1 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s 2 3 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" 4 target triple = "x86_64-apple-macosx10.8.0" 5 6 define i32 @add(i32 %arg) { 7 8 ; -- Same size registeres -- 9 ;CHECK: cost of 1 {{.*}} zext 10 %A = zext <4 x i1> undef to <4 x i32> 11 ;CHECK: cost of 2 {{.*}} sext 12 %B = sext <4 x i1> undef to <4 x i32> 13 ;CHECK: cost of 0 {{.*}} trunc 14 %C = trunc <4 x i32> undef to <4 x i1> 15 16 ; -- Different size registers -- 17 ;CHECK-NOT: cost of 1 {{.*}} zext 18 %D = zext <8 x i1> undef to <8 x i32> 19 ;CHECK-NOT: cost of 2 {{.*}} sext 20 %E = sext <8 x i1> undef to <8 x i32> 21 ;CHECK-NOT: cost of 2 {{.*}} trunc 22 %F = trunc <8 x i32> undef to <8 x i1> 23 24 ; -- scalars -- 25 26 ;CHECK: cost of 1 {{.*}} zext 27 %G = zext i1 undef to i32 28 ;CHECK: cost of 0 {{.*}} trunc 29 %H = trunc i32 undef to i1 30 31 ;CHECK: cost of 0 {{.*}} ret 32 ret i32 undef 33 } 34 35 define i32 @zext_sext(<8 x i1> %in) { 36 ;CHECK: cost of 6 {{.*}} zext 37 %Z = zext <8 x i1> %in to <8 x i32> 38 ;CHECK: cost of 9 {{.*}} sext 39 %S = sext <8 x i1> %in to <8 x i32> 40 41 ;CHECK: cost of 1 {{.*}} sext 42 %A = sext <8 x i16> undef to <8 x i32> 43 ;CHECK: cost of 1 {{.*}} zext 44 %B = zext <8 x i16> undef to <8 x i32> 45 ;CHECK: cost of 1 {{.*}} sext 46 %C = sext <4 x i32> undef to <4 x i64> 47 ;CHECK: cost of 6 {{.*}} sext 48 %C1 = sext <4 x i8> undef to <4 x i64> 49 ;CHECK: cost of 6 {{.*}} sext 50 %C2 = sext <4 x i16> undef to <4 x i64> 51 52 ;CHECK: cost of 1 {{.*}} zext 53 %D = zext <4 x i32> undef to <4 x i64> 54 ;CHECK: cost of 1 {{.*}} trunc 55 56 %E = trunc <4 x i64> undef to <4 x i32> 57 ;CHECK: cost of 1 {{.*}} trunc 58 %F = trunc <8 x i32> undef to <8 x i16> 59 60 ;CHECK: cost of 3 {{.*}} trunc 61 %G = trunc <8 x i64> undef to <8 x i32> 62 63 ret i32 undef 64 } 65 66 define i32 @masks8(<8 x i1> %in) { 67 ;CHECK: cost of 6 {{.*}} zext 68 %Z = zext <8 x i1> %in to <8 x i32> 69 ;CHECK: cost of 9 {{.*}} sext 70 %S = sext <8 x i1> %in to <8 x i32> 71 ret i32 undef 72 } 73 74 define i32 @masks4(<4 x i1> %in) { 75 ;CHECK: cost of 8 {{.*}} sext 76 %S = sext <4 x i1> %in to <4 x i64> 77 ret i32 undef 78 } 79 80 define void @sitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) { 81 ; CHECK: cost of 3 {{.*}} sitofp 82 %A1 = sitofp <4 x i1> %a to <4 x float> 83 ; CHECK: cost of 3 {{.*}} sitofp 84 %A2 = sitofp <4 x i1> %a to <4 x double> 85 86 ; CHECK: cost of 3 {{.*}} sitofp 87 %B1 = sitofp <4 x i8> %b to <4 x float> 88 ; CHECK: cost of 3 {{.*}} sitofp 89 %B2 = sitofp <4 x i8> %b to <4 x double> 90 91 ; CHECK: cost of 3 {{.*}} sitofp 92 %C1 = sitofp <4 x i16> %c to <4 x float> 93 ; CHECK: cost of 3 {{.*}} sitofp 94 %C2 = sitofp <4 x i16> %c to <4 x double> 95 96 ; CHECK: cost of 1 {{.*}} sitofp 97 %D1 = sitofp <4 x i32> %d to <4 x float> 98 ; CHECK: cost of 1 {{.*}} sitofp 99 %D2 = sitofp <4 x i32> %d to <4 x double> 100 ret void 101 } 102 103 define void @sitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) { 104 ; CHECK: cost of 8 {{.*}} sitofp 105 %A1 = sitofp <8 x i1> %a to <8 x float> 106 107 ; CHECK: cost of 8 {{.*}} sitofp 108 %B1 = sitofp <8 x i8> %b to <8 x float> 109 110 ; CHECK: cost of 5 {{.*}} sitofp 111 %C1 = sitofp <8 x i16> %c to <8 x float> 112 113 ; CHECK: cost of 1 {{.*}} sitofp 114 %D1 = sitofp <8 x i32> %d to <8 x float> 115 ret void 116 } 117 118 define void @uitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) { 119 ; CHECK: cost of 7 {{.*}} uitofp 120 %A1 = uitofp <4 x i1> %a to <4 x float> 121 ; CHECK: cost of 7 {{.*}} uitofp 122 %A2 = uitofp <4 x i1> %a to <4 x double> 123 124 ; CHECK: cost of 2 {{.*}} uitofp 125 %B1 = uitofp <4 x i8> %b to <4 x float> 126 ; CHECK: cost of 2 {{.*}} uitofp 127 %B2 = uitofp <4 x i8> %b to <4 x double> 128 129 ; CHECK: cost of 2 {{.*}} uitofp 130 %C1 = uitofp <4 x i16> %c to <4 x float> 131 ; CHECK: cost of 2 {{.*}} uitofp 132 %C2 = uitofp <4 x i16> %c to <4 x double> 133 134 ; CHECK: cost of 6 {{.*}} uitofp 135 %D1 = uitofp <4 x i32> %d to <4 x float> 136 ; CHECK: cost of 6 {{.*}} uitofp 137 %D2 = uitofp <4 x i32> %d to <4 x double> 138 ret void 139 } 140 141 define void @uitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) { 142 ; CHECK: cost of 6 {{.*}} uitofp 143 %A1 = uitofp <8 x i1> %a to <8 x float> 144 145 ; CHECK: cost of 5 {{.*}} uitofp 146 %B1 = uitofp <8 x i8> %b to <8 x float> 147 148 ; CHECK: cost of 5 {{.*}} uitofp 149 %C1 = uitofp <8 x i16> %c to <8 x float> 150 151 ; CHECK: cost of 9 {{.*}} uitofp 152 %D1 = uitofp <8 x i32> %d to <8 x float> 153 ret void 154 } 155
