1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE2 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41 4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1 5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX --check-prefix=AVX512F 7; 8; Just one 32-bit run to make sure we do reasonable things. 9; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=X32-SSE 10 11define <2 x double> @merge_2f64_f64_23(double* %ptr) nounwind uwtable noinline ssp { 12; SSE-LABEL: merge_2f64_f64_23: 13; SSE: # BB#0: 14; SSE-NEXT: movups 16(%rdi), %xmm0 15; SSE-NEXT: retq 16; 17; AVX-LABEL: merge_2f64_f64_23: 18; AVX: # BB#0: 19; AVX-NEXT: vmovups 16(%rdi), %xmm0 20; AVX-NEXT: retq 21; 22; X32-SSE-LABEL: merge_2f64_f64_23: 23; X32-SSE: # BB#0: 24; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 25; X32-SSE-NEXT: movups 16(%eax), %xmm0 26; X32-SSE-NEXT: retl 27 %ptr0 = getelementptr inbounds double, double* %ptr, i64 2 28 %ptr1 = getelementptr inbounds double, double* %ptr, i64 3 29 %val0 = load double, double* %ptr0 30 %val1 = load double, double* %ptr1 31 %res0 = insertelement <2 x double> undef, double %val0, i32 0 32 %res1 = insertelement <2 x double> %res0, double %val1, i32 1 33 ret <2 x double> %res1 34} 35 36define <2 x i64> @merge_2i64_i64_12(i64* %ptr) nounwind uwtable noinline ssp { 37; SSE-LABEL: merge_2i64_i64_12: 38; SSE: # BB#0: 39; SSE-NEXT: movups 8(%rdi), %xmm0 40; SSE-NEXT: retq 41; 42; AVX-LABEL: merge_2i64_i64_12: 43; AVX: # BB#0: 44; AVX-NEXT: vmovups 8(%rdi), %xmm0 45; AVX-NEXT: retq 46; 47; X32-SSE-LABEL: merge_2i64_i64_12: 48; X32-SSE: # BB#0: 49; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 50; X32-SSE-NEXT: movups 8(%eax), %xmm0 51; X32-SSE-NEXT: retl 52 %ptr0 = getelementptr inbounds i64, i64* %ptr, i64 1 53 %ptr1 = getelementptr inbounds i64, i64* %ptr, i64 2 54 %val0 = load i64, i64* %ptr0 55 %val1 = load i64, i64* %ptr1 56 %res0 = insertelement <2 x i64> undef, i64 %val0, i32 0 57 %res1 = insertelement <2 x i64> %res0, i64 %val1, i32 1 58 ret <2 x i64> %res1 59} 60 61define <4 x float> @merge_4f32_f32_2345(float* %ptr) nounwind uwtable noinline ssp { 62; SSE-LABEL: merge_4f32_f32_2345: 63; SSE: # BB#0: 64; SSE-NEXT: movups 8(%rdi), %xmm0 65; SSE-NEXT: retq 66; 67; AVX-LABEL: merge_4f32_f32_2345: 68; AVX: # BB#0: 69; AVX-NEXT: vmovups 8(%rdi), %xmm0 70; AVX-NEXT: retq 71; 72; X32-SSE-LABEL: merge_4f32_f32_2345: 73; X32-SSE: # BB#0: 74; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 75; X32-SSE-NEXT: movups 8(%eax), %xmm0 76; X32-SSE-NEXT: retl 77 %ptr0 = getelementptr inbounds float, float* %ptr, i64 2 78 %ptr1 = getelementptr inbounds float, float* %ptr, i64 3 79 %ptr2 = getelementptr inbounds float, float* %ptr, i64 4 80 %ptr3 = getelementptr inbounds float, float* %ptr, i64 5 81 %val0 = load float, float* %ptr0 82 %val1 = load float, float* %ptr1 83 %val2 = load float, float* %ptr2 84 %val3 = load float, float* %ptr3 85 %res0 = insertelement <4 x float> undef, float %val0, i32 0 86 %res1 = insertelement <4 x float> %res0, float %val1, i32 1 87 %res2 = insertelement <4 x float> %res1, float %val2, i32 2 88 %res3 = insertelement <4 x float> %res2, float %val3, i32 3 89 ret <4 x float> %res3 90} 91 92define <4 x float> @merge_4f32_f32_3zuu(float* %ptr) nounwind uwtable noinline ssp { 93; SSE-LABEL: merge_4f32_f32_3zuu: 94; SSE: # BB#0: 95; SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 96; SSE-NEXT: retq 97; 98; AVX-LABEL: merge_4f32_f32_3zuu: 99; AVX: # BB#0: 100; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero 101; AVX-NEXT: retq 102; 103; X32-SSE-LABEL: merge_4f32_f32_3zuu: 104; X32-SSE: # BB#0: 105; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 106; X32-SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 107; X32-SSE-NEXT: retl 108 %ptr0 = getelementptr inbounds float, float* %ptr, i64 3 109 %val0 = load float, float* %ptr0 110 %res0 = insertelement <4 x float> undef, float %val0, i32 0 111 %res1 = insertelement <4 x float> %res0, float 0.0, i32 1 112 ret <4 x float> %res1 113} 114 115define <4 x float> @merge_4f32_f32_34uu(float* %ptr) nounwind uwtable noinline ssp { 116; SSE-LABEL: merge_4f32_f32_34uu: 117; SSE: # BB#0: 118; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 119; SSE-NEXT: retq 120; 121; AVX-LABEL: merge_4f32_f32_34uu: 122; AVX: # BB#0: 123; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 124; AVX-NEXT: retq 125; 126; X32-SSE-LABEL: merge_4f32_f32_34uu: 127; X32-SSE: # BB#0: 128; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 129; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 130; X32-SSE-NEXT: retl 131 %ptr0 = getelementptr inbounds float, float* %ptr, i64 3 132 %ptr1 = getelementptr inbounds float, float* %ptr, i64 4 133 %val0 = load float, float* %ptr0 134 %val1 = load float, float* %ptr1 135 %res0 = insertelement <4 x float> undef, float %val0, i32 0 136 %res1 = insertelement <4 x float> %res0, float %val1, i32 1 137 ret <4 x float> %res1 138} 139 140define <4 x float> @merge_4f32_f32_34z6(float* %ptr) nounwind uwtable noinline ssp { 141; SSE2-LABEL: merge_4f32_f32_34z6: 142; SSE2: # BB#0: 143; SSE2-NEXT: movups 12(%rdi), %xmm0 144; SSE2-NEXT: xorps %xmm1, %xmm1 145; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0] 146; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2] 147; SSE2-NEXT: retq 148; 149; SSE41-LABEL: merge_4f32_f32_34z6: 150; SSE41: # BB#0: 151; SSE41-NEXT: movups 12(%rdi), %xmm1 152; SSE41-NEXT: xorps %xmm0, %xmm0 153; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm1[0,1],xmm0[2],xmm1[3] 154; SSE41-NEXT: retq 155; 156; AVX-LABEL: merge_4f32_f32_34z6: 157; AVX: # BB#0: 158; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0 159; AVX-NEXT: vblendps {{.*#+}} xmm0 = mem[0,1],xmm0[2],mem[3] 160; AVX-NEXT: retq 161; 162; X32-SSE-LABEL: merge_4f32_f32_34z6: 163; X32-SSE: # BB#0: 164; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 165; X32-SSE-NEXT: movups 12(%eax), %xmm1 166; X32-SSE-NEXT: xorps %xmm0, %xmm0 167; X32-SSE-NEXT: blendps {{.*#+}} xmm0 = xmm1[0,1],xmm0[2],xmm1[3] 168; X32-SSE-NEXT: retl 169 %ptr0 = getelementptr inbounds float, float* %ptr, i64 3 170 %ptr1 = getelementptr inbounds float, float* %ptr, i64 4 171 %ptr3 = getelementptr inbounds float, float* %ptr, i64 6 172 %val0 = load float, float* %ptr0 173 %val1 = load float, float* %ptr1 174 %val3 = load float, float* %ptr3 175 %res0 = insertelement <4 x float> zeroinitializer, float %val0, i32 0 176 %res1 = insertelement <4 x float> %res0, float %val1, i32 1 177 %res3 = insertelement <4 x float> %res1, float %val3, i32 3 178 ret <4 x float> %res3 179} 180 181define <4 x float> @merge_4f32_f32_45zz(float* %ptr) nounwind uwtable noinline ssp { 182; SSE-LABEL: merge_4f32_f32_45zz: 183; SSE: # BB#0: 184; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 185; SSE-NEXT: retq 186; 187; AVX-LABEL: merge_4f32_f32_45zz: 188; AVX: # BB#0: 189; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 190; AVX-NEXT: retq 191; 192; X32-SSE-LABEL: merge_4f32_f32_45zz: 193; X32-SSE: # BB#0: 194; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 195; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 196; X32-SSE-NEXT: retl 197 %ptr0 = getelementptr inbounds float, float* %ptr, i64 4 198 %ptr1 = getelementptr inbounds float, float* %ptr, i64 5 199 %val0 = load float, float* %ptr0 200 %val1 = load float, float* %ptr1 201 %res0 = insertelement <4 x float> zeroinitializer, float %val0, i32 0 202 %res1 = insertelement <4 x float> %res0, float %val1, i32 1 203 ret <4 x float> %res1 204} 205 206define <4 x float> @merge_4f32_f32_012u(float* %ptr) nounwind uwtable noinline ssp { 207; SSE2-LABEL: merge_4f32_f32_012u: 208; SSE2: # BB#0: 209; SSE2-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero 210; SSE2-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 211; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero 212; SSE2-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] 213; SSE2-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1] 214; SSE2-NEXT: retq 215; 216; SSE41-LABEL: merge_4f32_f32_012u: 217; SSE41: # BB#0: 218; SSE41-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 219; SSE41-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 220; SSE41-NEXT: retq 221; 222; AVX-LABEL: merge_4f32_f32_012u: 223; AVX: # BB#0: 224; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 225; AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 226; AVX-NEXT: retq 227; 228; X32-SSE-LABEL: merge_4f32_f32_012u: 229; X32-SSE: # BB#0: 230; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 231; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 232; X32-SSE-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 233; X32-SSE-NEXT: retl 234 %ptr0 = getelementptr inbounds float, float* %ptr, i64 0 235 %ptr1 = getelementptr inbounds float, float* %ptr, i64 1 236 %ptr2 = getelementptr inbounds float, float* %ptr, i64 2 237 %val0 = load float, float* %ptr0 238 %val1 = load float, float* %ptr1 239 %val2 = load float, float* %ptr2 240 %res0 = insertelement <4 x float> undef, float %val0, i32 0 241 %res1 = insertelement <4 x float> %res0, float %val1, i32 1 242 %res2 = insertelement <4 x float> %res1, float %val2, i32 2 243 %res3 = insertelement <4 x float> %res2, float undef, i32 3 244 ret <4 x float> %res3 245} 246 247define <4 x float> @merge_4f32_f32_019u(float* %ptr) nounwind uwtable noinline ssp { 248; SSE2-LABEL: merge_4f32_f32_019u: 249; SSE2: # BB#0: 250; SSE2-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero 251; SSE2-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 252; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero 253; SSE2-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] 254; SSE2-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1] 255; SSE2-NEXT: retq 256; 257; SSE41-LABEL: merge_4f32_f32_019u: 258; SSE41: # BB#0: 259; SSE41-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 260; SSE41-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 261; SSE41-NEXT: retq 262; 263; AVX-LABEL: merge_4f32_f32_019u: 264; AVX: # BB#0: 265; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 266; AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 267; AVX-NEXT: retq 268; 269; X32-SSE-LABEL: merge_4f32_f32_019u: 270; X32-SSE: # BB#0: 271; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 272; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 273; X32-SSE-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 274; X32-SSE-NEXT: retl 275 %ptr0 = getelementptr inbounds float, float* %ptr, i64 0 276 %ptr1 = getelementptr inbounds float, float* %ptr, i64 1 277 %ptr2 = getelementptr inbounds float, float* %ptr, i64 9 278 %val0 = load float, float* %ptr0 279 %val1 = load float, float* %ptr1 280 %val2 = load float, float* %ptr2 281 %res0 = insertelement <4 x float> undef, float %val0, i32 0 282 %res1 = insertelement <4 x float> %res0, float %val1, i32 1 283 %res2 = insertelement <4 x float> %res1, float %val2, i32 2 284 %res3 = insertelement <4 x float> %res2, float undef, i32 3 285 ret <4 x float> %res3 286} 287 288define <4 x i32> @merge_4i32_i32_23u5(i32* %ptr) nounwind uwtable noinline ssp { 289; SSE-LABEL: merge_4i32_i32_23u5: 290; SSE: # BB#0: 291; SSE-NEXT: movups 8(%rdi), %xmm0 292; SSE-NEXT: retq 293; 294; AVX-LABEL: merge_4i32_i32_23u5: 295; AVX: # BB#0: 296; AVX-NEXT: vmovups 8(%rdi), %xmm0 297; AVX-NEXT: retq 298; 299; X32-SSE-LABEL: merge_4i32_i32_23u5: 300; X32-SSE: # BB#0: 301; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 302; X32-SSE-NEXT: movups 8(%eax), %xmm0 303; X32-SSE-NEXT: retl 304 %ptr0 = getelementptr inbounds i32, i32* %ptr, i64 2 305 %ptr1 = getelementptr inbounds i32, i32* %ptr, i64 3 306 %ptr3 = getelementptr inbounds i32, i32* %ptr, i64 5 307 %val0 = load i32, i32* %ptr0 308 %val1 = load i32, i32* %ptr1 309 %val3 = load i32, i32* %ptr3 310 %res0 = insertelement <4 x i32> undef, i32 %val0, i32 0 311 %res1 = insertelement <4 x i32> %res0, i32 %val1, i32 1 312 %res3 = insertelement <4 x i32> %res1, i32 %val3, i32 3 313 ret <4 x i32> %res3 314} 315 316define <4 x i32> @merge_4i32_i32_3zuu(i32* %ptr) nounwind uwtable noinline ssp { 317; SSE-LABEL: merge_4i32_i32_3zuu: 318; SSE: # BB#0: 319; SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 320; SSE-NEXT: retq 321; 322; AVX-LABEL: merge_4i32_i32_3zuu: 323; AVX: # BB#0: 324; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 325; AVX-NEXT: retq 326; 327; X32-SSE-LABEL: merge_4i32_i32_3zuu: 328; X32-SSE: # BB#0: 329; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 330; X32-SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 331; X32-SSE-NEXT: retl 332 %ptr0 = getelementptr inbounds i32, i32* %ptr, i64 3 333 %val0 = load i32, i32* %ptr0 334 %res0 = insertelement <4 x i32> undef, i32 %val0, i32 0 335 %res1 = insertelement <4 x i32> %res0, i32 0, i32 1 336 ret <4 x i32> %res1 337} 338 339define <4 x i32> @merge_4i32_i32_34uu(i32* %ptr) nounwind uwtable noinline ssp { 340; SSE-LABEL: merge_4i32_i32_34uu: 341; SSE: # BB#0: 342; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 343; SSE-NEXT: retq 344; 345; AVX-LABEL: merge_4i32_i32_34uu: 346; AVX: # BB#0: 347; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 348; AVX-NEXT: retq 349; 350; X32-SSE-LABEL: merge_4i32_i32_34uu: 351; X32-SSE: # BB#0: 352; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 353; X32-SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 354; X32-SSE-NEXT: retl 355 %ptr0 = getelementptr inbounds i32, i32* %ptr, i64 3 356 %ptr1 = getelementptr inbounds i32, i32* %ptr, i64 4 357 %val0 = load i32, i32* %ptr0 358 %val1 = load i32, i32* %ptr1 359 %res0 = insertelement <4 x i32> undef, i32 %val0, i32 0 360 %res1 = insertelement <4 x i32> %res0, i32 %val1, i32 1 361 ret <4 x i32> %res1 362} 363 364define <4 x i32> @merge_4i32_i32_45zz(i32* %ptr) nounwind uwtable noinline ssp { 365; SSE-LABEL: merge_4i32_i32_45zz: 366; SSE: # BB#0: 367; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 368; SSE-NEXT: retq 369; 370; AVX-LABEL: merge_4i32_i32_45zz: 371; AVX: # BB#0: 372; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 373; AVX-NEXT: retq 374; 375; X32-SSE-LABEL: merge_4i32_i32_45zz: 376; X32-SSE: # BB#0: 377; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 378; X32-SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 379; X32-SSE-NEXT: retl 380 %ptr0 = getelementptr inbounds i32, i32* %ptr, i64 4 381 %ptr1 = getelementptr inbounds i32, i32* %ptr, i64 5 382 %val0 = load i32, i32* %ptr0 383 %val1 = load i32, i32* %ptr1 384 %res0 = insertelement <4 x i32> zeroinitializer, i32 %val0, i32 0 385 %res1 = insertelement <4 x i32> %res0, i32 %val1, i32 1 386 ret <4 x i32> %res1 387} 388 389define <8 x i16> @merge_8i16_i16_23u567u9(i16* %ptr) nounwind uwtable noinline ssp { 390; SSE-LABEL: merge_8i16_i16_23u567u9: 391; SSE: # BB#0: 392; SSE-NEXT: movups 4(%rdi), %xmm0 393; SSE-NEXT: retq 394; 395; AVX-LABEL: merge_8i16_i16_23u567u9: 396; AVX: # BB#0: 397; AVX-NEXT: vmovups 4(%rdi), %xmm0 398; AVX-NEXT: retq 399; 400; X32-SSE-LABEL: merge_8i16_i16_23u567u9: 401; X32-SSE: # BB#0: 402; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 403; X32-SSE-NEXT: movups 4(%eax), %xmm0 404; X32-SSE-NEXT: retl 405 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 2 406 %ptr1 = getelementptr inbounds i16, i16* %ptr, i64 3 407 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 5 408 %ptr4 = getelementptr inbounds i16, i16* %ptr, i64 6 409 %ptr5 = getelementptr inbounds i16, i16* %ptr, i64 7 410 %ptr7 = getelementptr inbounds i16, i16* %ptr, i64 9 411 %val0 = load i16, i16* %ptr0 412 %val1 = load i16, i16* %ptr1 413 %val3 = load i16, i16* %ptr3 414 %val4 = load i16, i16* %ptr4 415 %val5 = load i16, i16* %ptr5 416 %val7 = load i16, i16* %ptr7 417 %res0 = insertelement <8 x i16> undef, i16 %val0, i32 0 418 %res1 = insertelement <8 x i16> %res0, i16 %val1, i32 1 419 %res3 = insertelement <8 x i16> %res1, i16 %val3, i32 3 420 %res4 = insertelement <8 x i16> %res3, i16 %val4, i32 4 421 %res5 = insertelement <8 x i16> %res4, i16 %val5, i32 5 422 %res7 = insertelement <8 x i16> %res5, i16 %val7, i32 7 423 ret <8 x i16> %res7 424} 425 426define <8 x i16> @merge_8i16_i16_34uuuuuu(i16* %ptr) nounwind uwtable noinline ssp { 427; SSE-LABEL: merge_8i16_i16_34uuuuuu: 428; SSE: # BB#0: 429; SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 430; SSE-NEXT: retq 431; 432; AVX-LABEL: merge_8i16_i16_34uuuuuu: 433; AVX: # BB#0: 434; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 435; AVX-NEXT: retq 436; 437; X32-SSE-LABEL: merge_8i16_i16_34uuuuuu: 438; X32-SSE: # BB#0: 439; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 440; X32-SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 441; X32-SSE-NEXT: retl 442 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 3 443 %ptr1 = getelementptr inbounds i16, i16* %ptr, i64 4 444 %val0 = load i16, i16* %ptr0 445 %val1 = load i16, i16* %ptr1 446 %res0 = insertelement <8 x i16> undef, i16 %val0, i32 0 447 %res1 = insertelement <8 x i16> %res0, i16 %val1, i32 1 448 ret <8 x i16> %res1 449} 450 451define <8 x i16> @merge_8i16_i16_45u7zzzz(i16* %ptr) nounwind uwtable noinline ssp { 452; SSE-LABEL: merge_8i16_i16_45u7zzzz: 453; SSE: # BB#0: 454; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 455; SSE-NEXT: retq 456; 457; AVX-LABEL: merge_8i16_i16_45u7zzzz: 458; AVX: # BB#0: 459; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 460; AVX-NEXT: retq 461; 462; X32-SSE-LABEL: merge_8i16_i16_45u7zzzz: 463; X32-SSE: # BB#0: 464; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 465; X32-SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 466; X32-SSE-NEXT: retl 467 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 4 468 %ptr1 = getelementptr inbounds i16, i16* %ptr, i64 5 469 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 7 470 %val0 = load i16, i16* %ptr0 471 %val1 = load i16, i16* %ptr1 472 %val3 = load i16, i16* %ptr3 473 %res0 = insertelement <8 x i16> undef, i16 %val0, i32 0 474 %res1 = insertelement <8 x i16> %res0, i16 %val1, i32 1 475 %res3 = insertelement <8 x i16> %res1, i16 %val3, i32 3 476 %res4 = insertelement <8 x i16> %res3, i16 0, i32 4 477 %res5 = insertelement <8 x i16> %res4, i16 0, i32 5 478 %res6 = insertelement <8 x i16> %res5, i16 0, i32 6 479 %res7 = insertelement <8 x i16> %res6, i16 0, i32 7 480 ret <8 x i16> %res7 481} 482 483define <16 x i8> @merge_16i8_i8_01u3456789ABCDuF(i8* %ptr) nounwind uwtable noinline ssp { 484; SSE-LABEL: merge_16i8_i8_01u3456789ABCDuF: 485; SSE: # BB#0: 486; SSE-NEXT: movups (%rdi), %xmm0 487; SSE-NEXT: retq 488; 489; AVX-LABEL: merge_16i8_i8_01u3456789ABCDuF: 490; AVX: # BB#0: 491; AVX-NEXT: vmovups (%rdi), %xmm0 492; AVX-NEXT: retq 493; 494; X32-SSE-LABEL: merge_16i8_i8_01u3456789ABCDuF: 495; X32-SSE: # BB#0: 496; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 497; X32-SSE-NEXT: movups (%eax), %xmm0 498; X32-SSE-NEXT: retl 499 %ptr0 = getelementptr inbounds i8, i8* %ptr, i64 0 500 %ptr1 = getelementptr inbounds i8, i8* %ptr, i64 1 501 %ptr3 = getelementptr inbounds i8, i8* %ptr, i64 3 502 %ptr4 = getelementptr inbounds i8, i8* %ptr, i64 4 503 %ptr5 = getelementptr inbounds i8, i8* %ptr, i64 5 504 %ptr6 = getelementptr inbounds i8, i8* %ptr, i64 6 505 %ptr7 = getelementptr inbounds i8, i8* %ptr, i64 7 506 %ptr8 = getelementptr inbounds i8, i8* %ptr, i64 8 507 %ptr9 = getelementptr inbounds i8, i8* %ptr, i64 9 508 %ptrA = getelementptr inbounds i8, i8* %ptr, i64 10 509 %ptrB = getelementptr inbounds i8, i8* %ptr, i64 11 510 %ptrC = getelementptr inbounds i8, i8* %ptr, i64 12 511 %ptrD = getelementptr inbounds i8, i8* %ptr, i64 13 512 %ptrF = getelementptr inbounds i8, i8* %ptr, i64 15 513 %val0 = load i8, i8* %ptr0 514 %val1 = load i8, i8* %ptr1 515 %val3 = load i8, i8* %ptr3 516 %val4 = load i8, i8* %ptr4 517 %val5 = load i8, i8* %ptr5 518 %val6 = load i8, i8* %ptr6 519 %val7 = load i8, i8* %ptr7 520 %val8 = load i8, i8* %ptr8 521 %val9 = load i8, i8* %ptr9 522 %valA = load i8, i8* %ptrA 523 %valB = load i8, i8* %ptrB 524 %valC = load i8, i8* %ptrC 525 %valD = load i8, i8* %ptrD 526 %valF = load i8, i8* %ptrF 527 %res0 = insertelement <16 x i8> undef, i8 %val0, i32 0 528 %res1 = insertelement <16 x i8> %res0, i8 %val1, i32 1 529 %res3 = insertelement <16 x i8> %res1, i8 %val3, i32 3 530 %res4 = insertelement <16 x i8> %res3, i8 %val4, i32 4 531 %res5 = insertelement <16 x i8> %res4, i8 %val5, i32 5 532 %res6 = insertelement <16 x i8> %res5, i8 %val6, i32 6 533 %res7 = insertelement <16 x i8> %res6, i8 %val7, i32 7 534 %res8 = insertelement <16 x i8> %res7, i8 %val8, i32 8 535 %res9 = insertelement <16 x i8> %res8, i8 %val9, i32 9 536 %resA = insertelement <16 x i8> %res9, i8 %valA, i32 10 537 %resB = insertelement <16 x i8> %resA, i8 %valB, i32 11 538 %resC = insertelement <16 x i8> %resB, i8 %valC, i32 12 539 %resD = insertelement <16 x i8> %resC, i8 %valD, i32 13 540 %resF = insertelement <16 x i8> %resD, i8 %valF, i32 15 541 ret <16 x i8> %resF 542} 543 544define <16 x i8> @merge_16i8_i8_01u3uuzzuuuuuzzz(i8* %ptr) nounwind uwtable noinline ssp { 545; SSE-LABEL: merge_16i8_i8_01u3uuzzuuuuuzzz: 546; SSE: # BB#0: 547; SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 548; SSE-NEXT: retq 549; 550; AVX-LABEL: merge_16i8_i8_01u3uuzzuuuuuzzz: 551; AVX: # BB#0: 552; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 553; AVX-NEXT: retq 554; 555; X32-SSE-LABEL: merge_16i8_i8_01u3uuzzuuuuuzzz: 556; X32-SSE: # BB#0: 557; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 558; X32-SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 559; X32-SSE-NEXT: retl 560 %ptr0 = getelementptr inbounds i8, i8* %ptr, i64 0 561 %ptr1 = getelementptr inbounds i8, i8* %ptr, i64 1 562 %ptr3 = getelementptr inbounds i8, i8* %ptr, i64 3 563 %val0 = load i8, i8* %ptr0 564 %val1 = load i8, i8* %ptr1 565 %val3 = load i8, i8* %ptr3 566 %res0 = insertelement <16 x i8> undef, i8 %val0, i32 0 567 %res1 = insertelement <16 x i8> %res0, i8 %val1, i32 1 568 %res3 = insertelement <16 x i8> %res1, i8 %val3, i32 3 569 %res6 = insertelement <16 x i8> %res3, i8 0, i32 6 570 %res7 = insertelement <16 x i8> %res6, i8 0, i32 7 571 %resD = insertelement <16 x i8> %res7, i8 0, i32 13 572 %resE = insertelement <16 x i8> %resD, i8 0, i32 14 573 %resF = insertelement <16 x i8> %resE, i8 0, i32 15 574 ret <16 x i8> %resF 575} 576 577define <16 x i8> @merge_16i8_i8_0123uu67uuuuuzzz(i8* %ptr) nounwind uwtable noinline ssp { 578; SSE-LABEL: merge_16i8_i8_0123uu67uuuuuzzz: 579; SSE: # BB#0: 580; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 581; SSE-NEXT: retq 582; 583; AVX-LABEL: merge_16i8_i8_0123uu67uuuuuzzz: 584; AVX: # BB#0: 585; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 586; AVX-NEXT: retq 587; 588; X32-SSE-LABEL: merge_16i8_i8_0123uu67uuuuuzzz: 589; X32-SSE: # BB#0: 590; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 591; X32-SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 592; X32-SSE-NEXT: retl 593 %ptr0 = getelementptr inbounds i8, i8* %ptr, i64 0 594 %ptr1 = getelementptr inbounds i8, i8* %ptr, i64 1 595 %ptr2 = getelementptr inbounds i8, i8* %ptr, i64 2 596 %ptr3 = getelementptr inbounds i8, i8* %ptr, i64 3 597 %ptr6 = getelementptr inbounds i8, i8* %ptr, i64 6 598 %ptr7 = getelementptr inbounds i8, i8* %ptr, i64 7 599 %val0 = load i8, i8* %ptr0 600 %val1 = load i8, i8* %ptr1 601 %val2 = load i8, i8* %ptr2 602 %val3 = load i8, i8* %ptr3 603 %val6 = load i8, i8* %ptr6 604 %val7 = load i8, i8* %ptr7 605 %res0 = insertelement <16 x i8> undef, i8 %val0, i32 0 606 %res1 = insertelement <16 x i8> %res0, i8 %val1, i32 1 607 %res2 = insertelement <16 x i8> %res1, i8 %val2, i32 2 608 %res3 = insertelement <16 x i8> %res2, i8 %val3, i32 3 609 %res6 = insertelement <16 x i8> %res3, i8 %val6, i32 6 610 %res7 = insertelement <16 x i8> %res6, i8 %val7, i32 7 611 %resD = insertelement <16 x i8> %res7, i8 0, i32 13 612 %resE = insertelement <16 x i8> %resD, i8 0, i32 14 613 %resF = insertelement <16 x i8> %resE, i8 0, i32 15 614 ret <16 x i8> %resF 615} 616 617define void @merge_4i32_i32_combine(<4 x i32>* %dst, i32* %src) { 618; SSE-LABEL: merge_4i32_i32_combine: 619; SSE: # BB#0: 620; SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 621; SSE-NEXT: movaps %xmm0, (%rdi) 622; SSE-NEXT: retq 623; 624; AVX1-LABEL: merge_4i32_i32_combine: 625; AVX1: # BB#0: 626; AVX1-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 627; AVX1-NEXT: vmovaps %xmm0, (%rdi) 628; AVX1-NEXT: retq 629; 630; AVX2-LABEL: merge_4i32_i32_combine: 631; AVX2: # BB#0: 632; AVX2-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 633; AVX2-NEXT: vmovaps %xmm0, (%rdi) 634; AVX2-NEXT: retq 635; 636; AVX512F-LABEL: merge_4i32_i32_combine: 637; AVX512F: # BB#0: 638; AVX512F-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 639; AVX512F-NEXT: vmovdqa %xmm0, (%rdi) 640; AVX512F-NEXT: retq 641; 642; X32-SSE-LABEL: merge_4i32_i32_combine: 643; X32-SSE: # BB#0: 644; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 645; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %ecx 646; X32-SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 647; X32-SSE-NEXT: movaps %xmm0, (%eax) 648; X32-SSE-NEXT: retl 649 %1 = getelementptr i32, i32* %src, i32 0 650 %2 = load i32, i32* %1 651 %3 = insertelement <4 x i32> undef, i32 %2, i32 0 652 %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <4 x i32> zeroinitializer 653 %5 = lshr <4 x i32> %4, <i32 0, i32 undef, i32 undef, i32 undef> 654 %6 = and <4 x i32> %5, <i32 -1, i32 0, i32 0, i32 0> 655 store <4 x i32> %6, <4 x i32>* %dst 656 ret void 657} 658 659; 660; consecutive loads including any/all volatiles may not be combined 661; 662 663define <2 x i64> @merge_2i64_i64_12_volatile(i64* %ptr) nounwind uwtable noinline ssp { 664; SSE-LABEL: merge_2i64_i64_12_volatile: 665; SSE: # BB#0: 666; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 667; SSE-NEXT: movq {{.*#+}} xmm1 = mem[0],zero 668; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 669; SSE-NEXT: retq 670; 671; AVX-LABEL: merge_2i64_i64_12_volatile: 672; AVX: # BB#0: 673; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 674; AVX-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero 675; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 676; AVX-NEXT: retq 677; 678; X32-SSE-LABEL: merge_2i64_i64_12_volatile: 679; X32-SSE: # BB#0: 680; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 681; X32-SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 682; X32-SSE-NEXT: pinsrd $1, 12(%eax), %xmm0 683; X32-SSE-NEXT: pinsrd $2, 16(%eax), %xmm0 684; X32-SSE-NEXT: pinsrd $3, 20(%eax), %xmm0 685; X32-SSE-NEXT: retl 686 %ptr0 = getelementptr inbounds i64, i64* %ptr, i64 1 687 %ptr1 = getelementptr inbounds i64, i64* %ptr, i64 2 688 %val0 = load volatile i64, i64* %ptr0 689 %val1 = load volatile i64, i64* %ptr1 690 %res0 = insertelement <2 x i64> undef, i64 %val0, i32 0 691 %res1 = insertelement <2 x i64> %res0, i64 %val1, i32 1 692 ret <2 x i64> %res1 693} 694 695define <4 x float> @merge_4f32_f32_2345_volatile(float* %ptr) nounwind uwtable noinline ssp { 696; SSE2-LABEL: merge_4f32_f32_2345_volatile: 697; SSE2: # BB#0: 698; SSE2-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 699; SSE2-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero 700; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero 701; SSE2-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero 702; SSE2-NEXT: unpcklps {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1] 703; SSE2-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1] 704; SSE2-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1] 705; SSE2-NEXT: retq 706; 707; SSE41-LABEL: merge_4f32_f32_2345_volatile: 708; SSE41: # BB#0: 709; SSE41-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 710; SSE41-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] 711; SSE41-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 712; SSE41-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] 713; SSE41-NEXT: retq 714; 715; AVX-LABEL: merge_4f32_f32_2345_volatile: 716; AVX: # BB#0: 717; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero 718; AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] 719; AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 720; AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] 721; AVX-NEXT: retq 722; 723; X32-SSE-LABEL: merge_4f32_f32_2345_volatile: 724; X32-SSE: # BB#0: 725; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 726; X32-SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 727; X32-SSE-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] 728; X32-SSE-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] 729; X32-SSE-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] 730; X32-SSE-NEXT: retl 731 %ptr0 = getelementptr inbounds float, float* %ptr, i64 2 732 %ptr1 = getelementptr inbounds float, float* %ptr, i64 3 733 %ptr2 = getelementptr inbounds float, float* %ptr, i64 4 734 %ptr3 = getelementptr inbounds float, float* %ptr, i64 5 735 %val0 = load volatile float, float* %ptr0 736 %val1 = load float, float* %ptr1 737 %val2 = load float, float* %ptr2 738 %val3 = load float, float* %ptr3 739 %res0 = insertelement <4 x float> undef, float %val0, i32 0 740 %res1 = insertelement <4 x float> %res0, float %val1, i32 1 741 %res2 = insertelement <4 x float> %res1, float %val2, i32 2 742 %res3 = insertelement <4 x float> %res2, float %val3, i32 3 743 ret <4 x float> %res3 744} 745 746; 747; Non-consecutive test. 748; 749 750define <4 x float> @merge_4f32_f32_X0YY(float* %ptr0, float* %ptr1) nounwind uwtable noinline ssp { 751; SSE-LABEL: merge_4f32_f32_X0YY: 752; SSE: # BB#0: 753; SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 754; SSE-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero 755; SSE-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0,0,1,1] 756; SSE-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0] 757; SSE-NEXT: retq 758; 759; AVX-LABEL: merge_4f32_f32_X0YY: 760; AVX: # BB#0: 761; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero 762; AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero 763; AVX-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[0,0,1,1] 764; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0] 765; AVX-NEXT: retq 766; 767; X32-SSE-LABEL: merge_4f32_f32_X0YY: 768; X32-SSE: # BB#0: 769; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %eax 770; X32-SSE-NEXT: movl {{[0-9]+}}(%esp), %ecx 771; X32-SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 772; X32-SSE-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero 773; X32-SSE-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0,0,1,1] 774; X32-SSE-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0] 775; X32-SSE-NEXT: retl 776 %val0 = load float, float* %ptr0, align 4 777 %val1 = load float, float* %ptr1, align 4 778 %res0 = insertelement <4 x float> undef, float %val0, i32 0 779 %res1 = insertelement <4 x float> %res0, float 0.000000e+00, i32 1 780 %res2 = insertelement <4 x float> %res1, float %val1, i32 2 781 %res3 = insertelement <4 x float> %res2, float %val1, i32 3 782 ret <4 x float> %res3 783} 784