Skip to content

Commit 35cecac

Browse files
RKSimonRKSimon
committed
[X86] Split rr/rm CVT schedules on SNB/HSW/BDW
The folded load variants almost never require Port5 for length changing conversions (just for SNB ymm cases), and don't have an extra uop for the load. Confirmed with a mixture of Agner + uops.info comparisons.
1 parent 0c21ed4 commit 35cecac

File tree

15 files changed

+102
-96
lines changed

15 files changed

+102
-96
lines changed

‎llvm/lib/Target/X86/X86SchedBroadwell.td‎

Lines changed: 11 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -367,21 +367,26 @@ defm : BWWriteResPair<WriteCvtPD2IY, [BWPort1,BWPort5], 6, [1,1], 2, 6>;
367367
defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
368368

369369
defm : X86WriteRes<WriteCvtI2SS, [BWPort1,BWPort5], 4, [1,1], 2>;
370+
defm : X86WriteRes<WriteCvtI2PS, [BWPort1], 3, [1], 1>;
371+
defm : X86WriteRes<WriteCvtI2PSY, [BWPort1], 3, [1], 1>;
370372
defm : X86WriteRes<WriteCvtI2SSLd, [BWPort1,BWPort23], 9, [1,1], 2>;
371-
defm : BWWriteResPair<WriteCvtI2PS, [BWPort1], 3>;
372-
defm : BWWriteResPair<WriteCvtI2PSY, [BWPort1], 3, [1], 1, 6>;
373+
defm : X86WriteRes<WriteCvtI2PSLd, [BWPort1,BWPort23], 8, [1,1], 2>;
374+
defm : X86WriteRes<WriteCvtI2PSYLd, [BWPort1,BWPort23], 9, [1,1], 2>;
373375
defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
374376
defm : X86WriteRes<WriteCvtI2SD, [BWPort1,BWPort5], 4, [1,1], 2>;
377+
defm : X86WriteRes<WriteCvtI2PD, [BWPort1,BWPort5], 4, [1,1], 2>;
378+
defm : X86WriteRes<WriteCvtI2PDY, [BWPort1,BWPort5], 6, [1,1], 2>;
375379
defm : X86WriteRes<WriteCvtI2SDLd, [BWPort1,BWPort23], 9, [1,1], 2>;
376-
defm : BWWriteResPair<WriteCvtI2PD, [BWPort1,BWPort5], 4, [1,1], 2, 5>;
377-
defm : BWWriteResPair<WriteCvtI2PDY, [BWPort1,BWPort5], 6, [1,1], 2, 5>;
380+
defm : X86WriteRes<WriteCvtI2PDLd, [BWPort1,BWPort23], 9, [1,1], 2>;
381+
defm : X86WriteRes<WriteCvtI2PDYLd, [BWPort1,BWPort23],11, [1,1], 2>;
378382
defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
379383

380384
defm : X86WriteRes<WriteCvtSS2SD, [BWPort0,BWPort5], 2, [1,1], 2>;
381-
defm : X86WriteRes<WriteCvtSS2SDLd, [BWPort0,BWPort23], 6, [1,1], 2>;
382385
defm : X86WriteRes<WriteCvtPS2PD, [BWPort0,BWPort5], 2, [1,1], 2>;
386+
defm : X86WriteRes<WriteCvtPS2PDY, [BWPort0,BWPort5], 4, [1,1], 2>;
387+
defm : X86WriteRes<WriteCvtSS2SDLd, [BWPort0,BWPort23], 6, [1,1], 2>;
383388
defm : X86WriteRes<WriteCvtPS2PDLd, [BWPort0,BWPort23], 6, [1,1], 2>;
384-
defm : BWWriteResPair<WriteCvtPS2PDY, [BWPort0,BWPort5], 4, [1,1], 2, 5>;
389+
defm : X86WriteRes<WriteCvtPS2PDYLd, [BWPort0,BWPort23], 9, [1,1], 2>;
385390
defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
386391
defm : BWWriteResPair<WriteCvtSD2SS, [BWPort1,BWPort5], 4, [1,1], 2, 5>;
387392
defm : BWWriteResPair<WriteCvtPD2PS, [BWPort1,BWPort5], 4, [1,1], 2, 5>;

‎llvm/lib/Target/X86/X86SchedHaswell.td‎

Lines changed: 17 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -364,22 +364,30 @@ defm : HWWriteResPair<WriteCvtPS2IY, [HWPort1], 3, [1], 1, 7>;
364364
defm : HWWriteResPair<WriteCvtPS2IZ, [HWPort1], 3, [1], 1, 7>; // Unsupported = 1
365365

366366
defm : X86WriteRes<WriteCvtI2SD, [HWPort1,HWPort5], 4, [1,1], 2>;
367+
defm : X86WriteRes<WriteCvtI2PD, [HWPort1,HWPort5], 4, [1,1], 2>;
368+
defm : X86WriteRes<WriteCvtI2PDY, [HWPort1,HWPort5], 6, [1,1], 2>;
369+
defm : X86WriteRes<WriteCvtI2PDZ, [HWPort1,HWPort5], 6, [1,1], 2>; // Unsupported = 1
367370
defm : X86WriteRes<WriteCvtI2SDLd, [HWPort1,HWPort23], 9, [1,1], 2>;
368-
defm : HWWriteResPair<WriteCvtI2PD, [HWPort1,HWPort5], 4, [1,1], 2, 6>;
369-
defm : HWWriteResPair<WriteCvtI2PDY, [HWPort1,HWPort5], 6, [1,1], 2, 6>;
370-
defm : HWWriteResPair<WriteCvtI2PDZ, [HWPort1,HWPort5], 6, [1,1], 2, 6>; // Unsupported = 1
371+
defm : X86WriteRes<WriteCvtI2PDLd, [HWPort1,HWPort23],10, [1,1], 2>;
372+
defm : X86WriteRes<WriteCvtI2PDYLd, [HWPort1,HWPort23],12, [1,1], 2>;
373+
defm : X86WriteRes<WriteCvtI2PDZLd, [HWPort1,HWPort23],12, [1,1], 2>; // Unsupported = 1
371374
defm : X86WriteRes<WriteCvtI2SS, [HWPort1,HWPort5], 4, [1,1], 2>;
375+
defm : X86WriteRes<WriteCvtI2PS, [HWPort1], 3, [1], 1>;
376+
defm : X86WriteRes<WriteCvtI2PSY, [HWPort1], 3, [1], 1>;
377+
defm : X86WriteRes<WriteCvtI2PSZ, [HWPort1], 3, [1], 1>; // Unsupported = 1
372378
defm : X86WriteRes<WriteCvtI2SSLd, [HWPort1,HWPort23], 9, [1,1], 2>;
373-
defm : HWWriteResPair<WriteCvtI2PS, [HWPort1], 3, [1], 1, 6>;
374-
defm : HWWriteResPair<WriteCvtI2PSY, [HWPort1], 3, [1], 1, 7>;
375-
defm : HWWriteResPair<WriteCvtI2PSZ, [HWPort1], 3, [1], 1, 7>; // Unsupported = 1
379+
defm : X86WriteRes<WriteCvtI2PSLd, [HWPort1,HWPort23], 9, [1,1], 2>;
380+
defm : X86WriteRes<WriteCvtI2PSYLd, [HWPort1,HWPort23],10, [1,1], 2>;
381+
defm : X86WriteRes<WriteCvtI2PSZLd, [HWPort1,HWPort23],10, [1,1], 2>; // Unsupported = 1
376382

377383
defm : X86WriteRes<WriteCvtSS2SD, [HWPort0,HWPort5], 2, [1,1], 2>;
378-
defm : X86WriteRes<WriteCvtSS2SDLd, [HWPort0,HWPort23], 7, [1,1], 2>;
379384
defm : X86WriteRes<WriteCvtPS2PD, [HWPort0,HWPort5], 2, [1,1], 2>;
385+
defm : X86WriteRes<WriteCvtPS2PDY, [HWPort0,HWPort5], 4, [1,1], 2>;
386+
defm : X86WriteRes<WriteCvtPS2PDZ, [HWPort0,HWPort5], 4, [1,1], 2>; // Unsupported = 1
387+
defm : X86WriteRes<WriteCvtSS2SDLd, [HWPort0,HWPort23], 7, [1,1], 2>;
380388
defm : X86WriteRes<WriteCvtPS2PDLd, [HWPort0,HWPort23], 6, [1,1], 2>;
381-
defm : HWWriteResPair<WriteCvtPS2PDY, [HWPort0,HWPort5], 4, [1,1], 2, 6>;
382-
defm : HWWriteResPair<WriteCvtPS2PDZ, [HWPort0,HWPort5], 4, [1,1], 2, 6>; // Unsupported = 1
389+
defm : X86WriteRes<WriteCvtPS2PDYLd, [HWPort0,HWPort23],10, [1,1], 2>;
390+
defm : X86WriteRes<WriteCvtPS2PDZLd, [HWPort0,HWPort23],10, [1,1], 2>; // Unsupported = 1
383391
defm : HWWriteResPair<WriteCvtSD2SS, [HWPort1,HWPort5], 4, [1,1], 2, 5>;
384392
defm : HWWriteResPair<WriteCvtPD2PS, [HWPort1,HWPort5], 4, [1,1], 2, 6>;
385393
defm : HWWriteResPair<WriteCvtPD2PSY, [HWPort1,HWPort5], 6, [1,1], 2, 6>;
@@ -983,7 +991,6 @@ def HWWriteResGroup12 : SchedWriteRes<[HWPort1,HWPort23]> {
983991
let NumMicroOps = 2;
984992
let ReleaseAtCycles = [1,1];
985993
}
986-
def: InstRW<[HWWriteResGroup12], (instrs MMX_CVTPI2PSrm)>;
987994
def: InstRW<[HWWriteResGroup12], (instregex "P(DEP|EXT)(32|64)rm")>;
988995

989996
def HWWriteResGroup13 : SchedWriteRes<[HWPort5,HWPort23]> {
@@ -1349,13 +1356,6 @@ def HWWriteResGroup75 : SchedWriteRes<[HWPort1,HWPort23]> {
13491356
}
13501357
def: InstRW<[HWWriteResGroup75], (instregex "FICOM(P?)(16|32)m")>;
13511358

1352-
def HWWriteResGroup78_1 : SchedWriteRes<[HWPort1,HWPort5,HWPort23]> {
1353-
let Latency = 9;
1354-
let NumMicroOps = 3;
1355-
let ReleaseAtCycles = [1,1,1];
1356-
}
1357-
def: InstRW<[HWWriteResGroup78_1], (instrs MMX_CVTPI2PDrm)>;
1358-
13591359
def HWWriteResGroup80 : SchedWriteRes<[HWPort5,HWPort23,HWPort015]> {
13601360
let Latency = 9;
13611361
let NumMicroOps = 3;

‎llvm/lib/Target/X86/X86SchedSandyBridge.td‎

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -348,13 +348,14 @@ defm : X86WriteRes<WriteCvtI2PDLd, [SBPort1,SBPort5,SBPort23], 10, [1,1,1], 3>
348348
defm : X86WriteRes<WriteCvtI2PDYLd, [SBPort1,SBPort5,SBPort23], 10, [1,1,1], 3>;
349349
defm : X86WriteRes<WriteCvtI2PDZLd, [SBPort1,SBPort5,SBPort23], 10, [1,1,1], 3>; // Unsupported = 1
350350

351-
defm : SBWriteResPair<WriteCvtSS2SD, [SBPort0], 1, [1], 1, 6>;
351+
defm : X86WriteRes<WriteCvtSS2SD, [SBPort0,SBPort5], 1, [1,1], 2>;
352352
defm : X86WriteRes<WriteCvtPS2PD, [SBPort0,SBPort5], 2, [1,1], 2>;
353353
defm : X86WriteRes<WriteCvtPS2PDY, [SBPort0,SBPort5], 2, [1,1], 2>;
354354
defm : X86WriteRes<WriteCvtPS2PDZ, [SBPort0,SBPort5], 2, [1,1], 2>; // Unsupported = 1
355+
defm : X86WriteRes<WriteCvtSS2SDLd, [SBPort0,SBPort23], 7, [1,1], 2>;
355356
defm : X86WriteRes<WriteCvtPS2PDLd, [SBPort0,SBPort23], 7, [1,1], 2>;
356-
defm : X86WriteRes<WriteCvtPS2PDYLd, [SBPort0,SBPort23], 7, [1,1], 2>;
357-
defm : X86WriteRes<WriteCvtPS2PDZLd, [SBPort0,SBPort23], 7, [1,1], 2>; // Unsupported = 1
357+
defm : X86WriteRes<WriteCvtPS2PDYLd, [SBPort0,SBPort5,SBPort23], 7, [1,1,1], 3>;
358+
defm : X86WriteRes<WriteCvtPS2PDZLd, [SBPort0,SBPort5,SBPort23], 7, [1,1,1], 3>; // Unsupported = 1
358359
defm : SBWriteResPair<WriteCvtSD2SS, [SBPort1,SBPort5], 4, [1,1], 2, 6>;
359360
defm : SBWriteResPair<WriteCvtPD2PS, [SBPort1,SBPort5], 4, [1,1], 2, 6>;
360361
defm : SBWriteResPair<WriteCvtPD2PSY, [SBPort1,SBPort5], 4, [1,1], 2, 7>;

‎llvm/test/tools/llvm-mca/X86/Barcelona/resources-sse2.s‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -448,7 +448,7 @@ xorpd (%rax), %xmm2
448448
# CHECK-NEXT: 2 4 1.00 cvtsi2sd %rcx, %xmm2
449449
# CHECK-NEXT: 2 9 1.00 * cvtsi2sdl (%rax), %xmm2
450450
# CHECK-NEXT: 2 9 1.00 * cvtsi2sdq (%rax), %xmm2
451-
# CHECK-NEXT: 1 1 1.00 cvtss2sd %xmm0, %xmm2
451+
# CHECK-NEXT: 2 1 1.00 cvtss2sd %xmm0, %xmm2
452452
# CHECK-NEXT: 2 7 1.00 * cvtss2sd (%rax), %xmm2
453453
# CHECK-NEXT: 2 4 1.00 cvttpd2dq %xmm0, %xmm2
454454
# CHECK-NEXT: 3 10 1.00 * cvttpd2dq (%rax), %xmm2
@@ -687,7 +687,7 @@ xorpd (%rax), %xmm2
687687

688688
# CHECK: Resource pressure per iteration:
689689
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1]
690-
# CHECK-NEXT: - 172.00 75.83 117.33 17.00 101.83 67.00 67.00
690+
# CHECK-NEXT: - 172.00 75.83 117.33 17.00 102.83 67.00 67.00
691691

692692
# CHECK: Resource pressure by instruction:
693693
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions:
@@ -732,7 +732,7 @@ xorpd (%rax), %xmm2
732732
# CHECK-NEXT: - - - 1.00 - 1.00 - - cvtsi2sd %rcx, %xmm2
733733
# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 cvtsi2sdl (%rax), %xmm2
734734
# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 cvtsi2sdq (%rax), %xmm2
735-
# CHECK-NEXT: - - 1.00 - - - - - cvtss2sd %xmm0, %xmm2
735+
# CHECK-NEXT: - - 1.00 - - 1.00 - - cvtss2sd %xmm0, %xmm2
736736
# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 cvtss2sd (%rax), %xmm2
737737
# CHECK-NEXT: - - - 1.00 - 1.00 - - cvttpd2dq %xmm0, %xmm2
738738
# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 cvttpd2dq (%rax), %xmm2

‎llvm/test/tools/llvm-mca/X86/Broadwell/resources-avx1.s‎

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -1115,9 +1115,9 @@ vzeroupper
11151115
# CHECK-NEXT: 1 3 1.00 vcomiss %xmm0, %xmm1
11161116
# CHECK-NEXT: 2 8 1.00 * vcomiss (%rax), %xmm1
11171117
# CHECK-NEXT: 2 4 1.00 vcvtdq2pd %xmm0, %xmm2
1118-
# CHECK-NEXT: 3 9 1.00 * vcvtdq2pd (%rax), %xmm2
1118+
# CHECK-NEXT: 2 9 1.00 * vcvtdq2pd (%rax), %xmm2
11191119
# CHECK-NEXT: 2 6 1.00 vcvtdq2pd %xmm0, %ymm2
1120-
# CHECK-NEXT: 3 11 1.00 * vcvtdq2pd (%rax), %ymm2
1120+
# CHECK-NEXT: 2 11 1.00 * vcvtdq2pd (%rax), %ymm2
11211121
# CHECK-NEXT: 1 3 1.00 vcvtdq2ps %xmm0, %xmm2
11221122
# CHECK-NEXT: 2 8 1.00 * vcvtdq2ps (%rax), %xmm2
11231123
# CHECK-NEXT: 1 3 1.00 vcvtdq2ps %ymm0, %ymm2
@@ -1137,7 +1137,7 @@ vzeroupper
11371137
# CHECK-NEXT: 2 2 1.00 vcvtps2pd %xmm0, %xmm2
11381138
# CHECK-NEXT: 2 6 1.00 * vcvtps2pd (%rax), %xmm2
11391139
# CHECK-NEXT: 2 4 1.00 vcvtps2pd %xmm0, %ymm2
1140-
# CHECK-NEXT: 3 9 1.00 * vcvtps2pd (%rax), %ymm2
1140+
# CHECK-NEXT: 2 9 1.00 * vcvtps2pd (%rax), %ymm2
11411141
# CHECK-NEXT: 2 4 1.00 vcvtsd2si %xmm0, %ecx
11421142
# CHECK-NEXT: 2 4 1.00 vcvtsd2si %xmm0, %rcx
11431143
# CHECK-NEXT: 3 9 1.00 * vcvtsd2si (%rax), %ecx
@@ -1736,7 +1736,7 @@ vzeroupper
17361736

17371737
# CHECK: Resource pressure per iteration:
17381738
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9]
1739-
# CHECK-NEXT: - 257.00 216.25 247.25 173.17 173.17 38.00 424.25 3.25 12.67
1739+
# CHECK-NEXT: - 257.00 216.25 247.25 173.17 173.17 38.00 421.25 3.25 12.67
17401740

17411741
# CHECK: Resource pressure by instruction:
17421742
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] Instructions:
@@ -1825,9 +1825,9 @@ vzeroupper
18251825
# CHECK-NEXT: - - - 1.00 - - - - - - vcomiss %xmm0, %xmm1
18261826
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - vcomiss (%rax), %xmm1
18271827
# CHECK-NEXT: - - - 1.00 - - - 1.00 - - vcvtdq2pd %xmm0, %xmm2
1828-
# CHECK-NEXT: - - - 1.00 0.50 0.50 - 1.00 - - vcvtdq2pd (%rax), %xmm2
1828+
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - vcvtdq2pd (%rax), %xmm2
18291829
# CHECK-NEXT: - - - 1.00 - - - 1.00 - - vcvtdq2pd %xmm0, %ymm2
1830-
# CHECK-NEXT: - - - 1.00 0.50 0.50 - 1.00 - - vcvtdq2pd (%rax), %ymm2
1830+
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - vcvtdq2pd (%rax), %ymm2
18311831
# CHECK-NEXT: - - - 1.00 - - - - - - vcvtdq2ps %xmm0, %xmm2
18321832
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - vcvtdq2ps (%rax), %xmm2
18331833
# CHECK-NEXT: - - - 1.00 - - - - - - vcvtdq2ps %ymm0, %ymm2
@@ -1847,7 +1847,7 @@ vzeroupper
18471847
# CHECK-NEXT: - - 1.00 - - - - 1.00 - - vcvtps2pd %xmm0, %xmm2
18481848
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - vcvtps2pd (%rax), %xmm2
18491849
# CHECK-NEXT: - - 1.00 - - - - 1.00 - - vcvtps2pd %xmm0, %ymm2
1850-
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - 1.00 - - vcvtps2pd (%rax), %ymm2
1850+
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - vcvtps2pd (%rax), %ymm2
18511851
# CHECK-NEXT: - - 1.00 1.00 - - - - - - vcvtsd2si %xmm0, %ecx
18521852
# CHECK-NEXT: - - 1.00 1.00 - - - - - - vcvtsd2si %xmm0, %rcx
18531853
# CHECK-NEXT: - - 1.00 1.00 0.50 0.50 - - - - vcvtsd2si (%rax), %ecx

‎llvm/test/tools/llvm-mca/X86/Broadwell/resources-sse2.s‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -423,7 +423,7 @@ xorpd (%rax), %xmm2
423423
# CHECK-NEXT: 1 3 1.00 comisd %xmm0, %xmm1
424424
# CHECK-NEXT: 2 8 1.00 * comisd (%rax), %xmm1
425425
# CHECK-NEXT: 2 4 1.00 cvtdq2pd %xmm0, %xmm2
426-
# CHECK-NEXT: 3 9 1.00 * cvtdq2pd (%rax), %xmm2
426+
# CHECK-NEXT: 2 9 1.00 * cvtdq2pd (%rax), %xmm2
427427
# CHECK-NEXT: 1 3 1.00 cvtdq2ps %xmm0, %xmm2
428428
# CHECK-NEXT: 2 8 1.00 * cvtdq2ps (%rax), %xmm2
429429
# CHECK-NEXT: 2 4 1.00 cvtpd2dq %xmm0, %xmm2
@@ -433,7 +433,7 @@ xorpd (%rax), %xmm2
433433
# CHECK-NEXT: 2 4 1.00 cvtpd2ps %xmm0, %xmm2
434434
# CHECK-NEXT: 3 9 1.00 * cvtpd2ps (%rax), %xmm2
435435
# CHECK-NEXT: 2 4 1.00 cvtpi2pd %mm0, %xmm2
436-
# CHECK-NEXT: 3 9 1.00 * cvtpi2pd (%rax), %xmm2
436+
# CHECK-NEXT: 2 9 1.00 * cvtpi2pd (%rax), %xmm2
437437
# CHECK-NEXT: 1 3 1.00 cvtps2dq %xmm0, %xmm2
438438
# CHECK-NEXT: 2 8 1.00 * cvtps2dq (%rax), %xmm2
439439
# CHECK-NEXT: 2 2 1.00 cvtps2pd %xmm0, %xmm2
@@ -689,7 +689,7 @@ xorpd (%rax), %xmm2
689689

690690
# CHECK: Resource pressure per iteration:
691691
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9]
692-
# CHECK-NEXT: - 78.00 70.75 95.75 63.17 63.17 14.00 119.25 2.25 4.67
692+
# CHECK-NEXT: - 78.00 70.75 95.75 63.17 63.17 14.00 117.25 2.25 4.67
693693

694694
# CHECK: Resource pressure by instruction:
695695
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] Instructions:
@@ -709,7 +709,7 @@ xorpd (%rax), %xmm2
709709
# CHECK-NEXT: - - - 1.00 - - - - - - comisd %xmm0, %xmm1
710710
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - comisd (%rax), %xmm1
711711
# CHECK-NEXT: - - - 1.00 - - - 1.00 - - cvtdq2pd %xmm0, %xmm2
712-
# CHECK-NEXT: - - - 1.00 0.50 0.50 - 1.00 - - cvtdq2pd (%rax), %xmm2
712+
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - cvtdq2pd (%rax), %xmm2
713713
# CHECK-NEXT: - - - 1.00 - - - - - - cvtdq2ps %xmm0, %xmm2
714714
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - cvtdq2ps (%rax), %xmm2
715715
# CHECK-NEXT: - - - 1.00 - - - 1.00 - - cvtpd2dq %xmm0, %xmm2
@@ -719,7 +719,7 @@ xorpd (%rax), %xmm2
719719
# CHECK-NEXT: - - - 1.00 - - - 1.00 - - cvtpd2ps %xmm0, %xmm2
720720
# CHECK-NEXT: - - - 1.00 0.50 0.50 - 1.00 - - cvtpd2ps (%rax), %xmm2
721721
# CHECK-NEXT: - - - 1.00 - - - 1.00 - - cvtpi2pd %mm0, %xmm2
722-
# CHECK-NEXT: - - - 1.00 0.50 0.50 - 1.00 - - cvtpi2pd (%rax), %xmm2
722+
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - cvtpi2pd (%rax), %xmm2
723723
# CHECK-NEXT: - - - 1.00 - - - - - - cvtps2dq %xmm0, %xmm2
724724
# CHECK-NEXT: - - - 1.00 0.50 0.50 - - - - cvtps2dq (%rax), %xmm2
725725
# CHECK-NEXT: - - 1.00 - - - - 1.00 - - cvtps2pd %xmm0, %xmm2

0 commit comments

Comments
 (0)