1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 ; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mcpu=generic | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC 3 ; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mcpu=atom | FileCheck %s --check-prefix=CHECK --check-prefix=ATOM 4 ; RUN: llc < %s -mtriple=i386-intel-elfiamcu | FileCheck %s --check-prefix=MCU 5 6 ; PR5757 7 %0 = type { i64, i32 } 8 9 define i32 @test1(%0* %p, %0* %q, i1 %r) nounwind { 10 ; CHECK-LABEL: test1: 11 ; CHECK: ## %bb.0: 12 ; CHECK-NEXT: addq $8, %rdi 13 ; CHECK-NEXT: addq $8, %rsi 14 ; CHECK-NEXT: testb $1, %dl 15 ; CHECK-NEXT: cmovneq %rdi, %rsi 16 ; CHECK-NEXT: movl (%rsi), %eax 17 ; CHECK-NEXT: retq 18 ; 19 ; MCU-LABEL: test1: 20 ; MCU: # %bb.0: 21 ; MCU-NEXT: testb $1, %cl 22 ; MCU-NEXT: jne .LBB0_1 23 ; MCU-NEXT: # %bb.2: 24 ; MCU-NEXT: addl $8, %edx 25 ; MCU-NEXT: movl (%edx), %eax 26 ; MCU-NEXT: retl 27 ; MCU-NEXT: .LBB0_1: 28 ; MCU-NEXT: addl $8, %eax 29 ; MCU-NEXT: movl (%eax), %eax 30 ; MCU-NEXT: retl 31 %t0 = load %0, %0* %p 32 %t1 = load %0, %0* %q 33 %t4 = select i1 %r, %0 %t0, %0 %t1 34 %t5 = extractvalue %0 %t4, 1 35 ret i32 %t5 36 } 37 38 ; PR2139 39 define i32 @test2() nounwind { 40 ; GENERIC-LABEL: test2: 41 ; GENERIC: ## %bb.0: ## %entry 42 ; GENERIC-NEXT: pushq %rax 43 ; GENERIC-NEXT: callq _return_false 44 ; GENERIC-NEXT: xorl %ecx, %ecx 45 ; GENERIC-NEXT: testb $1, %al 46 ; GENERIC-NEXT: movl $-480, %eax ## imm = 0xFE20 47 ; GENERIC-NEXT: cmovnel %ecx, %eax 48 ; GENERIC-NEXT: shll $3, %eax 49 ; GENERIC-NEXT: cmpl $32768, %eax ## imm = 0x8000 50 ; GENERIC-NEXT: jge LBB1_1 51 ; GENERIC-NEXT: ## %bb.2: ## %bb91 52 ; GENERIC-NEXT: xorl %eax, %eax 53 ; GENERIC-NEXT: popq %rcx 54 ; GENERIC-NEXT: retq 55 ; GENERIC-NEXT: LBB1_1: ## %bb90 56 ; 57 ; ATOM-LABEL: test2: 58 ; ATOM: ## %bb.0: ## %entry 59 ; ATOM-NEXT: pushq %rax 60 ; ATOM-NEXT: callq _return_false 61 ; ATOM-NEXT: xorl %ecx, %ecx 62 ; ATOM-NEXT: movl $-480, %edx ## imm = 0xFE20 63 ; ATOM-NEXT: testb $1, %al 64 ; ATOM-NEXT: cmovnel %ecx, %edx 65 ; ATOM-NEXT: shll $3, %edx 66 ; ATOM-NEXT: cmpl $32768, %edx ## imm = 0x8000 67 ; ATOM-NEXT: jge LBB1_1 68 ; ATOM-NEXT: ## %bb.2: ## %bb91 69 ; ATOM-NEXT: xorl %eax, %eax 70 ; ATOM-NEXT: popq %rcx 71 ; ATOM-NEXT: retq 72 ; ATOM-NEXT: LBB1_1: ## %bb90 73 ; 74 ; MCU-LABEL: test2: 75 ; MCU: # %bb.0: # %entry 76 ; MCU-NEXT: calll return_false 77 ; MCU-NEXT: xorl %ecx, %ecx 78 ; MCU-NEXT: testb $1, %al 79 ; MCU-NEXT: jne .LBB1_2 80 ; MCU-NEXT: # %bb.1: # %entry 81 ; MCU-NEXT: movl $-480, %ecx # imm = 0xFE20 82 ; MCU-NEXT: .LBB1_2: # %entry 83 ; MCU-NEXT: shll $3, %ecx 84 ; MCU-NEXT: cmpl $32768, %ecx # imm = 0x8000 85 ; MCU-NEXT: jge .LBB1_3 86 ; MCU-NEXT: # %bb.4: # %bb91 87 ; MCU-NEXT: xorl %eax, %eax 88 ; MCU-NEXT: retl 89 ; MCU-NEXT: .LBB1_3: # %bb90 90 entry: 91 %tmp73 = tail call i1 @return_false() 92 %g.0 = select i1 %tmp73, i16 0, i16 -480 93 %tmp7778 = sext i16 %g.0 to i32 94 %tmp80 = shl i32 %tmp7778, 3 95 %tmp87 = icmp sgt i32 %tmp80, 32767 96 br i1 %tmp87, label %bb90, label %bb91 97 bb90: 98 unreachable 99 bb91: 100 ret i32 0 101 } 102 103 declare i1 @return_false() 104 105 ;; Select between two floating point constants. 106 define float @test3(i32 %x) nounwind readnone { 107 ; GENERIC-LABEL: test3: 108 ; GENERIC: ## %bb.0: ## %entry 109 ; GENERIC-NEXT: xorl %eax, %eax 110 ; GENERIC-NEXT: testl %edi, %edi 111 ; GENERIC-NEXT: sete %al 112 ; GENERIC-NEXT: leaq {{.*}}(%rip), %rcx 113 ; GENERIC-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 114 ; GENERIC-NEXT: retq 115 ; 116 ; ATOM-LABEL: test3: 117 ; ATOM: ## %bb.0: ## %entry 118 ; ATOM-NEXT: xorl %eax, %eax 119 ; ATOM-NEXT: leaq {{.*}}(%rip), %rcx 120 ; ATOM-NEXT: testl %edi, %edi 121 ; ATOM-NEXT: sete %al 122 ; ATOM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 123 ; ATOM-NEXT: retq 124 ; 125 ; MCU-LABEL: test3: 126 ; MCU: # %bb.0: # %entry 127 ; MCU-NEXT: xorl %ecx, %ecx 128 ; MCU-NEXT: testl %eax, %eax 129 ; MCU-NEXT: sete %cl 130 ; MCU-NEXT: flds {{\.LCPI.*}}(,%ecx,4) 131 ; MCU-NEXT: retl 132 entry: 133 %0 = icmp eq i32 %x, 0 134 %iftmp.0.0 = select i1 %0, float 4.200000e+01, float 2.300000e+01 135 ret float %iftmp.0.0 136 } 137 138 define signext i8 @test4(i8* nocapture %P, double %F) nounwind readonly { 139 ; CHECK-LABEL: test4: 140 ; CHECK: ## %bb.0: ## %entry 141 ; CHECK-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero 142 ; CHECK-NEXT: xorl %eax, %eax 143 ; CHECK-NEXT: ucomisd %xmm0, %xmm1 144 ; CHECK-NEXT: seta %al 145 ; CHECK-NEXT: movsbl (%rdi,%rax,4), %eax 146 ; CHECK-NEXT: retq 147 ; 148 ; MCU-LABEL: test4: 149 ; MCU: # %bb.0: # %entry 150 ; MCU-NEXT: movl %eax, %ecx 151 ; MCU-NEXT: fldl {{[0-9]+}}(%esp) 152 ; MCU-NEXT: flds {{\.LCPI.*}} 153 ; MCU-NEXT: fucompp 154 ; MCU-NEXT: fnstsw %ax 155 ; MCU-NEXT: xorl %edx, %edx 156 ; MCU-NEXT: # kill: def $ah killed $ah killed $ax 157 ; MCU-NEXT: sahf 158 ; MCU-NEXT: seta %dl 159 ; MCU-NEXT: movb (%ecx,%edx,4), %al 160 ; MCU-NEXT: retl 161 entry: 162 %0 = fcmp olt double %F, 4.200000e+01 163 %iftmp.0.0 = select i1 %0, i32 4, i32 0 164 %1 = getelementptr i8, i8* %P, i32 %iftmp.0.0 165 %2 = load i8, i8* %1, align 1 166 ret i8 %2 167 } 168 169 define void @test5(i1 %c, <2 x i16> %a, <2 x i16> %b, <2 x i16>* %p) nounwind { 170 ; CHECK-LABEL: test5: 171 ; CHECK: ## %bb.0: 172 ; CHECK-NEXT: testb $1, %dil 173 ; CHECK-NEXT: jne LBB4_2 174 ; CHECK-NEXT: ## %bb.1: 175 ; CHECK-NEXT: movdqa %xmm1, %xmm0 176 ; CHECK-NEXT: LBB4_2: 177 ; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] 178 ; CHECK-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7] 179 ; CHECK-NEXT: movd %xmm0, (%rsi) 180 ; CHECK-NEXT: retq 181 ; 182 ; MCU-LABEL: test5: 183 ; MCU: # %bb.0: 184 ; MCU-NEXT: pushl %esi 185 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %esi 186 ; MCU-NEXT: testb $1, %al 187 ; MCU-NEXT: jne .LBB4_2 188 ; MCU-NEXT: # %bb.1: 189 ; MCU-NEXT: movzwl {{[0-9]+}}(%esp), %ecx 190 ; MCU-NEXT: movzwl {{[0-9]+}}(%esp), %edx 191 ; MCU-NEXT: .LBB4_2: 192 ; MCU-NEXT: movw %cx, 2(%esi) 193 ; MCU-NEXT: movw %dx, (%esi) 194 ; MCU-NEXT: popl %esi 195 ; MCU-NEXT: retl 196 %x = select i1 %c, <2 x i16> %a, <2 x i16> %b 197 store <2 x i16> %x, <2 x i16>* %p 198 ret void 199 } 200 201 ; Verify that the fmul gets sunk into the one part of the diamond where it is needed. 202 define void @test6(i32 %C, <4 x float>* %A, <4 x float>* %B) nounwind { 203 ; CHECK-LABEL: test6: 204 ; CHECK: ## %bb.0: 205 ; CHECK-NEXT: testl %edi, %edi 206 ; CHECK-NEXT: je LBB5_1 207 ; CHECK-NEXT: ## %bb.2: 208 ; CHECK-NEXT: movaps (%rsi), %xmm0 209 ; CHECK-NEXT: movaps %xmm0, (%rsi) 210 ; CHECK-NEXT: retq 211 ; CHECK-NEXT: LBB5_1: 212 ; CHECK-NEXT: movaps (%rdx), %xmm0 213 ; CHECK-NEXT: mulps %xmm0, %xmm0 214 ; CHECK-NEXT: movaps %xmm0, (%rsi) 215 ; CHECK-NEXT: retq 216 ; 217 ; MCU-LABEL: test6: 218 ; MCU: # %bb.0: 219 ; MCU-NEXT: pushl %eax 220 ; MCU-NEXT: flds 12(%edx) 221 ; MCU-NEXT: fstps (%esp) # 4-byte Folded Spill 222 ; MCU-NEXT: flds 8(%edx) 223 ; MCU-NEXT: flds 4(%edx) 224 ; MCU-NEXT: flds (%ecx) 225 ; MCU-NEXT: flds 4(%ecx) 226 ; MCU-NEXT: flds 8(%ecx) 227 ; MCU-NEXT: flds 12(%ecx) 228 ; MCU-NEXT: fmul %st(0), %st(0) 229 ; MCU-NEXT: fxch %st(1) 230 ; MCU-NEXT: fmul %st(0), %st(0) 231 ; MCU-NEXT: fxch %st(2) 232 ; MCU-NEXT: fmul %st(0), %st(0) 233 ; MCU-NEXT: fxch %st(3) 234 ; MCU-NEXT: fmul %st(0), %st(0) 235 ; MCU-NEXT: testl %eax, %eax 236 ; MCU-NEXT: flds (%edx) 237 ; MCU-NEXT: je .LBB5_2 238 ; MCU-NEXT: # %bb.1: 239 ; MCU-NEXT: fstp %st(1) 240 ; MCU-NEXT: fstp %st(3) 241 ; MCU-NEXT: fstp %st(1) 242 ; MCU-NEXT: fstp %st(0) 243 ; MCU-NEXT: flds (%esp) # 4-byte Folded Reload 244 ; MCU-NEXT: fldz 245 ; MCU-NEXT: fldz 246 ; MCU-NEXT: fldz 247 ; MCU-NEXT: fxch %st(1) 248 ; MCU-NEXT: fxch %st(6) 249 ; MCU-NEXT: fxch %st(1) 250 ; MCU-NEXT: fxch %st(5) 251 ; MCU-NEXT: fxch %st(4) 252 ; MCU-NEXT: fxch %st(1) 253 ; MCU-NEXT: fxch %st(3) 254 ; MCU-NEXT: fxch %st(2) 255 ; MCU-NEXT: .LBB5_2: 256 ; MCU-NEXT: fstp %st(0) 257 ; MCU-NEXT: fstp %st(5) 258 ; MCU-NEXT: fstp %st(3) 259 ; MCU-NEXT: fxch %st(2) 260 ; MCU-NEXT: fstps 12(%edx) 261 ; MCU-NEXT: fxch %st(1) 262 ; MCU-NEXT: fstps 8(%edx) 263 ; MCU-NEXT: fstps 4(%edx) 264 ; MCU-NEXT: fstps (%edx) 265 ; MCU-NEXT: popl %eax 266 ; MCU-NEXT: retl 267 %tmp = load <4 x float>, <4 x float>* %A 268 %tmp3 = load <4 x float>, <4 x float>* %B 269 %tmp9 = fmul <4 x float> %tmp3, %tmp3 270 %tmp.upgrd.1 = icmp eq i32 %C, 0 271 %iftmp.38.0 = select i1 %tmp.upgrd.1, <4 x float> %tmp9, <4 x float> %tmp 272 store <4 x float> %iftmp.38.0, <4 x float>* %A 273 ret void 274 } 275 276 ; Select with fp80's 277 define x86_fp80 @test7(i32 %tmp8) nounwind { 278 ; GENERIC-LABEL: test7: 279 ; GENERIC: ## %bb.0: 280 ; GENERIC-NEXT: xorl %eax, %eax 281 ; GENERIC-NEXT: testl %edi, %edi 282 ; GENERIC-NEXT: setns %al 283 ; GENERIC-NEXT: shlq $4, %rax 284 ; GENERIC-NEXT: leaq {{.*}}(%rip), %rcx 285 ; GENERIC-NEXT: fldt (%rax,%rcx) 286 ; GENERIC-NEXT: retq 287 ; 288 ; ATOM-LABEL: test7: 289 ; ATOM: ## %bb.0: 290 ; ATOM-NEXT: xorl %eax, %eax 291 ; ATOM-NEXT: leaq {{.*}}(%rip), %rcx 292 ; ATOM-NEXT: testl %edi, %edi 293 ; ATOM-NEXT: setns %al 294 ; ATOM-NEXT: shlq $4, %rax 295 ; ATOM-NEXT: fldt (%rax,%rcx) 296 ; ATOM-NEXT: retq 297 ; 298 ; MCU-LABEL: test7: 299 ; MCU: # %bb.0: 300 ; MCU-NEXT: notl %eax 301 ; MCU-NEXT: shrl $27, %eax 302 ; MCU-NEXT: andl $-16, %eax 303 ; MCU-NEXT: fldt {{\.LCPI.*}}(%eax) 304 ; MCU-NEXT: retl 305 %tmp9 = icmp sgt i32 %tmp8, -1 306 %retval = select i1 %tmp9, x86_fp80 0xK4005B400000000000000, x86_fp80 0xK40078700000000000000 307 ret x86_fp80 %retval 308 } 309 310 ; widening select v6i32 and then a sub 311 define void @test8(i1 %c, <6 x i32>* %dst.addr, <6 x i32> %src1,<6 x i32> %src2) nounwind { 312 ; GENERIC-LABEL: test8: 313 ; GENERIC: ## %bb.0: 314 ; GENERIC-NEXT: testb $1, %dil 315 ; GENERIC-NEXT: jne LBB7_1 316 ; GENERIC-NEXT: ## %bb.2: 317 ; GENERIC-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 318 ; GENERIC-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero 319 ; GENERIC-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1] 320 ; GENERIC-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero 321 ; GENERIC-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 322 ; GENERIC-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1] 323 ; GENERIC-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 324 ; GENERIC-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero 325 ; GENERIC-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero 326 ; GENERIC-NEXT: jmp LBB7_3 327 ; GENERIC-NEXT: LBB7_1: 328 ; GENERIC-NEXT: movd %r9d, %xmm0 329 ; GENERIC-NEXT: movd %r8d, %xmm1 330 ; GENERIC-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1] 331 ; GENERIC-NEXT: movd %ecx, %xmm2 332 ; GENERIC-NEXT: movd %edx, %xmm0 333 ; GENERIC-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1] 334 ; GENERIC-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 335 ; GENERIC-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero 336 ; GENERIC-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero 337 ; GENERIC-NEXT: LBB7_3: 338 ; GENERIC-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1] 339 ; GENERIC-NEXT: pcmpeqd %xmm2, %xmm2 340 ; GENERIC-NEXT: paddd %xmm2, %xmm0 341 ; GENERIC-NEXT: paddd %xmm2, %xmm1 342 ; GENERIC-NEXT: movq %xmm1, 16(%rsi) 343 ; GENERIC-NEXT: movdqa %xmm0, (%rsi) 344 ; GENERIC-NEXT: retq 345 ; 346 ; ATOM-LABEL: test8: 347 ; ATOM: ## %bb.0: 348 ; ATOM-NEXT: testb $1, %dil 349 ; ATOM-NEXT: jne LBB7_1 350 ; ATOM-NEXT: ## %bb.2: 351 ; ATOM-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero 352 ; ATOM-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero 353 ; ATOM-NEXT: movd {{.*#+}} xmm3 = mem[0],zero,zero,zero 354 ; ATOM-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero 355 ; ATOM-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1] 356 ; ATOM-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1] 357 ; ATOM-NEXT: movd {{.*#+}} xmm3 = mem[0],zero,zero,zero 358 ; ATOM-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero 359 ; ATOM-NEXT: jmp LBB7_3 360 ; ATOM-NEXT: LBB7_1: 361 ; ATOM-NEXT: movd %r9d, %xmm1 362 ; ATOM-NEXT: movd %r8d, %xmm2 363 ; ATOM-NEXT: movd %ecx, %xmm3 364 ; ATOM-NEXT: movd %edx, %xmm0 365 ; ATOM-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1] 366 ; ATOM-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1] 367 ; ATOM-NEXT: movd {{.*#+}} xmm3 = mem[0],zero,zero,zero 368 ; ATOM-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero 369 ; ATOM-NEXT: LBB7_3: 370 ; ATOM-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0] 371 ; ATOM-NEXT: pcmpeqd %xmm2, %xmm2 372 ; ATOM-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1] 373 ; ATOM-NEXT: paddd %xmm2, %xmm0 374 ; ATOM-NEXT: paddd %xmm2, %xmm1 375 ; ATOM-NEXT: movq %xmm1, 16(%rsi) 376 ; ATOM-NEXT: movdqa %xmm0, (%rsi) 377 ; ATOM-NEXT: retq 378 ; 379 ; MCU-LABEL: test8: 380 ; MCU: # %bb.0: 381 ; MCU-NEXT: pushl %ebp 382 ; MCU-NEXT: pushl %ebx 383 ; MCU-NEXT: pushl %edi 384 ; MCU-NEXT: pushl %esi 385 ; MCU-NEXT: testb $1, %al 386 ; MCU-NEXT: jne .LBB7_1 387 ; MCU-NEXT: # %bb.2: 388 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %eax 389 ; MCU-NEXT: movl (%eax), %eax 390 ; MCU-NEXT: je .LBB7_5 391 ; MCU-NEXT: .LBB7_4: 392 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %ecx 393 ; MCU-NEXT: movl (%ecx), %ecx 394 ; MCU-NEXT: je .LBB7_8 395 ; MCU-NEXT: .LBB7_7: 396 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %esi 397 ; MCU-NEXT: movl (%esi), %esi 398 ; MCU-NEXT: je .LBB7_11 399 ; MCU-NEXT: .LBB7_10: 400 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %edi 401 ; MCU-NEXT: movl (%edi), %edi 402 ; MCU-NEXT: je .LBB7_14 403 ; MCU-NEXT: .LBB7_13: 404 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %ebx 405 ; MCU-NEXT: movl (%ebx), %ebx 406 ; MCU-NEXT: je .LBB7_17 407 ; MCU-NEXT: .LBB7_16: 408 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %ebp 409 ; MCU-NEXT: jmp .LBB7_18 410 ; MCU-NEXT: .LBB7_1: 411 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %eax 412 ; MCU-NEXT: movl (%eax), %eax 413 ; MCU-NEXT: jne .LBB7_4 414 ; MCU-NEXT: .LBB7_5: 415 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %ecx 416 ; MCU-NEXT: movl (%ecx), %ecx 417 ; MCU-NEXT: jne .LBB7_7 418 ; MCU-NEXT: .LBB7_8: 419 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %esi 420 ; MCU-NEXT: movl (%esi), %esi 421 ; MCU-NEXT: jne .LBB7_10 422 ; MCU-NEXT: .LBB7_11: 423 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %edi 424 ; MCU-NEXT: movl (%edi), %edi 425 ; MCU-NEXT: jne .LBB7_13 426 ; MCU-NEXT: .LBB7_14: 427 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %ebx 428 ; MCU-NEXT: movl (%ebx), %ebx 429 ; MCU-NEXT: jne .LBB7_16 430 ; MCU-NEXT: .LBB7_17: 431 ; MCU-NEXT: leal {{[0-9]+}}(%esp), %ebp 432 ; MCU-NEXT: .LBB7_18: 433 ; MCU-NEXT: movl (%ebp), %ebp 434 ; MCU-NEXT: decl %ebp 435 ; MCU-NEXT: decl %ebx 436 ; MCU-NEXT: decl %edi 437 ; MCU-NEXT: decl %esi 438 ; MCU-NEXT: decl %ecx 439 ; MCU-NEXT: decl %eax 440 ; MCU-NEXT: movl %eax, 20(%edx) 441 ; MCU-NEXT: movl %ecx, 16(%edx) 442 ; MCU-NEXT: movl %esi, 12(%edx) 443 ; MCU-NEXT: movl %edi, 8(%edx) 444 ; MCU-NEXT: movl %ebx, 4(%edx) 445 ; MCU-NEXT: movl %ebp, (%edx) 446 ; MCU-NEXT: popl %esi 447 ; MCU-NEXT: popl %edi 448 ; MCU-NEXT: popl %ebx 449 ; MCU-NEXT: popl %ebp 450 ; MCU-NEXT: retl 451 %x = select i1 %c, <6 x i32> %src1, <6 x i32> %src2 452 %val = sub <6 x i32> %x, < i32 1, i32 1, i32 1, i32 1, i32 1, i32 1 > 453 store <6 x i32> %val, <6 x i32>* %dst.addr 454 ret void 455 } 456 457 458 ;; Test integer select between values and constants. 459 460 define i64 @test9(i64 %x, i64 %y) nounwind readnone ssp noredzone { 461 ; GENERIC-LABEL: test9: 462 ; GENERIC: ## %bb.0: 463 ; GENERIC-NEXT: cmpq $1, %rdi 464 ; GENERIC-NEXT: sbbq %rax, %rax 465 ; GENERIC-NEXT: orq %rsi, %rax 466 ; GENERIC-NEXT: retq 467 ; 468 ; ATOM-LABEL: test9: 469 ; ATOM: ## %bb.0: 470 ; ATOM-NEXT: cmpq $1, %rdi 471 ; ATOM-NEXT: sbbq %rax, %rax 472 ; ATOM-NEXT: orq %rsi, %rax 473 ; ATOM-NEXT: nop 474 ; ATOM-NEXT: nop 475 ; ATOM-NEXT: retq 476 ; 477 ; MCU-LABEL: test9: 478 ; MCU: # %bb.0: 479 ; MCU-NEXT: orl %edx, %eax 480 ; MCU-NEXT: jne .LBB8_1 481 ; MCU-NEXT: # %bb.2: 482 ; MCU-NEXT: movl $-1, %eax 483 ; MCU-NEXT: movl $-1, %edx 484 ; MCU-NEXT: retl 485 ; MCU-NEXT: .LBB8_1: 486 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %eax 487 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %edx 488 ; MCU-NEXT: retl 489 %cmp = icmp ne i64 %x, 0 490 %cond = select i1 %cmp, i64 %y, i64 -1 491 ret i64 %cond 492 } 493 494 ;; Same as test9 495 define i64 @test9a(i64 %x, i64 %y) nounwind readnone ssp noredzone { 496 ; GENERIC-LABEL: test9a: 497 ; GENERIC: ## %bb.0: 498 ; GENERIC-NEXT: cmpq $1, %rdi 499 ; GENERIC-NEXT: sbbq %rax, %rax 500 ; GENERIC-NEXT: orq %rsi, %rax 501 ; GENERIC-NEXT: retq 502 ; 503 ; ATOM-LABEL: test9a: 504 ; ATOM: ## %bb.0: 505 ; ATOM-NEXT: cmpq $1, %rdi 506 ; ATOM-NEXT: sbbq %rax, %rax 507 ; ATOM-NEXT: orq %rsi, %rax 508 ; ATOM-NEXT: nop 509 ; ATOM-NEXT: nop 510 ; ATOM-NEXT: retq 511 ; 512 ; MCU-LABEL: test9a: 513 ; MCU: # %bb.0: 514 ; MCU-NEXT: orl %edx, %eax 515 ; MCU-NEXT: movl $-1, %eax 516 ; MCU-NEXT: movl $-1, %edx 517 ; MCU-NEXT: je .LBB9_2 518 ; MCU-NEXT: # %bb.1: 519 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %eax 520 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %edx 521 ; MCU-NEXT: .LBB9_2: 522 ; MCU-NEXT: retl 523 %cmp = icmp eq i64 %x, 0 524 %cond = select i1 %cmp, i64 -1, i64 %y 525 ret i64 %cond 526 } 527 528 define i64 @test9b(i64 %x, i64 %y) nounwind readnone ssp noredzone { 529 ; GENERIC-LABEL: test9b: 530 ; GENERIC: ## %bb.0: 531 ; GENERIC-NEXT: cmpq $1, %rdi 532 ; GENERIC-NEXT: sbbq %rax, %rax 533 ; GENERIC-NEXT: orq %rsi, %rax 534 ; GENERIC-NEXT: retq 535 ; 536 ; ATOM-LABEL: test9b: 537 ; ATOM: ## %bb.0: 538 ; ATOM-NEXT: cmpq $1, %rdi 539 ; ATOM-NEXT: sbbq %rax, %rax 540 ; ATOM-NEXT: orq %rsi, %rax 541 ; ATOM-NEXT: nop 542 ; ATOM-NEXT: nop 543 ; ATOM-NEXT: retq 544 ; 545 ; MCU-LABEL: test9b: 546 ; MCU: # %bb.0: 547 ; MCU-NEXT: movl %edx, %ecx 548 ; MCU-NEXT: xorl %edx, %edx 549 ; MCU-NEXT: orl %ecx, %eax 550 ; MCU-NEXT: sete %dl 551 ; MCU-NEXT: negl %edx 552 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %eax 553 ; MCU-NEXT: orl %edx, %eax 554 ; MCU-NEXT: orl {{[0-9]+}}(%esp), %edx 555 ; MCU-NEXT: retl 556 %cmp = icmp eq i64 %x, 0 557 %A = sext i1 %cmp to i64 558 %cond = or i64 %y, %A 559 ret i64 %cond 560 } 561 562 ;; Select between -1 and 1. 563 define i64 @test10(i64 %x, i64 %y) nounwind readnone ssp noredzone { 564 ; CHECK-LABEL: test10: 565 ; CHECK: ## %bb.0: 566 ; CHECK-NEXT: xorl %eax, %eax 567 ; CHECK-NEXT: testq %rdi, %rdi 568 ; CHECK-NEXT: setne %al 569 ; CHECK-NEXT: leaq -1(%rax,%rax), %rax 570 ; CHECK-NEXT: retq 571 ; 572 ; MCU-LABEL: test10: 573 ; MCU: # %bb.0: 574 ; MCU-NEXT: orl %edx, %eax 575 ; MCU-NEXT: movl $-1, %eax 576 ; MCU-NEXT: movl $-1, %edx 577 ; MCU-NEXT: je .LBB11_2 578 ; MCU-NEXT: # %bb.1: 579 ; MCU-NEXT: xorl %edx, %edx 580 ; MCU-NEXT: movl $1, %eax 581 ; MCU-NEXT: .LBB11_2: 582 ; MCU-NEXT: retl 583 %cmp = icmp eq i64 %x, 0 584 %cond = select i1 %cmp, i64 -1, i64 1 585 ret i64 %cond 586 } 587 588 define i64 @test11(i64 %x, i64 %y) nounwind readnone ssp noredzone { 589 ; CHECK-LABEL: test11: 590 ; CHECK: ## %bb.0: 591 ; CHECK-NEXT: cmpq $1, %rdi 592 ; CHECK-NEXT: sbbq %rax, %rax 593 ; CHECK-NEXT: notq %rax 594 ; CHECK-NEXT: orq %rsi, %rax 595 ; CHECK-NEXT: retq 596 ; 597 ; MCU-LABEL: test11: 598 ; MCU: # %bb.0: 599 ; MCU-NEXT: orl %edx, %eax 600 ; MCU-NEXT: je .LBB12_1 601 ; MCU-NEXT: # %bb.2: 602 ; MCU-NEXT: movl $-1, %eax 603 ; MCU-NEXT: movl $-1, %edx 604 ; MCU-NEXT: retl 605 ; MCU-NEXT: .LBB12_1: 606 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %eax 607 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %edx 608 ; MCU-NEXT: retl 609 %cmp = icmp eq i64 %x, 0 610 %cond = select i1 %cmp, i64 %y, i64 -1 611 ret i64 %cond 612 } 613 614 define i64 @test11a(i64 %x, i64 %y) nounwind readnone ssp noredzone { 615 ; CHECK-LABEL: test11a: 616 ; CHECK: ## %bb.0: 617 ; CHECK-NEXT: cmpq $1, %rdi 618 ; CHECK-NEXT: sbbq %rax, %rax 619 ; CHECK-NEXT: notq %rax 620 ; CHECK-NEXT: orq %rsi, %rax 621 ; CHECK-NEXT: retq 622 ; 623 ; MCU-LABEL: test11a: 624 ; MCU: # %bb.0: 625 ; MCU-NEXT: orl %edx, %eax 626 ; MCU-NEXT: movl $-1, %eax 627 ; MCU-NEXT: movl $-1, %edx 628 ; MCU-NEXT: jne .LBB13_2 629 ; MCU-NEXT: # %bb.1: 630 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %eax 631 ; MCU-NEXT: movl {{[0-9]+}}(%esp), %edx 632 ; MCU-NEXT: .LBB13_2: 633 ; MCU-NEXT: retl 634 %cmp = icmp ne i64 %x, 0 635 %cond = select i1 %cmp, i64 -1, i64 %y 636 ret i64 %cond 637 } 638 639 640 declare noalias i8* @_Znam(i64) noredzone 641 642 define noalias i8* @test12(i64 %count) nounwind ssp noredzone { 643 ; GENERIC-LABEL: test12: 644 ; GENERIC: ## %bb.0: ## %entry 645 ; GENERIC-NEXT: movl $4, %ecx 646 ; GENERIC-NEXT: movq %rdi, %rax 647 ; GENERIC-NEXT: mulq %rcx 648 ; GENERIC-NEXT: movq $-1, %rdi 649 ; GENERIC-NEXT: cmovnoq %rax, %rdi 650 ; GENERIC-NEXT: jmp __Znam ## TAILCALL 651 ; 652 ; ATOM-LABEL: test12: 653 ; ATOM: ## %bb.0: ## %entry 654 ; ATOM-NEXT: movq %rdi, %rax 655 ; ATOM-NEXT: movl $4, %ecx 656 ; ATOM-NEXT: movq $-1, %rdi 657 ; ATOM-NEXT: mulq %rcx 658 ; ATOM-NEXT: cmovnoq %rax, %rdi 659 ; ATOM-NEXT: jmp __Znam ## TAILCALL 660 ; 661 ; MCU-LABEL: test12: 662 ; MCU: # %bb.0: # %entry 663 ; MCU-NEXT: pushl %ebp 664 ; MCU-NEXT: pushl %ebx 665 ; MCU-NEXT: pushl %edi 666 ; MCU-NEXT: pushl %esi 667 ; MCU-NEXT: movl %edx, %ebx 668 ; MCU-NEXT: movl %eax, %ebp 669 ; MCU-NEXT: movl $4, %ecx 670 ; MCU-NEXT: mull %ecx 671 ; MCU-NEXT: movl %eax, %esi 672 ; MCU-NEXT: leal (%edx,%ebx,4), %edi 673 ; MCU-NEXT: movl %edi, %edx 674 ; MCU-NEXT: pushl $0 675 ; MCU-NEXT: pushl $4 676 ; MCU-NEXT: calll __udivdi3 677 ; MCU-NEXT: addl $8, %esp 678 ; MCU-NEXT: xorl %ebx, %edx 679 ; MCU-NEXT: xorl %ebp, %eax 680 ; MCU-NEXT: orl %edx, %eax 681 ; MCU-NEXT: movl $-1, %eax 682 ; MCU-NEXT: movl $-1, %edx 683 ; MCU-NEXT: jne .LBB14_2 684 ; MCU-NEXT: # %bb.1: # %entry 685 ; MCU-NEXT: movl %esi, %eax 686 ; MCU-NEXT: movl %edi, %edx 687 ; MCU-NEXT: .LBB14_2: # %entry 688 ; MCU-NEXT: popl %esi 689 ; MCU-NEXT: popl %edi 690 ; MCU-NEXT: popl %ebx 691 ; MCU-NEXT: popl %ebp 692 ; MCU-NEXT: jmp _Znam # TAILCALL 693 entry: 694 %A = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %count, i64 4) 695 %B = extractvalue { i64, i1 } %A, 1 696 %C = extractvalue { i64, i1 } %A, 0 697 %D = select i1 %B, i64 -1, i64 %C 698 %call = tail call noalias i8* @_Znam(i64 %D) nounwind noredzone 699 ret i8* %call 700 } 701 702 declare { i64, i1 } @llvm.umul.with.overflow.i64(i64, i64) nounwind readnone 703 704 define i32 @test13(i32 %a, i32 %b) nounwind { 705 ; GENERIC-LABEL: test13: 706 ; GENERIC: ## %bb.0: 707 ; GENERIC-NEXT: cmpl %esi, %edi 708 ; GENERIC-NEXT: sbbl %eax, %eax 709 ; GENERIC-NEXT: retq 710 ; 711 ; ATOM-LABEL: test13: 712 ; ATOM: ## %bb.0: 713 ; ATOM-NEXT: cmpl %esi, %edi 714 ; ATOM-NEXT: sbbl %eax, %eax 715 ; ATOM-NEXT: nop 716 ; ATOM-NEXT: nop 717 ; ATOM-NEXT: nop 718 ; ATOM-NEXT: nop 719 ; ATOM-NEXT: retq 720 ; 721 ; MCU-LABEL: test13: 722 ; MCU: # %bb.0: 723 ; MCU-NEXT: cmpl %edx, %eax 724 ; MCU-NEXT: sbbl %eax, %eax 725 ; MCU-NEXT: retl 726 %c = icmp ult i32 %a, %b 727 %d = sext i1 %c to i32 728 ret i32 %d 729 } 730 731 define i32 @test14(i32 %a, i32 %b) nounwind { 732 ; CHECK-LABEL: test14: 733 ; CHECK: ## %bb.0: 734 ; CHECK-NEXT: xorl %eax, %eax 735 ; CHECK-NEXT: cmpl %esi, %edi 736 ; CHECK-NEXT: setae %al 737 ; CHECK-NEXT: negl %eax 738 ; CHECK-NEXT: retq 739 ; 740 ; MCU-LABEL: test14: 741 ; MCU: # %bb.0: 742 ; MCU-NEXT: xorl %ecx, %ecx 743 ; MCU-NEXT: cmpl %edx, %eax 744 ; MCU-NEXT: setae %cl 745 ; MCU-NEXT: negl %ecx 746 ; MCU-NEXT: movl %ecx, %eax 747 ; MCU-NEXT: retl 748 %c = icmp uge i32 %a, %b 749 %d = sext i1 %c to i32 750 ret i32 %d 751 } 752 753 ; rdar://10961709 754 define i32 @test15(i32 %x) nounwind { 755 ; GENERIC-LABEL: test15: 756 ; GENERIC: ## %bb.0: ## %entry 757 ; GENERIC-NEXT: negl %edi 758 ; GENERIC-NEXT: sbbl %eax, %eax 759 ; GENERIC-NEXT: retq 760 ; 761 ; ATOM-LABEL: test15: 762 ; ATOM: ## %bb.0: ## %entry 763 ; ATOM-NEXT: negl %edi 764 ; ATOM-NEXT: sbbl %eax, %eax 765 ; ATOM-NEXT: nop 766 ; ATOM-NEXT: nop 767 ; ATOM-NEXT: nop 768 ; ATOM-NEXT: nop 769 ; ATOM-NEXT: retq 770 ; 771 ; MCU-LABEL: test15: 772 ; MCU: # %bb.0: # %entry 773 ; MCU-NEXT: negl %eax 774 ; MCU-NEXT: sbbl %eax, %eax 775 ; MCU-NEXT: retl 776 entry: 777 %cmp = icmp ne i32 %x, 0 778 %sub = sext i1 %cmp to i32 779 ret i32 %sub 780 } 781 782 define i64 @test16(i64 %x) nounwind uwtable readnone ssp { 783 ; GENERIC-LABEL: test16: 784 ; GENERIC: ## %bb.0: ## %entry 785 ; GENERIC-NEXT: negq %rdi 786 ; GENERIC-NEXT: sbbq %rax, %rax 787 ; GENERIC-NEXT: retq 788 ; 789 ; ATOM-LABEL: test16: 790 ; ATOM: ## %bb.0: ## %entry 791 ; ATOM-NEXT: negq %rdi 792 ; ATOM-NEXT: sbbq %rax, %rax 793 ; ATOM-NEXT: nop 794 ; ATOM-NEXT: nop 795 ; ATOM-NEXT: nop 796 ; ATOM-NEXT: nop 797 ; ATOM-NEXT: retq 798 ; 799 ; MCU-LABEL: test16: 800 ; MCU: # %bb.0: # %entry 801 ; MCU-NEXT: movl %eax, %ecx 802 ; MCU-NEXT: xorl %eax, %eax 803 ; MCU-NEXT: orl %edx, %ecx 804 ; MCU-NEXT: setne %al 805 ; MCU-NEXT: negl %eax 806 ; MCU-NEXT: movl %eax, %edx 807 ; MCU-NEXT: retl 808 entry: 809 %cmp = icmp ne i64 %x, 0 810 %conv1 = sext i1 %cmp to i64 811 ret i64 %conv1 812 } 813 814 define i16 @test17(i16 %x) nounwind { 815 ; GENERIC-LABEL: test17: 816 ; GENERIC: ## %bb.0: ## %entry 817 ; GENERIC-NEXT: negw %di 818 ; GENERIC-NEXT: sbbl %eax, %eax 819 ; GENERIC-NEXT: ## kill: def $ax killed $ax killed $eax 820 ; GENERIC-NEXT: retq 821 ; 822 ; ATOM-LABEL: test17: 823 ; ATOM: ## %bb.0: ## %entry 824 ; ATOM-NEXT: negw %di 825 ; ATOM-NEXT: sbbl %eax, %eax 826 ; ATOM-NEXT: ## kill: def $ax killed $ax killed $eax 827 ; ATOM-NEXT: nop 828 ; ATOM-NEXT: nop 829 ; ATOM-NEXT: nop 830 ; ATOM-NEXT: nop 831 ; ATOM-NEXT: retq 832 ; 833 ; MCU-LABEL: test17: 834 ; MCU: # %bb.0: # %entry 835 ; MCU-NEXT: negw %ax 836 ; MCU-NEXT: sbbl %eax, %eax 837 ; MCU-NEXT: # kill: def $ax killed $ax killed $eax 838 ; MCU-NEXT: retl 839 entry: 840 %cmp = icmp ne i16 %x, 0 841 %sub = sext i1 %cmp to i16 842 ret i16 %sub 843 } 844 845 define i8 @test18(i32 %x, i8 zeroext %a, i8 zeroext %b) nounwind { 846 ; GENERIC-LABEL: test18: 847 ; GENERIC: ## %bb.0: 848 ; GENERIC-NEXT: cmpl $15, %edi 849 ; GENERIC-NEXT: cmovgel %edx, %esi 850 ; GENERIC-NEXT: movl %esi, %eax 851 ; GENERIC-NEXT: retq 852 ; 853 ; ATOM-LABEL: test18: 854 ; ATOM: ## %bb.0: 855 ; ATOM-NEXT: cmpl $15, %edi 856 ; ATOM-NEXT: cmovgel %edx, %esi 857 ; ATOM-NEXT: movl %esi, %eax 858 ; ATOM-NEXT: nop 859 ; ATOM-NEXT: nop 860 ; ATOM-NEXT: retq 861 ; 862 ; MCU-LABEL: test18: 863 ; MCU: # %bb.0: 864 ; MCU-NEXT: cmpl $15, %eax 865 ; MCU-NEXT: jl .LBB20_2 866 ; MCU-NEXT: # %bb.1: 867 ; MCU-NEXT: movl %ecx, %edx 868 ; MCU-NEXT: .LBB20_2: 869 ; MCU-NEXT: movl %edx, %eax 870 ; MCU-NEXT: retl 871 %cmp = icmp slt i32 %x, 15 872 %sel = select i1 %cmp, i8 %a, i8 %b 873 ret i8 %sel 874 } 875 876 define i32 @trunc_select_miscompile(i32 %a, i1 zeroext %cc) { 877 ; CHECK-LABEL: trunc_select_miscompile: 878 ; CHECK: ## %bb.0: 879 ; CHECK-NEXT: orb $2, %sil 880 ; CHECK-NEXT: movl %esi, %ecx 881 ; CHECK-NEXT: shll %cl, %edi 882 ; CHECK-NEXT: movl %edi, %eax 883 ; CHECK-NEXT: retq 884 ; 885 ; MCU-LABEL: trunc_select_miscompile: 886 ; MCU: # %bb.0: 887 ; MCU-NEXT: orb $2, %dl 888 ; MCU-NEXT: movl %edx, %ecx 889 ; MCU-NEXT: shll %cl, %eax 890 ; MCU-NEXT: retl 891 %tmp1 = select i1 %cc, i32 3, i32 2 892 %tmp2 = shl i32 %a, %tmp1 893 ret i32 %tmp2 894 } 895 896 ; reproducer for pr29002 897 define void @clamp_i8(i32 %src, i8* %dst) { 898 ; GENERIC-LABEL: clamp_i8: 899 ; GENERIC: ## %bb.0: 900 ; GENERIC-NEXT: cmpl $127, %edi 901 ; GENERIC-NEXT: movl $127, %eax 902 ; GENERIC-NEXT: cmovlel %edi, %eax 903 ; GENERIC-NEXT: cmpl $-128, %eax 904 ; GENERIC-NEXT: movb $-128, %cl 905 ; GENERIC-NEXT: jl LBB22_2 906 ; GENERIC-NEXT: ## %bb.1: 907 ; GENERIC-NEXT: movl %eax, %ecx 908 ; GENERIC-NEXT: LBB22_2: 909 ; GENERIC-NEXT: movb %cl, (%rsi) 910 ; GENERIC-NEXT: retq 911 ; 912 ; ATOM-LABEL: clamp_i8: 913 ; ATOM: ## %bb.0: 914 ; ATOM-NEXT: cmpl $127, %edi 915 ; ATOM-NEXT: movl $127, %eax 916 ; ATOM-NEXT: movb $-128, %cl 917 ; ATOM-NEXT: cmovlel %edi, %eax 918 ; ATOM-NEXT: cmpl $-128, %eax 919 ; ATOM-NEXT: jl LBB22_2 920 ; ATOM-NEXT: ## %bb.1: 921 ; ATOM-NEXT: movl %eax, %ecx 922 ; ATOM-NEXT: LBB22_2: 923 ; ATOM-NEXT: movb %cl, (%rsi) 924 ; ATOM-NEXT: retq 925 ; 926 ; MCU-LABEL: clamp_i8: 927 ; MCU: # %bb.0: 928 ; MCU-NEXT: cmpl $127, %eax 929 ; MCU-NEXT: movl $127, %ecx 930 ; MCU-NEXT: jg .LBB22_2 931 ; MCU-NEXT: # %bb.1: 932 ; MCU-NEXT: movl %eax, %ecx 933 ; MCU-NEXT: .LBB22_2: 934 ; MCU-NEXT: cmpl $-128, %ecx 935 ; MCU-NEXT: movb $-128, %al 936 ; MCU-NEXT: jl .LBB22_4 937 ; MCU-NEXT: # %bb.3: 938 ; MCU-NEXT: movl %ecx, %eax 939 ; MCU-NEXT: .LBB22_4: 940 ; MCU-NEXT: movb %al, (%edx) 941 ; MCU-NEXT: retl 942 %cmp = icmp sgt i32 %src, 127 943 %sel1 = select i1 %cmp, i32 127, i32 %src 944 %cmp1 = icmp slt i32 %sel1, -128 945 %sel2 = select i1 %cmp1, i32 -128, i32 %sel1 946 %conv = trunc i32 %sel2 to i8 947 store i8 %conv, i8* %dst, align 2 948 ret void 949 } 950 951 ; reproducer for pr29002 952 define void @clamp(i32 %src, i16* %dst) { 953 ; GENERIC-LABEL: clamp: 954 ; GENERIC: ## %bb.0: 955 ; GENERIC-NEXT: cmpl $32767, %edi ## imm = 0x7FFF 956 ; GENERIC-NEXT: movl $32767, %eax ## imm = 0x7FFF 957 ; GENERIC-NEXT: cmovlel %edi, %eax 958 ; GENERIC-NEXT: cmpl $-32768, %eax ## imm = 0x8000 959 ; GENERIC-NEXT: movl $32768, %ecx ## imm = 0x8000 960 ; GENERIC-NEXT: cmovgel %eax, %ecx 961 ; GENERIC-NEXT: movw %cx, (%rsi) 962 ; GENERIC-NEXT: retq 963 ; 964 ; ATOM-LABEL: clamp: 965 ; ATOM: ## %bb.0: 966 ; ATOM-NEXT: cmpl $32767, %edi ## imm = 0x7FFF 967 ; ATOM-NEXT: movl $32767, %eax ## imm = 0x7FFF 968 ; ATOM-NEXT: movl $32768, %ecx ## imm = 0x8000 969 ; ATOM-NEXT: cmovlel %edi, %eax 970 ; ATOM-NEXT: cmpl $-32768, %eax ## imm = 0x8000 971 ; ATOM-NEXT: cmovgel %eax, %ecx 972 ; ATOM-NEXT: movw %cx, (%rsi) 973 ; ATOM-NEXT: retq 974 ; 975 ; MCU-LABEL: clamp: 976 ; MCU: # %bb.0: 977 ; MCU-NEXT: cmpl $32767, %eax # imm = 0x7FFF 978 ; MCU-NEXT: movl $32767, %ecx # imm = 0x7FFF 979 ; MCU-NEXT: jg .LBB23_2 980 ; MCU-NEXT: # %bb.1: 981 ; MCU-NEXT: movl %eax, %ecx 982 ; MCU-NEXT: .LBB23_2: 983 ; MCU-NEXT: cmpl $-32768, %ecx # imm = 0x8000 984 ; MCU-NEXT: movl $32768, %eax # imm = 0x8000 985 ; MCU-NEXT: jl .LBB23_4 986 ; MCU-NEXT: # %bb.3: 987 ; MCU-NEXT: movl %ecx, %eax 988 ; MCU-NEXT: .LBB23_4: 989 ; MCU-NEXT: movw %ax, (%edx) 990 ; MCU-NEXT: retl 991 %cmp = icmp sgt i32 %src, 32767 992 %sel1 = select i1 %cmp, i32 32767, i32 %src 993 %cmp1 = icmp slt i32 %sel1, -32768 994 %sel2 = select i1 %cmp1, i32 -32768, i32 %sel1 995 %conv = trunc i32 %sel2 to i16 996 store i16 %conv, i16* %dst, align 2 997 ret void 998 } 999 1000 define void @test19() { 1001 ; This is a massive reduction of an llvm-stress test case that generates 1002 ; interesting chains feeding setcc and eventually a f32 select operation. This 1003 ; is intended to exercise the SELECT formation in the DAG combine simplifying 1004 ; a simplified select_cc node. If it it regresses and is no longer triggering 1005 ; that code path, it can be deleted. 1006 ; 1007 ; CHECK-LABEL: test19: 1008 ; CHECK: ## %bb.0: ## %BB 1009 ; CHECK-NEXT: movl $-1, %eax 1010 ; CHECK-NEXT: movb $1, %cl 1011 ; CHECK-NEXT: .p2align 4, 0x90 1012 ; CHECK-NEXT: LBB24_1: ## %CF 1013 ; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1 1014 ; CHECK-NEXT: testb %cl, %cl 1015 ; CHECK-NEXT: jne LBB24_1 1016 ; CHECK-NEXT: ## %bb.2: ## %CF250 1017 ; CHECK-NEXT: ## in Loop: Header=BB24_1 Depth=1 1018 ; CHECK-NEXT: jne LBB24_1 1019 ; CHECK-NEXT: .p2align 4, 0x90 1020 ; CHECK-NEXT: LBB24_3: ## %CF242 1021 ; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1 1022 ; CHECK-NEXT: cmpl %eax, %eax 1023 ; CHECK-NEXT: ucomiss %xmm0, %xmm0 1024 ; CHECK-NEXT: jp LBB24_3 1025 ; CHECK-NEXT: ## %bb.4: ## %CF244 1026 ; CHECK-NEXT: retq 1027 ; 1028 ; MCU-LABEL: test19: 1029 ; MCU: # %bb.0: # %BB 1030 ; MCU-NEXT: movl $-1, %ecx 1031 ; MCU-NEXT: movb $1, %al 1032 ; MCU-NEXT: .p2align 4, 0x90 1033 ; MCU-NEXT: .LBB24_1: # %CF 1034 ; MCU-NEXT: # =>This Inner Loop Header: Depth=1 1035 ; MCU-NEXT: testb %al, %al 1036 ; MCU-NEXT: jne .LBB24_1 1037 ; MCU-NEXT: # %bb.2: # %CF250 1038 ; MCU-NEXT: # in Loop: Header=BB24_1 Depth=1 1039 ; MCU-NEXT: jne .LBB24_1 1040 ; MCU-NEXT: # %bb.3: # %CF242.preheader 1041 ; MCU-NEXT: fldz 1042 ; MCU-NEXT: .p2align 4, 0x90 1043 ; MCU-NEXT: .LBB24_4: # %CF242 1044 ; MCU-NEXT: # =>This Inner Loop Header: Depth=1 1045 ; MCU-NEXT: cmpl %eax, %ecx 1046 ; MCU-NEXT: fucom %st(0) 1047 ; MCU-NEXT: fnstsw %ax 1048 ; MCU-NEXT: # kill: def $ah killed $ah killed $ax 1049 ; MCU-NEXT: sahf 1050 ; MCU-NEXT: jp .LBB24_4 1051 ; MCU-NEXT: # %bb.5: # %CF244 1052 ; MCU-NEXT: fstp %st(0) 1053 ; MCU-NEXT: retl 1054 BB: 1055 br label %CF 1056 1057 CF: 1058 %Cmp10 = icmp ule i8 undef, undef 1059 br i1 %Cmp10, label %CF, label %CF250 1060 1061 CF250: 1062 %E12 = extractelement <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, i32 2 1063 %Cmp32 = icmp ugt i1 %Cmp10, false 1064 br i1 %Cmp32, label %CF, label %CF242 1065 1066 CF242: 1067 %Cmp38 = icmp uge i32 %E12, undef 1068 %FC = uitofp i1 %Cmp38 to float 1069 %Sl59 = select i1 %Cmp32, float %FC, float undef 1070 %Cmp60 = fcmp ugt float undef, undef 1071 br i1 %Cmp60, label %CF242, label %CF244 1072 1073 CF244: 1074 %B122 = fadd float %Sl59, undef 1075 ret void 1076 } 1077 1078 define i16 @select_xor_1(i16 %A, i8 %cond) { 1079 ; CHECK-LABEL: select_xor_1: 1080 ; CHECK: ## %bb.0: ## %entry 1081 ; CHECK-NEXT: movl %edi, %eax 1082 ; CHECK-NEXT: xorl $43, %eax 1083 ; CHECK-NEXT: testb $1, %sil 1084 ; CHECK-NEXT: cmovel %edi, %eax 1085 ; CHECK-NEXT: ## kill: def $ax killed $ax killed $eax 1086 ; CHECK-NEXT: retq 1087 ; 1088 ; MCU-LABEL: select_xor_1: 1089 ; MCU: # %bb.0: # %entry 1090 ; MCU-NEXT: andl $1, %edx 1091 ; MCU-NEXT: negl %edx 1092 ; MCU-NEXT: andl $43, %edx 1093 ; MCU-NEXT: xorl %edx, %eax 1094 ; MCU-NEXT: # kill: def $ax killed $ax killed $eax 1095 ; MCU-NEXT: retl 1096 entry: 1097 %and = and i8 %cond, 1 1098 %cmp10 = icmp eq i8 %and, 0 1099 %0 = xor i16 %A, 43 1100 %1 = select i1 %cmp10, i16 %A, i16 %0 1101 ret i16 %1 1102 } 1103 1104 ; Equivalent to above, but with icmp ne (and %cond, 1), 1 instead of 1105 ; icmp eq (and %cond, 1), 0 1106 define i16 @select_xor_1b(i16 %A, i8 %cond) { 1107 ; CHECK-LABEL: select_xor_1b: 1108 ; CHECK: ## %bb.0: ## %entry 1109 ; CHECK-NEXT: movl %edi, %eax 1110 ; CHECK-NEXT: xorl $43, %eax 1111 ; CHECK-NEXT: testb $1, %sil 1112 ; CHECK-NEXT: cmovel %edi, %eax 1113 ; CHECK-NEXT: ## kill: def $ax killed $ax killed $eax 1114 ; CHECK-NEXT: retq 1115 ; 1116 ; MCU-LABEL: select_xor_1b: 1117 ; MCU: # %bb.0: # %entry 1118 ; MCU-NEXT: testb $1, %dl 1119 ; MCU-NEXT: je .LBB26_2 1120 ; MCU-NEXT: # %bb.1: 1121 ; MCU-NEXT: xorl $43, %eax 1122 ; MCU-NEXT: .LBB26_2: # %entry 1123 ; MCU-NEXT: # kill: def $ax killed $ax killed $eax 1124 ; MCU-NEXT: retl 1125 entry: 1126 %and = and i8 %cond, 1 1127 %cmp10 = icmp ne i8 %and, 1 1128 %0 = xor i16 %A, 43 1129 %1 = select i1 %cmp10, i16 %A, i16 %0 1130 ret i16 %1 1131 } 1132 1133 define i32 @select_xor_2(i32 %A, i32 %B, i8 %cond) { 1134 ; CHECK-LABEL: select_xor_2: 1135 ; CHECK: ## %bb.0: ## %entry 1136 ; CHECK-NEXT: xorl %edi, %esi 1137 ; CHECK-NEXT: testb $1, %dl 1138 ; CHECK-NEXT: cmovel %edi, %esi 1139 ; CHECK-NEXT: movl %esi, %eax 1140 ; CHECK-NEXT: retq 1141 ; 1142 ; MCU-LABEL: select_xor_2: 1143 ; MCU: # %bb.0: # %entry 1144 ; MCU-NEXT: andl $1, %ecx 1145 ; MCU-NEXT: negl %ecx 1146 ; MCU-NEXT: andl %edx, %ecx 1147 ; MCU-NEXT: xorl %ecx, %eax 1148 ; MCU-NEXT: retl 1149 entry: 1150 %and = and i8 %cond, 1 1151 %cmp10 = icmp eq i8 %and, 0 1152 %0 = xor i32 %B, %A 1153 %1 = select i1 %cmp10, i32 %A, i32 %0 1154 ret i32 %1 1155 } 1156 1157 ; Equivalent to above, but with icmp ne (and %cond, 1), 1 instead of 1158 ; icmp eq (and %cond, 1), 0 1159 define i32 @select_xor_2b(i32 %A, i32 %B, i8 %cond) { 1160 ; CHECK-LABEL: select_xor_2b: 1161 ; CHECK: ## %bb.0: ## %entry 1162 ; CHECK-NEXT: xorl %edi, %esi 1163 ; CHECK-NEXT: testb $1, %dl 1164 ; CHECK-NEXT: cmovel %edi, %esi 1165 ; CHECK-NEXT: movl %esi, %eax 1166 ; CHECK-NEXT: retq 1167 ; 1168 ; MCU-LABEL: select_xor_2b: 1169 ; MCU: # %bb.0: # %entry 1170 ; MCU-NEXT: testb $1, %cl 1171 ; MCU-NEXT: je .LBB28_2 1172 ; MCU-NEXT: # %bb.1: 1173 ; MCU-NEXT: xorl %edx, %eax 1174 ; MCU-NEXT: .LBB28_2: # %entry 1175 ; MCU-NEXT: retl 1176 entry: 1177 %and = and i8 %cond, 1 1178 %cmp10 = icmp ne i8 %and, 1 1179 %0 = xor i32 %B, %A 1180 %1 = select i1 %cmp10, i32 %A, i32 %0 1181 ret i32 %1 1182 } 1183 1184 define i32 @select_or(i32 %A, i32 %B, i8 %cond) { 1185 ; CHECK-LABEL: select_or: 1186 ; CHECK: ## %bb.0: ## %entry 1187 ; CHECK-NEXT: orl %edi, %esi 1188 ; CHECK-NEXT: testb $1, %dl 1189 ; CHECK-NEXT: cmovel %edi, %esi 1190 ; CHECK-NEXT: movl %esi, %eax 1191 ; CHECK-NEXT: retq 1192 ; 1193 ; MCU-LABEL: select_or: 1194 ; MCU: # %bb.0: # %entry 1195 ; MCU-NEXT: andl $1, %ecx 1196 ; MCU-NEXT: negl %ecx 1197 ; MCU-NEXT: andl %edx, %ecx 1198 ; MCU-NEXT: orl %ecx, %eax 1199 ; MCU-NEXT: retl 1200 entry: 1201 %and = and i8 %cond, 1 1202 %cmp10 = icmp eq i8 %and, 0 1203 %0 = or i32 %B, %A 1204 %1 = select i1 %cmp10, i32 %A, i32 %0 1205 ret i32 %1 1206 } 1207 1208 ; Equivalent to above, but with icmp ne (and %cond, 1), 1 instead of 1209 ; icmp eq (and %cond, 1), 0 1210 define i32 @select_or_b(i32 %A, i32 %B, i8 %cond) { 1211 ; CHECK-LABEL: select_or_b: 1212 ; CHECK: ## %bb.0: ## %entry 1213 ; CHECK-NEXT: orl %edi, %esi 1214 ; CHECK-NEXT: testb $1, %dl 1215 ; CHECK-NEXT: cmovel %edi, %esi 1216 ; CHECK-NEXT: movl %esi, %eax 1217 ; CHECK-NEXT: retq 1218 ; 1219 ; MCU-LABEL: select_or_b: 1220 ; MCU: # %bb.0: # %entry 1221 ; MCU-NEXT: testb $1, %cl 1222 ; MCU-NEXT: je .LBB30_2 1223 ; MCU-NEXT: # %bb.1: 1224 ; MCU-NEXT: orl %edx, %eax 1225 ; MCU-NEXT: .LBB30_2: # %entry 1226 ; MCU-NEXT: retl 1227 entry: 1228 %and = and i8 %cond, 1 1229 %cmp10 = icmp ne i8 %and, 1 1230 %0 = or i32 %B, %A 1231 %1 = select i1 %cmp10, i32 %A, i32 %0 1232 ret i32 %1 1233 } 1234 1235 define i32 @select_or_1(i32 %A, i32 %B, i32 %cond) { 1236 ; CHECK-LABEL: select_or_1: 1237 ; CHECK: ## %bb.0: ## %entry 1238 ; CHECK-NEXT: orl %edi, %esi 1239 ; CHECK-NEXT: testb $1, %dl 1240 ; CHECK-NEXT: cmovel %edi, %esi 1241 ; CHECK-NEXT: movl %esi, %eax 1242 ; CHECK-NEXT: retq 1243 ; 1244 ; MCU-LABEL: select_or_1: 1245 ; MCU: # %bb.0: # %entry 1246 ; MCU-NEXT: andl $1, %ecx 1247 ; MCU-NEXT: negl %ecx 1248 ; MCU-NEXT: andl %edx, %ecx 1249 ; MCU-NEXT: orl %ecx, %eax 1250 ; MCU-NEXT: retl 1251 entry: 1252 %and = and i32 %cond, 1 1253 %cmp10 = icmp eq i32 %and, 0 1254 %0 = or i32 %B, %A 1255 %1 = select i1 %cmp10, i32 %A, i32 %0 1256 ret i32 %1 1257 } 1258 1259 ; Equivalent to above, but with icmp ne (and %cond, 1), 1 instead of 1260 ; icmp eq (and %cond, 1), 0 1261 define i32 @select_or_1b(i32 %A, i32 %B, i32 %cond) { 1262 ; CHECK-LABEL: select_or_1b: 1263 ; CHECK: ## %bb.0: ## %entry 1264 ; CHECK-NEXT: orl %edi, %esi 1265 ; CHECK-NEXT: testb $1, %dl 1266 ; CHECK-NEXT: cmovel %edi, %esi 1267 ; CHECK-NEXT: movl %esi, %eax 1268 ; CHECK-NEXT: retq 1269 ; 1270 ; MCU-LABEL: select_or_1b: 1271 ; MCU: # %bb.0: # %entry 1272 ; MCU-NEXT: testb $1, %cl 1273 ; MCU-NEXT: je .LBB32_2 1274 ; MCU-NEXT: # %bb.1: 1275 ; MCU-NEXT: orl %edx, %eax 1276 ; MCU-NEXT: .LBB32_2: # %entry 1277 ; MCU-NEXT: retl 1278 entry: 1279 %and = and i32 %cond, 1 1280 %cmp10 = icmp ne i32 %and, 1 1281 %0 = or i32 %B, %A 1282 %1 = select i1 %cmp10, i32 %A, i32 %0 1283 ret i32 %1 1284 } 1285