1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 ; RUN: llc < %s -mcpu=x86-64 | FileCheck %s --check-prefix=CHECK-ALL --check-prefix=CHECK-SSE --check-prefix=CHECK-NOSSSE3 3 ; RUN: llc < %s -mcpu=core2 | FileCheck %s --check-prefix=CHECK-ALL --check-prefix=CHECK-SSE --check-prefix=CHECK-SSSE3 4 ; RUN: llc < %s -mcpu=core-avx2 | FileCheck %s --check-prefix=CHECK-ALL --check-prefix=CHECK-AVX --check-prefix=CHECK-AVX2 5 ; RUN: llc < %s -mcpu=core-avx2 -x86-experimental-vector-widening-legalization | FileCheck %s --check-prefix=CHECK-ALL --check-prefix=CHECK-WIDE-AVX --check-prefix=CHECK-WIDE-AVX2 6 7 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" 8 target triple = "x86_64-unknown-linux-gnu" 9 10 declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>) 11 declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>) 12 declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>) 13 14 define <8 x i16> @test1(<8 x i16> %v) { 15 ; CHECK-NOSSSE3-LABEL: test1: 16 ; CHECK-NOSSSE3: # BB#0: # %entry 17 ; CHECK-NOSSSE3-NEXT: pxor %xmm1, %xmm1 18 ; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2 19 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] 20 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[1,0,3,2,4,5,6,7] 21 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,4,7,6] 22 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] 23 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[1,0,3,2,4,5,6,7] 24 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,4,7,6] 25 ; CHECK-NOSSSE3-NEXT: packuswb %xmm2, %xmm0 26 ; CHECK-NOSSSE3-NEXT: retq 27 ; 28 ; CHECK-SSSE3-LABEL: test1: 29 ; CHECK-SSSE3: # BB#0: # %entry 30 ; CHECK-SSSE3-NEXT: pshufb {{.*#+}} xmm0 = xmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14] 31 ; CHECK-SSSE3-NEXT: retq 32 ; 33 ; CHECK-AVX-LABEL: test1: 34 ; CHECK-AVX: # BB#0: # %entry 35 ; CHECK-AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14] 36 ; CHECK-AVX-NEXT: retq 37 ; 38 ; CHECK-WIDE-AVX-LABEL: test1: 39 ; CHECK-WIDE-AVX: # BB#0: # %entry 40 ; CHECK-WIDE-AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14] 41 ; CHECK-WIDE-AVX-NEXT: retq 42 entry: 43 %r = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %v) 44 ret <8 x i16> %r 45 } 46 47 define <4 x i32> @test2(<4 x i32> %v) { 48 ; CHECK-NOSSSE3-LABEL: test2: 49 ; CHECK-NOSSSE3: # BB#0: # %entry 50 ; CHECK-NOSSSE3-NEXT: pxor %xmm1, %xmm1 51 ; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2 52 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] 53 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7] 54 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4] 55 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] 56 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7] 57 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4] 58 ; CHECK-NOSSSE3-NEXT: packuswb %xmm2, %xmm0 59 ; CHECK-NOSSSE3-NEXT: retq 60 ; 61 ; CHECK-SSSE3-LABEL: test2: 62 ; CHECK-SSSE3: # BB#0: # %entry 63 ; CHECK-SSSE3-NEXT: pshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12] 64 ; CHECK-SSSE3-NEXT: retq 65 ; 66 ; CHECK-AVX-LABEL: test2: 67 ; CHECK-AVX: # BB#0: # %entry 68 ; CHECK-AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12] 69 ; CHECK-AVX-NEXT: retq 70 ; 71 ; CHECK-WIDE-AVX-LABEL: test2: 72 ; CHECK-WIDE-AVX: # BB#0: # %entry 73 ; CHECK-WIDE-AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12] 74 ; CHECK-WIDE-AVX-NEXT: retq 75 entry: 76 %r = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %v) 77 ret <4 x i32> %r 78 } 79 80 define <2 x i64> @test3(<2 x i64> %v) { 81 ; CHECK-NOSSSE3-LABEL: test3: 82 ; CHECK-NOSSSE3: # BB#0: # %entry 83 ; CHECK-NOSSSE3-NEXT: pxor %xmm1, %xmm1 84 ; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2 85 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] 86 ; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1] 87 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7] 88 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4] 89 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] 90 ; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1] 91 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7] 92 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4] 93 ; CHECK-NOSSSE3-NEXT: packuswb %xmm2, %xmm0 94 ; CHECK-NOSSSE3-NEXT: retq 95 ; 96 ; CHECK-SSSE3-LABEL: test3: 97 ; CHECK-SSSE3: # BB#0: # %entry 98 ; CHECK-SSSE3-NEXT: pshufb {{.*#+}} xmm0 = xmm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8] 99 ; CHECK-SSSE3-NEXT: retq 100 ; 101 ; CHECK-AVX-LABEL: test3: 102 ; CHECK-AVX: # BB#0: # %entry 103 ; CHECK-AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8] 104 ; CHECK-AVX-NEXT: retq 105 ; 106 ; CHECK-WIDE-AVX-LABEL: test3: 107 ; CHECK-WIDE-AVX: # BB#0: # %entry 108 ; CHECK-WIDE-AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8] 109 ; CHECK-WIDE-AVX-NEXT: retq 110 entry: 111 %r = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %v) 112 ret <2 x i64> %r 113 } 114 115 declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>) 116 declare <8 x i32> @llvm.bswap.v8i32(<8 x i32>) 117 declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>) 118 119 define <16 x i16> @test4(<16 x i16> %v) { 120 ; CHECK-NOSSSE3-LABEL: test4: 121 ; CHECK-NOSSSE3: # BB#0: # %entry 122 ; CHECK-NOSSSE3-NEXT: pxor %xmm2, %xmm2 123 ; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm3 124 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15] 125 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[1,0,3,2,4,5,6,7] 126 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,4,7,6] 127 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7] 128 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[1,0,3,2,4,5,6,7] 129 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,4,7,6] 130 ; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm0 131 ; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm3 132 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15] 133 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[1,0,3,2,4,5,6,7] 134 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,4,7,6] 135 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7] 136 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[1,0,3,2,4,5,6,7] 137 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,4,7,6] 138 ; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm1 139 ; CHECK-NOSSSE3-NEXT: retq 140 ; 141 ; CHECK-SSSE3-LABEL: test4: 142 ; CHECK-SSSE3: # BB#0: # %entry 143 ; CHECK-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14] 144 ; CHECK-SSSE3-NEXT: pshufb %xmm2, %xmm0 145 ; CHECK-SSSE3-NEXT: pshufb %xmm2, %xmm1 146 ; CHECK-SSSE3-NEXT: retq 147 ; 148 ; CHECK-AVX-LABEL: test4: 149 ; CHECK-AVX: # BB#0: # %entry 150 ; CHECK-AVX-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14,17,16,19,18,21,20,23,22,25,24,27,26,29,28,31,30] 151 ; CHECK-AVX-NEXT: retq 152 ; 153 ; CHECK-WIDE-AVX-LABEL: test4: 154 ; CHECK-WIDE-AVX: # BB#0: # %entry 155 ; CHECK-WIDE-AVX-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14,17,16,19,18,21,20,23,22,25,24,27,26,29,28,31,30] 156 ; CHECK-WIDE-AVX-NEXT: retq 157 entry: 158 %r = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %v) 159 ret <16 x i16> %r 160 } 161 162 define <8 x i32> @test5(<8 x i32> %v) { 163 ; CHECK-NOSSSE3-LABEL: test5: 164 ; CHECK-NOSSSE3: # BB#0: # %entry 165 ; CHECK-NOSSSE3-NEXT: pxor %xmm2, %xmm2 166 ; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm3 167 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15] 168 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,2,1,0,4,5,6,7] 169 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,6,5,4] 170 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7] 171 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7] 172 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4] 173 ; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm0 174 ; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm3 175 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15] 176 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,2,1,0,4,5,6,7] 177 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,6,5,4] 178 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7] 179 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[3,2,1,0,4,5,6,7] 180 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,6,5,4] 181 ; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm1 182 ; CHECK-NOSSSE3-NEXT: retq 183 ; 184 ; CHECK-SSSE3-LABEL: test5: 185 ; CHECK-SSSE3: # BB#0: # %entry 186 ; CHECK-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12] 187 ; CHECK-SSSE3-NEXT: pshufb %xmm2, %xmm0 188 ; CHECK-SSSE3-NEXT: pshufb %xmm2, %xmm1 189 ; CHECK-SSSE3-NEXT: retq 190 ; 191 ; CHECK-AVX-LABEL: test5: 192 ; CHECK-AVX: # BB#0: # %entry 193 ; CHECK-AVX-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12,19,18,17,16,23,22,21,20,27,26,25,24,31,30,29,28] 194 ; CHECK-AVX-NEXT: retq 195 ; 196 ; CHECK-WIDE-AVX-LABEL: test5: 197 ; CHECK-WIDE-AVX: # BB#0: # %entry 198 ; CHECK-WIDE-AVX-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12,19,18,17,16,23,22,21,20,27,26,25,24,31,30,29,28] 199 ; CHECK-WIDE-AVX-NEXT: retq 200 entry: 201 %r = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %v) 202 ret <8 x i32> %r 203 } 204 205 define <4 x i64> @test6(<4 x i64> %v) { 206 ; CHECK-NOSSSE3-LABEL: test6: 207 ; CHECK-NOSSSE3: # BB#0: # %entry 208 ; CHECK-NOSSSE3-NEXT: pxor %xmm2, %xmm2 209 ; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm3 210 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15] 211 ; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1] 212 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,2,1,0,4,5,6,7] 213 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,6,5,4] 214 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7] 215 ; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1] 216 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7] 217 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4] 218 ; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm0 219 ; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm3 220 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15] 221 ; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1] 222 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,2,1,0,4,5,6,7] 223 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,6,5,4] 224 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7] 225 ; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] 226 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[3,2,1,0,4,5,6,7] 227 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,6,5,4] 228 ; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm1 229 ; CHECK-NOSSSE3-NEXT: retq 230 ; 231 ; CHECK-SSSE3-LABEL: test6: 232 ; CHECK-SSSE3: # BB#0: # %entry 233 ; CHECK-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8] 234 ; CHECK-SSSE3-NEXT: pshufb %xmm2, %xmm0 235 ; CHECK-SSSE3-NEXT: pshufb %xmm2, %xmm1 236 ; CHECK-SSSE3-NEXT: retq 237 ; 238 ; CHECK-AVX-LABEL: test6: 239 ; CHECK-AVX: # BB#0: # %entry 240 ; CHECK-AVX-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,23,22,21,20,19,18,17,16,31,30,29,28,27,26,25,24] 241 ; CHECK-AVX-NEXT: retq 242 ; 243 ; CHECK-WIDE-AVX-LABEL: test6: 244 ; CHECK-WIDE-AVX: # BB#0: # %entry 245 ; CHECK-WIDE-AVX-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,23,22,21,20,19,18,17,16,31,30,29,28,27,26,25,24] 246 ; CHECK-WIDE-AVX-NEXT: retq 247 entry: 248 %r = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %v) 249 ret <4 x i64> %r 250 } 251 252 declare <4 x i16> @llvm.bswap.v4i16(<4 x i16>) 253 254 define <4 x i16> @test7(<4 x i16> %v) { 255 ; CHECK-NOSSSE3-LABEL: test7: 256 ; CHECK-NOSSSE3: # BB#0: # %entry 257 ; CHECK-NOSSSE3-NEXT: pxor %xmm1, %xmm1 258 ; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2 259 ; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] 260 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7] 261 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4] 262 ; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] 263 ; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7] 264 ; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4] 265 ; CHECK-NOSSSE3-NEXT: packuswb %xmm2, %xmm0 266 ; CHECK-NOSSSE3-NEXT: psrld $16, %xmm0 267 ; CHECK-NOSSSE3-NEXT: retq 268 ; 269 ; CHECK-SSSE3-LABEL: test7: 270 ; CHECK-SSSE3: # BB#0: # %entry 271 ; CHECK-SSSE3-NEXT: pshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12] 272 ; CHECK-SSSE3-NEXT: psrld $16, %xmm0 273 ; CHECK-SSSE3-NEXT: retq 274 ; 275 ; CHECK-AVX-LABEL: test7: 276 ; CHECK-AVX: # BB#0: # %entry 277 ; CHECK-AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12] 278 ; CHECK-AVX-NEXT: vpsrld $16, %xmm0, %xmm0 279 ; CHECK-AVX-NEXT: retq 280 ; 281 ; CHECK-WIDE-AVX-LABEL: test7: 282 ; CHECK-WIDE-AVX: # BB#0: # %entry 283 ; CHECK-WIDE-AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14] 284 ; CHECK-WIDE-AVX-NEXT: retq 285 entry: 286 %r = call <4 x i16> @llvm.bswap.v4i16(<4 x i16> %v) 287 ret <4 x i16> %r 288 } 289 290 ; 291 ; Double BSWAP -> Identity 292 ; 293 294 define <8 x i16> @identity_v8i16(<8 x i16> %v) { 295 ; CHECK-ALL-LABEL: identity_v8i16: 296 ; CHECK-ALL: # BB#0: # %entry 297 ; CHECK-ALL-NEXT: retq 298 entry: 299 %bs1 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %v) 300 %bs2 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %bs1) 301 ret <8 x i16> %bs2 302 } 303 304 define <4 x i32> @identity_v4i32(<4 x i32> %v) { 305 ; CHECK-ALL-LABEL: identity_v4i32: 306 ; CHECK-ALL: # BB#0: # %entry 307 ; CHECK-ALL-NEXT: retq 308 entry: 309 %bs1 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %v) 310 %bs2 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %bs1) 311 ret <4 x i32> %bs2 312 } 313 314 define <2 x i64> @identity_v2i64(<2 x i64> %v) { 315 ; CHECK-ALL-LABEL: identity_v2i64: 316 ; CHECK-ALL: # BB#0: # %entry 317 ; CHECK-ALL-NEXT: retq 318 entry: 319 %bs1 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %v) 320 %bs2 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %bs1) 321 ret <2 x i64> %bs2 322 } 323 324 define <16 x i16> @identity_v16i16(<16 x i16> %v) { 325 ; CHECK-ALL-LABEL: identity_v16i16: 326 ; CHECK-ALL: # BB#0: # %entry 327 ; CHECK-ALL-NEXT: retq 328 entry: 329 %bs1 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %v) 330 %bs2 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %bs1) 331 ret <16 x i16> %bs2 332 } 333 334 define <8 x i32> @identity_v8i32(<8 x i32> %v) { 335 ; CHECK-ALL-LABEL: identity_v8i32: 336 ; CHECK-ALL: # BB#0: # %entry 337 ; CHECK-ALL-NEXT: retq 338 entry: 339 %bs1 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %v) 340 %bs2 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %bs1) 341 ret <8 x i32> %bs2 342 } 343 344 define <4 x i64> @identity_v4i64(<4 x i64> %v) { 345 ; CHECK-ALL-LABEL: identity_v4i64: 346 ; CHECK-ALL: # BB#0: # %entry 347 ; CHECK-ALL-NEXT: retq 348 entry: 349 %bs1 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %v) 350 %bs2 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %bs1) 351 ret <4 x i64> %bs2 352 } 353 354 define <4 x i16> @identity_v4i16(<4 x i16> %v) { 355 ; CHECK-ALL-LABEL: identity_v4i16: 356 ; CHECK-ALL: # BB#0: # %entry 357 ; CHECK-ALL-NEXT: retq 358 entry: 359 %bs1 = call <4 x i16> @llvm.bswap.v4i16(<4 x i16> %v) 360 %bs2 = call <4 x i16> @llvm.bswap.v4i16(<4 x i16> %bs1) 361 ret <4 x i16> %bs2 362 } 363 364 ; 365 ; Constant Folding 366 ; 367 368 define <8 x i16> @fold_v8i16() { 369 ; CHECK-SSE-LABEL: fold_v8i16: 370 ; CHECK-SSE: # BB#0: # %entry 371 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm0 = [0,256,65535,512,65023,1024,64511,1536] 372 ; CHECK-SSE-NEXT: retq 373 ; 374 ; CHECK-AVX-LABEL: fold_v8i16: 375 ; CHECK-AVX: # BB#0: # %entry 376 ; CHECK-AVX-NEXT: vmovaps {{.*#+}} xmm0 = [0,256,65535,512,65023,1024,64511,1536] 377 ; CHECK-AVX-NEXT: retq 378 ; 379 ; CHECK-WIDE-AVX-LABEL: fold_v8i16: 380 ; CHECK-WIDE-AVX: # BB#0: # %entry 381 ; CHECK-WIDE-AVX-NEXT: vmovaps {{.*#+}} xmm0 = [0,256,65535,512,65023,1024,64511,1536] 382 ; CHECK-WIDE-AVX-NEXT: retq 383 entry: 384 %r = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> <i16 0, i16 1, i16 -1, i16 2, i16 -3, i16 4, i16 -5, i16 6>) 385 ret <8 x i16> %r 386 } 387 388 define <4 x i32> @fold_v4i32() { 389 ; CHECK-SSE-LABEL: fold_v4i32: 390 ; CHECK-SSE: # BB#0: # %entry 391 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm0 = [0,4294967295,33554432,4261412863] 392 ; CHECK-SSE-NEXT: retq 393 ; 394 ; CHECK-AVX-LABEL: fold_v4i32: 395 ; CHECK-AVX: # BB#0: # %entry 396 ; CHECK-AVX-NEXT: vmovaps {{.*#+}} xmm0 = [0,4294967295,33554432,4261412863] 397 ; CHECK-AVX-NEXT: retq 398 ; 399 ; CHECK-WIDE-AVX-LABEL: fold_v4i32: 400 ; CHECK-WIDE-AVX: # BB#0: # %entry 401 ; CHECK-WIDE-AVX-NEXT: vmovaps {{.*#+}} xmm0 = [0,4294967295,33554432,4261412863] 402 ; CHECK-WIDE-AVX-NEXT: retq 403 entry: 404 %r = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> <i32 0, i32 -1, i32 2, i32 -3>) 405 ret <4 x i32> %r 406 } 407 408 define <2 x i64> @fold_v2i64() { 409 ; CHECK-SSE-LABEL: fold_v2i64: 410 ; CHECK-SSE: # BB#0: # %entry 411 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm0 = [18374686479671623680,18446744073709551615] 412 ; CHECK-SSE-NEXT: retq 413 ; 414 ; CHECK-AVX-LABEL: fold_v2i64: 415 ; CHECK-AVX: # BB#0: # %entry 416 ; CHECK-AVX-NEXT: vmovaps {{.*#+}} xmm0 = [18374686479671623680,18446744073709551615] 417 ; CHECK-AVX-NEXT: retq 418 ; 419 ; CHECK-WIDE-AVX-LABEL: fold_v2i64: 420 ; CHECK-WIDE-AVX: # BB#0: # %entry 421 ; CHECK-WIDE-AVX-NEXT: vmovaps {{.*#+}} xmm0 = [18374686479671623680,18446744073709551615] 422 ; CHECK-WIDE-AVX-NEXT: retq 423 entry: 424 %r = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> <i64 255, i64 -1>) 425 ret <2 x i64> %r 426 } 427 428 define <16 x i16> @fold_v16i16() { 429 ; CHECK-SSE-LABEL: fold_v16i16: 430 ; CHECK-SSE: # BB#0: # %entry 431 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm0 = [0,256,65535,512,65023,1024,64511,1536] 432 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm1 = [63999,2048,63487,2560,62975,3072,62463,3584] 433 ; CHECK-SSE-NEXT: retq 434 ; 435 ; CHECK-AVX-LABEL: fold_v16i16: 436 ; CHECK-AVX: # BB#0: # %entry 437 ; CHECK-AVX-NEXT: vmovaps {{.*#+}} ymm0 = [0,256,65535,512,65023,1024,64511,1536,63999,2048,63487,2560,62975,3072,62463,3584] 438 ; CHECK-AVX-NEXT: retq 439 ; 440 ; CHECK-WIDE-AVX-LABEL: fold_v16i16: 441 ; CHECK-WIDE-AVX: # BB#0: # %entry 442 ; CHECK-WIDE-AVX-NEXT: vmovaps {{.*#+}} ymm0 = [0,256,65535,512,65023,1024,64511,1536,63999,2048,63487,2560,62975,3072,62463,3584] 443 ; CHECK-WIDE-AVX-NEXT: retq 444 entry: 445 %r = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> <i16 0, i16 1, i16 -1, i16 2, i16 -3, i16 4, i16 -5, i16 6, i16 -7, i16 8, i16 -9, i16 10, i16 -11, i16 12, i16 -13, i16 14>) 446 ret <16 x i16> %r 447 } 448 449 define <8 x i32> @fold_v8i32() { 450 ; CHECK-SSE-LABEL: fold_v8i32: 451 ; CHECK-SSE: # BB#0: # %entry 452 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm0 = [0,16777216,4294967295,33554432] 453 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm1 = [4261412863,67108864,4227858431,100663296] 454 ; CHECK-SSE-NEXT: retq 455 ; 456 ; CHECK-AVX-LABEL: fold_v8i32: 457 ; CHECK-AVX: # BB#0: # %entry 458 ; CHECK-AVX-NEXT: vmovaps {{.*#+}} ymm0 = [0,16777216,4294967295,33554432,4261412863,67108864,4227858431,100663296] 459 ; CHECK-AVX-NEXT: retq 460 ; 461 ; CHECK-WIDE-AVX-LABEL: fold_v8i32: 462 ; CHECK-WIDE-AVX: # BB#0: # %entry 463 ; CHECK-WIDE-AVX-NEXT: vmovaps {{.*#+}} ymm0 = [0,16777216,4294967295,33554432,4261412863,67108864,4227858431,100663296] 464 ; CHECK-WIDE-AVX-NEXT: retq 465 entry: 466 %r = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> <i32 0, i32 1, i32 -1, i32 2, i32 -3, i32 4, i32 -5, i32 6>) 467 ret <8 x i32> %r 468 } 469 470 define <4 x i64> @fold_v4i64() { 471 ; CHECK-SSE-LABEL: fold_v4i64: 472 ; CHECK-SSE: # BB#0: # %entry 473 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm0 = [18374686479671623680,18446744073709551615] 474 ; CHECK-SSE-NEXT: movaps {{.*#+}} xmm1 = [18446462598732840960,72056494526300160] 475 ; CHECK-SSE-NEXT: retq 476 ; 477 ; CHECK-AVX-LABEL: fold_v4i64: 478 ; CHECK-AVX: # BB#0: # %entry 479 ; CHECK-AVX-NEXT: vmovaps {{.*#+}} ymm0 = [18374686479671623680,18446744073709551615,18446462598732840960,72056494526300160] 480 ; CHECK-AVX-NEXT: retq 481 ; 482 ; CHECK-WIDE-AVX-LABEL: fold_v4i64: 483 ; CHECK-WIDE-AVX: # BB#0: # %entry 484 ; CHECK-WIDE-AVX-NEXT: vmovaps {{.*#+}} ymm0 = [18374686479671623680,18446744073709551615,18446462598732840960,72056494526300160] 485 ; CHECK-WIDE-AVX-NEXT: retq 486 entry: 487 %r = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> <i64 255, i64 -1, i64 65535, i64 16776960>) 488 ret <4 x i64> %r 489 } 490
