Home | History | Annotate | Download | only in X86
      1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
      2 ; RUN: llc < %s -mcpu=generic -mtriple=i386-apple-darwin -verify-machineinstrs -no-integrated-as | FileCheck %s
      3 
      4 ; There should be no stack manipulations between the inline asm and ret.
      5 define x86_fp80 @test1() {
      6 ; CHECK-LABEL: test1:
      7 ; CHECK:       ## %bb.0:
      8 ; CHECK-NEXT:    ## InlineAsm Start
      9 ; CHECK-NEXT:    fld0
     10 ; CHECK-NEXT:    ## InlineAsm End
     11 ; CHECK-NEXT:    retl
     12   %tmp85 = call x86_fp80 asm sideeffect "fld0", "={st(0)}"()
     13   ret x86_fp80 %tmp85
     14 }
     15 
     16 define double @test2() {
     17 ; CHECK-LABEL: test2:
     18 ; CHECK:       ## %bb.0:
     19 ; CHECK-NEXT:    ## InlineAsm Start
     20 ; CHECK-NEXT:    fld0
     21 ; CHECK-NEXT:    ## InlineAsm End
     22 ; CHECK-NEXT:    retl
     23   %tmp85 = call double asm sideeffect "fld0", "={st(0)}"()
     24   ret double %tmp85
     25 }
     26 
     27 ; Setting up argument in st(0) should be a single fld.
     28 ; Asm consumes stack, nothing should be popped.
     29 define void @test3(x86_fp80 %X) {
     30 ; CHECK-LABEL: test3:
     31 ; CHECK:       ## %bb.0:
     32 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
     33 ; CHECK-NEXT:    ## InlineAsm Start
     34 ; CHECK-NEXT:    frob
     35 ; CHECK-NEXT:    ## InlineAsm End
     36 ; CHECK-NEXT:    retl
     37   call void asm sideeffect "frob ", "{st(0)},~{st},~{dirflag},~{fpsr},~{flags}"( x86_fp80 %X)
     38   ret void
     39 }
     40 
     41 define void @test4(double %X) {
     42 ; CHECK-LABEL: test4:
     43 ; CHECK:       ## %bb.0:
     44 ; CHECK-NEXT:    fldl {{[0-9]+}}(%esp)
     45 ; CHECK-NEXT:    ## InlineAsm Start
     46 ; CHECK-NEXT:    frob
     47 ; CHECK-NEXT:    ## InlineAsm End
     48 ; CHECK-NEXT:    retl
     49   call void asm sideeffect "frob ", "{st(0)},~{st},~{dirflag},~{fpsr},~{flags}"( double %X)
     50   ret void
     51 }
     52 
     53 ; Same as test3/4, but using value from fadd.
     54 ; The fadd can be done in xmm or x87 regs - we don't test that.
     55 define void @test5(double %X) {
     56 ; CHECK-LABEL: test5:
     57 ; CHECK:       ## %bb.0:
     58 ; CHECK-NEXT:    fldl {{[0-9]+}}(%esp)
     59 ; CHECK-NEXT:    fadds LCPI4_0
     60 ; CHECK-NEXT:    ## InlineAsm Start
     61 ; CHECK-NEXT:    frob
     62 ; CHECK-NEXT:    ## InlineAsm End
     63 ; CHECK-NEXT:    retl
     64   %Y = fadd double %X, 123.0
     65   call void asm sideeffect "frob ", "{st(0)},~{st},~{dirflag},~{fpsr},~{flags}"( double %Y)
     66   ret void
     67 }
     68 
     69 define void @test6(double %A, double %B, double %C, double %D, double %E) nounwind {
     70 ; CHECK-LABEL: test6:
     71 ; CHECK:       ## %bb.0: ## %entry
     72 ; CHECK-NEXT:    fldl {{[0-9]+}}(%esp)
     73 ; CHECK-NEXT:    fldl {{[0-9]+}}(%esp)
     74 ; CHECK-NEXT:    fldl {{[0-9]+}}(%esp)
     75 ; CHECK-NEXT:    fldl {{[0-9]+}}(%esp)
     76 ; CHECK-NEXT:    fldl {{[0-9]+}}(%esp)
     77 ; CHECK-NEXT:    ## InlineAsm Start
     78 ; CHECK-NEXT:    foo %st(0) %st(0)
     79 ; CHECK-NEXT:    ## InlineAsm End
     80 ; CHECK-NEXT:    fstp %st(0)
     81 ; CHECK-NEXT:    ## InlineAsm Start
     82 ; CHECK-NEXT:    bar %st(1) %st(0)
     83 ; CHECK-NEXT:    ## InlineAsm End
     84 ; CHECK-NEXT:    fstp %st(1)
     85 ; CHECK-NEXT:    fstp %st(0)
     86 ; CHECK-NEXT:    ## InlineAsm Start
     87 ; CHECK-NEXT:    baz %st(1) %st(0)
     88 ; CHECK-NEXT:    ## InlineAsm End
     89 ; CHECK-NEXT:    fstp %st(0)
     90 ; CHECK-NEXT:    ## InlineAsm Start
     91 ; CHECK-NEXT:    baz %st(0)
     92 ; CHECK-NEXT:    ## InlineAsm End
     93 ; CHECK-NEXT:    fstp %st(0)
     94 ; CHECK-NEXT:    retl
     95 ; CHECK-NEXT:    ## -- End function
     96 entry:
     97 ; Uses the same value twice, should have one fstp after the asm.
     98   tail call void asm sideeffect "foo $0 $1", "f,f,~{dirflag},~{fpsr},~{flags}"( double %A, double %A ) nounwind
     99 ; Uses two different values, should be in st(0)/st(1) and both be popped.
    100   tail call void asm sideeffect "bar $0 $1", "f,f,~{dirflag},~{fpsr},~{flags}"( double %B, double %C ) nounwind
    101 ; Uses two different values, one of which isn't killed in this asm, it should not be popped after the asm.
    102   tail call void asm sideeffect "baz $0 $1", "f,f,~{dirflag},~{fpsr},~{flags}"( double %D, double %E ) nounwind
    103 ; This is the last use of %D, so it should be popped after.
    104   tail call void asm sideeffect "baz $0", "f,~{dirflag},~{fpsr},~{flags}"( double %D ) nounwind
    105   ret void
    106 }
    107 
    108 ; PR4185
    109 ; Passing a non-killed value to asm in {st}.
    110 ; Make sure it is duped before.
    111 ; asm kills st(0), so we shouldn't pop anything
    112 ; A valid alternative would be to remat the constant pool load before each
    113 ; inline asm.
    114 define void @testPR4185() {
    115 ; CHECK-LABEL: testPR4185:
    116 ; CHECK:       ## %bb.0: ## %return
    117 ; CHECK-NEXT:    flds LCPI6_0
    118 ; CHECK-NEXT:    fld %st(0)
    119 ; CHECK-NEXT:    ## InlineAsm Start
    120 ; CHECK-NEXT:    fistpl %st(0)
    121 ; CHECK-NEXT:    ## InlineAsm End
    122 ; CHECK-NEXT:    ## InlineAsm Start
    123 ; CHECK-NEXT:    fistpl %st(0)
    124 ; CHECK-NEXT:    ## InlineAsm End
    125 ; CHECK-NEXT:    retl
    126 return:
    127   call void asm sideeffect "fistpl $0", "{st},~{st}"(double 1.000000e+06)
    128   call void asm sideeffect "fistpl $0", "{st},~{st}"(double 1.000000e+06)
    129   ret void
    130 }
    131 
    132 ; Passing a non-killed value through asm in {st}.
    133 ; Make sure it is not duped before.
    134 ; Second asm kills st(0), so we shouldn't pop anything
    135 ; A valid alternative would be to remat the constant pool load before each inline asm.
    136 define void @testPR4185b() {
    137 ; CHECK-LABEL: testPR4185b:
    138 ; CHECK:       ## %bb.0: ## %return
    139 ; CHECK-NEXT:    flds LCPI7_0
    140 ; CHECK-NEXT:    ## InlineAsm Start
    141 ; CHECK-NEXT:    fistl %st(0)
    142 ; CHECK-NEXT:    ## InlineAsm End
    143 ; CHECK-NEXT:    ## InlineAsm Start
    144 ; CHECK-NEXT:    fistpl %st(0)
    145 ; CHECK-NEXT:    ## InlineAsm End
    146 ; CHECK-NEXT:    retl
    147 return:
    148 	call void asm sideeffect "fistl $0", "{st}"(double 1.000000e+06)
    149 	call void asm sideeffect "fistpl $0", "{st},~{st}"(double 1.000000e+06)
    150 	ret void
    151 }
    152 
    153 ; PR4459
    154 ; The return value from ceil must be duped before being consumed by asm.
    155 define void @testPR4459(x86_fp80 %a) {
    156 ; CHECK-LABEL: testPR4459:
    157 ; CHECK:       ## %bb.0: ## %entry
    158 ; CHECK-NEXT:    subl $28, %esp
    159 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
    160 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    161 ; CHECK-NEXT:    fstpt (%esp)
    162 ; CHECK-NEXT:    calll _ceil
    163 ; CHECK-NEXT:    fld %st(0)
    164 ; CHECK-NEXT:    fxch %st(1)
    165 ; CHECK-NEXT:    ## InlineAsm Start
    166 ; CHECK-NEXT:    fistpl %st(0)
    167 ; CHECK-NEXT:    ## InlineAsm End
    168 ; CHECK-NEXT:    fstpt (%esp)
    169 ; CHECK-NEXT:    calll _test3
    170 ; CHECK-NEXT:    addl $28, %esp
    171 ; CHECK-NEXT:    retl
    172 entry:
    173   %0 = call x86_fp80 @ceil(x86_fp80 %a)
    174   call void asm sideeffect "fistpl $0", "{st},~{st}"( x86_fp80 %0)
    175   call void @test3(x86_fp80 %0 )
    176   ret void
    177 }
    178 declare x86_fp80 @ceil(x86_fp80)
    179 
    180 ; PR4484
    181 ; test1 leaves a value on the stack that is needed after the asm.
    182 ; Load %a from stack after ceil
    183 ; Set up call to test.
    184 define void @testPR4484(x86_fp80 %a) {
    185 ; CHECK-LABEL: testPR4484:
    186 ; CHECK:       ## %bb.0: ## %entry
    187 ; CHECK-NEXT:    subl $28, %esp
    188 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
    189 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    190 ; CHECK-NEXT:    fstpt {{[0-9]+}}(%esp) ## 10-byte Folded Spill
    191 ; CHECK-NEXT:    calll _test1
    192 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp) ## 10-byte Folded Reload
    193 ; CHECK-NEXT:    ## InlineAsm Start
    194 ; CHECK-NEXT:    fistpl %st(0)
    195 ; CHECK-NEXT:    ## InlineAsm End
    196 ; CHECK-NEXT:    fstpt (%esp)
    197 ; CHECK-NEXT:    calll _test3
    198 ; CHECK-NEXT:    addl $28, %esp
    199 ; CHECK-NEXT:    retl
    200 entry:
    201   %0 = call x86_fp80 @test1()
    202   call void asm sideeffect "fistpl $0", "{st},~{st}"(x86_fp80 %a)
    203   call void @test3(x86_fp80 %0)
    204   ret void
    205 }
    206 
    207 ; PR4485
    208 define void @testPR4485(x86_fp80* %a) {
    209 ; CHECK-LABEL: testPR4485:
    210 ; CHECK:       ## %bb.0: ## %entry
    211 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
    212 ; CHECK-NEXT:    fldt (%eax)
    213 ; CHECK-NEXT:    flds LCPI10_0
    214 ; CHECK-NEXT:    fmul %st(0), %st(1)
    215 ; CHECK-NEXT:    flds LCPI10_1
    216 ; CHECK-NEXT:    fmul %st(0), %st(2)
    217 ; CHECK-NEXT:    fxch %st(2)
    218 ; CHECK-NEXT:    ## InlineAsm Start
    219 ; CHECK-NEXT:    fistpl %st(0)
    220 ; CHECK-NEXT:    ## InlineAsm End
    221 ; CHECK-NEXT:    fldt (%eax)
    222 ; CHECK-NEXT:    fmulp %st(1)
    223 ; CHECK-NEXT:    fmulp %st(1)
    224 ; CHECK-NEXT:    ## InlineAsm Start
    225 ; CHECK-NEXT:    fistpl %st(0)
    226 ; CHECK-NEXT:    ## InlineAsm End
    227 ; CHECK-NEXT:    retl
    228 entry:
    229   %0 = load x86_fp80, x86_fp80* %a, align 16
    230   %1 = fmul x86_fp80 %0, 0xK4006B400000000000000
    231   %2 = fmul x86_fp80 %1, 0xK4012F424000000000000
    232   tail call void asm sideeffect "fistpl $0", "{st},~{st}"(x86_fp80 %2)
    233   %3 = load x86_fp80, x86_fp80* %a, align 16
    234   %4 = fmul x86_fp80 %3, 0xK4006B400000000000000
    235   %5 = fmul x86_fp80 %4, 0xK4012F424000000000000
    236   tail call void asm sideeffect "fistpl $0", "{st},~{st}"(x86_fp80 %5)
    237   ret void
    238 }
    239 
    240 ; An input argument in a fixed position is implicitly popped by the asm only if
    241 ; the input argument is tied to an output register, or it is in the clobber list.
    242 ; The clobber list case is tested above.
    243 ;
    244 ; This doesn't implicitly pop the stack:
    245 ;
    246 ;   void fist1(long double x, int *p) {
    247 ;     asm volatile ("fistl %1" : : "t"(x), "m"(*p));
    248 ;   }
    249 define void @fist1(x86_fp80 %x, i32* %p) nounwind ssp {
    250 ; CHECK-LABEL: fist1:
    251 ; CHECK:       ## %bb.0: ## %entry
    252 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    253 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
    254 ; CHECK-NEXT:    ## InlineAsm Start
    255 ; CHECK-NEXT:    fistl (%eax)
    256 ; CHECK-NEXT:    ## InlineAsm End
    257 ; CHECK-NEXT:    fstp %st(0)
    258 ; CHECK-NEXT:    retl
    259 ; CHECK-NEXT:    ## -- End function
    260 entry:
    261   tail call void asm sideeffect "fistl $1", "{st},*m,~{memory},~{dirflag},~{fpsr},~{flags}"(x86_fp80 %x, i32* %p) nounwind
    262   ret void
    263 }
    264 
    265 ; Here, the input operand is tied to an output which means that is is
    266 ; implicitly popped (and then the output is implicitly pushed).
    267 ;
    268 ;   long double fist2(long double x, int *p) {
    269 ;     long double y;
    270 ;     asm ("fistl %1" : "=&t"(y) : "0"(x), "m"(*p) : "memory");
    271 ;     return y;
    272 ;   }
    273 define x86_fp80 @fist2(x86_fp80 %x, i32* %p) nounwind ssp {
    274 ; CHECK-LABEL: fist2:
    275 ; CHECK:       ## %bb.0: ## %entry
    276 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    277 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
    278 ; CHECK-NEXT:    ## InlineAsm Start
    279 ; CHECK-NEXT:    fistl (%eax)
    280 ; CHECK-NEXT:    ## InlineAsm End
    281 ; CHECK-NEXT:    retl
    282 ; CHECK-NEXT:    ## -- End function
    283 entry:
    284   %0 = tail call x86_fp80 asm "fistl $2", "=&{st},0,*m,~{memory},~{dirflag},~{fpsr},~{flags}"(x86_fp80 %x, i32* %p) nounwind
    285   ret x86_fp80 %0
    286 }
    287 
    288 ; An 'f' constraint is never implicitly popped:
    289 ;
    290 ;   void fucomp1(long double x, long double y) {
    291 ;     asm volatile ("fucomp %1" : : "t"(x), "f"(y) : "st");
    292 ;   }
    293 define void @fucomp1(x86_fp80 %x, x86_fp80 %y) nounwind ssp {
    294 ; CHECK-LABEL: fucomp1:
    295 ; CHECK:       ## %bb.0: ## %entry
    296 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    297 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    298 ; CHECK-NEXT:    fxch %st(1)
    299 ; CHECK-NEXT:    ## InlineAsm Start
    300 ; CHECK-NEXT:    fucomp %st(1)
    301 ; CHECK-NEXT:    ## InlineAsm End
    302 ; CHECK-NEXT:    fstp %st(0)
    303 ; CHECK-NEXT:    retl
    304 ; CHECK-NEXT:    ## -- End function
    305 entry:
    306   tail call void asm sideeffect "fucomp $1", "{st},f,~{st},~{dirflag},~{fpsr},~{flags}"(x86_fp80 %x, x86_fp80 %y) nounwind
    307   ret void
    308 }
    309 
    310 ; The 'u' constraint is only popped implicitly when clobbered:
    311 ;
    312 ;   void fucomp2(long double x, long double y) {
    313 ;     asm volatile ("fucomp %1" : : "t"(x), "u"(y) : "st");
    314 ;   }
    315 ;
    316 ;   void fucomp3(long double x, long double y) {
    317 ;     asm volatile ("fucompp %1" : : "t"(x), "u"(y) : "st", "st(1)");
    318 ;   }
    319 ;
    320 define void @fucomp2(x86_fp80 %x, x86_fp80 %y) nounwind ssp {
    321 ; CHECK-LABEL: fucomp2:
    322 ; CHECK:       ## %bb.0: ## %entry
    323 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    324 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    325 ; CHECK-NEXT:    fxch %st(1)
    326 ; CHECK-NEXT:    ## InlineAsm Start
    327 ; CHECK-NEXT:    fucomp %st(1)
    328 ; CHECK-NEXT:    ## InlineAsm End
    329 ; CHECK-NEXT:    fstp %st(0)
    330 ; CHECK-NEXT:    retl
    331 ; CHECK-NEXT:    ## -- End function
    332 entry:
    333   tail call void asm sideeffect "fucomp $1", "{st},{st(1)},~{st},~{dirflag},~{fpsr},~{flags}"(x86_fp80 %x, x86_fp80 %y) nounwind
    334   ret void
    335 }
    336 
    337 define void @fucomp3(x86_fp80 %x, x86_fp80 %y) nounwind ssp {
    338 ; CHECK-LABEL: fucomp3:
    339 ; CHECK:       ## %bb.0: ## %entry
    340 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    341 ; CHECK-NEXT:    fldt {{[0-9]+}}(%esp)
    342 ; CHECK-NEXT:    fxch %st(1)
    343 ; CHECK-NEXT:    ## InlineAsm Start
    344 ; CHECK-NEXT:    fucompp %st(1)
    345 ; CHECK-NEXT:    ## InlineAsm End
    346 ; CHECK-NEXT:    retl
    347 ; CHECK-NEXT:    ## -- End function
    348 entry:
    349   tail call void asm sideeffect "fucompp $1", "{st},{st(1)},~{st},~{st(1)},~{dirflag},~{fpsr},~{flags}"(x86_fp80 %x, x86_fp80 %y) nounwind
    350   ret void
    351 }
    352 
    353 ; One input, two outputs, one dead output.
    354 %complex = type { float, float }
    355 define float @sincos1(float %x) nounwind ssp {
    356 ; CHECK-LABEL: sincos1:
    357 ; CHECK:       ## %bb.0: ## %entry
    358 ; CHECK-NEXT:    flds {{[0-9]+}}(%esp)
    359 ; CHECK-NEXT:    ## InlineAsm Start
    360 ; CHECK-NEXT:    sincos
    361 ; CHECK-NEXT:    ## InlineAsm End
    362 ; CHECK-NEXT:    fstp %st(1)
    363 ; CHECK-NEXT:    retl
    364 ; CHECK-NEXT:    ## -- End function
    365 entry:
    366   %0 = tail call %complex asm "sincos", "={st},={st(1)},0,~{dirflag},~{fpsr},~{flags}"(float %x) nounwind
    367   %asmresult = extractvalue %complex %0, 0
    368   ret float %asmresult
    369 }
    370 
    371 ; Same thing, swapped output operands.
    372 define float @sincos2(float %x) nounwind ssp {
    373 ; CHECK-LABEL: sincos2:
    374 ; CHECK:       ## %bb.0: ## %entry
    375 ; CHECK-NEXT:    flds {{[0-9]+}}(%esp)
    376 ; CHECK-NEXT:    ## InlineAsm Start
    377 ; CHECK-NEXT:    sincos
    378 ; CHECK-NEXT:    ## InlineAsm End
    379 ; CHECK-NEXT:    fstp %st(1)
    380 ; CHECK-NEXT:    retl
    381 ; CHECK-NEXT:    ## -- End function
    382 entry:
    383   %0 = tail call %complex asm "sincos", "={st(1)},={st},1,~{dirflag},~{fpsr},~{flags}"(float %x) nounwind
    384   %asmresult = extractvalue %complex %0, 1
    385   ret float %asmresult
    386 }
    387 
    388 ; Clobber st(0) after it was live-out/dead from the previous asm.
    389 ; Load x, make a copy for the second asm.
    390 ; Discard dead result in st(0), bring x to the top.
    391 ; x is now in st(0) for the second asm
    392 ; Discard both results.
    393 define float @sincos3(float %x) nounwind ssp {
    394 ; CHECK-LABEL: sincos3:
    395 ; CHECK:       ## %bb.0: ## %entry
    396 ; CHECK-NEXT:    flds {{[0-9]+}}(%esp)
    397 ; CHECK-NEXT:    fld %st(0)
    398 ; CHECK-NEXT:    ## InlineAsm Start
    399 ; CHECK-NEXT:    sincos
    400 ; CHECK-NEXT:    ## InlineAsm End
    401 ; CHECK-NEXT:    fstp %st(0)
    402 ; CHECK-NEXT:    fxch %st(1)
    403 ; CHECK-NEXT:    ## InlineAsm Start
    404 ; CHECK-NEXT:    sincos
    405 ; CHECK-NEXT:    ## InlineAsm End
    406 ; CHECK-NEXT:    fstp %st(1)
    407 ; CHECK-NEXT:    fstp %st(0)
    408 ; CHECK-NEXT:    retl
    409 ; CHECK-NEXT:    ## -- End function
    410 entry:
    411   %0 = tail call %complex asm sideeffect "sincos", "={st(1)},={st},1,~{dirflag},~{fpsr},~{flags}"(float %x) nounwind
    412   %1 = tail call %complex asm sideeffect "sincos", "={st(1)},={st},1,~{dirflag},~{fpsr},~{flags}"(float %x) nounwind
    413   %asmresult = extractvalue %complex %0, 0
    414   ret float %asmresult
    415 }
    416 
    417 ; Pass the same value in two fixed stack slots.
    418 define i32 @PR10602() nounwind ssp {
    419 ; CHECK-LABEL: PR10602:
    420 ; CHECK:       ## %bb.0: ## %entry
    421 ; CHECK-NEXT:    flds LCPI19_0
    422 ; CHECK-NEXT:    fld %st(0)
    423 ; CHECK-NEXT:    fxch %st(1)
    424 ; CHECK-NEXT:    ## InlineAsm Start
    425 ; CHECK-NEXT:    fcomi %st(1), %st(0); pushf; pop %eax
    426 ; CHECK-NEXT:    ## InlineAsm End
    427 ; CHECK-NEXT:    fstp %st(0)
    428 ; CHECK-NEXT:    fstp %st(0)
    429 ; CHECK-NEXT:    retl
    430 ; CHECK-NEXT:    ## -- End function
    431 entry:
    432   %0 = tail call i32 asm "fcomi $2, $1; pushf; pop $0", "=r,{st},{st(1)},~{dirflag},~{fpsr},~{flags}"(double 2.000000e+00, double 2.000000e+00) nounwind
    433   ret i32 %0
    434 }
    435 
    436 ; <rdar://problem/16952634>
    437 ; X87 stackifier asserted when there was an ST register defined by an
    438 ; inline-asm instruction and the ST register was live across another
    439 ; inline-asm instruction.
    440 ;
    441 ; INLINEASM $frndint [sideeffect] [attdialect], $0:[regdef], %st0<imp-def,tied5>, $1:[reguse tiedto:$0], %st0<tied3>, $2:[clobber], early-clobber implicit dead %eflags
    442 ; INLINEASM $fldcw $0 [sideeffect] [mayload] [attdialect], $0:[mem], undef %eax, 1, %noreg, 0, %noreg, $1:[clobber], early-clobber implicit dead %eflags
    443 ; %fp0 = COPY %st0
    444 
    445 %struct.fpu_t = type { [8 x x86_fp80], x86_fp80, %struct.anon1, %struct.anon2, i32, i8, [15 x i8] }
    446 %struct.anon1 = type { i32, i32, i32 }
    447 %struct.anon2 = type { i32, i32, i32, i32 }
    448 
    449 @fpu = external global %struct.fpu_t, align 16
    450 
    451 ; Function Attrs: ssp
    452 define void @test_live_st(i32 %a1) {
    453 ; CHECK-LABEL: test_live_st:
    454 ; CHECK:       ## %bb.0: ## %entry
    455 ; CHECK-NEXT:    subl $12, %esp
    456 ; CHECK-NEXT:    .cfi_def_cfa_offset 16
    457 ; CHECK-NEXT:    fldt (%eax)
    458 ; CHECK-NEXT:    cmpl $1, {{[0-9]+}}(%esp)
    459 ; CHECK-NEXT:    jne LBB20_2
    460 ; CHECK-NEXT:  ## %bb.1: ## %sw.bb4.i
    461 ; CHECK-NEXT:    ## InlineAsm Start
    462 ; CHECK-NEXT:    frndint
    463 ; CHECK-NEXT:    ## InlineAsm End
    464 ; CHECK-NEXT:    ## InlineAsm Start
    465 ; CHECK-NEXT:    fldcw (%eax)
    466 ; CHECK-NEXT:    ## InlineAsm End
    467 ; CHECK-NEXT:  LBB20_2: ## %_Z5tointRKe.exit
    468 ; CHECK-NEXT:    fnstcw {{[0-9]+}}(%esp)
    469 ; CHECK-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
    470 ; CHECK-NEXT:    movw $3199, {{[0-9]+}}(%esp) ## imm = 0xC7F
    471 ; CHECK-NEXT:    fldcw {{[0-9]+}}(%esp)
    472 ; CHECK-NEXT:    movw %ax, {{[0-9]+}}(%esp)
    473 ; CHECK-NEXT:    fistpl {{[0-9]+}}(%esp)
    474 ; CHECK-NEXT:    fldcw {{[0-9]+}}(%esp)
    475 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
    476 ; CHECK-NEXT:    movl %eax, {{[0-9]+}}(%esp)
    477 ; CHECK-NEXT:    fildl {{[0-9]+}}(%esp)
    478 ; CHECK-NEXT:    movl L_fpu$non_lazy_ptr, %eax
    479 ; CHECK-NEXT:    fstpt 128(%eax)
    480 ; CHECK-NEXT:    addl $12, %esp
    481 ; CHECK-NEXT:    retl
    482 entry:
    483   %0 = load x86_fp80, x86_fp80* undef, align 16
    484   %cond = icmp eq i32 %a1, 1
    485   br i1 %cond, label %sw.bb4.i, label %_Z5tointRKe.exit
    486 
    487 sw.bb4.i:
    488   %1 = call x86_fp80 asm sideeffect "frndint", "={st},0,~{dirflag},~{fpsr},~{flags}"(x86_fp80 %0)
    489   call void asm sideeffect "fldcw $0", "*m,~{dirflag},~{fpsr},~{flags}"(i32* undef)
    490   br label %_Z5tointRKe.exit
    491 
    492 _Z5tointRKe.exit:
    493   %result.0.i = phi x86_fp80 [ %1, %sw.bb4.i ], [ %0, %entry ]
    494   %conv.i1814 = fptosi x86_fp80 %result.0.i to i32
    495   %conv626 = sitofp i32 %conv.i1814 to x86_fp80
    496   store x86_fp80 %conv626, x86_fp80* getelementptr inbounds (%struct.fpu_t, %struct.fpu_t* @fpu, i32 0, i32 1)
    497   br label %return
    498 
    499 return:
    500   ret void
    501 }
    502 
    503 ; Check that x87 stackifier is correctly rewriting FP registers to ST registers.
    504 define double @test_operand_rewrite() {
    505 ; CHECK-LABEL: test_operand_rewrite:
    506 ; CHECK:       ## %bb.0: ## %entry
    507 ; CHECK-NEXT:    ## InlineAsm Start
    508 ; CHECK-NEXT:    foo %st(0), %st(1)
    509 ; CHECK-NEXT:    ## InlineAsm End
    510 ; CHECK-NEXT:    fsubp %st(1)
    511 ; CHECK-NEXT:    retl
    512 entry:
    513   %0 = tail call { double, double } asm sideeffect "foo $0, $1", "={st},={st(1)},~{dirflag},~{fpsr},~{flags}"()
    514   %asmresult = extractvalue { double, double } %0, 0
    515   %asmresult1 = extractvalue { double, double } %0, 1
    516   %sub = fsub double %asmresult, %asmresult1
    517   ret double %sub
    518 }
    519