| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 1 | ; RUN: llc < %s -march=x86-64 -mcpu=x86-64 | FileCheck %s -check-prefix=SSE2 |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 2 | ; RUN: llc < %s -march=x86-64 -mcpu=corei7 | FileCheck %s -check-prefix=SSSE3 |
| 3 | ; RUN: llc < %s -march=x86-64 -mcpu=core-avx2 | FileCheck %s -check-prefix=AVX2 |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 4 | |
| 5 | define <4 x i32> @test1(<4 x i32> %a) nounwind { |
| 6 | ; SSE2: test1: |
| 7 | ; SSE2: movdqa |
| 8 | ; SSE2-NEXT: psrad $31 |
| 9 | ; SSE2-NEXT: padd |
| 10 | ; SSE2-NEXT: pxor |
| 11 | ; SSE2-NEXT: ret |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 12 | |
| 13 | ; SSSE3: test1: |
| 14 | ; SSSE3: pabsd |
| 15 | ; SSSE3-NEXT: ret |
| 16 | |
| 17 | ; AVX2: test1: |
| 18 | ; AVX2: vpabsd |
| 19 | ; AVX2-NEXT: ret |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 20 | %tmp1neg = sub <4 x i32> zeroinitializer, %a |
| 21 | %b = icmp sgt <4 x i32> %a, <i32 -1, i32 -1, i32 -1, i32 -1> |
| 22 | %abs = select <4 x i1> %b, <4 x i32> %a, <4 x i32> %tmp1neg |
| 23 | ret <4 x i32> %abs |
| 24 | } |
| 25 | |
| 26 | define <4 x i32> @test2(<4 x i32> %a) nounwind { |
| 27 | ; SSE2: test2: |
| 28 | ; SSE2: movdqa |
| 29 | ; SSE2-NEXT: psrad $31 |
| 30 | ; SSE2-NEXT: padd |
| 31 | ; SSE2-NEXT: pxor |
| 32 | ; SSE2-NEXT: ret |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 33 | |
| 34 | ; SSSE3: test2: |
| 35 | ; SSSE3: pabsd |
| 36 | ; SSSE3-NEXT: ret |
| 37 | |
| 38 | ; AVX2: test2: |
| 39 | ; AVX2: vpabsd |
| 40 | ; AVX2-NEXT: ret |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 41 | %tmp1neg = sub <4 x i32> zeroinitializer, %a |
| 42 | %b = icmp sge <4 x i32> %a, zeroinitializer |
| 43 | %abs = select <4 x i1> %b, <4 x i32> %a, <4 x i32> %tmp1neg |
| 44 | ret <4 x i32> %abs |
| 45 | } |
| 46 | |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 47 | define <8 x i16> @test3(<8 x i16> %a) nounwind { |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 48 | ; SSE2: test3: |
| 49 | ; SSE2: movdqa |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 50 | ; SSE2-NEXT: psraw $15 |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 51 | ; SSE2-NEXT: padd |
| 52 | ; SSE2-NEXT: pxor |
| 53 | ; SSE2-NEXT: ret |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 54 | |
| 55 | ; SSSE3: test3: |
| 56 | ; SSSE3: pabsw |
| 57 | ; SSSE3-NEXT: ret |
| 58 | |
| 59 | ; AVX2: test3: |
| 60 | ; AVX2: vpabsw |
| 61 | ; AVX2-NEXT: ret |
| 62 | %tmp1neg = sub <8 x i16> zeroinitializer, %a |
| 63 | %b = icmp sgt <8 x i16> %a, zeroinitializer |
| 64 | %abs = select <8 x i1> %b, <8 x i16> %a, <8 x i16> %tmp1neg |
| 65 | ret <8 x i16> %abs |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 66 | } |
| 67 | |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 68 | define <16 x i8> @test4(<16 x i8> %a) nounwind { |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 69 | ; SSE2: test4: |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 70 | ; SSE2: pxor |
| 71 | ; SSE2-NEXT: pcmpgtb |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 72 | ; SSE2-NEXT: padd |
| 73 | ; SSE2-NEXT: pxor |
| 74 | ; SSE2-NEXT: ret |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 75 | |
| 76 | ; SSSE3: test4: |
| 77 | ; SSSE3: pabsb |
| 78 | ; SSSE3-NEXT: ret |
| 79 | |
| 80 | ; AVX2: test4: |
| 81 | ; AVX2: vpabsb |
| 82 | ; AVX2-NEXT: ret |
| 83 | %tmp1neg = sub <16 x i8> zeroinitializer, %a |
| 84 | %b = icmp slt <16 x i8> %a, zeroinitializer |
| 85 | %abs = select <16 x i1> %b, <16 x i8> %tmp1neg, <16 x i8> %a |
| 86 | ret <16 x i8> %abs |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 87 | } |
| 88 | |
| 89 | define <4 x i32> @test5(<4 x i32> %a) nounwind { |
| 90 | ; SSE2: test5: |
| 91 | ; SSE2: movdqa |
| 92 | ; SSE2-NEXT: psrad $31 |
| 93 | ; SSE2-NEXT: padd |
| 94 | ; SSE2-NEXT: pxor |
| 95 | ; SSE2-NEXT: ret |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 96 | |
| 97 | ; SSSE3: test5: |
| 98 | ; SSSE3: pabsd |
| 99 | ; SSSE3-NEXT: ret |
| 100 | |
| 101 | ; AVX2: test5: |
| 102 | ; AVX2: vpabsd |
| 103 | ; AVX2-NEXT: ret |
| Benjamin Kramer | d56ffc7 | 2013-04-26 09:19:19 +0000 | [diff] [blame] | 104 | %tmp1neg = sub <4 x i32> zeroinitializer, %a |
| 105 | %b = icmp sle <4 x i32> %a, zeroinitializer |
| 106 | %abs = select <4 x i1> %b, <4 x i32> %tmp1neg, <4 x i32> %a |
| 107 | ret <4 x i32> %abs |
| 108 | } |
| Benjamin Kramer | aec9053 | 2013-04-26 12:05:21 +0000 | [diff] [blame^] | 109 | |
| 110 | define <8 x i32> @test6(<8 x i32> %a) nounwind { |
| 111 | ; SSSE3: test6: |
| 112 | ; SSSE3: pabsd |
| 113 | ; SSSE3: pabsd |
| 114 | ; SSSE3-NEXT: ret |
| 115 | |
| 116 | ; AVX2: test6: |
| 117 | ; AVX2: vpabsd %ymm |
| 118 | ; AVX2-NEXT: ret |
| 119 | %tmp1neg = sub <8 x i32> zeroinitializer, %a |
| 120 | %b = icmp sgt <8 x i32> %a, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1> |
| 121 | %abs = select <8 x i1> %b, <8 x i32> %a, <8 x i32> %tmp1neg |
| 122 | ret <8 x i32> %abs |
| 123 | } |
| 124 | |
| 125 | define <8 x i32> @test7(<8 x i32> %a) nounwind { |
| 126 | ; SSSE3: test7: |
| 127 | ; SSSE3: pabsd |
| 128 | ; SSSE3: pabsd |
| 129 | ; SSSE3-NEXT: ret |
| 130 | |
| 131 | ; AVX2: test7: |
| 132 | ; AVX2: vpabsd %ymm |
| 133 | ; AVX2-NEXT: ret |
| 134 | %tmp1neg = sub <8 x i32> zeroinitializer, %a |
| 135 | %b = icmp sge <8 x i32> %a, zeroinitializer |
| 136 | %abs = select <8 x i1> %b, <8 x i32> %a, <8 x i32> %tmp1neg |
| 137 | ret <8 x i32> %abs |
| 138 | } |
| 139 | |
| 140 | define <16 x i16> @test8(<16 x i16> %a) nounwind { |
| 141 | ; SSSE3: test8: |
| 142 | ; SSSE3: pabsw |
| 143 | ; SSSE3: pabsw |
| 144 | ; SSSE3-NEXT: ret |
| 145 | |
| 146 | ; AVX2: test8: |
| 147 | ; AVX2: vpabsw %ymm |
| 148 | ; AVX2-NEXT: ret |
| 149 | %tmp1neg = sub <16 x i16> zeroinitializer, %a |
| 150 | %b = icmp sgt <16 x i16> %a, zeroinitializer |
| 151 | %abs = select <16 x i1> %b, <16 x i16> %a, <16 x i16> %tmp1neg |
| 152 | ret <16 x i16> %abs |
| 153 | } |
| 154 | |
| 155 | define <32 x i8> @test9(<32 x i8> %a) nounwind { |
| 156 | ; SSSE3: test9: |
| 157 | ; SSSE3: pabsb |
| 158 | ; SSSE3: pabsb |
| 159 | ; SSSE3-NEXT: ret |
| 160 | |
| 161 | ; AVX2: test9: |
| 162 | ; AVX2: vpabsb %ymm |
| 163 | ; AVX2-NEXT: ret |
| 164 | %tmp1neg = sub <32 x i8> zeroinitializer, %a |
| 165 | %b = icmp slt <32 x i8> %a, zeroinitializer |
| 166 | %abs = select <32 x i1> %b, <32 x i8> %tmp1neg, <32 x i8> %a |
| 167 | ret <32 x i8> %abs |
| 168 | } |
| 169 | |
| 170 | define <8 x i32> @test10(<8 x i32> %a) nounwind { |
| 171 | ; SSSE3: test10: |
| 172 | ; SSSE3: pabsd |
| 173 | ; SSSE3: pabsd |
| 174 | ; SSSE3-NEXT: ret |
| 175 | |
| 176 | ; AVX2: test10: |
| 177 | ; AVX2: vpabsd %ymm |
| 178 | ; AVX2-NEXT: ret |
| 179 | %tmp1neg = sub <8 x i32> zeroinitializer, %a |
| 180 | %b = icmp sle <8 x i32> %a, zeroinitializer |
| 181 | %abs = select <8 x i1> %b, <8 x i32> %tmp1neg, <8 x i32> %a |
| 182 | ret <8 x i32> %abs |
| 183 | } |