1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 ; RUN: llc < %s -mtriple=x86_64-linux -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,CHECK-LIN 3 ; RUN: llc < %s -mtriple=x86_64-win32 -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,CHECK-WIN 4 5 define <8 x i32> @a(<8 x i16> %a) nounwind { 6 ; CHECK-LIN-LABEL: a: 7 ; CHECK-LIN: # %bb.0: 8 ; CHECK-LIN-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3] 9 ; CHECK-LIN-NEXT: psrad $16, %xmm2 10 ; CHECK-LIN-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] 11 ; CHECK-LIN-NEXT: psrad $16, %xmm1 12 ; CHECK-LIN-NEXT: movdqa %xmm2, %xmm0 13 ; CHECK-LIN-NEXT: retq 14 ; 15 ; CHECK-WIN-LABEL: a: 16 ; CHECK-WIN: # %bb.0: 17 ; CHECK-WIN-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 18 ; CHECK-WIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3] 19 ; CHECK-WIN-NEXT: psrad $16, %xmm0 20 ; CHECK-WIN-NEXT: movq {{.*#+}} xmm1 = mem[0],zero 21 ; CHECK-WIN-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3] 22 ; CHECK-WIN-NEXT: psrad $16, %xmm1 23 ; CHECK-WIN-NEXT: retq 24 %c = sext <8 x i16> %a to <8 x i32> 25 ret <8 x i32> %c 26 } 27 28 define <3 x i32> @b(<3 x i16> %a) nounwind { 29 ; CHECK-LIN-LABEL: b: 30 ; CHECK-LIN: # %bb.0: 31 ; CHECK-LIN-NEXT: movd %edi, %xmm0 32 ; CHECK-LIN-NEXT: pinsrw $1, %esi, %xmm0 33 ; CHECK-LIN-NEXT: pinsrw $2, %edx, %xmm0 34 ; CHECK-LIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3] 35 ; CHECK-LIN-NEXT: psrad $16, %xmm0 36 ; CHECK-LIN-NEXT: retq 37 ; 38 ; CHECK-WIN-LABEL: b: 39 ; CHECK-WIN: # %bb.0: 40 ; CHECK-WIN-NEXT: # kill: def $r8w killed $r8w def $r8d 41 ; CHECK-WIN-NEXT: # kill: def $dx killed $dx def $edx 42 ; CHECK-WIN-NEXT: # kill: def $cx killed $cx def $ecx 43 ; CHECK-WIN-NEXT: movd %ecx, %xmm0 44 ; CHECK-WIN-NEXT: pinsrw $1, %edx, %xmm0 45 ; CHECK-WIN-NEXT: pinsrw $2, %r8d, %xmm0 46 ; CHECK-WIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3] 47 ; CHECK-WIN-NEXT: psrad $16, %xmm0 48 ; CHECK-WIN-NEXT: retq 49 %c = sext <3 x i16> %a to <3 x i32> 50 ret <3 x i32> %c 51 } 52 53 define <1 x i32> @c(<1 x i16> %a) nounwind { 54 ; CHECK-LIN-LABEL: c: 55 ; CHECK-LIN: # %bb.0: 56 ; CHECK-LIN-NEXT: movd %edi, %xmm0 57 ; CHECK-LIN-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,2,1,4,5,6,7] 58 ; CHECK-LIN-NEXT: psrad $16, %xmm0 59 ; CHECK-LIN-NEXT: movd %xmm0, %eax 60 ; CHECK-LIN-NEXT: retq 61 ; 62 ; CHECK-WIN-LABEL: c: 63 ; CHECK-WIN: # %bb.0: 64 ; CHECK-WIN-NEXT: # kill: def $cx killed $cx def $ecx 65 ; CHECK-WIN-NEXT: movd %ecx, %xmm0 66 ; CHECK-WIN-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,2,1,4,5,6,7] 67 ; CHECK-WIN-NEXT: psrad $16, %xmm0 68 ; CHECK-WIN-NEXT: movd %xmm0, %eax 69 ; CHECK-WIN-NEXT: retq 70 %c = sext <1 x i16> %a to <1 x i32> 71 ret <1 x i32> %c 72 } 73 74 define <8 x i32> @d(<8 x i16> %a) nounwind { 75 ; CHECK-LIN-LABEL: d: 76 ; CHECK-LIN: # %bb.0: 77 ; CHECK-LIN-NEXT: movdqa %xmm0, %xmm1 78 ; CHECK-LIN-NEXT: pxor %xmm2, %xmm2 79 ; CHECK-LIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3] 80 ; CHECK-LIN-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7] 81 ; CHECK-LIN-NEXT: retq 82 ; 83 ; CHECK-WIN-LABEL: d: 84 ; CHECK-WIN: # %bb.0: 85 ; CHECK-WIN-NEXT: movdqa (%rcx), %xmm1 86 ; CHECK-WIN-NEXT: pxor %xmm2, %xmm2 87 ; CHECK-WIN-NEXT: movdqa %xmm1, %xmm0 88 ; CHECK-WIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3] 89 ; CHECK-WIN-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7] 90 ; CHECK-WIN-NEXT: retq 91 %c = zext <8 x i16> %a to <8 x i32> 92 ret <8 x i32> %c 93 } 94 95 define <3 x i32> @e(<3 x i16> %a) nounwind { 96 ; CHECK-LIN-LABEL: e: 97 ; CHECK-LIN: # %bb.0: 98 ; CHECK-LIN-NEXT: movd %edi, %xmm0 99 ; CHECK-LIN-NEXT: pinsrw $1, %esi, %xmm0 100 ; CHECK-LIN-NEXT: pinsrw $2, %edx, %xmm0 101 ; CHECK-LIN-NEXT: pxor %xmm1, %xmm1 102 ; CHECK-LIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] 103 ; CHECK-LIN-NEXT: retq 104 ; 105 ; CHECK-WIN-LABEL: e: 106 ; CHECK-WIN: # %bb.0: 107 ; CHECK-WIN-NEXT: # kill: def $r8w killed $r8w def $r8d 108 ; CHECK-WIN-NEXT: # kill: def $dx killed $dx def $edx 109 ; CHECK-WIN-NEXT: # kill: def $cx killed $cx def $ecx 110 ; CHECK-WIN-NEXT: movd %ecx, %xmm0 111 ; CHECK-WIN-NEXT: pinsrw $1, %edx, %xmm0 112 ; CHECK-WIN-NEXT: pinsrw $2, %r8d, %xmm0 113 ; CHECK-WIN-NEXT: pxor %xmm1, %xmm1 114 ; CHECK-WIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] 115 ; CHECK-WIN-NEXT: retq 116 %c = zext <3 x i16> %a to <3 x i32> 117 ret <3 x i32> %c 118 } 119 120 define <1 x i32> @f(<1 x i16> %a) nounwind { 121 ; CHECK-LIN-LABEL: f: 122 ; CHECK-LIN: # %bb.0: 123 ; CHECK-LIN-NEXT: movd %edi, %xmm0 124 ; CHECK-LIN-NEXT: pxor %xmm1, %xmm1 125 ; CHECK-LIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] 126 ; CHECK-LIN-NEXT: movd %xmm0, %eax 127 ; CHECK-LIN-NEXT: retq 128 ; 129 ; CHECK-WIN-LABEL: f: 130 ; CHECK-WIN: # %bb.0: 131 ; CHECK-WIN-NEXT: # kill: def $cx killed $cx def $ecx 132 ; CHECK-WIN-NEXT: movd %ecx, %xmm0 133 ; CHECK-WIN-NEXT: pxor %xmm1, %xmm1 134 ; CHECK-WIN-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] 135 ; CHECK-WIN-NEXT: movd %xmm0, %eax 136 ; CHECK-WIN-NEXT: retq 137 %c = zext <1 x i16> %a to <1 x i32> 138 ret <1 x i32> %c 139 } 140 141 define <8 x i16> @g(<8 x i32> %a) nounwind { 142 ; CHECK-LIN-LABEL: g: 143 ; CHECK-LIN: # %bb.0: 144 ; CHECK-LIN-NEXT: movdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15] 145 ; CHECK-LIN-NEXT: pshufb %xmm2, %xmm1 146 ; CHECK-LIN-NEXT: pshufb %xmm2, %xmm0 147 ; CHECK-LIN-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 148 ; CHECK-LIN-NEXT: retq 149 ; 150 ; CHECK-WIN-LABEL: g: 151 ; CHECK-WIN: # %bb.0: 152 ; CHECK-WIN-NEXT: movdqa (%rcx), %xmm0 153 ; CHECK-WIN-NEXT: movdqa (%rdx), %xmm1 154 ; CHECK-WIN-NEXT: movdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15] 155 ; CHECK-WIN-NEXT: pshufb %xmm2, %xmm1 156 ; CHECK-WIN-NEXT: pshufb %xmm2, %xmm0 157 ; CHECK-WIN-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 158 ; CHECK-WIN-NEXT: retq 159 %c = trunc <8 x i32> %a to <8 x i16> 160 ret <8 x i16> %c 161 } 162 163 define <3 x i16> @h(<3 x i32> %a) nounwind { 164 ; CHECK-LIN-LABEL: h: 165 ; CHECK-LIN: # %bb.0: 166 ; CHECK-LIN-NEXT: movd %xmm0, %eax 167 ; CHECK-LIN-NEXT: pextrw $2, %xmm0, %edx 168 ; CHECK-LIN-NEXT: pextrw $4, %xmm0, %ecx 169 ; CHECK-LIN-NEXT: # kill: def $ax killed $ax killed $eax 170 ; CHECK-LIN-NEXT: # kill: def $dx killed $dx killed $edx 171 ; CHECK-LIN-NEXT: # kill: def $cx killed $cx killed $ecx 172 ; CHECK-LIN-NEXT: retq 173 ; 174 ; CHECK-WIN-LABEL: h: 175 ; CHECK-WIN: # %bb.0: 176 ; CHECK-WIN-NEXT: movdqa (%rcx), %xmm0 177 ; CHECK-WIN-NEXT: movd %xmm0, %eax 178 ; CHECK-WIN-NEXT: pextrw $2, %xmm0, %edx 179 ; CHECK-WIN-NEXT: pextrw $4, %xmm0, %ecx 180 ; CHECK-WIN-NEXT: # kill: def $ax killed $ax killed $eax 181 ; CHECK-WIN-NEXT: # kill: def $dx killed $dx killed $edx 182 ; CHECK-WIN-NEXT: # kill: def $cx killed $cx killed $ecx 183 ; CHECK-WIN-NEXT: retq 184 %c = trunc <3 x i32> %a to <3 x i16> 185 ret <3 x i16> %c 186 } 187 188 define <1 x i16> @i(<1 x i32> %a) nounwind { 189 ; CHECK-LIN-LABEL: i: 190 ; CHECK-LIN: # %bb.0: 191 ; CHECK-LIN-NEXT: movl %edi, %eax 192 ; CHECK-LIN-NEXT: retq 193 ; 194 ; CHECK-WIN-LABEL: i: 195 ; CHECK-WIN: # %bb.0: 196 ; CHECK-WIN-NEXT: movl %ecx, %eax 197 ; CHECK-WIN-NEXT: retq 198 %c = trunc <1 x i32> %a to <1 x i16> 199 ret <1 x i16> %c 200 } 201 202 ; PR6438 203 define void @__OpenCL_math_kernel4_kernel() nounwind { 204 ; CHECK-LABEL: __OpenCL_math_kernel4_kernel: 205 ; CHECK: # %bb.0: 206 ; CHECK-NEXT: retq 207 %tmp12.i = and <4 x i32> zeroinitializer, <i32 2139095040, i32 2139095040, i32 2139095040, i32 2139095040> ; <<4 x i32>> [#uses=1] 208 %cmp13.i = icmp eq <4 x i32> %tmp12.i, <i32 2139095040, i32 2139095040, i32 2139095040, i32 2139095040> ; <<4 x i1>> [#uses=2] 209 %cmp.ext14.i = sext <4 x i1> %cmp13.i to <4 x i32> ; <<4 x i32>> [#uses=0] 210 %tmp2110.i = and <4 x i1> %cmp13.i, zeroinitializer ; <<4 x i1>> [#uses=0] 211 ret void 212 } 213
