 x float> %ny, <4 x i32> <i32 0, i32 1, i32 4, i32 5> 1851 ret <4 x float> %r 1852} 1853 1854; length-changing shuffle and extra use are ok 1855 1856define <4 x float> @fneg_shuf_widen_use1(<2 x float> %x, <2 x float> %y) { 1857; CHECK-LABEL: @fneg_shuf_widen_use1( 1858; CHECK-NEXT: [[NX:%.*]] = fneg nnan <2 x float> [[X:%.*]] 1859; CHECK-NEXT: call void @use(<2 x float> [[NX]]) 1860; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <2 x float> [[X]], <2 x float> [[Y:%.*]], <4 x i32> <i32 undef, i32 1, i32 2, i32 3> 1861; CHECK-NEXT: [[R:%.*]] = fneg nnan <4 x float> [[TMP1]] 1862; CHECK-NEXT: ret <4 x float> [[R]] 1863; 1864 %nx = fneg nnan <2 x float> %x 1865 call void @use(<2 x float> %nx) 1866 %ny = fneg nnan <2 x float> %y 1867 %r = shufflevector <2 x float> %nx, <2 x float> %ny, <4 x i32> <i32 poison, i32 1, i32 2, i32 3> 1868 ret <4 x float> %r 1869} 1870 1871; length-changing shuffle and extra use still ok 1872 1873define <2 x float> @fneg_shuf_narrow_use2(<4 x float> %x, <4 x float> %y) { 1874; CHECK-LABEL: @fneg_shuf_narrow_use2( 1875; CHECK-NEXT: [[NY:%.*]] = fneg nnan nsz <4 x float> [[Y:%.*]] 1876; CHECK-NEXT: call void @use4(<4 x float> [[NY]]) 1877; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> [[Y]], <2 x i32> <i32 3, i32 5> 1878; CHECK-NEXT: [[R:%.*]] = fneg nnan nsz <2 x float> [[TMP1]] 1879; CHECK-NEXT: ret <2 x float> [[R]] 1880; 1881 %nx = fneg nsz nnan <4 x float> %x 1882 %ny = fneg nsz nnan <4 x float> %y 1883 call void @use4(<4 x float> %ny) 1884 %r = shufflevector <4 x float> %nx, <4 x float> %ny, <2 x i32> <i32 3, i32 5> 1885 ret <2 x float> %r 1886} 1887 1888; negative test - too many extra uses 1889 1890define <2 x float> @fneg_shuf_use3(<2 x float> %x, <2 x float> %y) { 1891; CHECK-LABEL: @fneg_shuf_use3( 1892; CHECK-NEXT: [[NX:%.*]] = fneg nnan <2 x float> [[X:%.*]] 1893; CHECK-NEXT: call void @use(<2 x float> [[NX]]) 1894; CHECK-NEXT: [[NY:%.*]] = fneg nnan <2 x float> [[Y:%.*]] 1895; CHECK-NEXT: call void @use(<2 x float> [[NY]]) 1896; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[NX]], <2 x float> [[NY]], <2 x i32> <i32 1, i32 2> 1897; CHECK-NEXT: ret <2 x float> [[R]] 1898; 1899 %nx = fneg nnan <2 x float> %x 1900 call void @use(<2 x float> %nx) 1901 %ny = fneg nnan <2 x float> %y 1902 call void @use(<2 x float> %ny) 1903 %r = shufflevector <2 x float> %nx, <2 x float> %ny, <2 x i32> <i32 1, i32 2> 1904 ret <2 x float> %r 1905} 1906