1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+popcnt | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC 3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=slm | FileCheck %s --check-prefix=CHECK --check-prefix=SLM 4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=goldmont | FileCheck %s --check-prefix=CHECK --check-prefix=SLM 5 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=sandybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY 6 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=ivybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY 7 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL 8 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL 9 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE 10 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=knl | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL 11 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2 12 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1 13 14 define i16 @test_ctpop_i16(i16 zeroext %a0, i16 *%a1) { 15 ; GENERIC-LABEL: test_ctpop_i16: 16 ; GENERIC: # %bb.0: 17 ; GENERIC-NEXT: popcntw (%rsi), %cx # sched: [9:1.00] 18 ; GENERIC-NEXT: popcntw %di, %ax # sched: [3:1.00] 19 ; GENERIC-NEXT: orl %ecx, %eax # sched: [1:0.33] 20 ; GENERIC-NEXT: # kill: def $ax killed $ax killed $eax 21 ; GENERIC-NEXT: retq # sched: [1:1.00] 22 ; 23 ; SLM-LABEL: test_ctpop_i16: 24 ; SLM: # %bb.0: 25 ; SLM-NEXT: popcntw (%rsi), %cx # sched: [6:1.00] 26 ; SLM-NEXT: popcntw %di, %ax # sched: [3:1.00] 27 ; SLM-NEXT: orl %ecx, %eax # sched: [1:0.50] 28 ; SLM-NEXT: # kill: def $ax killed $ax killed $eax 29 ; SLM-NEXT: retq # sched: [4:1.00] 30 ; 31 ; SANDY-LABEL: test_ctpop_i16: 32 ; SANDY: # %bb.0: 33 ; SANDY-NEXT: popcntw (%rsi), %cx # sched: [9:1.00] 34 ; SANDY-NEXT: popcntw %di, %ax # sched: [3:1.00] 35 ; SANDY-NEXT: orl %ecx, %eax # sched: [1:0.33] 36 ; SANDY-NEXT: # kill: def $ax killed $ax killed $eax 37 ; SANDY-NEXT: retq # sched: [1:1.00] 38 ; 39 ; HASWELL-LABEL: test_ctpop_i16: 40 ; HASWELL: # %bb.0: 41 ; HASWELL-NEXT: popcntw (%rsi), %cx # sched: [8:1.00] 42 ; HASWELL-NEXT: popcntw %di, %ax # sched: [3:1.00] 43 ; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25] 44 ; HASWELL-NEXT: # kill: def $ax killed $ax killed $eax 45 ; HASWELL-NEXT: retq # sched: [7:1.00] 46 ; 47 ; BROADWELL-LABEL: test_ctpop_i16: 48 ; BROADWELL: # %bb.0: 49 ; BROADWELL-NEXT: popcntw (%rsi), %cx # sched: [8:1.00] 50 ; BROADWELL-NEXT: popcntw %di, %ax # sched: [3:1.00] 51 ; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25] 52 ; BROADWELL-NEXT: # kill: def $ax killed $ax killed $eax 53 ; BROADWELL-NEXT: retq # sched: [7:1.00] 54 ; 55 ; SKYLAKE-LABEL: test_ctpop_i16: 56 ; SKYLAKE: # %bb.0: 57 ; SKYLAKE-NEXT: popcntw (%rsi), %cx # sched: [8:1.00] 58 ; SKYLAKE-NEXT: popcntw %di, %ax # sched: [3:1.00] 59 ; SKYLAKE-NEXT: orl %ecx, %eax # sched: [1:0.25] 60 ; SKYLAKE-NEXT: # kill: def $ax killed $ax killed $eax 61 ; SKYLAKE-NEXT: retq # sched: [7:1.00] 62 ; 63 ; BTVER2-LABEL: test_ctpop_i16: 64 ; BTVER2: # %bb.0: 65 ; BTVER2-NEXT: popcntw (%rsi), %cx # sched: [4:1.00] 66 ; BTVER2-NEXT: popcntw %di, %ax # sched: [1:0.50] 67 ; BTVER2-NEXT: orl %ecx, %eax # sched: [1:0.50] 68 ; BTVER2-NEXT: # kill: def $ax killed $ax killed $eax 69 ; BTVER2-NEXT: retq # sched: [4:1.00] 70 ; 71 ; ZNVER1-LABEL: test_ctpop_i16: 72 ; ZNVER1: # %bb.0: 73 ; ZNVER1-NEXT: popcntw (%rsi), %cx # sched: [5:0.50] 74 ; ZNVER1-NEXT: popcntw %di, %ax # sched: [1:0.25] 75 ; ZNVER1-NEXT: orl %ecx, %eax # sched: [1:0.25] 76 ; ZNVER1-NEXT: # kill: def $ax killed $ax killed $eax 77 ; ZNVER1-NEXT: retq # sched: [1:0.50] 78 %1 = load i16, i16 *%a1 79 %2 = tail call i16 @llvm.ctpop.i16( i16 %1 ) 80 %3 = tail call i16 @llvm.ctpop.i16( i16 %a0 ) 81 %4 = or i16 %2, %3 82 ret i16 %4 83 } 84 declare i16 @llvm.ctpop.i16(i16) 85 86 define i32 @test_ctpop_i32(i32 %a0, i32 *%a1) { 87 ; GENERIC-LABEL: test_ctpop_i32: 88 ; GENERIC: # %bb.0: 89 ; GENERIC-NEXT: popcntl (%rsi), %ecx # sched: [9:1.00] 90 ; GENERIC-NEXT: popcntl %edi, %eax # sched: [3:1.00] 91 ; GENERIC-NEXT: orl %ecx, %eax # sched: [1:0.33] 92 ; GENERIC-NEXT: retq # sched: [1:1.00] 93 ; 94 ; SLM-LABEL: test_ctpop_i32: 95 ; SLM: # %bb.0: 96 ; SLM-NEXT: popcntl (%rsi), %ecx # sched: [6:1.00] 97 ; SLM-NEXT: popcntl %edi, %eax # sched: [3:1.00] 98 ; SLM-NEXT: orl %ecx, %eax # sched: [1:0.50] 99 ; SLM-NEXT: retq # sched: [4:1.00] 100 ; 101 ; SANDY-LABEL: test_ctpop_i32: 102 ; SANDY: # %bb.0: 103 ; SANDY-NEXT: popcntl (%rsi), %ecx # sched: [9:1.00] 104 ; SANDY-NEXT: popcntl %edi, %eax # sched: [3:1.00] 105 ; SANDY-NEXT: orl %ecx, %eax # sched: [1:0.33] 106 ; SANDY-NEXT: retq # sched: [1:1.00] 107 ; 108 ; HASWELL-LABEL: test_ctpop_i32: 109 ; HASWELL: # %bb.0: 110 ; HASWELL-NEXT: popcntl (%rsi), %ecx # sched: [8:1.00] 111 ; HASWELL-NEXT: popcntl %edi, %eax # sched: [3:1.00] 112 ; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25] 113 ; HASWELL-NEXT: retq # sched: [7:1.00] 114 ; 115 ; BROADWELL-LABEL: test_ctpop_i32: 116 ; BROADWELL: # %bb.0: 117 ; BROADWELL-NEXT: popcntl (%rsi), %ecx # sched: [8:1.00] 118 ; BROADWELL-NEXT: popcntl %edi, %eax # sched: [3:1.00] 119 ; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25] 120 ; BROADWELL-NEXT: retq # sched: [7:1.00] 121 ; 122 ; SKYLAKE-LABEL: test_ctpop_i32: 123 ; SKYLAKE: # %bb.0: 124 ; SKYLAKE-NEXT: popcntl (%rsi), %ecx # sched: [8:1.00] 125 ; SKYLAKE-NEXT: popcntl %edi, %eax # sched: [3:1.00] 126 ; SKYLAKE-NEXT: orl %ecx, %eax # sched: [1:0.25] 127 ; SKYLAKE-NEXT: retq # sched: [7:1.00] 128 ; 129 ; BTVER2-LABEL: test_ctpop_i32: 130 ; BTVER2: # %bb.0: 131 ; BTVER2-NEXT: popcntl (%rsi), %ecx # sched: [4:1.00] 132 ; BTVER2-NEXT: popcntl %edi, %eax # sched: [1:0.50] 133 ; BTVER2-NEXT: orl %ecx, %eax # sched: [1:0.50] 134 ; BTVER2-NEXT: retq # sched: [4:1.00] 135 ; 136 ; ZNVER1-LABEL: test_ctpop_i32: 137 ; ZNVER1: # %bb.0: 138 ; ZNVER1-NEXT: popcntl (%rsi), %ecx # sched: [5:0.50] 139 ; ZNVER1-NEXT: popcntl %edi, %eax # sched: [1:0.25] 140 ; ZNVER1-NEXT: orl %ecx, %eax # sched: [1:0.25] 141 ; ZNVER1-NEXT: retq # sched: [1:0.50] 142 %1 = load i32, i32 *%a1 143 %2 = tail call i32 @llvm.ctpop.i32( i32 %1 ) 144 %3 = tail call i32 @llvm.ctpop.i32( i32 %a0 ) 145 %4 = or i32 %2, %3 146 ret i32 %4 147 } 148 declare i32 @llvm.ctpop.i32(i32) 149 150 define i64 @test_ctpop_i64(i64 %a0, i64 *%a1) { 151 ; GENERIC-LABEL: test_ctpop_i64: 152 ; GENERIC: # %bb.0: 153 ; GENERIC-NEXT: popcntq (%rsi), %rcx # sched: [9:1.00] 154 ; GENERIC-NEXT: popcntq %rdi, %rax # sched: [3:1.00] 155 ; GENERIC-NEXT: orq %rcx, %rax # sched: [1:0.33] 156 ; GENERIC-NEXT: retq # sched: [1:1.00] 157 ; 158 ; SLM-LABEL: test_ctpop_i64: 159 ; SLM: # %bb.0: 160 ; SLM-NEXT: popcntq (%rsi), %rcx # sched: [6:1.00] 161 ; SLM-NEXT: popcntq %rdi, %rax # sched: [3:1.00] 162 ; SLM-NEXT: orq %rcx, %rax # sched: [1:0.50] 163 ; SLM-NEXT: retq # sched: [4:1.00] 164 ; 165 ; SANDY-LABEL: test_ctpop_i64: 166 ; SANDY: # %bb.0: 167 ; SANDY-NEXT: popcntq (%rsi), %rcx # sched: [9:1.00] 168 ; SANDY-NEXT: popcntq %rdi, %rax # sched: [3:1.00] 169 ; SANDY-NEXT: orq %rcx, %rax # sched: [1:0.33] 170 ; SANDY-NEXT: retq # sched: [1:1.00] 171 ; 172 ; HASWELL-LABEL: test_ctpop_i64: 173 ; HASWELL: # %bb.0: 174 ; HASWELL-NEXT: popcntq (%rsi), %rcx # sched: [8:1.00] 175 ; HASWELL-NEXT: popcntq %rdi, %rax # sched: [3:1.00] 176 ; HASWELL-NEXT: orq %rcx, %rax # sched: [1:0.25] 177 ; HASWELL-NEXT: retq # sched: [7:1.00] 178 ; 179 ; BROADWELL-LABEL: test_ctpop_i64: 180 ; BROADWELL: # %bb.0: 181 ; BROADWELL-NEXT: popcntq (%rsi), %rcx # sched: [8:1.00] 182 ; BROADWELL-NEXT: popcntq %rdi, %rax # sched: [3:1.00] 183 ; BROADWELL-NEXT: orq %rcx, %rax # sched: [1:0.25] 184 ; BROADWELL-NEXT: retq # sched: [7:1.00] 185 ; 186 ; SKYLAKE-LABEL: test_ctpop_i64: 187 ; SKYLAKE: # %bb.0: 188 ; SKYLAKE-NEXT: popcntq (%rsi), %rcx # sched: [8:1.00] 189 ; SKYLAKE-NEXT: popcntq %rdi, %rax # sched: [3:1.00] 190 ; SKYLAKE-NEXT: orq %rcx, %rax # sched: [1:0.25] 191 ; SKYLAKE-NEXT: retq # sched: [7:1.00] 192 ; 193 ; BTVER2-LABEL: test_ctpop_i64: 194 ; BTVER2: # %bb.0: 195 ; BTVER2-NEXT: popcntq (%rsi), %rcx # sched: [4:1.00] 196 ; BTVER2-NEXT: popcntq %rdi, %rax # sched: [1:0.50] 197 ; BTVER2-NEXT: orq %rcx, %rax # sched: [1:0.50] 198 ; BTVER2-NEXT: retq # sched: [4:1.00] 199 ; 200 ; ZNVER1-LABEL: test_ctpop_i64: 201 ; ZNVER1: # %bb.0: 202 ; ZNVER1-NEXT: popcntq (%rsi), %rcx # sched: [5:0.50] 203 ; ZNVER1-NEXT: popcntq %rdi, %rax # sched: [1:0.25] 204 ; ZNVER1-NEXT: orq %rcx, %rax # sched: [1:0.25] 205 ; ZNVER1-NEXT: retq # sched: [1:0.50] 206 %1 = load i64, i64 *%a1 207 %2 = tail call i64 @llvm.ctpop.i64( i64 %1 ) 208 %3 = tail call i64 @llvm.ctpop.i64( i64 %a0 ) 209 %4 = or i64 %2, %3 210 ret i64 %4 211 } 212 declare i64 @llvm.ctpop.i64(i64) 213
