brintos / llvm-project-archived public Read only

1; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py2; RUN: opt -passes="print<cost-model>" -cost-kind=all 2>&1 -disable-output -mtriple=thumbv8.1m.main-none-eabi -mattr=+mve < %s | FileCheck %s --check-prefixes=CHECK,CHECK-UF23; RUN: opt -passes="print<cost-model>" -cost-kind=all 2>&1 -disable-output -mtriple=thumbv8.1m.main-none-eabi -mattr=+mve.fp < %s | FileCheck %s --check-prefixes=CHECK,CHECK-UF24; RUN: opt -passes="print<cost-model>" -cost-kind=all 2>&1 -disable-output -mtriple=thumbv8.1m.main-none-eabi -mattr=+mve -mve-max-interleave-factor=4 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-UF45; RUN: opt -passes="print<cost-model>" -cost-kind=all 2>&1 -disable-output -mtriple=thumbv8.1m.main-none-eabi -mattr=+mve.fp -mve-max-interleave-factor=4 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-UF46 7define void @vld2(ptr %p) {8; CHECK-LABEL: 'vld2'9; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v4i8 = load <4 x i8>, ptr %p, align 410; CHECK-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v4i8_0 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 0, i32 2>11; CHECK-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v4i8_1 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 1, i32 3>12; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v8i8 = load <8 x i8>, ptr %p, align 813; CHECK-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v8i8_0 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>14; CHECK-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v8i8_1 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>15; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v16i8 = load <16 x i8>, ptr %p, align 816; CHECK-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v16i8_0 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>17; CHECK-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v16i8_1 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>18; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v32i8 = load <32 x i8>, ptr %p, align 3219; CHECK-NEXT:  Cost Model: Found costs of RThru:258 CodeSize:129 Lat:258 SizeLat:258 for: %v32i8_0 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>20; CHECK-NEXT:  Cost Model: Found costs of RThru:258 CodeSize:129 Lat:258 SizeLat:258 for: %v32i8_1 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>21; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v4i16 = load <4 x i16>, ptr %p, align 822; CHECK-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v4i16_0 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 0, i32 2>23; CHECK-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v4i16_1 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 1, i32 3>24; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v8i16 = load <8 x i16>, ptr %p, align 825; CHECK-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v8i16_0 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>26; CHECK-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v8i16_1 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>27; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v16i16 = load <16 x i16>, ptr %p, align 3228; CHECK-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v16i16_0 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>29; CHECK-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v16i16_1 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>30; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v32i16 = load <32 x i16>, ptr %p, align 6431; CHECK-NEXT:  Cost Model: Found costs of RThru:258 CodeSize:129 Lat:258 SizeLat:258 for: %v32i16_0 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>32; CHECK-NEXT:  Cost Model: Found costs of RThru:258 CodeSize:129 Lat:258 SizeLat:258 for: %v32i16_1 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>33; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v4i32 = load <4 x i32>, ptr %p, align 834; CHECK-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v4i32_0 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 0, i32 2>35; CHECK-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v4i32_1 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 1, i32 3>36; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v8i32 = load <8 x i32>, ptr %p, align 3237; CHECK-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v8i32_0 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>38; CHECK-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v8i32_1 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>39; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v16i32 = load <16 x i32>, ptr %p, align 6440; CHECK-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v16i32_0 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>41; CHECK-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v16i32_1 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>42; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v32i32 = load <32 x i32>, ptr %p, align 12843; CHECK-NEXT:  Cost Model: Found costs of RThru:260 CodeSize:130 Lat:260 SizeLat:260 for: %v32i32_0 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>44; CHECK-NEXT:  Cost Model: Found costs of RThru:260 CodeSize:130 Lat:260 SizeLat:260 for: %v32i32_1 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>45; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v2i64 = load <4 x i64>, ptr %p, align 3246; CHECK-NEXT:  Cost Model: Found costs of RThru:192 CodeSize:96 Lat:192 SizeLat:192 for: %v2i64_0 = shufflevector <4 x i64> %v2i64, <4 x i64> undef, <2 x i32> <i32 0, i32 2>47; CHECK-NEXT:  Cost Model: Found costs of RThru:192 CodeSize:96 Lat:192 SizeLat:192 for: %v2i64_1 = shufflevector <4 x i64> %v2i64, <4 x i64> undef, <2 x i32> <i32 1, i32 3>48; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v4i64 = load <8 x i64>, ptr %p, align 6449; CHECK-NEXT:  Cost Model: Found costs of RThru:384 CodeSize:192 Lat:384 SizeLat:384 for: %v4i64_0 = shufflevector <8 x i64> %v4i64, <8 x i64> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>50; CHECK-NEXT:  Cost Model: Found costs of RThru:384 CodeSize:192 Lat:384 SizeLat:384 for: %v4i64_1 = shufflevector <8 x i64> %v4i64, <8 x i64> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>51; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v8i64 = load <16 x i64>, ptr %p, align 12852; CHECK-NEXT:  Cost Model: Found costs of RThru:768 CodeSize:384 Lat:768 SizeLat:768 for: %v8i64_0 = shufflevector <16 x i64> %v8i64, <16 x i64> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>53; CHECK-NEXT:  Cost Model: Found costs of RThru:768 CodeSize:384 Lat:768 SizeLat:768 for: %v8i64_1 = shufflevector <16 x i64> %v8i64, <16 x i64> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>54; CHECK-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:4 SizeLat:1 for: %v16i64 = load <32 x i64>, ptr %p, align 25655; CHECK-NEXT:  Cost Model: Found costs of RThru:1536 CodeSize:768 Lat:1536 SizeLat:1536 for: %v16i64_0 = shufflevector <32 x i64> %v16i64, <32 x i64> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>56; CHECK-NEXT:  Cost Model: Found costs of RThru:1536 CodeSize:768 Lat:1536 SizeLat:1536 for: %v16i64_1 = shufflevector <32 x i64> %v16i64, <32 x i64> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>57; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void58;59  %v4i8 = load <4 x i8>, ptr %p60  %v4i8_0 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 0, i32 2>61  %v4i8_1 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 1, i32 3>62  %v8i8 = load <8 x i8>, ptr %p63  %v8i8_0 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>64  %v8i8_1 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>65  %v16i8 = load <16 x i8>, ptr %p66  %v16i8_0 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>67  %v16i8_1 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>68  %v32i8 = load <32 x i8>, ptr %p69  %v32i8_0 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>70  %v32i8_1 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>71 72  %v4i16 = load <4 x i16>, ptr %p73  %v4i16_0 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 0, i32 2>74  %v4i16_1 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 1, i32 3>75  %v8i16 = load <8 x i16>, ptr %p76  %v8i16_0 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>77  %v8i16_1 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>78  %v16i16 = load <16 x i16>, ptr %p79  %v16i16_0 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>80  %v16i16_1 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>81  %v32i16 = load <32 x i16>, ptr %p82  %v32i16_0 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>83  %v32i16_1 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>84 85  %v4i32 = load <4 x i32>, ptr %p86  %v4i32_0 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 0, i32 2>87  %v4i32_1 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 1, i32 3>88  %v8i32 = load <8 x i32>, ptr %p89  %v8i32_0 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>90  %v8i32_1 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>91  %v16i32 = load <16 x i32>, ptr %p92  %v16i32_0 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>93  %v16i32_1 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>94  %v32i32 = load <32 x i32>, ptr %p95  %v32i32_0 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>96  %v32i32_1 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>97 98  %v2i64 = load <4 x i64>, ptr %p99  %v2i64_0 = shufflevector <4 x i64> %v2i64, <4 x i64> undef, <2 x i32> <i32 0, i32 2>100  %v2i64_1 = shufflevector <4 x i64> %v2i64, <4 x i64> undef, <2 x i32> <i32 1, i32 3>101  %v4i64 = load <8 x i64>, ptr %p102  %v4i64_0 = shufflevector <8 x i64> %v4i64, <8 x i64> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>103  %v4i64_1 = shufflevector <8 x i64> %v4i64, <8 x i64> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>104  %v8i64 = load <16 x i64>, ptr %p105  %v8i64_0 = shufflevector <16 x i64> %v8i64, <16 x i64> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>106  %v8i64_1 = shufflevector <16 x i64> %v8i64, <16 x i64> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>107  %v16i64 = load <32 x i64>, ptr %p108  %v16i64_0 = shufflevector <32 x i64> %v16i64, <32 x i64> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>109  %v16i64_1 = shufflevector <32 x i64> %v16i64, <32 x i64> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>110 111  ret void112}113 114 115define void @vld3(ptr %p) {116; CHECK-LABEL: 'vld3'117; CHECK-NEXT:  Cost Model: Found costs of RThru:50 CodeSize:1 Lat:4 SizeLat:1 for: %v2i8 = load <6 x i8>, ptr %p, align 8118; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i8_0 = shufflevector <6 x i8> %v2i8, <6 x i8> undef, <2 x i32> <i32 0, i32 3>119; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i8_1 = shufflevector <6 x i8> %v2i8, <6 x i8> undef, <2 x i32> <i32 1, i32 4>120; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i8_2 = shufflevector <6 x i8> %v2i8, <6 x i8> undef, <2 x i32> <i32 2, i32 5>121; CHECK-NEXT:  Cost Model: Found costs of RThru:98 CodeSize:1 Lat:4 SizeLat:1 for: %v4i8 = load <12 x i8>, ptr %p, align 16122; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i8_0 = shufflevector <12 x i8> %v4i8, <12 x i8> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>123; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i8_1 = shufflevector <12 x i8> %v4i8, <12 x i8> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>124; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i8_2 = shufflevector <12 x i8> %v4i8, <12 x i8> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>125; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v8i8 = load <24 x i8>, ptr %p, align 32126; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i8_0 = shufflevector <24 x i8> %v8i8, <24 x i8> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>127; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i8_1 = shufflevector <24 x i8> %v8i8, <24 x i8> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>128; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i8_2 = shufflevector <24 x i8> %v8i8, <24 x i8> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>129; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v16i8 = load <48 x i8>, ptr %p, align 64130; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i8_0 = shufflevector <48 x i8> %v16i8, <48 x i8> undef, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>131; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i8_1 = shufflevector <48 x i8> %v16i8, <48 x i8> undef, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>132; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i8_2 = shufflevector <48 x i8> %v16i8, <48 x i8> undef, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>133; CHECK-NEXT:  Cost Model: Found costs of RThru:50 CodeSize:1 Lat:4 SizeLat:1 for: %v2i16 = load <6 x i16>, ptr %p, align 16134; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i16_0 = shufflevector <6 x i16> %v2i16, <6 x i16> undef, <2 x i32> <i32 0, i32 3>135; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i16_1 = shufflevector <6 x i16> %v2i16, <6 x i16> undef, <2 x i32> <i32 1, i32 4>136; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i16_2 = shufflevector <6 x i16> %v2i16, <6 x i16> undef, <2 x i32> <i32 2, i32 5>137; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v4i16 = load <12 x i16>, ptr %p, align 32138; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i16_0 = shufflevector <12 x i16> %v4i16, <12 x i16> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>139; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i16_1 = shufflevector <12 x i16> %v4i16, <12 x i16> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>140; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i16_2 = shufflevector <12 x i16> %v4i16, <12 x i16> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>141; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v8i16 = load <24 x i16>, ptr %p, align 64142; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i16_0 = shufflevector <24 x i16> %v8i16, <24 x i16> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>143; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i16_1 = shufflevector <24 x i16> %v8i16, <24 x i16> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>144; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i16_2 = shufflevector <24 x i16> %v8i16, <24 x i16> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>145; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v16i16 = load <48 x i16>, ptr %p, align 128146; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i16_0 = shufflevector <48 x i16> %v16i16, <48 x i16> undef, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>147; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i16_1 = shufflevector <48 x i16> %v16i16, <48 x i16> undef, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>148; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i16_2 = shufflevector <48 x i16> %v16i16, <48 x i16> undef, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>149; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v2i32 = load <6 x i32>, ptr %p, align 32150; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i32_0 = shufflevector <6 x i32> %v2i32, <6 x i32> undef, <2 x i32> <i32 0, i32 3>151; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i32_1 = shufflevector <6 x i32> %v2i32, <6 x i32> undef, <2 x i32> <i32 1, i32 4>152; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v2i32_2 = shufflevector <6 x i32> %v2i32, <6 x i32> undef, <2 x i32> <i32 2, i32 5>153; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v4i32 = load <12 x i32>, ptr %p, align 64154; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i32_0 = shufflevector <12 x i32> %v4i32, <12 x i32> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>155; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i32_1 = shufflevector <12 x i32> %v4i32, <12 x i32> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>156; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v4i32_2 = shufflevector <12 x i32> %v4i32, <12 x i32> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>157; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v8i32 = load <24 x i32>, ptr %p, align 128158; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i32_0 = shufflevector <24 x i32> %v8i32, <24 x i32> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>159; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i32_1 = shufflevector <24 x i32> %v8i32, <24 x i32> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>160; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v8i32_2 = shufflevector <24 x i32> %v8i32, <24 x i32> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>161; CHECK-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:4 SizeLat:1 for: %v16i32 = load <48 x i32>, ptr %p, align 256162; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i32_0 = shufflevector <48 x i32> %v16i32, <48 x i32> undef, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>163; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i32_1 = shufflevector <48 x i32> %v16i32, <48 x i32> undef, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>164; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v16i32_2 = shufflevector <48 x i32> %v16i32, <48 x i32> undef, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>165; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v2i64 = load <6 x i64>, ptr %p, align 64166; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v2i64_0 = shufflevector <6 x i64> %v2i64, <6 x i64> undef, <2 x i32> <i32 0, i32 3>167; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v2i64_1 = shufflevector <6 x i64> %v2i64, <6 x i64> undef, <2 x i32> <i32 1, i32 4>168; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v2i64_2 = shufflevector <6 x i64> %v2i64, <6 x i64> undef, <2 x i32> <i32 2, i32 5>169; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v4i64 = load <12 x i64>, ptr %p, align 128170; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v4i64_0 = shufflevector <12 x i64> %v4i64, <12 x i64> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>171; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v4i64_1 = shufflevector <12 x i64> %v4i64, <12 x i64> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>172; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v4i64_2 = shufflevector <12 x i64> %v4i64, <12 x i64> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>173; CHECK-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:4 SizeLat:1 for: %v8i64 = load <24 x i64>, ptr %p, align 256174; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v8i64_0 = shufflevector <24 x i64> %v8i64, <24 x i64> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>175; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v8i64_1 = shufflevector <24 x i64> %v8i64, <24 x i64> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>176; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v8i64_2 = shufflevector <24 x i64> %v8i64, <24 x i64> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>177; CHECK-NEXT:  Cost Model: Found costs of RThru:64 CodeSize:1 Lat:4 SizeLat:1 for: %v16i64 = load <48 x i64>, ptr %p, align 512178; CHECK-NEXT:  Cost Model: Found costs of RThru:2048 CodeSize:1024 Lat:2048 SizeLat:2048 for: %v16i64_0 = shufflevector <48 x i64> %v16i64, <48 x i64> undef, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>179; CHECK-NEXT:  Cost Model: Found costs of RThru:2048 CodeSize:1024 Lat:2048 SizeLat:2048 for: %v16i64_1 = shufflevector <48 x i64> %v16i64, <48 x i64> undef, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>180; CHECK-NEXT:  Cost Model: Found costs of RThru:2048 CodeSize:1024 Lat:2048 SizeLat:2048 for: %v16i64_2 = shufflevector <48 x i64> %v16i64, <48 x i64> undef, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>181; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void182;183  %v2i8 = load <6 x i8>, ptr %p184  %v2i8_0 = shufflevector <6 x i8> %v2i8, <6 x i8> undef, <2 x i32> <i32 0, i32 3>185  %v2i8_1 = shufflevector <6 x i8> %v2i8, <6 x i8> undef, <2 x i32> <i32 1, i32 4>186  %v2i8_2 = shufflevector <6 x i8> %v2i8, <6 x i8> undef, <2 x i32> <i32 2, i32 5>187  %v4i8 = load <12 x i8>, ptr %p188  %v4i8_0 = shufflevector <12 x i8> %v4i8, <12 x i8> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>189  %v4i8_1 = shufflevector <12 x i8> %v4i8, <12 x i8> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>190  %v4i8_2 = shufflevector <12 x i8> %v4i8, <12 x i8> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>191  %v8i8 = load <24 x i8>, ptr %p192  %v8i8_0 = shufflevector <24 x i8> %v8i8, <24 x i8> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>193  %v8i8_1 = shufflevector <24 x i8> %v8i8, <24 x i8> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>194  %v8i8_2 = shufflevector <24 x i8> %v8i8, <24 x i8> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>195  %v16i8 = load <48 x i8>, ptr %p196  %v16i8_0 = shufflevector <48 x i8> %v16i8, <48 x i8> undef, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>197  %v16i8_1 = shufflevector <48 x i8> %v16i8, <48 x i8> undef, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>198  %v16i8_2 = shufflevector <48 x i8> %v16i8, <48 x i8> undef, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>199 200  %v2i16 = load <6 x i16>, ptr %p201  %v2i16_0 = shufflevector <6 x i16> %v2i16, <6 x i16> undef, <2 x i32> <i32 0, i32 3>202  %v2i16_1 = shufflevector <6 x i16> %v2i16, <6 x i16> undef, <2 x i32> <i32 1, i32 4>203  %v2i16_2 = shufflevector <6 x i16> %v2i16, <6 x i16> undef, <2 x i32> <i32 2, i32 5>204  %v4i16 = load <12 x i16>, ptr %p205  %v4i16_0 = shufflevector <12 x i16> %v4i16, <12 x i16> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>206  %v4i16_1 = shufflevector <12 x i16> %v4i16, <12 x i16> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>207  %v4i16_2 = shufflevector <12 x i16> %v4i16, <12 x i16> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>208  %v8i16 = load <24 x i16>, ptr %p209  %v8i16_0 = shufflevector <24 x i16> %v8i16, <24 x i16> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>210  %v8i16_1 = shufflevector <24 x i16> %v8i16, <24 x i16> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>211  %v8i16_2 = shufflevector <24 x i16> %v8i16, <24 x i16> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>212  %v16i16 = load <48 x i16>, ptr %p213  %v16i16_0 = shufflevector <48 x i16> %v16i16, <48 x i16> undef, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>214  %v16i16_1 = shufflevector <48 x i16> %v16i16, <48 x i16> undef, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>215  %v16i16_2 = shufflevector <48 x i16> %v16i16, <48 x i16> undef, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>216 217  %v2i32 = load <6 x i32>, ptr %p218  %v2i32_0 = shufflevector <6 x i32> %v2i32, <6 x i32> undef, <2 x i32> <i32 0, i32 3>219  %v2i32_1 = shufflevector <6 x i32> %v2i32, <6 x i32> undef, <2 x i32> <i32 1, i32 4>220  %v2i32_2 = shufflevector <6 x i32> %v2i32, <6 x i32> undef, <2 x i32> <i32 2, i32 5>221  %v4i32 = load <12 x i32>, ptr %p222  %v4i32_0 = shufflevector <12 x i32> %v4i32, <12 x i32> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>223  %v4i32_1 = shufflevector <12 x i32> %v4i32, <12 x i32> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>224  %v4i32_2 = shufflevector <12 x i32> %v4i32, <12 x i32> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>225  %v8i32 = load <24 x i32>, ptr %p226  %v8i32_0 = shufflevector <24 x i32> %v8i32, <24 x i32> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>227  %v8i32_1 = shufflevector <24 x i32> %v8i32, <24 x i32> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>228  %v8i32_2 = shufflevector <24 x i32> %v8i32, <24 x i32> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>229  %v16i32 = load <48 x i32>, ptr %p230  %v16i32_0 = shufflevector <48 x i32> %v16i32, <48 x i32> undef, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>231  %v16i32_1 = shufflevector <48 x i32> %v16i32, <48 x i32> undef, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>232  %v16i32_2 = shufflevector <48 x i32> %v16i32, <48 x i32> undef, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>233 234  %v2i64 = load <6 x i64>, ptr %p235  %v2i64_0 = shufflevector <6 x i64> %v2i64, <6 x i64> undef, <2 x i32> <i32 0, i32 3>236  %v2i64_1 = shufflevector <6 x i64> %v2i64, <6 x i64> undef, <2 x i32> <i32 1, i32 4>237  %v2i64_2 = shufflevector <6 x i64> %v2i64, <6 x i64> undef, <2 x i32> <i32 2, i32 5>238  %v4i64 = load <12 x i64>, ptr %p239  %v4i64_0 = shufflevector <12 x i64> %v4i64, <12 x i64> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>240  %v4i64_1 = shufflevector <12 x i64> %v4i64, <12 x i64> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>241  %v4i64_2 = shufflevector <12 x i64> %v4i64, <12 x i64> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>242  %v8i64 = load <24 x i64>, ptr %p243  %v8i64_0 = shufflevector <24 x i64> %v8i64, <24 x i64> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>244  %v8i64_1 = shufflevector <24 x i64> %v8i64, <24 x i64> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>245  %v8i64_2 = shufflevector <24 x i64> %v8i64, <24 x i64> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>246  %v16i64 = load <48 x i64>, ptr %p247  %v16i64_0 = shufflevector <48 x i64> %v16i64, <48 x i64> undef, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>248  %v16i64_1 = shufflevector <48 x i64> %v16i64, <48 x i64> undef, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>249  %v16i64_2 = shufflevector <48 x i64> %v16i64, <48 x i64> undef, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>250 251  ret void252}253 254define void @vld4(ptr %p) {255; CHECK-UF2-LABEL: 'vld4'256; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v2i8 = load <8 x i8>, ptr %p, align 8257; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i8_0 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 0, i32 4>258; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i8_1 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 1, i32 5>259; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i8_2 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 2, i32 6>260; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i8_3 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 3, i32 7>261; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v4i8 = load <16 x i8>, ptr %p, align 8262; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i8_0 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>263; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i8_1 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>264; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i8_2 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>265; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i8_3 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>266; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v8i8 = load <32 x i8>, ptr %p, align 32267; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i8_0 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>268; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i8_1 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>269; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i8_2 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>270; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i8_3 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>271; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v16i8 = load <64 x i8>, ptr %p, align 64272; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i8_0 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>273; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i8_1 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>274; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i8_2 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>275; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i8_3 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>276; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v2i16 = load <8 x i16>, ptr %p, align 8277; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i16_0 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 0, i32 4>278; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i16_1 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 1, i32 5>279; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i16_2 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 2, i32 6>280; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i16_3 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 3, i32 7>281; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v4i16 = load <16 x i16>, ptr %p, align 32282; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i16_0 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>283; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i16_1 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>284; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i16_2 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>285; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i16_3 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>286; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v8i16 = load <32 x i16>, ptr %p, align 64287; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i16_0 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>288; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i16_1 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>289; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i16_2 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>290; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i16_3 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>291; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v16i16 = load <64 x i16>, ptr %p, align 128292; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i16_0 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>293; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i16_1 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>294; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i16_2 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>295; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i16_3 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>296; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v2i32 = load <8 x i32>, ptr %p, align 32297; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i32_0 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 0, i32 4>298; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i32_1 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 1, i32 5>299; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i32_2 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 2, i32 6>300; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:160 CodeSize:80 Lat:160 SizeLat:160 for: %v2i32_3 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 3, i32 7>301; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v4i32 = load <16 x i32>, ptr %p, align 64302; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i32_0 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>303; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i32_1 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>304; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i32_2 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>305; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v4i32_3 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>306; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v8i32 = load <32 x i32>, ptr %p, align 128307; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i32_0 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>308; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i32_1 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>309; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i32_2 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>310; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v8i32_3 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>311; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:4 SizeLat:1 for: %v16i32 = load <64 x i32>, ptr %p, align 256312; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i32_0 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>313; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i32_1 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>314; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i32_2 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>315; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v16i32_3 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>316; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v2i64 = load <8 x i64>, ptr %p, align 64317; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v2i64_0 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 0, i32 4>318; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v2i64_1 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 1, i32 5>319; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v2i64_2 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 2, i32 6>320; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v2i64_3 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 3, i32 7>321; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v4i64 = load <16 x i64>, ptr %p, align 128322; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v4i64_0 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>323; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v4i64_1 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>324; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v4i64_2 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>325; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v4i64_3 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>326; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:4 SizeLat:1 for: %v8i64 = load <32 x i64>, ptr %p, align 256327; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v8i64_0 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>328; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v8i64_1 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>329; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v8i64_2 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>330; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v8i64_3 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>331; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:64 CodeSize:1 Lat:4 SizeLat:1 for: %v16i64 = load <64 x i64>, ptr %p, align 512332; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2560 CodeSize:1280 Lat:2560 SizeLat:2560 for: %v16i64_0 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>333; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2560 CodeSize:1280 Lat:2560 SizeLat:2560 for: %v16i64_1 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>334; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2560 CodeSize:1280 Lat:2560 SizeLat:2560 for: %v16i64_2 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>335; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2560 CodeSize:1280 Lat:2560 SizeLat:2560 for: %v16i64_3 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>336; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void337;338; CHECK-UF4-LABEL: 'vld4'339; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v2i8 = load <8 x i8>, ptr %p, align 8340; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i8_0 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 0, i32 4>341; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i8_1 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 1, i32 5>342; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i8_2 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 2, i32 6>343; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i8_3 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 3, i32 7>344; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v4i8 = load <16 x i8>, ptr %p, align 8345; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i8_0 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>346; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i8_1 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>347; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i8_2 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>348; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i8_3 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>349; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v8i8 = load <32 x i8>, ptr %p, align 32350; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v8i8_0 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>351; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v8i8_1 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>352; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v8i8_2 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>353; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v8i8_3 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>354; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v16i8 = load <64 x i8>, ptr %p, align 64355; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:258 CodeSize:129 Lat:258 SizeLat:258 for: %v16i8_0 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>356; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:258 CodeSize:129 Lat:258 SizeLat:258 for: %v16i8_1 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>357; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:258 CodeSize:129 Lat:258 SizeLat:258 for: %v16i8_2 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>358; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:258 CodeSize:129 Lat:258 SizeLat:258 for: %v16i8_3 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>359; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:1 for: %v2i16 = load <8 x i16>, ptr %p, align 8360; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i16_0 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 0, i32 4>361; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i16_1 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 1, i32 5>362; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i16_2 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 2, i32 6>363; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i16_3 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 3, i32 7>364; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v4i16 = load <16 x i16>, ptr %p, align 32365; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i16_0 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>366; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i16_1 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>367; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i16_2 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>368; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i16_3 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>369; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v8i16 = load <32 x i16>, ptr %p, align 64370; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v8i16_0 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>371; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v8i16_1 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>372; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v8i16_2 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>373; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:130 CodeSize:65 Lat:130 SizeLat:130 for: %v8i16_3 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>374; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v16i16 = load <64 x i16>, ptr %p, align 128375; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:260 CodeSize:130 Lat:260 SizeLat:260 for: %v16i16_0 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>376; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:260 CodeSize:130 Lat:260 SizeLat:260 for: %v16i16_1 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>377; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:260 CodeSize:130 Lat:260 SizeLat:260 for: %v16i16_2 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>378; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:260 CodeSize:130 Lat:260 SizeLat:260 for: %v16i16_3 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>379; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:1 for: %v2i32 = load <8 x i32>, ptr %p, align 32380; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i32_0 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 0, i32 4>381; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i32_1 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 1, i32 5>382; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i32_2 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 2, i32 6>383; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:34 CodeSize:17 Lat:34 SizeLat:34 for: %v2i32_3 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 3, i32 7>384; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v4i32 = load <16 x i32>, ptr %p, align 64385; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i32_0 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>386; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i32_1 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>387; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i32_2 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>388; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:33 Lat:66 SizeLat:66 for: %v4i32_3 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>389; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v8i32 = load <32 x i32>, ptr %p, align 128390; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:132 CodeSize:66 Lat:132 SizeLat:132 for: %v8i32_0 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>391; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:132 CodeSize:66 Lat:132 SizeLat:132 for: %v8i32_1 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>392; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:132 CodeSize:66 Lat:132 SizeLat:132 for: %v8i32_2 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>393; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:132 CodeSize:66 Lat:132 SizeLat:132 for: %v8i32_3 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>394; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:4 SizeLat:1 for: %v16i32 = load <64 x i32>, ptr %p, align 256395; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:264 CodeSize:132 Lat:264 SizeLat:264 for: %v16i32_0 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>396; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:264 CodeSize:132 Lat:264 SizeLat:264 for: %v16i32_1 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>397; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:264 CodeSize:132 Lat:264 SizeLat:264 for: %v16i32_2 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>398; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:264 CodeSize:132 Lat:264 SizeLat:264 for: %v16i32_3 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>399; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:4 SizeLat:1 for: %v2i64 = load <8 x i64>, ptr %p, align 64400; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v2i64_0 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 0, i32 4>401; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v2i64_1 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 1, i32 5>402; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v2i64_2 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 2, i32 6>403; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:320 CodeSize:160 Lat:320 SizeLat:320 for: %v2i64_3 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 3, i32 7>404; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:4 SizeLat:1 for: %v4i64 = load <16 x i64>, ptr %p, align 128405; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v4i64_0 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>406; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v4i64_1 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>407; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v4i64_2 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>408; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:640 CodeSize:320 Lat:640 SizeLat:640 for: %v4i64_3 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>409; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:4 SizeLat:1 for: %v8i64 = load <32 x i64>, ptr %p, align 256410; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v8i64_0 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>411; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v8i64_1 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>412; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v8i64_2 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>413; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:1280 CodeSize:640 Lat:1280 SizeLat:1280 for: %v8i64_3 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>414; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:64 CodeSize:1 Lat:4 SizeLat:1 for: %v16i64 = load <64 x i64>, ptr %p, align 512415; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2560 CodeSize:1280 Lat:2560 SizeLat:2560 for: %v16i64_0 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>416; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2560 CodeSize:1280 Lat:2560 SizeLat:2560 for: %v16i64_1 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>417; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2560 CodeSize:1280 Lat:2560 SizeLat:2560 for: %v16i64_2 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>418; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2560 CodeSize:1280 Lat:2560 SizeLat:2560 for: %v16i64_3 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>419; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void420;421  %v2i8 = load <8 x i8>, ptr %p422  %v2i8_0 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 0, i32 4>423  %v2i8_1 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 1, i32 5>424  %v2i8_2 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 2, i32 6>425  %v2i8_3 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 3, i32 7>426  %v4i8 = load <16 x i8>, ptr %p427  %v4i8_0 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>428  %v4i8_1 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>429  %v4i8_2 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>430  %v4i8_3 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>431  %v8i8 = load <32 x i8>, ptr %p432  %v8i8_0 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>433  %v8i8_1 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>434  %v8i8_2 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>435  %v8i8_3 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>436  %v16i8 = load <64 x i8>, ptr %p437  %v16i8_0 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>438  %v16i8_1 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>439  %v16i8_2 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>440  %v16i8_3 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>441 442  %v2i16 = load <8 x i16>, ptr %p443  %v2i16_0 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 0, i32 4>444  %v2i16_1 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 1, i32 5>445  %v2i16_2 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 2, i32 6>446  %v2i16_3 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 3, i32 7>447  %v4i16 = load <16 x i16>, ptr %p448  %v4i16_0 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>449  %v4i16_1 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>450  %v4i16_2 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>451  %v4i16_3 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>452  %v8i16 = load <32 x i16>, ptr %p453  %v8i16_0 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>454  %v8i16_1 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>455  %v8i16_2 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>456  %v8i16_3 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>457  %v16i16 = load <64 x i16>, ptr %p458  %v16i16_0 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>459  %v16i16_1 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>460  %v16i16_2 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>461  %v16i16_3 = shufflevector <64 x i16> %v16i16, <64 x i16> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>462 463  %v2i32 = load <8 x i32>, ptr %p464  %v2i32_0 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 0, i32 4>465  %v2i32_1 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 1, i32 5>466  %v2i32_2 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 2, i32 6>467  %v2i32_3 = shufflevector <8 x i32> %v2i32, <8 x i32> undef, <2 x i32> <i32 3, i32 7>468  %v4i32 = load <16 x i32>, ptr %p469  %v4i32_0 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>470  %v4i32_1 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>471  %v4i32_2 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>472  %v4i32_3 = shufflevector <16 x i32> %v4i32, <16 x i32> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>473  %v8i32 = load <32 x i32>, ptr %p474  %v8i32_0 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>475  %v8i32_1 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>476  %v8i32_2 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>477  %v8i32_3 = shufflevector <32 x i32> %v8i32, <32 x i32> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>478  %v16i32 = load <64 x i32>, ptr %p479  %v16i32_0 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>480  %v16i32_1 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>481  %v16i32_2 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>482  %v16i32_3 = shufflevector <64 x i32> %v16i32, <64 x i32> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>483 484  %v2i64 = load <8 x i64>, ptr %p485  %v2i64_0 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 0, i32 4>486  %v2i64_1 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 1, i32 5>487  %v2i64_2 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 2, i32 6>488  %v2i64_3 = shufflevector <8 x i64> %v2i64, <8 x i64> undef, <2 x i32> <i32 3, i32 7>489  %v4i64 = load <16 x i64>, ptr %p490  %v4i64_0 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>491  %v4i64_1 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>492  %v4i64_2 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>493  %v4i64_3 = shufflevector <16 x i64> %v4i64, <16 x i64> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>494  %v8i64 = load <32 x i64>, ptr %p495  %v8i64_0 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>496  %v8i64_1 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>497  %v8i64_2 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>498  %v8i64_3 = shufflevector <32 x i64> %v8i64, <32 x i64> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>499  %v16i64 = load <64 x i64>, ptr %p500  %v16i64_0 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>501  %v16i64_1 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>502  %v16i64_2 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>503  %v16i64_3 = shufflevector <64 x i64> %v16i64, <64 x i64> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>504 505  ret void506}507 508define void @vst2(ptr %p) {509; CHECK-LABEL: 'vst2'510; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v4i8 = shufflevector <2 x i8> undef, <2 x i8> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>511; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <4 x i8> %v4i8, ptr %p, align 4512; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v8i8 = shufflevector <4 x i8> undef, <4 x i8> undef, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>513; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i8> %v8i8, ptr %p, align 8514; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v16i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>515; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i8> %v16i8, ptr %p, align 8516; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %v32i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>517; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i8> %v32i8, ptr %p, align 32518; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v4i16 = shufflevector <2 x i16> undef, <2 x i16> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>519; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <4 x i16> %v4i16, ptr %p, align 8520; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v8i16 = shufflevector <4 x i16> undef, <4 x i16> undef, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>521; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i16> %v8i16, ptr %p, align 8522; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %v16i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>523; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i16> %v16i16, ptr %p, align 32524; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:4 Lat:8 SizeLat:8 for: %v32i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>525; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i16> %v32i16, ptr %p, align 64526; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v4i32 = shufflevector <2 x i32> undef, <2 x i32> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>527; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <4 x i32> %v4i32, ptr %p, align 8528; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %v8i32 = shufflevector <4 x i32> undef, <4 x i32> undef, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>529; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i32> %v8i32, ptr %p, align 32530; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:4 Lat:8 SizeLat:8 for: %v16i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>531; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i32> %v16i32, ptr %p, align 64532; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:8 Lat:16 SizeLat:16 for: %v32i32 = shufflevector <16 x i32> undef, <16 x i32> undef, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>533; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i32> %v32i32, ptr %p, align 128534; CHECK-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v4i64 = shufflevector <2 x i64> undef, <2 x i64> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>535; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <4 x i64> %v4i64, ptr %p, align 32536; CHECK-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v8i64 = shufflevector <4 x i64> undef, <4 x i64> undef, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>537; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i64> %v8i64, ptr %p, align 64538; CHECK-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v16i64 = shufflevector <8 x i64> undef, <8 x i64> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>539; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i64> %v16i64, ptr %p, align 128540; CHECK-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v32i64 = shufflevector <16 x i64> undef, <16 x i64> undef, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>541; CHECK-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i64> %v32i64, ptr %p, align 256542; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void543;544  %v4i8 = shufflevector <2 x i8> undef, <2 x i8> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>545  store <4 x i8> %v4i8, ptr %p546  %v8i8 = shufflevector <4 x i8> undef, <4 x i8> undef, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>547  store <8 x i8> %v8i8, ptr %p548  %v16i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>549  store <16 x i8> %v16i8, ptr %p550  %v32i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>551  store <32 x i8> %v32i8, ptr %p552 553  %v4i16 = shufflevector <2 x i16> undef, <2 x i16> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>554  store <4 x i16> %v4i16, ptr %p555  %v8i16 = shufflevector <4 x i16> undef, <4 x i16> undef, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>556  store <8 x i16> %v8i16, ptr %p557  %v16i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>558  store <16 x i16> %v16i16, ptr %p559  %v32i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>560  store <32 x i16> %v32i16, ptr %p561 562  %v4i32 = shufflevector <2 x i32> undef, <2 x i32> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>563  store <4 x i32> %v4i32, ptr %p564  %v8i32 = shufflevector <4 x i32> undef, <4 x i32> undef, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>565  store <8 x i32> %v8i32, ptr %p566  %v16i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>567  store <16 x i32> %v16i32, ptr %p568  %v32i32 = shufflevector <16 x i32> undef, <16 x i32> undef, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>569  store <32 x i32> %v32i32, ptr %p570 571  %v4i64 = shufflevector <2 x i64> undef, <2 x i64> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>572  store <4 x i64> %v4i64, ptr %p573  %v8i64 = shufflevector <4 x i64> undef, <4 x i64> undef, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>574  store <8 x i64> %v8i64, ptr %p575  %v16i64 = shufflevector <8 x i64> undef, <8 x i64> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>576  store <16 x i64> %v16i64, ptr %p577  %v32i64 = shufflevector <16 x i64> undef, <16 x i64> undef, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>578  store <32 x i64> %v32i64, ptr %p579 580  ret void581}582 583 584define void @vst3(ptr %p) {585; CHECK-LABEL: 'vst3'586; CHECK-NEXT:  Cost Model: Found costs of RThru:96 CodeSize:48 Lat:96 SizeLat:96 for: %v8i8 = shufflevector <4 x i8> undef, <4 x i8> undef, <6 x i32> <i32 0, i32 2, i32 4, i32 1, i32 3, i32 5>587; CHECK-NEXT:  Cost Model: Found costs of RThru:50 CodeSize:1 Lat:1 SizeLat:1 for: store <6 x i8> %v8i8, ptr %p, align 8588; CHECK-NEXT:  Cost Model: Found costs of RThru:192 CodeSize:96 Lat:192 SizeLat:192 for: %v16i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>589; CHECK-NEXT:  Cost Model: Found costs of RThru:98 CodeSize:1 Lat:1 SizeLat:1 for: store <12 x i8> %v16i8, ptr %p, align 16590; CHECK-NEXT:  Cost Model: Found costs of RThru:384 CodeSize:192 Lat:384 SizeLat:384 for: %v32i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>591; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <24 x i8> %v32i8, ptr %p, align 32592; CHECK-NEXT:  Cost Model: Found costs of RThru:768 CodeSize:384 Lat:768 SizeLat:768 for: %v64i8 = shufflevector <32 x i8> undef, <32 x i8> undef, <48 x i32> <i32 0, i32 16, i32 32, i32 1, i32 17, i32 33, i32 2, i32 18, i32 34, i32 3, i32 19, i32 35, i32 4, i32 20, i32 36, i32 5, i32 21, i32 37, i32 6, i32 22, i32 38, i32 7, i32 23, i32 39, i32 8, i32 24, i32 40, i32 9, i32 25, i32 41, i32 10, i32 26, i32 42, i32 11, i32 27, i32 43, i32 12, i32 28, i32 44, i32 13, i32 29, i32 45, i32 14, i32 30, i32 46, i32 15, i32 31, i32 47>593; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <48 x i8> %v64i8, ptr %p, align 64594; CHECK-NEXT:  Cost Model: Found costs of RThru:96 CodeSize:48 Lat:96 SizeLat:96 for: %v8i16 = shufflevector <4 x i16> undef, <4 x i16> undef, <6 x i32> <i32 0, i32 2, i32 4, i32 1, i32 3, i32 5>595; CHECK-NEXT:  Cost Model: Found costs of RThru:50 CodeSize:1 Lat:1 SizeLat:1 for: store <6 x i16> %v8i16, ptr %p, align 16596; CHECK-NEXT:  Cost Model: Found costs of RThru:192 CodeSize:96 Lat:192 SizeLat:192 for: %v16i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>597; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <12 x i16> %v16i16, ptr %p, align 32598; CHECK-NEXT:  Cost Model: Found costs of RThru:384 CodeSize:192 Lat:384 SizeLat:384 for: %v32i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>599; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <24 x i16> %v32i16, ptr %p, align 64600; CHECK-NEXT:  Cost Model: Found costs of RThru:768 CodeSize:384 Lat:768 SizeLat:768 for: %v64i16 = shufflevector <32 x i16> undef, <32 x i16> undef, <48 x i32> <i32 0, i32 16, i32 32, i32 1, i32 17, i32 33, i32 2, i32 18, i32 34, i32 3, i32 19, i32 35, i32 4, i32 20, i32 36, i32 5, i32 21, i32 37, i32 6, i32 22, i32 38, i32 7, i32 23, i32 39, i32 8, i32 24, i32 40, i32 9, i32 25, i32 41, i32 10, i32 26, i32 42, i32 11, i32 27, i32 43, i32 12, i32 28, i32 44, i32 13, i32 29, i32 45, i32 14, i32 30, i32 46, i32 15, i32 31, i32 47>601; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <48 x i16> %v64i16, ptr %p, align 128602; CHECK-NEXT:  Cost Model: Found costs of RThru:96 CodeSize:48 Lat:96 SizeLat:96 for: %v8i32 = shufflevector <4 x i32> undef, <4 x i32> undef, <6 x i32> <i32 0, i32 2, i32 4, i32 1, i32 3, i32 5>603; CHECK-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <6 x i32> %v8i32, ptr %p, align 32604; CHECK-NEXT:  Cost Model: Found costs of RThru:192 CodeSize:96 Lat:192 SizeLat:192 for: %v16i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>605; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <12 x i32> %v16i32, ptr %p, align 64606; CHECK-NEXT:  Cost Model: Found costs of RThru:384 CodeSize:192 Lat:384 SizeLat:384 for: %v32i32 = shufflevector <16 x i32> undef, <16 x i32> undef, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>607; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <24 x i32> %v32i32, ptr %p, align 128608; CHECK-NEXT:  Cost Model: Found costs of RThru:768 CodeSize:384 Lat:768 SizeLat:768 for: %v64i32 = shufflevector <32 x i32> undef, <32 x i32> undef, <48 x i32> <i32 0, i32 16, i32 32, i32 1, i32 17, i32 33, i32 2, i32 18, i32 34, i32 3, i32 19, i32 35, i32 4, i32 20, i32 36, i32 5, i32 21, i32 37, i32 6, i32 22, i32 38, i32 7, i32 23, i32 39, i32 8, i32 24, i32 40, i32 9, i32 25, i32 41, i32 10, i32 26, i32 42, i32 11, i32 27, i32 43, i32 12, i32 28, i32 44, i32 13, i32 29, i32 45, i32 14, i32 30, i32 46, i32 15, i32 31, i32 47>609; CHECK-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:1 SizeLat:1 for: store <48 x i32> %v64i32, ptr %p, align 256610; CHECK-NEXT:  Cost Model: Found costs of RThru:192 CodeSize:96 Lat:192 SizeLat:192 for: %v8i64 = shufflevector <4 x i64> undef, <4 x i64> undef, <6 x i32> <i32 0, i32 2, i32 4, i32 1, i32 3, i32 5>611; CHECK-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <6 x i64> %v8i64, ptr %p, align 64612; CHECK-NEXT:  Cost Model: Found costs of RThru:384 CodeSize:192 Lat:384 SizeLat:384 for: %v16i64 = shufflevector <8 x i64> undef, <8 x i64> undef, <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>613; CHECK-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <12 x i64> %v16i64, ptr %p, align 128614; CHECK-NEXT:  Cost Model: Found costs of RThru:768 CodeSize:384 Lat:768 SizeLat:768 for: %v32i64 = shufflevector <16 x i64> undef, <16 x i64> undef, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>615; CHECK-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:1 SizeLat:1 for: store <24 x i64> %v32i64, ptr %p, align 256616; CHECK-NEXT:  Cost Model: Found costs of RThru:1536 CodeSize:768 Lat:1536 SizeLat:1536 for: %v64i64 = shufflevector <32 x i64> undef, <32 x i64> undef, <48 x i32> <i32 0, i32 16, i32 32, i32 1, i32 17, i32 33, i32 2, i32 18, i32 34, i32 3, i32 19, i32 35, i32 4, i32 20, i32 36, i32 5, i32 21, i32 37, i32 6, i32 22, i32 38, i32 7, i32 23, i32 39, i32 8, i32 24, i32 40, i32 9, i32 25, i32 41, i32 10, i32 26, i32 42, i32 11, i32 27, i32 43, i32 12, i32 28, i32 44, i32 13, i32 29, i32 45, i32 14, i32 30, i32 46, i32 15, i32 31, i32 47>617; CHECK-NEXT:  Cost Model: Found costs of RThru:64 CodeSize:1 Lat:1 SizeLat:1 for: store <48 x i64> %v64i64, ptr %p, align 512618; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void619;620  %v8i8 = shufflevector <4 x i8> undef, <4 x i8> undef, <6 x i32> <i32 0, i32 2, i32 4, i32 1, i32 3, i32 5>621  store <6 x i8> %v8i8, ptr %p622  %v16i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>623  store <12 x i8> %v16i8, ptr %p624  %v32i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>625  store <24 x i8> %v32i8, ptr %p626  %v64i8 = shufflevector <32 x i8> undef, <32 x i8> undef, <48 x i32> <i32 0, i32 16, i32 32, i32 1, i32 17, i32 33, i32 2, i32 18, i32 34, i32 3, i32 19, i32 35, i32 4, i32 20, i32 36, i32 5, i32 21, i32 37, i32 6, i32 22, i32 38, i32 7, i32 23, i32 39, i32 8, i32 24, i32 40, i32 9, i32 25, i32 41, i32 10, i32 26, i32 42, i32 11, i32 27, i32 43, i32 12, i32 28, i32 44, i32 13, i32 29, i32 45, i32 14, i32 30, i32 46, i32 15, i32 31, i32 47>627  store <48 x i8> %v64i8, ptr %p628 629  %v8i16 = shufflevector <4 x i16> undef, <4 x i16> undef, <6 x i32> <i32 0, i32 2, i32 4, i32 1, i32 3, i32 5>630  store <6 x i16> %v8i16, ptr %p631  %v16i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>632  store <12 x i16> %v16i16, ptr %p633  %v32i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>634  store <24 x i16> %v32i16, ptr %p635  %v64i16 = shufflevector <32 x i16> undef, <32 x i16> undef, <48 x i32> <i32 0, i32 16, i32 32, i32 1, i32 17, i32 33, i32 2, i32 18, i32 34, i32 3, i32 19, i32 35, i32 4, i32 20, i32 36, i32 5, i32 21, i32 37, i32 6, i32 22, i32 38, i32 7, i32 23, i32 39, i32 8, i32 24, i32 40, i32 9, i32 25, i32 41, i32 10, i32 26, i32 42, i32 11, i32 27, i32 43, i32 12, i32 28, i32 44, i32 13, i32 29, i32 45, i32 14, i32 30, i32 46, i32 15, i32 31, i32 47>636  store <48 x i16> %v64i16, ptr %p637 638  %v8i32 = shufflevector <4 x i32> undef, <4 x i32> undef, <6 x i32> <i32 0, i32 2, i32 4, i32 1, i32 3, i32 5>639  store <6 x i32> %v8i32, ptr %p640  %v16i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>641  store <12 x i32> %v16i32, ptr %p642  %v32i32 = shufflevector <16 x i32> undef, <16 x i32> undef, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>643  store <24 x i32> %v32i32, ptr %p644  %v64i32 = shufflevector <32 x i32> undef, <32 x i32> undef, <48 x i32> <i32 0, i32 16, i32 32, i32 1, i32 17, i32 33, i32 2, i32 18, i32 34, i32 3, i32 19, i32 35, i32 4, i32 20, i32 36, i32 5, i32 21, i32 37, i32 6, i32 22, i32 38, i32 7, i32 23, i32 39, i32 8, i32 24, i32 40, i32 9, i32 25, i32 41, i32 10, i32 26, i32 42, i32 11, i32 27, i32 43, i32 12, i32 28, i32 44, i32 13, i32 29, i32 45, i32 14, i32 30, i32 46, i32 15, i32 31, i32 47>645  store <48 x i32> %v64i32, ptr %p646 647  %v8i64 = shufflevector <4 x i64> undef, <4 x i64> undef, <6 x i32> <i32 0, i32 2, i32 4, i32 1, i32 3, i32 5>648  store <6 x i64> %v8i64, ptr %p649  %v16i64 = shufflevector <8 x i64> undef, <8 x i64> undef, <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>650  store <12 x i64> %v16i64, ptr %p651  %v32i64 = shufflevector <16 x i64> undef, <16 x i64> undef, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>652  store <24 x i64> %v32i64, ptr %p653  %v64i64 = shufflevector <32 x i64> undef, <32 x i64> undef, <48 x i32> <i32 0, i32 16, i32 32, i32 1, i32 17, i32 33, i32 2, i32 18, i32 34, i32 3, i32 19, i32 35, i32 4, i32 20, i32 36, i32 5, i32 21, i32 37, i32 6, i32 22, i32 38, i32 7, i32 23, i32 39, i32 8, i32 24, i32 40, i32 9, i32 25, i32 41, i32 10, i32 26, i32 42, i32 11, i32 27, i32 43, i32 12, i32 28, i32 44, i32 13, i32 29, i32 45, i32 14, i32 30, i32 46, i32 15, i32 31, i32 47>654  store <48 x i64> %v64i64, ptr %p655 656  ret void657}658 659 660define void @vst4(ptr %p) {661; CHECK-UF2-LABEL: 'vst4'662; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v8i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>663; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i8> %v8i8, ptr %p, align 8664; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v16i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>665; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i8> %v16i8, ptr %p, align 8666; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v32i8 = shufflevector <32 x i8> undef, <32 x i8> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>667; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i8> %v32i8, ptr %p, align 32668; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v64i8 = shufflevector <64 x i8> undef, <64 x i8> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>669; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <64 x i8> %v64i8, ptr %p, align 64670; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v8i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>671; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i16> %v8i16, ptr %p, align 8672; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v16i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>673; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i16> %v16i16, ptr %p, align 32674; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v32i16 = shufflevector <32 x i16> undef, <32 x i16> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>675; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i16> %v32i16, ptr %p, align 64676; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v64i16 = shufflevector <64 x i16> undef, <64 x i16> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>677; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <64 x i16> %v64i16, ptr %p, align 128678; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:128 CodeSize:64 Lat:128 SizeLat:128 for: %v8i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>679; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i32> %v8i32, ptr %p, align 32680; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v16i32 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>681; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i32> %v16i32, ptr %p, align 64682; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v32i32 = shufflevector <32 x i32> undef, <32 x i32> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>683; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i32> %v32i32, ptr %p, align 128684; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v64i32 = shufflevector <64 x i32> undef, <64 x i32> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>685; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:1 SizeLat:1 for: store <64 x i32> %v64i32, ptr %p, align 256686; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v8i64 = shufflevector <8 x i64> undef, <8 x i64> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>687; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i64> %v8i64, ptr %p, align 64688; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v16i64 = shufflevector <16 x i64> undef, <16 x i64> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>689; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i64> %v16i64, ptr %p, align 128690; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v32i64 = shufflevector <32 x i64> undef, <32 x i64> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>691; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i64> %v32i64, ptr %p, align 256692; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:2048 CodeSize:1024 Lat:2048 SizeLat:2048 for: %v64i64 = shufflevector <64 x i64> undef, <64 x i64> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>693; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:64 CodeSize:1 Lat:1 SizeLat:1 for: store <64 x i64> %v64i64, ptr %p, align 512694; CHECK-UF2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void695;696; CHECK-UF4-LABEL: 'vst4'697; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v8i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>698; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i8> %v8i8, ptr %p, align 8699; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v16i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>700; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i8> %v16i8, ptr %p, align 8701; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %v32i8 = shufflevector <32 x i8> undef, <32 x i8> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>702; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i8> %v32i8, ptr %p, align 32703; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:4 Lat:8 SizeLat:8 for: %v64i8 = shufflevector <64 x i8> undef, <64 x i8> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>704; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <64 x i8> %v64i8, ptr %p, align 64705; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %v8i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>706; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i16> %v8i16, ptr %p, align 8707; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %v16i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>708; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i16> %v16i16, ptr %p, align 32709; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:4 Lat:8 SizeLat:8 for: %v32i16 = shufflevector <32 x i16> undef, <32 x i16> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>710; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i16> %v32i16, ptr %p, align 64711; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:8 Lat:16 SizeLat:16 for: %v64i16 = shufflevector <64 x i16> undef, <64 x i16> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>712; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <64 x i16> %v64i16, ptr %p, align 128713; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %v8i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>714; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i32> %v8i32, ptr %p, align 32715; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:4 Lat:8 SizeLat:8 for: %v16i32 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>716; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i32> %v16i32, ptr %p, align 64717; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:8 Lat:16 SizeLat:16 for: %v32i32 = shufflevector <32 x i32> undef, <32 x i32> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>718; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i32> %v32i32, ptr %p, align 128719; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:16 Lat:32 SizeLat:32 for: %v64i32 = shufflevector <64 x i32> undef, <64 x i32> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>720; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:1 SizeLat:1 for: store <64 x i32> %v64i32, ptr %p, align 256721; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:256 CodeSize:128 Lat:256 SizeLat:256 for: %v8i64 = shufflevector <8 x i64> undef, <8 x i64> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>722; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:1 Lat:1 SizeLat:1 for: store <8 x i64> %v8i64, ptr %p, align 64723; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:512 CodeSize:256 Lat:512 SizeLat:512 for: %v16i64 = shufflevector <16 x i64> undef, <16 x i64> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>724; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:16 CodeSize:1 Lat:1 SizeLat:1 for: store <16 x i64> %v16i64, ptr %p, align 128725; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:1024 CodeSize:512 Lat:1024 SizeLat:1024 for: %v32i64 = shufflevector <32 x i64> undef, <32 x i64> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>726; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:32 CodeSize:1 Lat:1 SizeLat:1 for: store <32 x i64> %v32i64, ptr %p, align 256727; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:2048 CodeSize:1024 Lat:2048 SizeLat:2048 for: %v64i64 = shufflevector <64 x i64> undef, <64 x i64> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>728; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:64 CodeSize:1 Lat:1 SizeLat:1 for: store <64 x i64> %v64i64, ptr %p, align 512729; CHECK-UF4-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void730;731  %v8i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>732  store <8 x i8> %v8i8, ptr %p733  %v16i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>734  store <16 x i8> %v16i8, ptr %p735  %v32i8 = shufflevector <32 x i8> undef, <32 x i8> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>736  store <32 x i8> %v32i8, ptr %p737  %v64i8 = shufflevector <64 x i8> undef, <64 x i8> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>738  store <64 x i8> %v64i8, ptr %p739 740  %v8i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>741  store <8 x i16> %v8i16, ptr %p742  %v16i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>743  store <16 x i16> %v16i16, ptr %p744  %v32i16 = shufflevector <32 x i16> undef, <32 x i16> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>745  store <32 x i16> %v32i16, ptr %p746  %v64i16 = shufflevector <64 x i16> undef, <64 x i16> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>747  store <64 x i16> %v64i16, ptr %p748 749  %v8i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>750  store <8 x i32> %v8i32, ptr %p751  %v16i32 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>752  store <16 x i32> %v16i32, ptr %p753  %v32i32 = shufflevector <32 x i32> undef, <32 x i32> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>754  store <32 x i32> %v32i32, ptr %p755  %v64i32 = shufflevector <64 x i32> undef, <64 x i32> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>756  store <64 x i32> %v64i32, ptr %p757 758  %v8i64 = shufflevector <8 x i64> undef, <8 x i64> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>759  store <8 x i64> %v8i64, ptr %p760  %v16i64 = shufflevector <16 x i64> undef, <16 x i64> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>761  store <16 x i64> %v16i64, ptr %p762  %v32i64 = shufflevector <32 x i64> undef, <32 x i64> undef, <32 x i32> <i32 0, i32 8, i32 16, i32 24, i32 1, i32 9, i32 17, i32 25, i32 2, i32 10, i32 18, i32 26, i32 3, i32 11, i32 19, i32 27, i32 4, i32 12, i32 20, i32 28, i32 5, i32 13, i32 21, i32 29, i32 6, i32 14, i32 22, i32 30, i32 7, i32 15, i32 23, i32 31>763  store <32 x i64> %v32i64, ptr %p764  %v64i64 = shufflevector <64 x i64> undef, <64 x i64> undef, <64 x i32> <i32 0, i32 16, i32 32, i32 48, i32 1, i32 17, i32 33, i32 49, i32 2, i32 18, i32 34, i32 50, i32 3, i32 19, i32 35, i32 51, i32 4, i32 20, i32 36, i32 52, i32 5, i32 21, i32 37, i32 53, i32 6, i32 22, i32 38, i32 54, i32 7, i32 23, i32 39, i32 55, i32 8, i32 24, i32 40, i32 56, i32 9, i32 25, i32 41, i32 57, i32 10, i32 26, i32 42, i32 58, i32 11, i32 27, i32 43, i32 59, i32 12, i32 28, i32 44, i32 60, i32 13, i32 29, i32 45, i32 61, i32 14, i32 30, i32 46, i32 62, i32 15, i32 31, i32 47, i32 63>765  store <64 x i64> %v64i64, ptr %p766 767  ret void768}769