1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -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:%.*]], i32 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:%.*]], i32 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:%.*]], i32 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 <4 x float> @test7(<4 x float> %x) { 63; CHECK-LABEL: @test7( 64; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 65; CHECK-NEXT: ret <4 x float> [[R]] 66; 67 %r = shufflevector <4 x float> %x, <4 x float> undef, <4 x i32> < i32 0, i32 1, i32 6, i32 7 > 68 ret <4 x float> %r 69} 70 71; This should turn into a single shuffle. 72define <4 x float> @test8(<4 x float> %x, <4 x float> %y) { 73; CHECK-LABEL: @test8( 74; CHECK-NEXT: [[T134:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> [[Y:%.*]], <4 x i32> <i32 1, i32 undef, i32 3, i32 4> 75; CHECK-NEXT: ret <4 x float> [[T134]] 76; 77 %t4 = extractelement <4 x float> %x, i32 1 78 %t2 = extractelement <4 x float> %x, i32 3 79 %t1 = extractelement <4 x float> %y, i32 0 80 %t128 = insertelement <4 x float> undef, float %t4, i32 0 81 %t130 = insertelement <4 x float> %t128, float undef, i32 1 82 %t132 = insertelement <4 x float> %t130, float %t2, i32 2 83 %t134 = insertelement <4 x float> %t132, float %t1, i32 3 84 ret <4 x float> %t134 85} 86 87; Test fold of two shuffles where the first shuffle vectors inputs are a 88; different length then the second. 89define <4 x i8> @test9(<16 x i8> %t6) { 90; CHECK-LABEL: @test9( 91; CHECK-NEXT: [[T9:%.*]] = shufflevector <16 x i8> [[T6:%.*]], <16 x i8> undef, <4 x i32> <i32 13, i32 9, i32 4, i32 13> 92; CHECK-NEXT: ret <4 x i8> [[T9]] 93; 94 %t7 = shufflevector <16 x i8> %t6, <16 x i8> undef, <4 x i32> < i32 13, i32 9, i32 4, i32 13 > 95 %t9 = shufflevector <4 x i8> %t7, <4 x i8> undef, <4 x i32> < i32 3, i32 1, i32 2, i32 0 > 96 ret <4 x i8> %t9 97} 98 99; Same as test9, but make sure that "undef" mask values are not confused with 100; mask values of 2*N, where N is the mask length. These shuffles should not 101; be folded (because [8,9,4,8] may not be a mask supported by the target). 102 103define <4 x i8> @test9a(<16 x i8> %t6) { 104; CHECK-LABEL: @test9a( 105; CHECK-NEXT: [[T7:%.*]] = shufflevector <16 x i8> [[T6:%.*]], <16 x i8> undef, <4 x i32> <i32 undef, i32 9, i32 4, i32 8> 106; CHECK-NEXT: [[T9:%.*]] = shufflevector <4 x i8> [[T7]], <4 x i8> undef, <4 x i32> <i32 3, i32 1, i32 2, i32 undef> 107; CHECK-NEXT: ret <4 x i8> [[T9]] 108; 109 %t7 = shufflevector <16 x i8> %t6, <16 x i8> undef, <4 x i32> < i32 undef, i32 9, i32 4, i32 8 > 110 %t9 = shufflevector <4 x i8> %t7, <4 x i8> undef, <4 x i32> < i32 3, i32 1, i32 2, i32 0 > 111 ret <4 x i8> %t9 112} 113 114; Test fold of two shuffles where the first shuffle vectors inputs are a 115; different length then the second. 116define <4 x i8> @test9b(<4 x i8> %t6, <4 x i8> %t7) { 117; CHECK-LABEL: @test9b( 118; CHECK-NEXT: [[T9:%.*]] = shufflevector <4 x i8> [[T6:%.*]], <4 x i8> [[T7:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 5> 119; CHECK-NEXT: ret <4 x i8> [[T9]] 120; 121 %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> 122 %t9 = shufflevector <8 x i8> %t1, <8 x i8> undef, <4 x i32> <i32 0, i32 1, i32 4, i32 5> 123 ret <4 x i8> %t9 124} 125 126; Redundant vector splats should be removed. Radar 8597790. 127define <4 x i32> @test10(<4 x i32> %t5) { 128; CHECK-LABEL: @test10( 129; CHECK-NEXT: [[T7:%.*]] = shufflevector <4 x i32> [[T5:%.*]], <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 130; CHECK-NEXT: ret <4 x i32> [[T7]] 131; 132 %t6 = shufflevector <4 x i32> %t5, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef> 133 %t7 = shufflevector <4 x i32> %t6, <4 x i32> undef, <4 x i32> zeroinitializer 134 ret <4 x i32> %t7 135} 136 137; Test fold of two shuffles where the two shufflevector inputs's op1 are the same. 138 139define <8 x i8> @test11(<16 x i8> %t6) { 140; CHECK-LABEL: @test11( 141; 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> 142; CHECK-NEXT: ret <8 x i8> [[T3]] 143; 144 %t1 = shufflevector <16 x i8> %t6, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 145 %t2 = shufflevector <16 x i8> %t6, <16 x i8> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7> 146 %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> 147 ret <8 x i8> %t3 148} 149 150; Test fold of two shuffles where the first shufflevector's inputs are the same as the second. 151 152define <8 x i8> @test12(<8 x i8> %t6, <8 x i8> %t2) { 153; CHECK-LABEL: @test12( 154; 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> 155; CHECK-NEXT: ret <8 x i8> [[T3]] 156; 157 %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> 158 %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> 159 ret <8 x i8> %t3 160} 161 162; Test fold of two shuffles where the first shufflevector's inputs are the same as the second. 163 164define <8 x i8> @test12a(<8 x i8> %t6, <8 x i8> %t2) { 165; CHECK-LABEL: @test12a( 166; 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> 167; CHECK-NEXT: ret <8 x i8> [[T3]] 168; 169 %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> 170 %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> 171 ret <8 x i8> %t3 172} 173 174; The mask length of the 1st shuffle can be reduced to eliminate the 2nd shuffle. 175 176define <2 x i8> @extract_subvector_of_shuffle(<2 x i8> %x, <2 x i8> %y) { 177; CHECK-LABEL: @extract_subvector_of_shuffle( 178; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <2 x i32> <i32 0, i32 2> 179; CHECK-NEXT: ret <2 x i8> [[EXTRACT_SUBV]] 180; 181 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <3 x i32> <i32 0, i32 2, i32 0> 182 %extract_subv = shufflevector <3 x i8> %shuf, <3 x i8> undef, <2 x i32> <i32 0, i32 1> 183 ret <2 x i8> %extract_subv 184} 185 186; Undef elements in either mask are ok. Undefs from the 2nd shuffle mask should propagate to the new shuffle. 187; The type of the inputs does not have to match the output type. 188 189define <4 x i8> @extract_subvector_of_shuffle_undefs_types(<2 x i8> %x, <2 x i8> %y) { 190; CHECK-LABEL: @extract_subvector_of_shuffle_undefs_types( 191; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <4 x i32> <i32 undef, i32 2, i32 0, i32 undef> 192; CHECK-NEXT: ret <4 x i8> [[EXTRACT_SUBV]] 193; 194 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <5 x i32> <i32 undef, i32 2, i32 0, i32 1, i32 0> 195 %extract_subv = shufflevector <5 x i8> %shuf, <5 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 196 ret <4 x i8> %extract_subv 197} 198 199; Extra uses are not ok - we only do the transform when we can eliminate an instruction. 200 201declare void @use_v5i8(<5 x i8>) 202 203define <4 x i8> @extract_subvector_of_shuffle_extra_use(<2 x i8> %x, <2 x i8> %y) { 204; CHECK-LABEL: @extract_subvector_of_shuffle_extra_use( 205; 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> 206; CHECK-NEXT: call void @use_v5i8(<5 x i8> [[SHUF]]) 207; CHECK-NEXT: [[EXTRACT_SUBV:%.*]] = shufflevector <5 x i8> [[SHUF]], <5 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 208; CHECK-NEXT: ret <4 x i8> [[EXTRACT_SUBV]] 209; 210 %shuf = shufflevector <2 x i8> %x, <2 x i8> %y, <5 x i32> <i32 undef, i32 2, i32 0, i32 1, i32 0> 211 call void @use_v5i8(<5 x i8> %shuf) 212 %extract_subv = shufflevector <5 x i8> %shuf, <5 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 213 ret <4 x i8> %extract_subv 214} 215 216define <2 x i8> @test13a(i8 %x1, i8 %x2) { 217; CHECK-LABEL: @test13a( 218; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i8> undef, i8 [[X1:%.*]], i32 1 219; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i8> [[TMP1]], i8 [[X2:%.*]], i32 0 220; CHECK-NEXT: [[TMP3:%.*]] = add <2 x i8> [[TMP2]], <i8 7, i8 5> 221; CHECK-NEXT: ret <2 x i8> [[TMP3]] 222; 223 %A = insertelement <2 x i8> undef, i8 %x1, i32 0 224 %B = insertelement <2 x i8> %A, i8 %x2, i32 1 225 %C = add <2 x i8> %B, <i8 5, i8 7> 226 %D = shufflevector <2 x i8> %C, <2 x i8> undef, <2 x i32> <i32 1, i32 0> 227 ret <2 x i8> %D 228} 229 230; Increasing length of vector ops is not a good canonicalization. 231 232define <3 x i32> @add_wider(i32 %y, i32 %z) { 233; CHECK-LABEL: @add_wider( 234; CHECK-NEXT: [[I0:%.*]] = insertelement <2 x i32> undef, i32 [[Y:%.*]], i32 0 235; CHECK-NEXT: [[I1:%.*]] = insertelement <2 x i32> [[I0]], i32 [[Z:%.*]], i32 1 236; CHECK-NEXT: [[A:%.*]] = add <2 x i32> [[I1]], <i32 255, i32 255> 237; CHECK-NEXT: [[EXT:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> undef, <3 x i32> <i32 0, i32 1, i32 undef> 238; CHECK-NEXT: ret <3 x i32> [[EXT]] 239; 240 %i0 = insertelement <2 x i32> undef, i32 %y, i32 0 241 %i1 = insertelement <2 x i32> %i0, i32 %z, i32 1 242 %a = add <2 x i32> %i1, <i32 255, i32 255> 243 %ext = shufflevector <2 x i32> %a, <2 x i32> undef, <3 x i32> <i32 0, i32 1, i32 undef> 244 ret <3 x i32> %ext 245} 246 247; Increasing length of vector ops must be safe from illegal undef propagation. 248 249define <3 x i32> @div_wider(i32 %y, i32 %z) { 250; CHECK-LABEL: @div_wider( 251; CHECK-NEXT: [[I0:%.*]] = insertelement <2 x i32> undef, i32 [[Y:%.*]], i32 0 252; CHECK-NEXT: [[I1:%.*]] = insertelement <2 x i32> [[I0]], i32 [[Z:%.*]], i32 1 253; CHECK-NEXT: [[A:%.*]] = sdiv <2 x i32> [[I1]], <i32 255, i32 255> 254; CHECK-NEXT: [[EXT:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> undef, <3 x i32> <i32 0, i32 1, i32 undef> 255; CHECK-NEXT: ret <3 x i32> [[EXT]] 256; 257 %i0 = insertelement <2 x i32> undef, i32 %y, i32 0 258 %i1 = insertelement <2 x i32> %i0, i32 %z, i32 1 259 %a = sdiv <2 x i32> %i1, <i32 255, i32 255> 260 %ext = shufflevector <2 x i32> %a, <2 x i32> undef, <3 x i32> <i32 0, i32 1, i32 undef> 261 ret <3 x i32> %ext 262} 263 264; Increasing length of insertelements (no math ops) is a good canonicalization. 265 266define <3 x i8> @fold_inselts_with_widening_shuffle(i8 %x, i8 %y) { 267; CHECK-LABEL: @fold_inselts_with_widening_shuffle( 268; CHECK-NEXT: [[TMP1:%.*]] = insertelement <3 x i8> undef, i8 [[X:%.*]], i32 0 269; CHECK-NEXT: [[TMP2:%.*]] = insertelement <3 x i8> [[TMP1]], i8 [[Y:%.*]], i32 1 270; CHECK-NEXT: ret <3 x i8> [[TMP2]] 271; 272 %ins0 = insertelement <2 x i8> undef, i8 %x, i32 0 273 %ins1 = insertelement <2 x i8> %ins0, i8 %y, i32 1 274 %widen = shufflevector <2 x i8> %ins1, <2 x i8> undef, <3 x i32> <i32 0, i32 1, i32 undef> 275 ret <3 x i8> %widen 276} 277 278define <2 x i8> @test13b(i8 %x) { 279; CHECK-LABEL: @test13b( 280; CHECK-NEXT: [[B:%.*]] = insertelement <2 x i8> undef, i8 [[X:%.*]], i32 1 281; CHECK-NEXT: ret <2 x i8> [[B]] 282; 283 %A = insertelement <2 x i8> undef, i8 %x, i32 0 284 %B = shufflevector <2 x i8> %A, <2 x i8> undef, <2 x i32> <i32 undef, i32 0> 285 ret <2 x i8> %B 286} 287 288define <2 x i8> @test13c(i8 %x1, i8 %x2) { 289; CHECK-LABEL: @test13c( 290; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i8> undef, i8 [[X1:%.*]], i32 0 291; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i8> [[TMP1]], i8 [[X2:%.*]], i32 1 292; CHECK-NEXT: ret <2 x i8> [[TMP2]] 293; 294 %A = insertelement <4 x i8> undef, i8 %x1, i32 0 295 %B = insertelement <4 x i8> %A, i8 %x2, i32 2 296 %C = shufflevector <4 x i8> %B, <4 x i8> undef, <2 x i32> <i32 0, i32 2> 297 ret <2 x i8> %C 298} 299 300define void @test14(i16 %conv10) { 301; CHECK-LABEL: @test14( 302; CHECK-NEXT: store <4 x i16> <i16 undef, i16 undef, i16 undef, i16 23>, <4 x i16>* undef, align 8 303; CHECK-NEXT: ret void 304; 305 %t = alloca <4 x i16>, align 8 306 %vecinit6 = insertelement <4 x i16> undef, i16 23, i32 3 307 store <4 x i16> %vecinit6, <4 x i16>* undef 308 %t1 = load <4 x i16>, <4 x i16>* undef 309 %vecinit11 = insertelement <4 x i16> undef, i16 %conv10, i32 3 310 %div = udiv <4 x i16> %t1, %vecinit11 311 store <4 x i16> %div, <4 x i16>* %t 312 %t4 = load <4 x i16>, <4 x i16>* %t 313 %t5 = shufflevector <4 x i16> %t4, <4 x i16> undef, <2 x i32> <i32 2, i32 0> 314 %cmp = icmp ule <2 x i16> %t5, undef 315 %sext = sext <2 x i1> %cmp to <2 x i16> 316 ret void 317} 318 319; Check that sequences of insert/extract element are 320; collapsed into valid shuffle instruction with correct shuffle indexes. 321 322define <4 x float> @test15a(<4 x float> %LHS, <4 x float> %RHS) { 323; CHECK-LABEL: @test15a( 324; CHECK-NEXT: [[T4:%.*]] = shufflevector <4 x float> [[LHS:%.*]], <4 x float> [[RHS:%.*]], <4 x i32> <i32 4, i32 0, i32 6, i32 6> 325; CHECK-NEXT: ret <4 x float> [[T4]] 326; 327 %t1 = extractelement <4 x float> %LHS, i32 0 328 %t2 = insertelement <4 x float> %RHS, float %t1, i32 1 329 %t3 = extractelement <4 x float> %RHS, i32 2 330 %t4 = insertelement <4 x float> %t2, float %t3, i32 3 331 ret <4 x float> %t4 332} 333 334define <4 x float> @test15b(<4 x float> %LHS, <4 x float> %RHS) { 335; CHECK-LABEL: @test15b( 336; CHECK-NEXT: [[T5:%.*]] = shufflevector <4 x float> [[LHS:%.*]], <4 x float> [[RHS:%.*]], <4 x i32> <i32 4, i32 3, i32 6, i32 6> 337; CHECK-NEXT: ret <4 x float> [[T5]] 338; 339 %t0 = extractelement <4 x float> %LHS, i32 3 340 %t1 = insertelement <4 x float> %RHS, float %t0, i32 0 341 %t2 = extractelement <4 x float> %t1, i32 0 342 %t3 = insertelement <4 x float> %RHS, float %t2, i32 1 343 %t4 = extractelement <4 x float> %RHS, i32 2 344 %t5 = insertelement <4 x float> %t3, float %t4, i32 3 345 ret <4 x float> %t5 346} 347 348define <1 x i32> @test16a(i32 %ele) { 349; CHECK-LABEL: @test16a( 350; CHECK-NEXT: ret <1 x i32> <i32 2> 351; 352 %t0 = insertelement <2 x i32> <i32 1, i32 undef>, i32 %ele, i32 1 353 %t1 = shl <2 x i32> %t0, <i32 1, i32 1> 354 %t2 = shufflevector <2 x i32> %t1, <2 x i32> undef, <1 x i32> <i32 0> 355 ret <1 x i32> %t2 356} 357 358define <4 x i8> @test16b(i8 %ele) { 359; CHECK-LABEL: @test16b( 360; CHECK-NEXT: ret <4 x i8> <i8 2, i8 2, i8 2, i8 2> 361; 362 %t0 = insertelement <8 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 undef, i8 1>, i8 %ele, i32 6 363 %t1 = shl <8 x i8> %t0, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1> 364 %t2 = shufflevector <8 x i8> %t1, <8 x i8> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 4> 365 ret <4 x i8> %t2 366} 367 368; If composition of two shuffles is identity, shuffles can be removed. 369define <4 x i32> @shuffle_17ident(<4 x i32> %v) { 370; CHECK-LABEL: @shuffle_17ident( 371; CHECK-NEXT: ret <4 x i32> [[V:%.*]] 372; 373 %shuffle = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 374 %shuffle2 = shufflevector <4 x i32> %shuffle, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 0, i32 1, i32 2> 375 ret <4 x i32> %shuffle2 376} 377 378; swizzle can be put after operation 379define <4 x i32> @shuffle_17and(<4 x i32> %v1, <4 x i32> %v2) { 380; CHECK-LABEL: @shuffle_17and( 381; CHECK-NEXT: [[TMP1:%.*]] = and <4 x i32> [[V1:%.*]], [[V2:%.*]] 382; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 383; CHECK-NEXT: ret <4 x i32> [[R]] 384; 385 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 386 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 387 %r = and <4 x i32> %t1, %t2 388 ret <4 x i32> %r 389} 390 391declare void @use(<2 x float>) 392 393; One extra use is ok to transform. 394 395define <2 x float> @shuffle_fadd_multiuse(<2 x float> %v1, <2 x float> %v2) { 396; CHECK-LABEL: @shuffle_fadd_multiuse( 397; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V1:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 398; CHECK-NEXT: [[TMP1:%.*]] = fadd <2 x float> [[V1]], [[V2:%.*]] 399; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 400; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 401; CHECK-NEXT: ret <2 x float> [[R]] 402; 403 %t1 = shufflevector <2 x float> %v1, <2 x float> undef, <2 x i32> <i32 1, i32 0> 404 %t2 = shufflevector <2 x float> %v2, <2 x float> undef, <2 x i32> <i32 1, i32 0> 405 %r = fadd <2 x float> %t1, %t2 406 call void @use(<2 x float> %t1) 407 ret <2 x float> %r 408} 409 410define <2 x float> @shuffle_fdiv_multiuse(<2 x float> %v1, <2 x float> %v2) { 411; CHECK-LABEL: @shuffle_fdiv_multiuse( 412; CHECK-NEXT: [[T2:%.*]] = shufflevector <2 x float> [[V2:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 413; CHECK-NEXT: [[TMP1:%.*]] = fdiv <2 x float> [[V1:%.*]], [[V2]] 414; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 415; CHECK-NEXT: call void @use(<2 x float> [[T2]]) 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 = fdiv <2 x float> %t1, %t2 421 call void @use(<2 x float> %t2) 422 ret <2 x float> %r 423} 424 425; But 2 extra uses would require an extra instruction. 426 427define <2 x float> @shuffle_fsub_multiuse(<2 x float> %v1, <2 x float> %v2) { 428; CHECK-LABEL: @shuffle_fsub_multiuse( 429; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V1:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 430; CHECK-NEXT: [[T2:%.*]] = shufflevector <2 x float> [[V2:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 431; CHECK-NEXT: [[R:%.*]] = fsub <2 x float> [[T1]], [[T2]] 432; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 433; CHECK-NEXT: call void @use(<2 x float> [[T2]]) 434; CHECK-NEXT: ret <2 x float> [[R]] 435; 436 %t1 = shufflevector <2 x float> %v1, <2 x float> undef, <2 x i32> <i32 1, i32 0> 437 %t2 = shufflevector <2 x float> %v2, <2 x float> undef, <2 x i32> <i32 1, i32 0> 438 %r = fsub <2 x float> %t1, %t2 439 call void @use(<2 x float> %t1) 440 call void @use(<2 x float> %t2) 441 ret <2 x float> %r 442} 443 444define <4 x i32> @shuffle_17add(<4 x i32> %v1, <4 x i32> %v2) { 445; CHECK-LABEL: @shuffle_17add( 446; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[V1:%.*]], [[V2:%.*]] 447; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 448; CHECK-NEXT: ret <4 x i32> [[R]] 449; 450 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 451 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 452 %r = add <4 x i32> %t1, %t2 453 ret <4 x i32> %r 454} 455 456define <4 x i32> @shuffle_17addnsw(<4 x i32> %v1, <4 x i32> %v2) { 457; CHECK-LABEL: @shuffle_17addnsw( 458; CHECK-NEXT: [[TMP1:%.*]] = add nsw <4 x i32> [[V1:%.*]], [[V2:%.*]] 459; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 460; CHECK-NEXT: ret <4 x i32> [[R]] 461; 462 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 463 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 464 %r = add nsw <4 x i32> %t1, %t2 465 ret <4 x i32> %r 466} 467 468define <4 x i32> @shuffle_17addnuw(<4 x i32> %v1, <4 x i32> %v2) { 469; CHECK-LABEL: @shuffle_17addnuw( 470; CHECK-NEXT: [[TMP1:%.*]] = add nuw <4 x i32> [[V1:%.*]], [[V2:%.*]] 471; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 472; CHECK-NEXT: ret <4 x i32> [[R]] 473; 474 %t1 = shufflevector <4 x i32> %v1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 475 %t2 = shufflevector <4 x i32> %v2, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 476 %r = add nuw <4 x i32> %t1, %t2 477 ret <4 x i32> %r 478} 479 480define <4 x float> @shuffle_17fsub_fast(<4 x float> %v1, <4 x float> %v2) { 481; CHECK-LABEL: @shuffle_17fsub_fast( 482; CHECK-NEXT: [[TMP1:%.*]] = fsub fast <4 x float> [[V1:%.*]], [[V2:%.*]] 483; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 484; CHECK-NEXT: ret <4 x float> [[R]] 485; 486 %t1 = shufflevector <4 x float> %v1, <4 x float> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 487 %t2 = shufflevector <4 x float> %v2, <4 x float> zeroinitializer, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 488 %r = fsub fast <4 x float> %t1, %t2 489 ret <4 x float> %r 490} 491 492define <4 x i32> @add_const(<4 x i32> %v) { 493; CHECK-LABEL: @add_const( 494; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[V:%.*]], <i32 44, i32 41, i32 42, i32 43> 495; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 496; CHECK-NEXT: ret <4 x i32> [[R]] 497; 498 %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0> 499 %r = add <4 x i32> %t1, <i32 41, i32 42, i32 43, i32 44> 500 ret <4 x i32> %r 501} 502 503define <4 x i32> @sub_const(<4 x i32> %v) { 504; CHECK-LABEL: @sub_const( 505; CHECK-NEXT: [[TMP1:%.*]] = sub <4 x i32> <i32 44, i32 43, i32 42, i32 41>, [[V:%.*]] 506; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 507; CHECK-NEXT: ret <4 x i32> [[R]] 508; 509 %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 510 %r = sub <4 x i32> <i32 41, i32 42, i32 43, i32 44>, %t1 511 ret <4 x i32> %r 512} 513 514; Math before shuffle requires an extra shuffle. 515 516define <2 x float> @fadd_const_multiuse(<2 x float> %v) { 517; CHECK-LABEL: @fadd_const_multiuse( 518; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 0> 519; CHECK-NEXT: [[R:%.*]] = fadd <2 x float> [[T1]], <float 4.100000e+01, float 4.200000e+01> 520; CHECK-NEXT: call void @use(<2 x float> [[T1]]) 521; CHECK-NEXT: ret <2 x float> [[R]] 522; 523 %t1 = shufflevector <2 x float> %v, <2 x float> undef, <2 x i32> <i32 1, i32 0> 524 %r = fadd <2 x float> %t1, <float 41.0, float 42.0> 525 call void @use(<2 x float> %t1) 526 ret <2 x float> %r 527} 528 529; Math before splat allows replacing constant elements with undef lanes. 530 531define <4 x i32> @mul_const_splat(<4 x i32> %v) { 532; CHECK-LABEL: @mul_const_splat( 533; CHECK-NEXT: [[TMP1:%.*]] = mul <4 x i32> [[V:%.*]], <i32 undef, i32 42, i32 undef, i32 undef> 534; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 535; CHECK-NEXT: ret <4 x i32> [[R]] 536; 537 %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 538 %r = mul <4 x i32> <i32 42, i32 42, i32 42, i32 42>, %t1 539 ret <4 x i32> %r 540} 541 542; Take 2 elements of a vector and shift each of those by a different amount 543 544define <4 x i32> @lshr_const_half_splat(<4 x i32> %v) { 545; CHECK-LABEL: @lshr_const_half_splat( 546; CHECK-NEXT: [[TMP1:%.*]] = lshr <4 x i32> <i32 undef, i32 8, i32 9, i32 undef>, [[V:%.*]] 547; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2> 548; CHECK-NEXT: ret <4 x i32> [[R]] 549; 550 %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2> 551 %r = lshr <4 x i32> <i32 8, i32 8, i32 9, i32 9>, %t1 552 ret <4 x i32> %r 553} 554 555; We can't change this because there's no pre-shuffle version of the fmul constant. 556 557define <2 x float> @fmul_const_invalid_constant(<2 x float> %v) { 558; CHECK-LABEL: @fmul_const_invalid_constant( 559; CHECK-NEXT: [[T1:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> undef, <2 x i32> zeroinitializer 560; CHECK-NEXT: [[R:%.*]] = fmul <2 x float> [[T1]], <float 4.100000e+01, float 4.200000e+01> 561; CHECK-NEXT: ret <2 x float> [[R]] 562; 563 %t1 = shufflevector <2 x float> %v, <2 x float> undef, <2 x i32> <i32 0, i32 0> 564 %r = fmul <2 x float> %t1, <float 41.0, float 42.0> 565 ret <2 x float> %r 566} 567 568; Reduce the width of the binop by moving it ahead of a shuffle. 569 570define <4 x i8> @widening_shuffle_add_1(<2 x i8> %x) { 571; CHECK-LABEL: @widening_shuffle_add_1( 572; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i8> [[X:%.*]], <i8 42, i8 43> 573; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i8> [[TMP1]], <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 574; CHECK-NEXT: ret <4 x i8> [[R]] 575; 576 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 577 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 578 ret <4 x i8> %r 579} 580 581; Reduce the width of the binop by moving it ahead of a shuffle. 582 583define <4 x i8> @widening_shuffle_add_2(<2 x i8> %x) { 584; CHECK-LABEL: @widening_shuffle_add_2( 585; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i8> [[X:%.*]], <i8 43, i8 42> 586; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i8> [[TMP1]], <2 x i8> undef, <4 x i32> <i32 1, i32 0, i32 undef, i32 undef> 587; CHECK-NEXT: ret <4 x i8> [[R]] 588; 589 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 1, i32 0, i32 undef, i32 undef> 590 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 591 ret <4 x i8> %r 592} 593 594; Negative test - widening shuffles have the same mask/constant constraint as non-size-changing shuffles. 595 596define <4 x i8> @widening_shuffle_add_invalid_constant(<2 x i8> %x) { 597; CHECK-LABEL: @widening_shuffle_add_invalid_constant( 598; CHECK-NEXT: [[WIDEX:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> undef, <4 x i32> <i32 1, i32 1, i32 undef, i32 undef> 599; CHECK-NEXT: [[R:%.*]] = add <4 x i8> [[WIDEX]], <i8 42, i8 43, i8 44, i8 45> 600; CHECK-NEXT: ret <4 x i8> [[R]] 601; 602 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 1, i32 1, i32 undef, i32 undef> 603 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 604 ret <4 x i8> %r 605} 606 607; Negative test - widening shuffles have an additional constraint: they must not extend with anything but undefs. 608 609define <4 x i8> @widening_shuffle_add_invalid_mask(<2 x i8> %x) { 610; CHECK-LABEL: @widening_shuffle_add_invalid_mask( 611; CHECK-NEXT: [[WIDEX:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 0> 612; CHECK-NEXT: [[R:%.*]] = add <4 x i8> [[WIDEX]], <i8 42, i8 43, i8 44, i8 45> 613; CHECK-NEXT: ret <4 x i8> [[R]] 614; 615 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 0> 616 %r = add <4 x i8> %widex, <i8 42, i8 43, i8 44, i8 45> 617 ret <4 x i8> %r 618} 619 620; A binop that produces undef in the high lanes can be moved before the shuffle. 621; This is ok because 'shl C, undef --> undef'. 622 623define <4 x i16> @widening_shuffle_shl_constant_op0(<2 x i16> %v) { 624; CHECK-LABEL: @widening_shuffle_shl_constant_op0( 625; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i16> <i16 42, i16 -42>, [[V:%.*]] 626; CHECK-NEXT: [[BO:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 627; CHECK-NEXT: ret <4 x i16> [[BO]] 628; 629 %shuf = shufflevector <2 x i16> %v, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 630 %bo = shl <4 x i16> <i16 42, i16 -42, i16 -1, i16 -1>, %shuf 631 ret <4 x i16> %bo 632} 633 634; A binop that produces undef in the high lanes can be moved before the shuffle. 635; This is ok because 'shl undef, 0 --> undef'. 636 637define <4 x i16> @widening_shuffle_shl_constant_op1(<2 x i16> %v) { 638; CHECK-LABEL: @widening_shuffle_shl_constant_op1( 639; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i16> [[V:%.*]], <i16 2, i16 4> 640; CHECK-NEXT: [[BO:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 641; CHECK-NEXT: ret <4 x i16> [[BO]] 642; 643 %shuf = shufflevector <2 x i16> %v, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 644 %bo = shl <4 x i16> %shuf, <i16 2, i16 4, i16 0, i16 0> 645 ret <4 x i16> %bo 646} 647 648; A binop that does not produce undef in the high lanes can not be moved before the shuffle. 649; This is not ok because 'shl undef, 1 (or 2)' --> 0' but moving the shuffle results in undef instead. 650 651define <4 x i16> @widening_shuffle_shl_constant_op1_non0(<2 x i16> %v) { 652; CHECK-LABEL: @widening_shuffle_shl_constant_op1_non0( 653; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <2 x i16> [[V:%.*]], <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 654; CHECK-NEXT: [[BO:%.*]] = shl <4 x i16> [[SHUF]], <i16 2, i16 4, i16 1, i16 2> 655; CHECK-NEXT: ret <4 x i16> [[BO]] 656; 657 %shuf = shufflevector <2 x i16> %v, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 658 %bo = shl <4 x i16> %shuf, <i16 2, i16 4, i16 1, i16 2> 659 ret <4 x i16> %bo 660} 661 662; A binop that does not produce undef in the high lanes can not be moved before the shuffle. 663; This is not ok because 'or -1, undef --> -1' but moving the shuffle results in undef instead. 664 665define <4 x i16> @widening_shuffle_or(<2 x i16> %v) { 666; CHECK-LABEL: @widening_shuffle_or( 667; CHECK-NEXT: [[SHUF:%.*]] = shufflevector <2 x i16> [[V:%.*]], <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 668; CHECK-NEXT: [[BO:%.*]] = or <4 x i16> [[SHUF]], <i16 42, i16 -42, i16 -1, i16 -1> 669; CHECK-NEXT: ret <4 x i16> [[BO]] 670; 671 %shuf = shufflevector <2 x i16> %v, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 672 %bo = or <4 x i16> %shuf, <i16 42, i16 -42, i16 -1, i16 -1> 673 ret <4 x i16> %bo 674} 675 676define <4 x i32> @shuffle_17add2(<4 x i32> %v) { 677; CHECK-LABEL: @shuffle_17add2( 678; CHECK-NEXT: [[TMP1:%.*]] = shl <4 x i32> [[V:%.*]], <i32 1, i32 1, i32 1, i32 1> 679; CHECK-NEXT: ret <4 x i32> [[TMP1]] 680; 681 %t1 = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 682 %t2 = add <4 x i32> %t1, %t1 683 %r = shufflevector <4 x i32> %t2, <4 x i32> zeroinitializer, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 684 ret <4 x i32> %r 685} 686 687define <4 x i32> @shuffle_17mulsplat(<4 x i32> %v) { 688; CHECK-LABEL: @shuffle_17mulsplat( 689; CHECK-NEXT: [[TMP1:%.*]] = mul <4 x i32> [[V:%.*]], [[V]] 690; CHECK-NEXT: [[M1:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> zeroinitializer 691; CHECK-NEXT: ret <4 x i32> [[M1]] 692; 693 %s1 = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer 694 %m1 = mul <4 x i32> %s1, %s1 695 %s2 = shufflevector <4 x i32> %m1, <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 1, i32 1, i32 1> 696 ret <4 x i32> %s2 697} 698 699; Do not reorder shuffle and binop if LHS of shuffles are of different size 700define <2 x i32> @pr19717(<4 x i32> %in0, <2 x i32> %in1) { 701; CHECK-LABEL: @pr19717( 702; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[IN0:%.*]], <4 x i32> undef, <2 x i32> zeroinitializer 703; CHECK-NEXT: [[SHUFFLE4:%.*]] = shufflevector <2 x i32> [[IN1:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 704; CHECK-NEXT: [[MUL:%.*]] = mul <2 x i32> [[SHUFFLE]], [[SHUFFLE4]] 705; CHECK-NEXT: ret <2 x i32> [[MUL]] 706; 707 %shuffle = shufflevector <4 x i32> %in0, <4 x i32> %in0, <2 x i32> zeroinitializer 708 %shuffle4 = shufflevector <2 x i32> %in1, <2 x i32> %in1, <2 x i32> zeroinitializer 709 %mul = mul <2 x i32> %shuffle, %shuffle4 710 ret <2 x i32> %mul 711} 712 713define <4 x i16> @pr19717a(<8 x i16> %in0, <8 x i16> %in1) { 714; CHECK-LABEL: @pr19717a( 715; CHECK-NEXT: [[TMP1:%.*]] = mul <8 x i16> [[IN0:%.*]], [[IN1:%.*]] 716; CHECK-NEXT: [[MUL:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> undef, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 717; CHECK-NEXT: ret <4 x i16> [[MUL]] 718; 719 %shuffle = shufflevector <8 x i16> %in0, <8 x i16> %in0, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 720 %shuffle1 = shufflevector <8 x i16> %in1, <8 x i16> %in1, <4 x i32> <i32 5, i32 5, i32 5, i32 5> 721 %mul = mul <4 x i16> %shuffle, %shuffle1 722 ret <4 x i16> %mul 723} 724 725define <8 x i8> @pr19730(<16 x i8> %in0) { 726; CHECK-LABEL: @pr19730( 727; 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> 728; 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> 729; CHECK-NEXT: ret <8 x i8> [[SHUFFLE1]] 730; 731 %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> 732 %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> 733 ret <8 x i8> %shuffle1 734} 735 736define i32 @pr19737(<4 x i32> %in0) { 737; CHECK-LABEL: @pr19737( 738; CHECK-NEXT: [[TMP1:%.*]] = extractelement <4 x i32> [[IN0:%.*]], i32 0 739; CHECK-NEXT: ret i32 [[TMP1]] 740; 741 %shuffle.i = shufflevector <4 x i32> zeroinitializer, <4 x i32> %in0, <4 x i32> <i32 0, i32 4, i32 2, i32 6> 742 %neg.i = xor <4 x i32> %shuffle.i, <i32 -1, i32 -1, i32 -1, i32 -1> 743 %and.i = and <4 x i32> %in0, %neg.i 744 %rv = extractelement <4 x i32> %and.i, i32 0 745 ret i32 %rv 746} 747 748; In PR20059 ( http://llvm.org/pr20059 ), shufflevector operations are reordered/removed 749; for an srem operation. This is not a valid optimization because it may cause a trap 750; on div-by-zero. 751 752define <4 x i32> @pr20059(<4 x i32> %p1, <4 x i32> %p2) { 753; CHECK-LABEL: @pr20059( 754; CHECK-NEXT: [[SPLAT1:%.*]] = shufflevector <4 x i32> [[P1:%.*]], <4 x i32> undef, <4 x i32> zeroinitializer 755; CHECK-NEXT: [[SPLAT2:%.*]] = shufflevector <4 x i32> [[P2:%.*]], <4 x i32> undef, <4 x i32> zeroinitializer 756; CHECK-NEXT: [[RETVAL:%.*]] = srem <4 x i32> [[SPLAT1]], [[SPLAT2]] 757; CHECK-NEXT: ret <4 x i32> [[RETVAL]] 758; 759 %splat1 = shufflevector <4 x i32> %p1, <4 x i32> undef, <4 x i32> zeroinitializer 760 %splat2 = shufflevector <4 x i32> %p2, <4 x i32> undef, <4 x i32> zeroinitializer 761 %retval = srem <4 x i32> %splat1, %splat2 762 ret <4 x i32> %retval 763} 764 765define <4 x i32> @pr20114(<4 x i32> %__mask) { 766; CHECK-LABEL: @pr20114( 767; CHECK-NEXT: [[MASK01_I:%.*]] = shufflevector <4 x i32> [[__MASK:%.*]], <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 1, i32 1> 768; 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>) 769; CHECK-NEXT: ret <4 x i32> [[MASKED_NEW_I_I_I]] 770; 771 %mask01.i = shufflevector <4 x i32> %__mask, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 1, i32 1> 772 %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 773 ret <4 x i32> %masked_new.i.i.i 774} 775 776define <2 x i32*> @pr23113(<4 x i32*> %A) { 777; CHECK-LABEL: @pr23113( 778; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32*> [[A:%.*]], <4 x i32*> undef, <2 x i32> <i32 0, i32 1> 779; CHECK-NEXT: ret <2 x i32*> [[TMP1]] 780; 781 %1 = shufflevector <4 x i32*> %A, <4 x i32*> undef, <2 x i32> <i32 0, i32 1> 782 ret <2 x i32*> %1 783} 784 785; Unused lanes in the new binop should not kill the entire op (although it may simplify anyway as shown here). 786 787define <2 x i32> @PR37648(<2 x i32> %x) { 788; CHECK-LABEL: @PR37648( 789; CHECK-NEXT: ret <2 x i32> zeroinitializer 790; 791 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 792 %r = urem <2 x i32> %splat, <i32 1, i32 1> 793 ret <2 x i32> %r 794} 795 796; Test shuffle followed by binop with splat constant for all 18 binop opcodes. 797; Test with constant as operand 0 and operand 1 for non-commutative opcodes. 798 799define <2 x i32> @add_splat_constant(<2 x i32> %x) { 800; CHECK-LABEL: @add_splat_constant( 801; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 42, i32 undef> 802; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 803; CHECK-NEXT: ret <2 x i32> [[R]] 804; 805 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 806 %r = add <2 x i32> %splat, <i32 42, i32 42> 807 ret <2 x i32> %r 808} 809 810define <2 x i32> @sub_splat_constant0(<2 x i32> %x) { 811; CHECK-LABEL: @sub_splat_constant0( 812; CHECK-NEXT: [[TMP1:%.*]] = sub <2 x i32> <i32 42, i32 undef>, [[X:%.*]] 813; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 814; CHECK-NEXT: ret <2 x i32> [[R]] 815; 816 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 817 %r = sub <2 x i32> <i32 42, i32 42>, %splat 818 ret <2 x i32> %r 819} 820 821define <2 x i32> @sub_splat_constant1(<2 x i32> %x) { 822; CHECK-LABEL: @sub_splat_constant1( 823; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 -42, i32 undef> 824; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 825; CHECK-NEXT: ret <2 x i32> [[R]] 826; 827 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 828 %r = sub <2 x i32> %splat, <i32 42, i32 42> 829 ret <2 x i32> %r 830} 831 832define <2 x i32> @mul_splat_constant(<2 x i32> %x) { 833; CHECK-LABEL: @mul_splat_constant( 834; CHECK-NEXT: [[TMP1:%.*]] = mul <2 x i32> [[X:%.*]], <i32 42, i32 undef> 835; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 836; CHECK-NEXT: ret <2 x i32> [[R]] 837; 838 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 839 %r = mul <2 x i32> %splat, <i32 42, i32 42> 840 ret <2 x i32> %r 841} 842 843define <2 x i32> @shl_splat_constant0(<2 x i32> %x) { 844; CHECK-LABEL: @shl_splat_constant0( 845; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> <i32 5, i32 undef>, [[X:%.*]] 846; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 847; CHECK-NEXT: ret <2 x i32> [[R]] 848; 849 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 850 %r = shl <2 x i32> <i32 5, i32 5>, %splat 851 ret <2 x i32> %r 852} 853 854define <2 x i32> @shl_splat_constant1(<2 x i32> %x) { 855; CHECK-LABEL: @shl_splat_constant1( 856; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> [[X:%.*]], <i32 5, i32 0> 857; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 858; CHECK-NEXT: ret <2 x i32> [[R]] 859; 860 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 861 %r = shl <2 x i32> %splat, <i32 5, i32 5> 862 ret <2 x i32> %r 863} 864 865define <2 x i32> @ashr_splat_constant0(<2 x i32> %x) { 866; CHECK-LABEL: @ashr_splat_constant0( 867; CHECK-NEXT: [[TMP1:%.*]] = ashr <2 x i32> <i32 5, i32 undef>, [[X:%.*]] 868; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 869; CHECK-NEXT: ret <2 x i32> [[R]] 870; 871 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 872 %r = ashr <2 x i32> <i32 5, i32 5>, %splat 873 ret <2 x i32> %r 874} 875 876define <2 x i32> @ashr_splat_constant1(<2 x i32> %x) { 877; CHECK-LABEL: @ashr_splat_constant1( 878; CHECK-NEXT: [[TMP1:%.*]] = ashr <2 x i32> [[X:%.*]], <i32 5, i32 0> 879; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 880; CHECK-NEXT: ret <2 x i32> [[R]] 881; 882 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 883 %r = ashr <2 x i32> %splat, <i32 5, i32 5> 884 ret <2 x i32> %r 885} 886 887define <2 x i32> @lshr_splat_constant0(<2 x i32> %x) { 888; CHECK-LABEL: @lshr_splat_constant0( 889; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i32> <i32 5, i32 undef>, [[X:%.*]] 890; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 891; CHECK-NEXT: ret <2 x i32> [[R]] 892; 893 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 894 %r = lshr <2 x i32> <i32 5, i32 5>, %splat 895 ret <2 x i32> %r 896} 897 898define <2 x i32> @lshr_splat_constant1(<2 x i32> %x) { 899; CHECK-LABEL: @lshr_splat_constant1( 900; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i32> [[X:%.*]], <i32 5, i32 0> 901; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 902; CHECK-NEXT: ret <2 x i32> [[R]] 903; 904 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 905 %r = lshr <2 x i32> %splat, <i32 5, i32 5> 906 ret <2 x i32> %r 907} 908 909define <2 x i32> @urem_splat_constant0(<2 x i32> %x) { 910; CHECK-LABEL: @urem_splat_constant0( 911; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 912; CHECK-NEXT: [[R:%.*]] = urem <2 x i32> <i32 42, i32 42>, [[SPLAT]] 913; CHECK-NEXT: ret <2 x i32> [[R]] 914; 915 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 916 %r = urem <2 x i32> <i32 42, i32 42>, %splat 917 ret <2 x i32> %r 918} 919 920define <2 x i32> @urem_splat_constant1(<2 x i32> %x) { 921; CHECK-LABEL: @urem_splat_constant1( 922; CHECK-NEXT: [[TMP1:%.*]] = urem <2 x i32> [[X:%.*]], <i32 42, i32 1> 923; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 924; CHECK-NEXT: ret <2 x i32> [[R]] 925; 926 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 927 %r = urem <2 x i32> %splat, <i32 42, i32 42> 928 ret <2 x i32> %r 929} 930 931define <2 x i32> @srem_splat_constant0(<2 x i32> %x) { 932; CHECK-LABEL: @srem_splat_constant0( 933; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 934; CHECK-NEXT: [[R:%.*]] = srem <2 x i32> <i32 42, i32 42>, [[SPLAT]] 935; CHECK-NEXT: ret <2 x i32> [[R]] 936; 937 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 938 %r = srem <2 x i32> <i32 42, i32 42>, %splat 939 ret <2 x i32> %r 940} 941 942define <2 x i32> @srem_splat_constant1(<2 x i32> %x) { 943; CHECK-LABEL: @srem_splat_constant1( 944; CHECK-NEXT: [[TMP1:%.*]] = srem <2 x i32> [[X:%.*]], <i32 42, i32 1> 945; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 946; CHECK-NEXT: ret <2 x i32> [[R]] 947; 948 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 949 %r = srem <2 x i32> %splat, <i32 42, i32 42> 950 ret <2 x i32> %r 951} 952 953define <2 x i32> @udiv_splat_constant0(<2 x i32> %x) { 954; CHECK-LABEL: @udiv_splat_constant0( 955; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 956; CHECK-NEXT: [[R:%.*]] = udiv <2 x i32> <i32 42, i32 42>, [[SPLAT]] 957; CHECK-NEXT: ret <2 x i32> [[R]] 958; 959 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 960 %r = udiv <2 x i32> <i32 42, i32 42>, %splat 961 ret <2 x i32> %r 962} 963 964define <2 x i32> @udiv_splat_constant1(<2 x i32> %x) { 965; CHECK-LABEL: @udiv_splat_constant1( 966; CHECK-NEXT: [[TMP1:%.*]] = udiv <2 x i32> [[X:%.*]], <i32 42, i32 1> 967; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 968; CHECK-NEXT: ret <2 x i32> [[R]] 969; 970 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 971 %r = udiv <2 x i32> %splat, <i32 42, i32 42> 972 ret <2 x i32> %r 973} 974 975define <2 x i32> @sdiv_splat_constant0(<2 x i32> %x) { 976; CHECK-LABEL: @sdiv_splat_constant0( 977; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer 978; CHECK-NEXT: [[R:%.*]] = sdiv <2 x i32> <i32 42, i32 42>, [[SPLAT]] 979; CHECK-NEXT: ret <2 x i32> [[R]] 980; 981 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 982 %r = sdiv <2 x i32> <i32 42, i32 42>, %splat 983 ret <2 x i32> %r 984} 985 986define <2 x i32> @sdiv_splat_constant1(<2 x i32> %x) { 987; CHECK-LABEL: @sdiv_splat_constant1( 988; CHECK-NEXT: [[TMP1:%.*]] = sdiv <2 x i32> [[X:%.*]], <i32 42, i32 1> 989; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 990; CHECK-NEXT: ret <2 x i32> [[R]] 991; 992 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 993 %r = sdiv <2 x i32> %splat, <i32 42, i32 42> 994 ret <2 x i32> %r 995} 996 997define <2 x i32> @and_splat_constant(<2 x i32> %x) { 998; CHECK-LABEL: @and_splat_constant( 999; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i32> [[X:%.*]], <i32 42, i32 undef> 1000; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 1001; CHECK-NEXT: ret <2 x i32> [[R]] 1002; 1003 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 1004 %r = and <2 x i32> %splat, <i32 42, i32 42> 1005 ret <2 x i32> %r 1006} 1007 1008; AND does not fold to undef for undef operands, we cannot move it 1009; across a shuffle with undef masks. 1010define <4 x i16> @and_constant_mask_undef(<4 x i16> %add) { 1011; CHECK-LABEL: @and_constant_mask_undef( 1012; CHECK-NEXT: entry: 1013; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[ADD:%.*]], <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1014; CHECK-NEXT: [[AND:%.*]] = and <4 x i16> [[SHUFFLE]], <i16 0, i16 0, i16 -1, i16 -1> 1015; CHECK-NEXT: ret <4 x i16> [[AND]] 1016; 1017entry: 1018 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1019 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 -1, i16 -1> 1020 ret <4 x i16> %and 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_2(<4 x i16> %add) { 1026; CHECK-LABEL: @and_constant_mask_undef_2( 1027; CHECK-NEXT: entry: 1028; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[ADD:%.*]], <4 x i16> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 undef> 1029; CHECK-NEXT: [[AND:%.*]] = and <4 x i16> [[SHUFFLE]], <i16 -1, i16 -1, i16 -1, i16 0> 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 1, i32 1, i32 1, i32 undef> 1034 %and = and <4 x i16> %shuffle, <i16 -1, i16 -1, i16 -1, i16 -0> 1035 ret <4 x i16> %and 1036} 1037 1038; We can move the AND across the shuffle, as -1 (AND identity value) is used for undef lanes. 1039define <4 x i16> @and_constant_mask_undef_3(<4 x i16> %add) { 1040; CHECK-LABEL: @and_constant_mask_undef_3( 1041; CHECK-NEXT: entry: 1042; CHECK-NEXT: ret <4 x i16> <i16 0, i16 0, i16 0, i16 undef> 1043; 1044entry: 1045 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 0, i32 1, i32 1, i32 undef> 1046 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 0, i16 -1> 1047 ret <4 x i16> %and 1048} 1049 1050; We can move the AND across the shuffle, as -1 (AND identity value) is used for undef lanes. 1051define <4 x i16> @and_constant_mask_undef_4(<4 x i16> %add) { 1052; CHECK-LABEL: @and_constant_mask_undef_4( 1053; CHECK-NEXT: entry: 1054; CHECK-NEXT: [[TMP0:%.*]] = and <4 x i16> [[ADD:%.*]], <i16 9, i16 20, i16 undef, i16 undef> 1055; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> undef, <4 x i32> <i32 0, i32 1, i32 1, i32 undef> 1056; CHECK-NEXT: ret <4 x i16> [[AND]] 1057; 1058entry: 1059 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 0, i32 1, i32 1, i32 undef> 1060 %and = and <4 x i16> %shuffle, <i16 9, i16 20, i16 20, i16 -1> 1061 ret <4 x i16> %and 1062} 1063 1064define <4 x i16> @and_constant_mask_not_undef(<4 x i16> %add) { 1065; CHECK-LABEL: @and_constant_mask_not_undef( 1066; CHECK-NEXT: entry: 1067; CHECK-NEXT: [[TMP0:%.*]] = and <4 x i16> [[ADD:%.*]], <i16 undef, i16 -1, i16 0, i16 0> 1068; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> undef, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1069; CHECK-NEXT: ret <4 x i16> [[AND]] 1070; 1071entry: 1072 %shuffle = shufflevector <4 x i16> %add, <4 x i16> undef, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1073 %and = and <4 x i16> %shuffle, <i16 0, i16 0, i16 -1, i16 -1> 1074 ret <4 x i16> %and 1075} 1076 1077; OR does not fold to undef for undef operands, we cannot move it 1078; across a shuffle with undef masks. 1079define <4 x i16> @or_constant_mask_undef(<4 x i16> %in) { 1080; CHECK-LABEL: @or_constant_mask_undef( 1081; CHECK-NEXT: entry: 1082; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1083; CHECK-NEXT: [[OR:%.*]] = or <4 x i16> [[SHUFFLE]], <i16 -1, i16 -1, i16 0, i16 0> 1084; CHECK-NEXT: ret <4 x i16> [[OR]] 1085; 1086entry: 1087 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1088 %or = or <4 x i16> %shuffle, <i16 -1, i16 -1, i16 0, i16 0> 1089 ret <4 x i16> %or 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_2(<4 x i16> %in) { 1095; CHECK-LABEL: @or_constant_mask_undef_2( 1096; CHECK-NEXT: entry: 1097; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> undef, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef> 1098; CHECK-NEXT: [[OR:%.*]] = or <4 x i16> [[SHUFFLE]], <i16 -1, i16 0, i16 0, i16 -1> 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 1, i32 1, i32 undef> 1103 %or = or <4 x i16> %shuffle, <i16 -1, i16 0, i16 0, i16 -1> 1104 ret <4 x i16> %or 1105} 1106 1107; We can move the OR across the shuffle, as 0 (OR identity value) is used for undef lanes. 1108define <4 x i16> @or_constant_mask_undef_3(<4 x i16> %in) { 1109; CHECK-LABEL: @or_constant_mask_undef_3( 1110; CHECK-NEXT: entry: 1111; CHECK-NEXT: ret <4 x i16> <i16 undef, i16 -1, i16 -1, i16 undef> 1112; 1113entry: 1114 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef> 1115 %or = or <4 x i16> %shuffle, <i16 0, i16 -1, i16 -1, i16 0> 1116 ret <4 x i16> %or 1117} 1118 1119; We can move the OR across the shuffle, as 0 (OR identity value) is used for undef lanes. 1120define <4 x i16> @or_constant_mask_undef_4(<4 x i16> %in) { 1121; CHECK-LABEL: @or_constant_mask_undef_4( 1122; CHECK-NEXT: entry: 1123; CHECK-NEXT: [[TMP0:%.*]] = or <4 x i16> [[IN:%.*]], <i16 undef, i16 99, i16 undef, i16 undef> 1124; CHECK-NEXT: [[OR:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> undef, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef> 1125; CHECK-NEXT: ret <4 x i16> [[OR]] 1126; 1127entry: 1128 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef> 1129 %or = or <4 x i16> %shuffle, <i16 0, i16 99, i16 99, i16 0> 1130 ret <4 x i16> %or 1131} 1132 1133define <4 x i16> @or_constant_mask_not_undef(<4 x i16> %in) { 1134; CHECK-LABEL: @or_constant_mask_not_undef( 1135; CHECK-NEXT: entry: 1136; CHECK-NEXT: [[TMP0:%.*]] = or <4 x i16> [[IN:%.*]], <i16 undef, i16 -1, i16 0, i16 0> 1137; CHECK-NEXT: [[AND:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> undef, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1138; CHECK-NEXT: ret <4 x i16> [[AND]] 1139; 1140entry: 1141 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 2, i32 3, i32 1, i32 1> 1142 %and = or <4 x i16> %shuffle, <i16 0, i16 0, i16 -1, i16 -1> 1143 ret <4 x i16> %and 1144} 1145 1146define <4 x i16> @shl_constant_mask_undef(<4 x i16> %in) { 1147; CHECK-LABEL: @shl_constant_mask_undef( 1148; CHECK-NEXT: entry: 1149; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> undef, <4 x i32> <i32 0, i32 undef, i32 1, i32 1> 1150; CHECK-NEXT: [[SHL:%.*]] = shl <4 x i16> [[SHUFFLE]], <i16 10, i16 3, i16 0, i16 0> 1151; CHECK-NEXT: ret <4 x i16> [[SHL]] 1152; 1153entry: 1154 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 0, i32 undef, i32 1, i32 1> 1155 %shl = shl <4 x i16> %shuffle, <i16 10, i16 3, i16 0, i16 0> 1156 ret <4 x i16> %shl 1157} 1158 1159define <4 x i16> @add_constant_mask_undef(<4 x i16> %in) { 1160; CHECK-LABEL: @add_constant_mask_undef( 1161; CHECK-NEXT: entry: 1162; CHECK-NEXT: [[ADD:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1163; CHECK-NEXT: ret <4 x i16> [[ADD]] 1164; 1165entry: 1166 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1167 %add = add <4 x i16> %shuffle, <i16 10, i16 3, i16 0, i16 0> 1168 ret <4 x i16> %add 1169} 1170 1171define <4 x i16> @add_constant_mask_undef_2(<4 x i16> %in) { 1172; CHECK-LABEL: @add_constant_mask_undef_2( 1173; CHECK-NEXT: entry: 1174; CHECK-NEXT: [[TMP0:%.*]] = add <4 x i16> [[IN:%.*]], <i16 undef, i16 0, i16 3, i16 undef> 1175; CHECK-NEXT: [[ADD:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> undef, <4 x i32> <i32 undef, i32 2, i32 1, i32 1> 1176; CHECK-NEXT: ret <4 x i16> [[ADD]] 1177; 1178entry: 1179 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 undef, i32 2, i32 1, i32 1> 1180 %add = add <4 x i16> %shuffle, <i16 10, i16 3, i16 0, i16 0> 1181 ret <4 x i16> %add 1182} 1183 1184define <4 x i16> @sub_constant_mask_undef(<4 x i16> %in) { 1185; CHECK-LABEL: @sub_constant_mask_undef( 1186; CHECK-NEXT: entry: 1187; CHECK-NEXT: [[SUB:%.*]] = shufflevector <4 x i16> [[IN:%.*]], <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1188; CHECK-NEXT: ret <4 x i16> [[SUB]] 1189; 1190entry: 1191 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 1> 1192 %sub = sub <4 x i16> %shuffle, <i16 10, i16 3, i16 0, i16 0> 1193 ret <4 x i16> %sub 1194} 1195 1196define <4 x i16> @sub_constant_mask_undef_2(<4 x i16> %in) { 1197; CHECK-LABEL: @sub_constant_mask_undef_2( 1198; CHECK-NEXT: entry: 1199; CHECK-NEXT: [[TMP0:%.*]] = add <4 x i16> [[IN:%.*]], <i16 undef, i16 0, i16 -10, i16 undef> 1200; CHECK-NEXT: [[SUB:%.*]] = shufflevector <4 x i16> [[TMP0]], <4 x i16> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 undef> 1201; CHECK-NEXT: ret <4 x i16> [[SUB]] 1202; 1203entry: 1204 %shuffle = shufflevector <4 x i16> %in, <4 x i16> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 undef> 1205 %sub = sub <4 x i16> %shuffle, <i16 0, i16 0, i16 10, i16 99> 1206 ret <4 x i16> %sub 1207} 1208 1209define <2 x i32> @or_splat_constant(<2 x i32> %x) { 1210; CHECK-LABEL: @or_splat_constant( 1211; CHECK-NEXT: [[TMP1:%.*]] = or <2 x i32> [[X:%.*]], <i32 42, i32 undef> 1212; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 1213; CHECK-NEXT: ret <2 x i32> [[R]] 1214; 1215 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 1216 %r = or <2 x i32> %splat, <i32 42, i32 42> 1217 ret <2 x i32> %r 1218} 1219 1220define <2 x i32> @xor_splat_constant(<2 x i32> %x) { 1221; CHECK-LABEL: @xor_splat_constant( 1222; CHECK-NEXT: [[TMP1:%.*]] = xor <2 x i32> [[X:%.*]], <i32 42, i32 undef> 1223; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer 1224; CHECK-NEXT: ret <2 x i32> [[R]] 1225; 1226 %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer 1227 %r = xor <2 x i32> %splat, <i32 42, i32 42> 1228 ret <2 x i32> %r 1229} 1230 1231define <2 x float> @fadd_splat_constant(<2 x float> %x) { 1232; CHECK-LABEL: @fadd_splat_constant( 1233; CHECK-NEXT: [[TMP1:%.*]] = fadd <2 x float> [[X:%.*]], <float 4.200000e+01, float undef> 1234; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1235; CHECK-NEXT: ret <2 x float> [[R]] 1236; 1237 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1238 %r = fadd <2 x float> %splat, <float 42.0, float 42.0> 1239 ret <2 x float> %r 1240} 1241 1242define <2 x float> @fsub_splat_constant0(<2 x float> %x) { 1243; CHECK-LABEL: @fsub_splat_constant0( 1244; CHECK-NEXT: [[TMP1:%.*]] = fsub <2 x float> <float 4.200000e+01, float undef>, [[X:%.*]] 1245; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1246; CHECK-NEXT: ret <2 x float> [[R]] 1247; 1248 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1249 %r = fsub <2 x float> <float 42.0, float 42.0>, %splat 1250 ret <2 x float> %r 1251} 1252 1253define <2 x float> @fsub_splat_constant1(<2 x float> %x) { 1254; CHECK-LABEL: @fsub_splat_constant1( 1255; CHECK-NEXT: [[TMP1:%.*]] = fadd <2 x float> [[X:%.*]], <float -4.200000e+01, float undef> 1256; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1257; CHECK-NEXT: ret <2 x float> [[R]] 1258; 1259 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1260 %r = fsub <2 x float> %splat, <float 42.0, float 42.0> 1261 ret <2 x float> %r 1262} 1263 1264define <2 x float> @fneg(<2 x float> %x) { 1265; CHECK-LABEL: @fneg( 1266; CHECK-NEXT: [[TMP1:%.*]] = fneg <2 x float> [[X:%.*]] 1267; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1268; CHECK-NEXT: ret <2 x float> [[R]] 1269; 1270 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1271 %r = fsub <2 x float> <float -0.0, float -0.0>, %splat 1272 ret <2 x float> %r 1273} 1274 1275define <2 x float> @fmul_splat_constant(<2 x float> %x) { 1276; CHECK-LABEL: @fmul_splat_constant( 1277; CHECK-NEXT: [[TMP1:%.*]] = fmul <2 x float> [[X:%.*]], <float 4.200000e+01, float undef> 1278; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1279; CHECK-NEXT: ret <2 x float> [[R]] 1280; 1281 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1282 %r = fmul <2 x float> %splat, <float 42.0, float 42.0> 1283 ret <2 x float> %r 1284} 1285 1286define <2 x float> @fdiv_splat_constant0(<2 x float> %x) { 1287; CHECK-LABEL: @fdiv_splat_constant0( 1288; CHECK-NEXT: [[TMP1:%.*]] = fdiv <2 x float> <float 4.200000e+01, float undef>, [[X:%.*]] 1289; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1290; CHECK-NEXT: ret <2 x float> [[R]] 1291; 1292 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1293 %r = fdiv <2 x float> <float 42.0, float 42.0>, %splat 1294 ret <2 x float> %r 1295} 1296 1297define <2 x float> @fdiv_splat_constant1(<2 x float> %x) { 1298; CHECK-LABEL: @fdiv_splat_constant1( 1299; CHECK-NEXT: [[TMP1:%.*]] = fdiv <2 x float> [[X:%.*]], <float 4.200000e+01, float undef> 1300; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1301; CHECK-NEXT: ret <2 x float> [[R]] 1302; 1303 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1304 %r = fdiv <2 x float> %splat, <float 42.0, float 42.0> 1305 ret <2 x float> %r 1306} 1307 1308define <2 x float> @frem_splat_constant0(<2 x float> %x) { 1309; CHECK-LABEL: @frem_splat_constant0( 1310; CHECK-NEXT: [[TMP1:%.*]] = frem <2 x float> <float 4.200000e+01, float undef>, [[X:%.*]] 1311; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1312; CHECK-NEXT: ret <2 x float> [[R]] 1313; 1314 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1315 %r = frem <2 x float> <float 42.0, float 42.0>, %splat 1316 ret <2 x float> %r 1317} 1318 1319define <2 x float> @frem_splat_constant1(<2 x float> %x) { 1320; CHECK-LABEL: @frem_splat_constant1( 1321; CHECK-NEXT: [[TMP1:%.*]] = frem <2 x float> [[X:%.*]], <float 4.200000e+01, float undef> 1322; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer 1323; CHECK-NEXT: ret <2 x float> [[R]] 1324; 1325 %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1326 %r = frem <2 x float> %splat, <float 42.0, float 42.0> 1327 ret <2 x float> %r 1328} 1329 1330; Equivalent shuffle masks, but only one is a narrowing op. 1331 1332define <2 x i1> @PR40734(<1 x i1> %x, <4 x i1> %y) { 1333; CHECK-LABEL: @PR40734( 1334; CHECK-NEXT: [[WIDEN:%.*]] = shufflevector <1 x i1> zeroinitializer, <1 x i1> [[X:%.*]], <2 x i32> <i32 0, i32 1> 1335; CHECK-NEXT: [[NARROW:%.*]] = shufflevector <4 x i1> [[Y:%.*]], <4 x i1> undef, <2 x i32> <i32 0, i32 1> 1336; CHECK-NEXT: [[R:%.*]] = and <2 x i1> [[WIDEN]], [[NARROW]] 1337; CHECK-NEXT: ret <2 x i1> [[R]] 1338; 1339 %widen = shufflevector <1 x i1> zeroinitializer, <1 x i1> %x, <2 x i32> <i32 0, i32 1> 1340 %narrow = shufflevector <4 x i1> %y, <4 x i1> undef, <2 x i32> <i32 0, i32 1> 1341 %r = and <2 x i1> %widen, %narrow 1342 ret <2 x i1> %r 1343} 1344 1345; Negative test - do not transform non-power-of-2 unless we know the backend handles these sequences identically. 1346 1347define <7 x i8> @insert_subvector_shuffles(<3 x i8> %x, <3 x i8> %y) { 1348; CHECK-LABEL: @insert_subvector_shuffles( 1349; 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> 1350; 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> 1351; 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> 1352; CHECK-NEXT: ret <7 x i8> [[S3]] 1353; 1354 %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> 1355 %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> 1356 %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> 1357 ret <7 x i8> %s3 1358} 1359 1360define <8 x i8> @insert_subvector_shuffles_pow2elts(<2 x i8> %x, <2 x i8> %y) { 1361; CHECK-LABEL: @insert_subvector_shuffles_pow2elts( 1362; 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> 1363; CHECK-NEXT: ret <8 x i8> [[S3]] 1364; 1365 %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> 1366 %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> 1367 %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> 1368 ret <8 x i8> %s3 1369} 1370 1371; The last shuffle may change the vector type. 1372; Negative test - do not transform non-power-of-2 unless we know the backend handles these sequences identically. 1373 1374define <2 x i8> @insert_subvector_shuffles_narrowing(<3 x i8> %x, <3 x i8> %y) { 1375; CHECK-LABEL: @insert_subvector_shuffles_narrowing( 1376; 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> 1377; 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> 1378; CHECK-NEXT: [[S3:%.*]] = shufflevector <7 x i8> [[S1]], <7 x i8> [[S2]], <2 x i32> <i32 0, i32 8> 1379; CHECK-NEXT: ret <2 x i8> [[S3]] 1380; 1381 %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> 1382 %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> 1383 %s3 = shufflevector <7 x i8> %s1, <7 x i8> %s2, <2 x i32> <i32 0, i32 8> 1384 ret <2 x i8> %s3 1385} 1386 1387define <2 x i8> @insert_subvector_shuffles_narrowing_pow2elts(<4 x i8> %x, <4 x i8> %y) { 1388; CHECK-LABEL: @insert_subvector_shuffles_narrowing_pow2elts( 1389; CHECK-NEXT: [[S3:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]], <2 x i32> <i32 0, i32 4> 1390; CHECK-NEXT: ret <2 x i8> [[S3]] 1391; 1392 %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> 1393 %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> 1394 %s3 = shufflevector <8 x i8> %s1, <8 x i8> %s2, <2 x i32> <i32 0, i32 8> 1395 ret <2 x i8> %s3 1396} 1397 1398; Similar to above, but this reduces to a widen with undefs of 'x'. 1399 1400define <4 x double> @insert_subvector_shuffles_identity(<2 x double> %x) { 1401; CHECK-LABEL: @insert_subvector_shuffles_identity( 1402; CHECK-NEXT: [[S3:%.*]] = shufflevector <2 x double> [[X:%.*]], <2 x double> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 1403; CHECK-NEXT: ret <4 x double> [[S3]] 1404; 1405 %s1 = shufflevector <2 x double> %x, <2 x double> undef, <4 x i32> <i32 undef, i32 1, i32 undef, i32 undef> 1406 %s2 = shufflevector <2 x double> %x, <2 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef> 1407 %s3 = shufflevector <4 x double> %s2, <4 x double> %s1, <4 x i32> <i32 0, i32 5, i32 undef, i32 undef> 1408 ret <4 x double> %s3 1409} 1410 1411; Negative test - not identity with padding (although this could be folded with better analysis). 1412 1413define <4 x double> @not_insert_subvector_shuffle(<2 x double> %x) { 1414; CHECK-LABEL: @not_insert_subvector_shuffle( 1415; CHECK-NEXT: [[S1:%.*]] = shufflevector <2 x double> [[X:%.*]], <2 x double> undef, <4 x i32> <i32 undef, i32 1, i32 undef, i32 1> 1416; CHECK-NEXT: [[S2:%.*]] = shufflevector <2 x double> [[X]], <2 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef> 1417; CHECK-NEXT: [[S3:%.*]] = shufflevector <4 x double> [[S2]], <4 x double> [[S1]], <4 x i32> <i32 0, i32 5, i32 7, 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 1> 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 7, i32 undef> 1423 ret <4 x double> %s3 1424} 1425 1426; Negative test - operands are not the same size (although this could be partly folded with better analysis). 1427 1428define <4 x double> @not_insert_subvector_shuffles_with_same_size(<2 x double> %x, <3 x double> %y) { 1429; CHECK-LABEL: @not_insert_subvector_shuffles_with_same_size( 1430; CHECK-NEXT: [[S1:%.*]] = shufflevector <2 x double> [[X:%.*]], <2 x double> undef, <4 x i32> <i32 undef, i32 1, i32 undef, i32 undef> 1431; CHECK-NEXT: [[S2:%.*]] = shufflevector <3 x double> [[Y:%.*]], <3 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 undef, 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 undef> 1436 %s2 = shufflevector <3 x double> %y, <3 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 undef, i32 undef> 1438 ret <4 x double> %s3 1439} 1440 1441; Demanded vector elements may not be able to simplify a shuffle mask 1442; before we try to narrow it. This used to crash. 1443 1444define <4 x float> @insert_subvector_crash_invalid_mask_elt(<2 x float> %x, <4 x float>* %p) { 1445; CHECK-LABEL: @insert_subvector_crash_invalid_mask_elt( 1446; CHECK-NEXT: [[WIDEN:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 1447; CHECK-NEXT: [[I:%.*]] = shufflevector <2 x float> [[X]], <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 1448; CHECK-NEXT: store <4 x float> [[I]], <4 x float>* [[P:%.*]], align 16 1449; CHECK-NEXT: ret <4 x float> [[WIDEN]] 1450; 1451 %widen = shufflevector <2 x float> %x, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 1452 %ext2 = extractelement <2 x float> %x, i32 0 1453 %I = insertelement <4 x float> %widen, float %ext2, i16 0 1454 store <4 x float> %I, <4 x float>* %p 1455 ret <4 x float> %widen 1456} 1457 1458define <4 x i32> @splat_assoc_add(<4 x i32> %x, <4 x i32> %y) { 1459; CHECK-LABEL: @splat_assoc_add( 1460; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[X:%.*]], <i32 317426, i32 undef, i32 undef, i32 undef> 1461; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> zeroinitializer 1462; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[TMP2]], [[Y:%.*]] 1463; CHECK-NEXT: ret <4 x i32> [[R]] 1464; 1465 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> zeroinitializer 1466 %a = add <4 x i32> %y, <i32 317426, i32 317426, i32 317426, i32 317426> 1467 %r = add <4 x i32> %splatx, %a 1468 ret <4 x i32> %r 1469} 1470 1471; Undefs in splat mask are replaced with defined splat index 1472 1473define <4 x i32> @splat_assoc_add_undef_mask_elts(<4 x i32> %x, <4 x i32> %y) { 1474; CHECK-LABEL: @splat_assoc_add_undef_mask_elts( 1475; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[X:%.*]], <i32 42, i32 undef, i32 undef, i32 undef> 1476; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <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> <i32 0, i32 0, i32 undef, i32 undef> 1481 %a = add <4 x i32> %y, <i32 42, i32 42, i32 42, i32 42> 1482 %r = add <4 x i32> %splatx, %a 1483 ret <4 x i32> %r 1484} 1485 1486; Undefs in splat mask are replaced with defined splat index 1487 1488define <4 x i32> @splat_assoc_add_undef_mask_elt_at_splat_index(<4 x i32> %x, <4 x i32> %y) { 1489; CHECK-LABEL: @splat_assoc_add_undef_mask_elt_at_splat_index( 1490; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[X:%.*]], <i32 42, i32 undef, i32 undef, i32 undef> 1491; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> zeroinitializer 1492; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[TMP2]], [[Y:%.*]] 1493; CHECK-NEXT: ret <4 x i32> [[R]] 1494; 1495 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> <i32 undef, i32 0, i32 0, i32 0> 1496 %a = add <4 x i32> %y, <i32 42, i32 42, i32 42, i32 42> 1497 %r = add <4 x i32> %splatx, %a 1498 ret <4 x i32> %r 1499} 1500 1501define <4 x i32> @splat_assoc_add_undef_constant_elts(<4 x i32> %x, <4 x i32> %y) { 1502; CHECK-LABEL: @splat_assoc_add_undef_constant_elts( 1503; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> zeroinitializer 1504; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 42, i32 undef, i32 undef, i32 42> 1505; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1506; CHECK-NEXT: ret <4 x i32> [[R]] 1507; 1508 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> zeroinitializer 1509 %a = add <4 x i32> %y, <i32 42, i32 undef, i32 undef, i32 42> 1510 %r = add <4 x i32> %splatx, %a 1511 ret <4 x i32> %r 1512} 1513 1514define <4 x i32> @splat_assoc_add_undef_constant_elt_at_splat_index(<4 x i32> %x, <4 x i32> %y) { 1515; CHECK-LABEL: @splat_assoc_add_undef_constant_elt_at_splat_index( 1516; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> zeroinitializer 1517; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 undef, i32 42, i32 undef, i32 42> 1518; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1519; CHECK-NEXT: ret <4 x i32> [[R]] 1520; 1521 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> zeroinitializer 1522 %a = add <4 x i32> %y, <i32 undef, i32 42, i32 undef, i32 42> 1523 %r = add <4 x i32> %splatx, %a 1524 ret <4 x i32> %r 1525} 1526 1527define <4 x i32> @splat_assoc_add_undef_mask_elts_undef_constant_elts(<4 x i32> %x, <4 x i32> %y) { 1528; CHECK-LABEL: @splat_assoc_add_undef_mask_elts_undef_constant_elts( 1529; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 0, i32 undef> 1530; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 42, i32 undef, i32 undef, i32 42> 1531; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1532; CHECK-NEXT: ret <4 x i32> [[R]] 1533; 1534 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 0, i32 undef> 1535 %a = add <4 x i32> %y, <i32 42, i32 undef, i32 undef, i32 42> 1536 %r = add <4 x i32> %splatx, %a 1537 ret <4 x i32> %r 1538} 1539 1540define <4 x i32> @splat_assoc_add_undef_mask_elt_at_splat_index_undef_constant_elts(<4 x i32> %x, <4 x i32> %y) { 1541; CHECK-LABEL: @splat_assoc_add_undef_mask_elt_at_splat_index_undef_constant_elts( 1542; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> <i32 undef, i32 0, i32 0, i32 0> 1543; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 42, i32 undef, i32 undef, i32 42> 1544; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1545; CHECK-NEXT: ret <4 x i32> [[R]] 1546; 1547 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> <i32 undef, i32 0, i32 0, i32 0> 1548 %a = add <4 x i32> %y, <i32 42, i32 undef, i32 undef, i32 42> 1549 %r = add <4 x i32> %splatx, %a 1550 ret <4 x i32> %r 1551} 1552 1553define <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) { 1554; CHECK-LABEL: @splat_assoc_add_undef_mask_elt_at_splat_index_undef_constant_elt_at_splat_index( 1555; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> <i32 undef, i32 0, i32 0, i32 0> 1556; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 undef, i32 42, i32 undef, i32 42> 1557; CHECK-NEXT: [[R:%.*]] = add <4 x i32> [[SPLATX]], [[A]] 1558; CHECK-NEXT: ret <4 x i32> [[R]] 1559; 1560 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> <i32 undef, i32 0, i32 0, i32 0> 1561 %a = add <4 x i32> %y, <i32 undef, i32 42, i32 undef, i32 42> 1562 %r = add <4 x i32> %splatx, %a 1563 ret <4 x i32> %r 1564} 1565 1566; Non-zero splat index; commute operands; FMF intersect 1567 1568define <2 x float> @splat_assoc_fmul(<2 x float> %x, <2 x float> %y) { 1569; CHECK-LABEL: @splat_assoc_fmul( 1570; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc nsz <2 x float> [[X:%.*]], <float undef, float 3.000000e+00> 1571; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> <i32 1, i32 1> 1572; CHECK-NEXT: [[R:%.*]] = fmul reassoc nsz <2 x float> [[TMP2]], [[Y:%.*]] 1573; CHECK-NEXT: ret <2 x float> [[R]] 1574; 1575 %splatx = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> <i32 1, i32 1> 1576 %a = fmul reassoc nsz <2 x float> %y, <float 3.0, float 3.0> 1577 %r = fmul reassoc nsz nnan <2 x float> %a, %splatx 1578 ret <2 x float> %r 1579} 1580 1581; Two splat shuffles; drop poison-generating flags 1582 1583define <3 x i8> @splat_assoc_mul(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1584; CHECK-LABEL: @splat_assoc_mul( 1585; CHECK-NEXT: [[TMP1:%.*]] = mul <3 x i8> [[Z:%.*]], [[X:%.*]] 1586; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <3 x i8> [[TMP1]], <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1587; CHECK-NEXT: [[R:%.*]] = mul <3 x i8> [[TMP2]], [[Y:%.*]] 1588; CHECK-NEXT: ret <3 x i8> [[R]] 1589; 1590 %splatx = shufflevector <3 x i8> %x, <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1591 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1592 %a = mul nsw <3 x i8> %y, %splatz 1593 %r = mul <3 x i8> %a, %splatx 1594 ret <3 x i8> %r 1595} 1596 1597define <3 x i8> @splat_assoc_mul_undef_elt1(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1598; CHECK-LABEL: @splat_assoc_mul_undef_elt1( 1599; CHECK-NEXT: [[TMP1:%.*]] = mul <3 x i8> [[Z:%.*]], [[X:%.*]] 1600; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <3 x i8> [[TMP1]], <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1601; CHECK-NEXT: [[R:%.*]] = mul <3 x i8> [[TMP2]], [[Y:%.*]] 1602; CHECK-NEXT: ret <3 x i8> [[R]] 1603; 1604 %splatx = shufflevector <3 x i8> %x, <3 x i8> undef, <3 x i32> <i32 undef, i32 2, i32 2> 1605 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1606 %a = mul nsw <3 x i8> %y, %splatz 1607 %r = mul nsw nuw <3 x i8> %a, %splatx 1608 ret <3 x i8> %r 1609} 1610 1611define <3 x i8> @splat_assoc_mul_undef_elt2(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1612; CHECK-LABEL: @splat_assoc_mul_undef_elt2( 1613; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <3 x i8> [[X:%.*]], <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1614; CHECK-NEXT: [[SPLATZ:%.*]] = shufflevector <3 x i8> [[Z:%.*]], <3 x i8> undef, <3 x i32> <i32 undef, i32 2, i32 2> 1615; CHECK-NEXT: [[A:%.*]] = mul nsw <3 x i8> [[SPLATZ]], [[Y:%.*]] 1616; CHECK-NEXT: [[R:%.*]] = mul nuw nsw <3 x i8> [[A]], [[SPLATX]] 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 undef, i32 2, i32 2> 1621 %a = mul nsw <3 x i8> %y, %splatz 1622 %r = mul nsw nuw <3 x i8> %a, %splatx 1623 ret <3 x i8> %r 1624} 1625 1626define <3 x i8> @splat_assoc_mul_undef_elt_at_splat_index1(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1627; CHECK-LABEL: @splat_assoc_mul_undef_elt_at_splat_index1( 1628; CHECK-NEXT: [[TMP1:%.*]] = mul <3 x i8> [[Z:%.*]], [[X:%.*]] 1629; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <3 x i8> [[TMP1]], <3 x i8> undef, <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 2, i32 2, i32 undef> 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_elt_at_splat_index2(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1641; CHECK-LABEL: @splat_assoc_mul_undef_elt_at_splat_index2( 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 2, i32 2, i32 undef> 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 2, i32 2, i32 undef> 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 1655; Negative test - mismatched splat elements 1656 1657define <3 x i8> @splat_assoc_or(<3 x i8> %x, <3 x i8> %y, <3 x i8> %z) { 1658; CHECK-LABEL: @splat_assoc_or( 1659; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <3 x i8> [[X:%.*]], <3 x i8> undef, <3 x i32> <i32 1, i32 1, i32 1> 1660; CHECK-NEXT: [[SPLATZ:%.*]] = shufflevector <3 x i8> [[Z:%.*]], <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1661; CHECK-NEXT: [[A:%.*]] = or <3 x i8> [[SPLATZ]], [[Y:%.*]] 1662; CHECK-NEXT: [[R:%.*]] = or <3 x i8> [[A]], [[SPLATX]] 1663; CHECK-NEXT: ret <3 x i8> [[R]] 1664; 1665 %splatx = shufflevector <3 x i8> %x, <3 x i8> undef, <3 x i32> <i32 1, i32 1, i32 1> 1666 %splatz = shufflevector <3 x i8> %z, <3 x i8> undef, <3 x i32> <i32 2, i32 2, i32 2> 1667 %a = or <3 x i8> %y, %splatz 1668 %r = or <3 x i8> %a, %splatx 1669 ret <3 x i8> %r 1670} 1671 1672; Negative test - not associative 1673 1674define <2 x float> @splat_assoc_fdiv(<2 x float> %x, <2 x float> %y) { 1675; CHECK-LABEL: @splat_assoc_fdiv( 1676; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> undef, <2 x i32> zeroinitializer 1677; CHECK-NEXT: [[A:%.*]] = fdiv reassoc nsz <2 x float> [[Y:%.*]], <float 3.000000e+00, float 3.000000e+00> 1678; CHECK-NEXT: [[R:%.*]] = fdiv reassoc nsz <2 x float> [[A]], [[SPLATX]] 1679; CHECK-NEXT: ret <2 x float> [[R]] 1680; 1681 %splatx = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer 1682 %a = fdiv reassoc nsz <2 x float> %y, <float 3.0, float 3.0> 1683 %r = fdiv reassoc nsz <2 x float> %a, %splatx 1684 ret <2 x float> %r 1685} 1686 1687; Negative test - extra use 1688 1689define <2 x float> @splat_assoc_fadd(<2 x float> %x, <2 x float> %y) { 1690; CHECK-LABEL: @splat_assoc_fadd( 1691; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> undef, <2 x i32> <i32 1, i32 1> 1692; CHECK-NEXT: [[A:%.*]] = fadd fast <2 x float> [[Y:%.*]], <float 3.000000e+00, float 3.000000e+00> 1693; CHECK-NEXT: call void @use(<2 x float> [[A]]) 1694; CHECK-NEXT: [[R:%.*]] = fadd fast <2 x float> [[A]], [[SPLATX]] 1695; CHECK-NEXT: ret <2 x float> [[R]] 1696; 1697 %splatx = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> <i32 1, i32 1> 1698 %a = fadd fast <2 x float> %y, <float 3.0, float 3.0> 1699 call void @use(<2 x float> %a) 1700 %r = fadd fast <2 x float> %a, %splatx 1701 ret <2 x float> %r 1702} 1703 1704; Negative test - narrowing splat 1705 1706define <3 x i32> @splat_assoc_and(<4 x i32> %x, <3 x i32> %y) { 1707; CHECK-LABEL: @splat_assoc_and( 1708; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <3 x i32> zeroinitializer 1709; CHECK-NEXT: [[A:%.*]] = and <3 x i32> [[Y:%.*]], <i32 42, i32 42, i32 42> 1710; CHECK-NEXT: [[R:%.*]] = and <3 x i32> [[SPLATX]], [[A]] 1711; CHECK-NEXT: ret <3 x i32> [[R]] 1712; 1713 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <3 x i32> zeroinitializer 1714 %a = and <3 x i32> %y, <i32 42, i32 42, i32 42> 1715 %r = and <3 x i32> %splatx, %a 1716 ret <3 x i32> %r 1717} 1718 1719; Negative test - widening splat 1720 1721define <5 x i32> @splat_assoc_xor(<4 x i32> %x, <5 x i32> %y) { 1722; CHECK-LABEL: @splat_assoc_xor( 1723; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <5 x i32> zeroinitializer 1724; CHECK-NEXT: [[A:%.*]] = xor <5 x i32> [[Y:%.*]], <i32 42, i32 42, i32 42, i32 42, i32 42> 1725; CHECK-NEXT: [[R:%.*]] = xor <5 x i32> [[SPLATX]], [[A]] 1726; CHECK-NEXT: ret <5 x i32> [[R]] 1727; 1728 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <5 x i32> zeroinitializer 1729 %a = xor <5 x i32> %y, <i32 42, i32 42, i32 42, i32 42, i32 42> 1730 %r = xor <5 x i32> %splatx, %a 1731 ret <5 x i32> %r 1732} 1733 1734; Negative test - opcode mismatch 1735 1736define <4 x i32> @splat_assoc_add_mul(<4 x i32> %x, <4 x i32> %y) { 1737; CHECK-LABEL: @splat_assoc_add_mul( 1738; CHECK-NEXT: [[SPLATX:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> zeroinitializer 1739; CHECK-NEXT: [[A:%.*]] = add <4 x i32> [[Y:%.*]], <i32 42, i32 42, i32 42, i32 42> 1740; CHECK-NEXT: [[R:%.*]] = mul <4 x i32> [[SPLATX]], [[A]] 1741; CHECK-NEXT: ret <4 x i32> [[R]] 1742; 1743 %splatx = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> zeroinitializer 1744 %a = add <4 x i32> %y, <i32 42, i32 42, i32 42, i32 42> 1745 %r = mul <4 x i32> %splatx, %a 1746 ret <4 x i32> %r 1747} 1748 1749 1750; Do not crash on constant expressions. 1751 1752define <4 x i32> @PR46872(<4 x i32> %x) { 1753; CHECK-LABEL: @PR46872( 1754; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> <i32 undef, i32 0, i32 1, i32 1> 1755; 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>) 1756; CHECK-NEXT: ret <4 x i32> [[A]] 1757; 1758 %s = shufflevector <4 x i32> %x, <4 x i32> undef, <4 x i32> <i32 undef, i32 0, i32 1, i32 1> 1759 %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>) 1760 ret <4 x i32> %a 1761} 1762 1763