1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX --check-prefix=AVX512F 5; 6; Just one 32-bit run to make sure we do reasonable things. 7; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32-AVX 8 9define <4 x double> @merge_4f64_2f64_23(<2 x double>* %ptr) nounwind uwtable noinline ssp { 10; AVX-LABEL: merge_4f64_2f64_23: 11; AVX: # BB#0: 12; AVX-NEXT: vmovups 32(%rdi), %ymm0 13; AVX-NEXT: retq 14; 15; X32-AVX-LABEL: merge_4f64_2f64_23: 16; X32-AVX: # BB#0: 17; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 18; X32-AVX-NEXT: vmovups 32(%eax), %ymm0 19; X32-AVX-NEXT: retl 20 %ptr0 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 2 21 %ptr1 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 3 22 %val0 = load <2 x double>, <2 x double>* %ptr0 23 %val1 = load <2 x double>, <2 x double>* %ptr1 24 %res = shufflevector <2 x double> %val0, <2 x double> %val1, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 25 ret <4 x double> %res 26} 27 28define <4 x double> @merge_4f64_2f64_2z(<2 x double>* %ptr) nounwind uwtable noinline ssp { 29; AVX-LABEL: merge_4f64_2f64_2z: 30; AVX: # BB#0: 31; AVX-NEXT: vmovaps 32(%rdi), %xmm0 32; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 33; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 34; AVX-NEXT: retq 35; 36; X32-AVX-LABEL: merge_4f64_2f64_2z: 37; X32-AVX: # BB#0: 38; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 39; X32-AVX-NEXT: vmovaps 32(%eax), %xmm0 40; X32-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 41; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 42; X32-AVX-NEXT: retl 43 %ptr0 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 2 44 %val0 = load <2 x double>, <2 x double>* %ptr0 45 %res = shufflevector <2 x double> %val0, <2 x double> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 46 ret <4 x double> %res 47} 48 49define <4 x double> @merge_4f64_f64_2345(double* %ptr) nounwind uwtable noinline ssp { 50; AVX-LABEL: merge_4f64_f64_2345: 51; AVX: # BB#0: 52; AVX-NEXT: vmovups 16(%rdi), %ymm0 53; AVX-NEXT: retq 54; 55; X32-AVX-LABEL: merge_4f64_f64_2345: 56; X32-AVX: # BB#0: 57; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 58; X32-AVX-NEXT: vmovups 16(%eax), %ymm0 59; X32-AVX-NEXT: retl 60 %ptr0 = getelementptr inbounds double, double* %ptr, i64 2 61 %ptr1 = getelementptr inbounds double, double* %ptr, i64 3 62 %ptr2 = getelementptr inbounds double, double* %ptr, i64 4 63 %ptr3 = getelementptr inbounds double, double* %ptr, i64 5 64 %val0 = load double, double* %ptr0 65 %val1 = load double, double* %ptr1 66 %val2 = load double, double* %ptr2 67 %val3 = load double, double* %ptr3 68 %res0 = insertelement <4 x double> undef, double %val0, i32 0 69 %res1 = insertelement <4 x double> %res0, double %val1, i32 1 70 %res2 = insertelement <4 x double> %res1, double %val2, i32 2 71 %res3 = insertelement <4 x double> %res2, double %val3, i32 3 72 ret <4 x double> %res3 73} 74 75define <4 x double> @merge_4f64_f64_3zuu(double* %ptr) nounwind uwtable noinline ssp { 76; AVX-LABEL: merge_4f64_f64_3zuu: 77; AVX: # BB#0: 78; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 79; AVX-NEXT: retq 80; 81; X32-AVX-LABEL: merge_4f64_f64_3zuu: 82; X32-AVX: # BB#0: 83; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 84; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 85; X32-AVX-NEXT: retl 86 %ptr0 = getelementptr inbounds double, double* %ptr, i64 3 87 %val0 = load double, double* %ptr0 88 %res0 = insertelement <4 x double> undef, double %val0, i32 0 89 %res1 = insertelement <4 x double> %res0, double 0.0, i32 1 90 ret <4 x double> %res1 91} 92 93define <4 x double> @merge_4f64_f64_34uu(double* %ptr) nounwind uwtable noinline ssp { 94; AVX-LABEL: merge_4f64_f64_34uu: 95; AVX: # BB#0: 96; AVX-NEXT: vmovups 24(%rdi), %xmm0 97; AVX-NEXT: retq 98; 99; X32-AVX-LABEL: merge_4f64_f64_34uu: 100; X32-AVX: # BB#0: 101; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 102; X32-AVX-NEXT: vmovups 24(%eax), %xmm0 103; X32-AVX-NEXT: retl 104 %ptr0 = getelementptr inbounds double, double* %ptr, i64 3 105 %ptr1 = getelementptr inbounds double, double* %ptr, i64 4 106 %val0 = load double, double* %ptr0 107 %val1 = load double, double* %ptr1 108 %res0 = insertelement <4 x double> undef, double %val0, i32 0 109 %res1 = insertelement <4 x double> %res0, double %val1, i32 1 110 ret <4 x double> %res1 111} 112 113define <4 x double> @merge_4f64_f64_45zz(double* %ptr) nounwind uwtable noinline ssp { 114; AVX-LABEL: merge_4f64_f64_45zz: 115; AVX: # BB#0: 116; AVX-NEXT: vmovups 32(%rdi), %xmm0 117; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 118; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 119; AVX-NEXT: retq 120; 121; X32-AVX-LABEL: merge_4f64_f64_45zz: 122; X32-AVX: # BB#0: 123; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 124; X32-AVX-NEXT: vmovups 32(%eax), %xmm0 125; X32-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 126; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 127; X32-AVX-NEXT: retl 128 %ptr0 = getelementptr inbounds double, double* %ptr, i64 4 129 %ptr1 = getelementptr inbounds double, double* %ptr, i64 5 130 %val0 = load double, double* %ptr0 131 %val1 = load double, double* %ptr1 132 %res0 = insertelement <4 x double> zeroinitializer, double %val0, i32 0 133 %res1 = insertelement <4 x double> %res0, double %val1, i32 1 134 ret <4 x double> %res1 135} 136 137define <4 x double> @merge_4f64_f64_34z6(double* %ptr) nounwind uwtable noinline ssp { 138; AVX-LABEL: merge_4f64_f64_34z6: 139; AVX: # BB#0: 140; AVX-NEXT: vxorpd %ymm0, %ymm0, %ymm0 141; AVX-NEXT: vblendpd {{.*#+}} ymm0 = mem[0,1],ymm0[2],mem[3] 142; AVX-NEXT: retq 143; 144; X32-AVX-LABEL: merge_4f64_f64_34z6: 145; X32-AVX: # BB#0: 146; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 147; X32-AVX-NEXT: vxorpd %ymm0, %ymm0, %ymm0 148; X32-AVX-NEXT: vblendpd {{.*#+}} ymm0 = mem[0,1],ymm0[2],mem[3] 149; X32-AVX-NEXT: retl 150 %ptr0 = getelementptr inbounds double, double* %ptr, i64 3 151 %ptr1 = getelementptr inbounds double, double* %ptr, i64 4 152 %ptr3 = getelementptr inbounds double, double* %ptr, i64 6 153 %val0 = load double, double* %ptr0 154 %val1 = load double, double* %ptr1 155 %val3 = load double, double* %ptr3 156 %res0 = insertelement <4 x double> undef, double %val0, i32 0 157 %res1 = insertelement <4 x double> %res0, double %val1, i32 1 158 %res2 = insertelement <4 x double> %res1, double 0.0, i32 2 159 %res3 = insertelement <4 x double> %res2, double %val3, i32 3 160 ret <4 x double> %res3 161} 162 163define <4 x i64> @merge_4i64_2i64_3z(<2 x i64>* %ptr) nounwind uwtable noinline ssp { 164; AVX-LABEL: merge_4i64_2i64_3z: 165; AVX: # BB#0: 166; AVX-NEXT: vmovaps 48(%rdi), %xmm0 167; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 168; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 169; AVX-NEXT: retq 170; 171; X32-AVX-LABEL: merge_4i64_2i64_3z: 172; X32-AVX: # BB#0: 173; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 174; X32-AVX-NEXT: vmovaps 48(%eax), %xmm0 175; X32-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 176; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 177; X32-AVX-NEXT: retl 178 %ptr0 = getelementptr inbounds <2 x i64>, <2 x i64>* %ptr, i64 3 179 %val0 = load <2 x i64>, <2 x i64>* %ptr0 180 %res = shufflevector <2 x i64> %val0, <2 x i64> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 181 ret <4 x i64> %res 182} 183 184define <4 x i64> @merge_4i64_i64_1234(i64* %ptr) nounwind uwtable noinline ssp { 185; AVX-LABEL: merge_4i64_i64_1234: 186; AVX: # BB#0: 187; AVX-NEXT: vmovups 8(%rdi), %ymm0 188; AVX-NEXT: retq 189; 190; X32-AVX-LABEL: merge_4i64_i64_1234: 191; X32-AVX: # BB#0: 192; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 193; X32-AVX-NEXT: vmovups 8(%eax), %ymm0 194; X32-AVX-NEXT: retl 195 %ptr0 = getelementptr inbounds i64, i64* %ptr, i64 1 196 %ptr1 = getelementptr inbounds i64, i64* %ptr, i64 2 197 %ptr2 = getelementptr inbounds i64, i64* %ptr, i64 3 198 %ptr3 = getelementptr inbounds i64, i64* %ptr, i64 4 199 %val0 = load i64, i64* %ptr0 200 %val1 = load i64, i64* %ptr1 201 %val2 = load i64, i64* %ptr2 202 %val3 = load i64, i64* %ptr3 203 %res0 = insertelement <4 x i64> undef, i64 %val0, i32 0 204 %res1 = insertelement <4 x i64> %res0, i64 %val1, i32 1 205 %res2 = insertelement <4 x i64> %res1, i64 %val2, i32 2 206 %res3 = insertelement <4 x i64> %res2, i64 %val3, i32 3 207 ret <4 x i64> %res3 208} 209 210define <4 x i64> @merge_4i64_i64_1zzu(i64* %ptr) nounwind uwtable noinline ssp { 211; AVX-LABEL: merge_4i64_i64_1zzu: 212; AVX: # BB#0: 213; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 214; AVX-NEXT: retq 215; 216; X32-AVX-LABEL: merge_4i64_i64_1zzu: 217; X32-AVX: # BB#0: 218; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 219; X32-AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 220; X32-AVX-NEXT: retl 221 %ptr0 = getelementptr inbounds i64, i64* %ptr, i64 1 222 %val0 = load i64, i64* %ptr0 223 %res0 = insertelement <4 x i64> undef, i64 %val0, i32 0 224 %res1 = insertelement <4 x i64> %res0, i64 0, i32 1 225 %res2 = insertelement <4 x i64> %res1, i64 0, i32 1 226 ret <4 x i64> %res2 227} 228 229define <4 x i64> @merge_4i64_i64_23zz(i64* %ptr) nounwind uwtable noinline ssp { 230; AVX-LABEL: merge_4i64_i64_23zz: 231; AVX: # BB#0: 232; AVX-NEXT: vmovups 16(%rdi), %xmm0 233; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 234; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 235; AVX-NEXT: retq 236; 237; X32-AVX-LABEL: merge_4i64_i64_23zz: 238; X32-AVX: # BB#0: 239; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 240; X32-AVX-NEXT: vmovups 16(%eax), %xmm0 241; X32-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 242; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 243; X32-AVX-NEXT: retl 244 %ptr0 = getelementptr inbounds i64, i64* %ptr, i64 2 245 %ptr1 = getelementptr inbounds i64, i64* %ptr, i64 3 246 %val0 = load i64, i64* %ptr0 247 %val1 = load i64, i64* %ptr1 248 %res0 = insertelement <4 x i64> zeroinitializer, i64 %val0, i32 0 249 %res1 = insertelement <4 x i64> %res0, i64 %val1, i32 1 250 ret <4 x i64> %res1 251} 252 253define <8 x float> @merge_8f32_2f32_23z5(<2 x float>* %ptr) nounwind uwtable noinline ssp { 254; AVX1-LABEL: merge_8f32_2f32_23z5: 255; AVX1: # BB#0: 256; AVX1-NEXT: vmovupd 16(%rdi), %xmm0 257; AVX1-NEXT: vxorpd %xmm1, %xmm1, %xmm1 258; AVX1-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] 259; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 260; AVX1-NEXT: retq 261; 262; AVX2-LABEL: merge_8f32_2f32_23z5: 263; AVX2: # BB#0: 264; AVX2-NEXT: vmovupd 16(%rdi), %xmm0 265; AVX2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 266; AVX2-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] 267; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 268; AVX2-NEXT: retq 269; 270; AVX512F-LABEL: merge_8f32_2f32_23z5: 271; AVX512F: # BB#0: 272; AVX512F-NEXT: vmovups 16(%rdi), %xmm0 273; AVX512F-NEXT: vxorps %xmm1, %xmm1, %xmm1 274; AVX512F-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] 275; AVX512F-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 276; AVX512F-NEXT: retq 277; 278; X32-AVX-LABEL: merge_8f32_2f32_23z5: 279; X32-AVX: # BB#0: 280; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 281; X32-AVX-NEXT: vxorpd %ymm0, %ymm0, %ymm0 282; X32-AVX-NEXT: vblendpd {{.*#+}} ymm0 = mem[0,1],ymm0[2],mem[3] 283; X32-AVX-NEXT: retl 284 %ptr0 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 2 285 %ptr1 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 3 286 %ptr3 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 5 287 %val0 = load <2 x float>, <2 x float>* %ptr0 288 %val1 = load <2 x float>, <2 x float>* %ptr1 289 %val3 = load <2 x float>, <2 x float>* %ptr3 290 %res01 = shufflevector <2 x float> %val0, <2 x float> %val1, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 291 %res23 = shufflevector <2 x float> zeroinitializer, <2 x float> %val3, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 292 %res = shufflevector <4 x float> %res01, <4 x float> %res23, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 293 ret <8 x float> %res 294} 295 296define <8 x float> @merge_8f32_4f32_z2(<4 x float>* %ptr) nounwind uwtable noinline ssp { 297; AVX-LABEL: merge_8f32_4f32_z2: 298; AVX: # BB#0: 299; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0 300; AVX-NEXT: vinsertf128 $1, 32(%rdi), %ymm0, %ymm0 301; AVX-NEXT: retq 302; 303; X32-AVX-LABEL: merge_8f32_4f32_z2: 304; X32-AVX: # BB#0: 305; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 306; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0 307; X32-AVX-NEXT: vinsertf128 $1, 32(%eax), %ymm0, %ymm0 308; X32-AVX-NEXT: retl 309 %ptr1 = getelementptr inbounds <4 x float>, <4 x float>* %ptr, i64 2 310 %val1 = load <4 x float>, <4 x float>* %ptr1 311 %res = shufflevector <4 x float> zeroinitializer, <4 x float> %val1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 312 ret <8 x float> %res 313} 314 315define <8 x float> @merge_8f32_f32_12zzuuzz(float* %ptr) nounwind uwtable noinline ssp { 316; AVX-LABEL: merge_8f32_f32_12zzuuzz: 317; AVX: # BB#0: 318; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 319; AVX-NEXT: retq 320; 321; X32-AVX-LABEL: merge_8f32_f32_12zzuuzz: 322; X32-AVX: # BB#0: 323; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 324; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 325; X32-AVX-NEXT: retl 326 %ptr0 = getelementptr inbounds float, float* %ptr, i64 1 327 %ptr1 = getelementptr inbounds float, float* %ptr, i64 2 328 %val0 = load float, float* %ptr0 329 %val1 = load float, float* %ptr1 330 %res0 = insertelement <8 x float> undef, float %val0, i32 0 331 %res1 = insertelement <8 x float> %res0, float %val1, i32 1 332 %res2 = insertelement <8 x float> %res1, float 0.0, i32 2 333 %res3 = insertelement <8 x float> %res2, float 0.0, i32 3 334 %res6 = insertelement <8 x float> %res3, float 0.0, i32 6 335 %res7 = insertelement <8 x float> %res6, float 0.0, i32 7 336 ret <8 x float> %res7 337} 338 339define <8 x float> @merge_8f32_f32_1u3u5zu8(float* %ptr) nounwind uwtable noinline ssp { 340; AVX-LABEL: merge_8f32_f32_1u3u5zu8: 341; AVX: # BB#0: 342; AVX-NEXT: vxorps %ymm0, %ymm0, %ymm0 343; AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7] 344; AVX-NEXT: retq 345; 346; X32-AVX-LABEL: merge_8f32_f32_1u3u5zu8: 347; X32-AVX: # BB#0: 348; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 349; X32-AVX-NEXT: vxorps %ymm0, %ymm0, %ymm0 350; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7] 351; X32-AVX-NEXT: retl 352 %ptr0 = getelementptr inbounds float, float* %ptr, i64 1 353 %ptr2 = getelementptr inbounds float, float* %ptr, i64 3 354 %ptr4 = getelementptr inbounds float, float* %ptr, i64 5 355 %ptr7 = getelementptr inbounds float, float* %ptr, i64 8 356 %val0 = load float, float* %ptr0 357 %val2 = load float, float* %ptr2 358 %val4 = load float, float* %ptr4 359 %val7 = load float, float* %ptr7 360 %res0 = insertelement <8 x float> undef, float %val0, i32 0 361 %res2 = insertelement <8 x float> %res0, float %val2, i32 2 362 %res4 = insertelement <8 x float> %res2, float %val4, i32 4 363 %res5 = insertelement <8 x float> %res4, float 0.0, i32 5 364 %res7 = insertelement <8 x float> %res5, float %val7, i32 7 365 ret <8 x float> %res7 366} 367 368define <8 x i32> @merge_8i32_4i32_z3(<4 x i32>* %ptr) nounwind uwtable noinline ssp { 369; AVX-LABEL: merge_8i32_4i32_z3: 370; AVX: # BB#0: 371; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0 372; AVX-NEXT: vinsertf128 $1, 48(%rdi), %ymm0, %ymm0 373; AVX-NEXT: retq 374; 375; X32-AVX-LABEL: merge_8i32_4i32_z3: 376; X32-AVX: # BB#0: 377; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 378; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0 379; X32-AVX-NEXT: vinsertf128 $1, 48(%eax), %ymm0, %ymm0 380; X32-AVX-NEXT: retl 381 %ptr1 = getelementptr inbounds <4 x i32>, <4 x i32>* %ptr, i64 3 382 %val1 = load <4 x i32>, <4 x i32>* %ptr1 383 %res = shufflevector <4 x i32> zeroinitializer, <4 x i32> %val1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 384 ret <8 x i32> %res 385} 386 387define <8 x i32> @merge_8i32_i32_56zz9uzz(i32* %ptr) nounwind uwtable noinline ssp { 388; AVX1-LABEL: merge_8i32_i32_56zz9uzz: 389; AVX1: # BB#0: 390; AVX1-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 391; AVX1-NEXT: vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero 392; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 393; AVX1-NEXT: retq 394; 395; AVX2-LABEL: merge_8i32_i32_56zz9uzz: 396; AVX2: # BB#0: 397; AVX2-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 398; AVX2-NEXT: vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero 399; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 400; AVX2-NEXT: retq 401; 402; AVX512F-LABEL: merge_8i32_i32_56zz9uzz: 403; AVX512F: # BB#0: 404; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 405; AVX512F-NEXT: vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero 406; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 407; AVX512F-NEXT: retq 408; 409; X32-AVX-LABEL: merge_8i32_i32_56zz9uzz: 410; X32-AVX: # BB#0: 411; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 412; X32-AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 413; X32-AVX-NEXT: vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero 414; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 415; X32-AVX-NEXT: retl 416 %ptr0 = getelementptr inbounds i32, i32* %ptr, i64 5 417 %ptr1 = getelementptr inbounds i32, i32* %ptr, i64 6 418 %ptr4 = getelementptr inbounds i32, i32* %ptr, i64 9 419 %val0 = load i32, i32* %ptr0 420 %val1 = load i32, i32* %ptr1 421 %val4 = load i32, i32* %ptr4 422 %res0 = insertelement <8 x i32> undef, i32 %val0, i32 0 423 %res1 = insertelement <8 x i32> %res0, i32 %val1, i32 1 424 %res2 = insertelement <8 x i32> %res1, i32 0, i32 2 425 %res3 = insertelement <8 x i32> %res2, i32 0, i32 3 426 %res4 = insertelement <8 x i32> %res3, i32 %val4, i32 4 427 %res6 = insertelement <8 x i32> %res4, i32 0, i32 6 428 %res7 = insertelement <8 x i32> %res6, i32 0, i32 7 429 ret <8 x i32> %res7 430} 431 432define <8 x i32> @merge_8i32_i32_1u3u5zu8(i32* %ptr) nounwind uwtable noinline ssp { 433; AVX1-LABEL: merge_8i32_i32_1u3u5zu8: 434; AVX1: # BB#0: 435; AVX1-NEXT: vxorps %ymm0, %ymm0, %ymm0 436; AVX1-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7] 437; AVX1-NEXT: retq 438; 439; AVX2-LABEL: merge_8i32_i32_1u3u5zu8: 440; AVX2: # BB#0: 441; AVX2-NEXT: vpxor %ymm0, %ymm0, %ymm0 442; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7] 443; AVX2-NEXT: retq 444; 445; AVX512F-LABEL: merge_8i32_i32_1u3u5zu8: 446; AVX512F: # BB#0: 447; AVX512F-NEXT: vpxor %ymm0, %ymm0, %ymm0 448; AVX512F-NEXT: vpblendd {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7] 449; AVX512F-NEXT: retq 450; 451; X32-AVX-LABEL: merge_8i32_i32_1u3u5zu8: 452; X32-AVX: # BB#0: 453; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 454; X32-AVX-NEXT: vxorps %ymm0, %ymm0, %ymm0 455; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7] 456; X32-AVX-NEXT: retl 457 %ptr0 = getelementptr inbounds i32, i32* %ptr, i64 1 458 %ptr2 = getelementptr inbounds i32, i32* %ptr, i64 3 459 %ptr4 = getelementptr inbounds i32, i32* %ptr, i64 5 460 %ptr7 = getelementptr inbounds i32, i32* %ptr, i64 8 461 %val0 = load i32, i32* %ptr0 462 %val2 = load i32, i32* %ptr2 463 %val4 = load i32, i32* %ptr4 464 %val7 = load i32, i32* %ptr7 465 %res0 = insertelement <8 x i32> undef, i32 %val0, i32 0 466 %res2 = insertelement <8 x i32> %res0, i32 %val2, i32 2 467 %res4 = insertelement <8 x i32> %res2, i32 %val4, i32 4 468 %res5 = insertelement <8 x i32> %res4, i32 0, i32 5 469 %res7 = insertelement <8 x i32> %res5, i32 %val7, i32 7 470 ret <8 x i32> %res7 471} 472 473define <16 x i16> @merge_16i16_i16_89zzzuuuuuuuuuuuz(i16* %ptr) nounwind uwtable noinline ssp { 474; AVX-LABEL: merge_16i16_i16_89zzzuuuuuuuuuuuz: 475; AVX: # BB#0: 476; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 477; AVX-NEXT: retq 478; 479; X32-AVX-LABEL: merge_16i16_i16_89zzzuuuuuuuuuuuz: 480; X32-AVX: # BB#0: 481; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 482; X32-AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 483; X32-AVX-NEXT: retl 484 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 8 485 %ptr1 = getelementptr inbounds i16, i16* %ptr, i64 9 486 %val0 = load i16, i16* %ptr0 487 %val1 = load i16, i16* %ptr1 488 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0 489 %res1 = insertelement <16 x i16> %res0, i16 %val1, i16 1 490 %res2 = insertelement <16 x i16> %res1, i16 0, i16 2 491 %res3 = insertelement <16 x i16> %res2, i16 0, i16 3 492 %res4 = insertelement <16 x i16> %res3, i16 0, i16 4 493 %resF = insertelement <16 x i16> %res4, i16 0, i16 15 494 ret <16 x i16> %resF 495} 496 497define <16 x i16> @merge_16i16_i16_45u7uuuuuuuuuuuu(i16* %ptr) nounwind uwtable noinline ssp { 498; AVX-LABEL: merge_16i16_i16_45u7uuuuuuuuuuuu: 499; AVX: # BB#0: 500; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 501; AVX-NEXT: retq 502; 503; X32-AVX-LABEL: merge_16i16_i16_45u7uuuuuuuuuuuu: 504; X32-AVX: # BB#0: 505; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 506; X32-AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 507; X32-AVX-NEXT: retl 508 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 4 509 %ptr1 = getelementptr inbounds i16, i16* %ptr, i64 5 510 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 7 511 %val0 = load i16, i16* %ptr0 512 %val1 = load i16, i16* %ptr1 513 %val3 = load i16, i16* %ptr3 514 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0 515 %res1 = insertelement <16 x i16> %res0, i16 %val1, i16 1 516 %res3 = insertelement <16 x i16> %res1, i16 %val3, i16 3 517 ret <16 x i16> %res3 518} 519 520define <16 x i16> @merge_16i16_i16_0uu3uuuuuuuuCuEF(i16* %ptr) nounwind uwtable noinline ssp { 521; AVX-LABEL: merge_16i16_i16_0uu3uuuuuuuuCuEF: 522; AVX: # BB#0: 523; AVX-NEXT: vmovups (%rdi), %ymm0 524; AVX-NEXT: retq 525; 526; X32-AVX-LABEL: merge_16i16_i16_0uu3uuuuuuuuCuEF: 527; X32-AVX: # BB#0: 528; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 529; X32-AVX-NEXT: vmovups (%eax), %ymm0 530; X32-AVX-NEXT: retl 531 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0 532 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3 533 %ptrC = getelementptr inbounds i16, i16* %ptr, i64 12 534 %ptrE = getelementptr inbounds i16, i16* %ptr, i64 14 535 %ptrF = getelementptr inbounds i16, i16* %ptr, i64 15 536 %val0 = load i16, i16* %ptr0 537 %val3 = load i16, i16* %ptr3 538 %valC = load i16, i16* %ptrC 539 %valE = load i16, i16* %ptrE 540 %valF = load i16, i16* %ptrF 541 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0 542 %res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3 543 %resC = insertelement <16 x i16> %res3, i16 %valC, i16 12 544 %resE = insertelement <16 x i16> %resC, i16 %valE, i16 14 545 %resF = insertelement <16 x i16> %resE, i16 %valF, i16 15 546 ret <16 x i16> %resF 547} 548 549define <16 x i16> @merge_16i16_i16_0uu3zzuuuuuzCuEF(i16* %ptr) nounwind uwtable noinline ssp { 550; AVX1-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF: 551; AVX1: # BB#0: 552; AVX1-NEXT: vmovaps {{.*#+}} ymm0 = [65535,0,0,65535,0,0,0,0,0,0,0,0,65535,0,65535,65535] 553; AVX1-NEXT: vandps (%rdi), %ymm0, %ymm0 554; AVX1-NEXT: retq 555; 556; AVX2-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF: 557; AVX2: # BB#0: 558; AVX2-NEXT: vmovups (%rdi), %ymm0 559; AVX2-NEXT: vandps {{.*}}(%rip), %ymm0, %ymm0 560; AVX2-NEXT: retq 561; 562; AVX512F-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF: 563; AVX512F: # BB#0: 564; AVX512F-NEXT: vmovups (%rdi), %ymm0 565; AVX512F-NEXT: vandps {{.*}}(%rip), %ymm0, %ymm0 566; AVX512F-NEXT: retq 567; 568; X32-AVX-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF: 569; X32-AVX: # BB#0: 570; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 571; X32-AVX-NEXT: vmovaps {{.*#+}} ymm0 = [65535,0,0,65535,0,0,0,0,0,0,0,0,65535,0,65535,65535] 572; X32-AVX-NEXT: vandps (%eax), %ymm0, %ymm0 573; X32-AVX-NEXT: retl 574 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0 575 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3 576 %ptrC = getelementptr inbounds i16, i16* %ptr, i64 12 577 %ptrE = getelementptr inbounds i16, i16* %ptr, i64 14 578 %ptrF = getelementptr inbounds i16, i16* %ptr, i64 15 579 %val0 = load i16, i16* %ptr0 580 %val3 = load i16, i16* %ptr3 581 %valC = load i16, i16* %ptrC 582 %valE = load i16, i16* %ptrE 583 %valF = load i16, i16* %ptrF 584 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0 585 %res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3 586 %res4 = insertelement <16 x i16> %res3, i16 0, i16 4 587 %res5 = insertelement <16 x i16> %res4, i16 0, i16 5 588 %resC = insertelement <16 x i16> %res5, i16 %valC, i16 12 589 %resD = insertelement <16 x i16> %resC, i16 0, i16 13 590 %resE = insertelement <16 x i16> %resD, i16 %valE, i16 14 591 %resF = insertelement <16 x i16> %resE, i16 %valF, i16 15 592 ret <16 x i16> %resF 593} 594 595define <32 x i8> @merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu(i8* %ptr) nounwind uwtable noinline ssp { 596; AVX-LABEL: merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu: 597; AVX: # BB#0: 598; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 599; AVX-NEXT: retq 600; 601; X32-AVX-LABEL: merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu: 602; X32-AVX: # BB#0: 603; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 604; X32-AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 605; X32-AVX-NEXT: retl 606 %ptr0 = getelementptr inbounds i8, i8* %ptr, i64 4 607 %ptr1 = getelementptr inbounds i8, i8* %ptr, i64 5 608 %ptr3 = getelementptr inbounds i8, i8* %ptr, i64 7 609 %val0 = load i8, i8* %ptr0 610 %val1 = load i8, i8* %ptr1 611 %val3 = load i8, i8* %ptr3 612 %res0 = insertelement <32 x i8> undef, i8 %val0, i8 0 613 %res1 = insertelement <32 x i8> %res0, i8 %val1, i8 1 614 %res3 = insertelement <32 x i8> %res1, i8 %val3, i8 3 615 ret <32 x i8> %res3 616} 617 618define <32 x i8> @merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu(i8* %ptr) nounwind uwtable noinline ssp { 619; AVX-LABEL: merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu: 620; AVX: # BB#0: 621; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 622; AVX-NEXT: retq 623; 624; X32-AVX-LABEL: merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu: 625; X32-AVX: # BB#0: 626; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 627; X32-AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero 628; X32-AVX-NEXT: retl 629 %ptr0 = getelementptr inbounds i8, i8* %ptr, i64 2 630 %ptr1 = getelementptr inbounds i8, i8* %ptr, i64 3 631 %ptr3 = getelementptr inbounds i8, i8* %ptr, i64 5 632 %val0 = load i8, i8* %ptr0 633 %val1 = load i8, i8* %ptr1 634 %val3 = load i8, i8* %ptr3 635 %res0 = insertelement <32 x i8> undef, i8 %val0, i8 0 636 %res1 = insertelement <32 x i8> %res0, i8 %val1, i8 1 637 %res3 = insertelement <32 x i8> %res1, i8 %val3, i8 3 638 %resE = insertelement <32 x i8> %res3, i8 0, i8 14 639 %resF = insertelement <32 x i8> %resE, i8 0, i8 15 640 %resG = insertelement <32 x i8> %resF, i8 0, i8 16 641 %resH = insertelement <32 x i8> %resG, i8 0, i8 17 642 ret <32 x i8> %resH 643} 644 645; 646; consecutive loads including any/all volatiles may not be combined 647; 648 649define <4 x double> @merge_4f64_f64_34uz_volatile(double* %ptr) nounwind uwtable noinline ssp { 650; AVX1-LABEL: merge_4f64_f64_34uz_volatile: 651; AVX1: # BB#0: 652; AVX1-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 653; AVX1-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0] 654; AVX1-NEXT: vxorpd %xmm1, %xmm1, %xmm1 655; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 656; AVX1-NEXT: retq 657; 658; AVX2-LABEL: merge_4f64_f64_34uz_volatile: 659; AVX2: # BB#0: 660; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 661; AVX2-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0] 662; AVX2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 663; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 664; AVX2-NEXT: retq 665; 666; AVX512F-LABEL: merge_4f64_f64_34uz_volatile: 667; AVX512F: # BB#0: 668; AVX512F-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 669; AVX512F-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0] 670; AVX512F-NEXT: vxorps %xmm1, %xmm1, %xmm1 671; AVX512F-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 672; AVX512F-NEXT: retq 673; 674; X32-AVX-LABEL: merge_4f64_f64_34uz_volatile: 675; X32-AVX: # BB#0: 676; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 677; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero 678; X32-AVX-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0] 679; X32-AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1 680; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 681; X32-AVX-NEXT: retl 682 %ptr0 = getelementptr inbounds double, double* %ptr, i64 3 683 %ptr1 = getelementptr inbounds double, double* %ptr, i64 4 684 %val0 = load volatile double, double* %ptr0 685 %val1 = load volatile double, double* %ptr1 686 %res0 = insertelement <4 x double> undef, double %val0, i32 0 687 %res1 = insertelement <4 x double> %res0, double %val1, i32 1 688 %res3 = insertelement <4 x double> %res1, double 0.0, i32 3 689 ret <4 x double> %res3 690} 691 692define <16 x i16> @merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile(i16* %ptr) nounwind uwtable noinline ssp { 693; AVX1-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile: 694; AVX1: # BB#0: 695; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0 696; AVX1-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1 697; AVX1-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1 698; AVX1-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0 699; AVX1-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0 700; AVX1-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0 701; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 702; AVX1-NEXT: retq 703; 704; AVX2-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile: 705; AVX2: # BB#0: 706; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 707; AVX2-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1 708; AVX2-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1 709; AVX2-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0 710; AVX2-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0 711; AVX2-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0 712; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 713; AVX2-NEXT: retq 714; 715; AVX512F-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile: 716; AVX512F: # BB#0: 717; AVX512F-NEXT: vpxor %xmm0, %xmm0, %xmm0 718; AVX512F-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1 719; AVX512F-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1 720; AVX512F-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0 721; AVX512F-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0 722; AVX512F-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0 723; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 724; AVX512F-NEXT: retq 725; 726; X32-AVX-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile: 727; X32-AVX: # BB#0: 728; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 729; X32-AVX-NEXT: vpxor %xmm0, %xmm0, %xmm0 730; X32-AVX-NEXT: vpinsrw $0, (%eax), %xmm0, %xmm1 731; X32-AVX-NEXT: vpinsrw $3, 6(%eax), %xmm1, %xmm1 732; X32-AVX-NEXT: vpinsrw $4, 24(%eax), %xmm0, %xmm0 733; X32-AVX-NEXT: vpinsrw $6, 28(%eax), %xmm0, %xmm0 734; X32-AVX-NEXT: vpinsrw $7, 30(%eax), %xmm0, %xmm0 735; X32-AVX-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 736; X32-AVX-NEXT: retl 737 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0 738 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3 739 %ptrC = getelementptr inbounds i16, i16* %ptr, i64 12 740 %ptrE = getelementptr inbounds i16, i16* %ptr, i64 14 741 %ptrF = getelementptr inbounds i16, i16* %ptr, i64 15 742 %val0 = load volatile i16, i16* %ptr0 743 %val3 = load i16, i16* %ptr3 744 %valC = load i16, i16* %ptrC 745 %valE = load i16, i16* %ptrE 746 %valF = load volatile i16, i16* %ptrF 747 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0 748 %res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3 749 %res4 = insertelement <16 x i16> %res3, i16 0, i16 4 750 %res5 = insertelement <16 x i16> %res4, i16 0, i16 5 751 %resC = insertelement <16 x i16> %res5, i16 %valC, i16 12 752 %resD = insertelement <16 x i16> %resC, i16 0, i16 13 753 %resE = insertelement <16 x i16> %resD, i16 %valE, i16 14 754 %resF = insertelement <16 x i16> %resE, i16 %valF, i16 15 755 ret <16 x i16> %resF 756} 757