Skip to content

Commit 6850033

Browse files
petar-avramovicpetar-avramovic
committed
AMDGPU/GlobalISel: Legalize s64->s16 G_SITOFP/G_UITOFP
Add widenScalar for TypeIdx == 0 for G_SITOFP/G_UITOFP. Legailize, using widenScalar, as s64->s32 G_SITOFP/G_UITOFP followed by s32->s16 G_FPTRUNC. Differential Revision: https://reviews.llvm.org/D83880
1 parent 7358a11 commit 6850033

File tree

4 files changed

+467
-8
lines changed

4 files changed

+467
-8
lines changed

‎llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

Lines changed: 12 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -1908,21 +1908,25 @@ LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
19081908
Observer.changedInstr(MI);
19091909
return Legalized;
19101910
case TargetOpcode::G_SITOFP:
1911-
if (TypeIdx != 1)
1912-
return UnableToLegalize;
19131911
Observer.changingInstr(MI);
1914-
widenScalarSrc(MI, WideTy, 1, TargetOpcode::G_SEXT);
1912+
1913+
if (TypeIdx == 0)
1914+
widenScalarDst(MI, WideTy, 0, TargetOpcode::G_FPTRUNC);
1915+
else
1916+
widenScalarSrc(MI, WideTy, 1, TargetOpcode::G_SEXT);
1917+
19151918
Observer.changedInstr(MI);
19161919
return Legalized;
1917-
19181920
case TargetOpcode::G_UITOFP:
1919-
if (TypeIdx != 1)
1920-
return UnableToLegalize;
19211921
Observer.changingInstr(MI);
1922-
widenScalarSrc(MI, WideTy, 1, TargetOpcode::G_ZEXT);
1922+
1923+
if (TypeIdx == 0)
1924+
widenScalarDst(MI, WideTy, 0, TargetOpcode::G_FPTRUNC);
1925+
else
1926+
widenScalarSrc(MI, WideTy, 1, TargetOpcode::G_ZEXT);
1927+
19231928
Observer.changedInstr(MI);
19241929
return Legalized;
1925-
19261930
case TargetOpcode::G_LOAD:
19271931
case TargetOpcode::G_SEXTLOAD:
19281932
case TargetOpcode::G_ZEXTLOAD:

‎llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -648,6 +648,7 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
648648
if (ST.has16BitInsts())
649649
IToFP.legalFor({{S16, S16}});
650650
IToFP.clampScalar(1, S32, S64)
651+
.minScalar(0, S32)
651652
.scalarize(0)
652653
.widenScalarToNextPow2(1);
653654

‎llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir

Lines changed: 259 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -535,3 +535,262 @@ body: |
535535
%2:_(s32) = G_SITOFP %1
536536
$vgpr0 = COPY %2
537537
...
538+
539+
---
540+
name: test_sitofp_s64_to_s16
541+
body: |
542+
bb.0:
543+
liveins: $vgpr0_vgpr1
544+
545+
; GFX6-LABEL: name: test_sitofp_s64_to_s16
546+
; GFX6: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
547+
; GFX6: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 63
548+
; GFX6: [[ASHR:%[0-9]+]]:_(s64) = G_ASHR [[COPY]], [[C]](s32)
549+
; GFX6: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
550+
; GFX6: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ASHR]](s64)
551+
; GFX6: [[UADDO:%[0-9]+]]:_(s32), [[UADDO1:%[0-9]+]]:_(s1) = G_UADDO [[UV]], [[UV2]]
552+
; GFX6: [[UADDE:%[0-9]+]]:_(s32), [[UADDE1:%[0-9]+]]:_(s1) = G_UADDE [[UV1]], [[UV3]], [[UADDO1]]
553+
; GFX6: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[UADDO]](s32), [[UADDE]](s32)
554+
; GFX6: [[XOR:%[0-9]+]]:_(s64) = G_XOR [[MV]], [[ASHR]]
555+
; GFX6: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
556+
; GFX6: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
557+
; GFX6: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[XOR]](s64)
558+
; GFX6: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 190
559+
; GFX6: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[C3]], [[CTLZ_ZERO_UNDEF]]
560+
; GFX6: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[XOR]](s64), [[C2]]
561+
; GFX6: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[ICMP]](s1), [[SUB]], [[C1]]
562+
; GFX6: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 9223372036854775807
563+
; GFX6: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[XOR]], [[CTLZ_ZERO_UNDEF]](s32)
564+
; GFX6: [[AND:%[0-9]+]]:_(s64) = G_AND [[SHL]], [[C4]]
565+
; GFX6: [[C5:%[0-9]+]]:_(s64) = G_CONSTANT i64 1099511627775
566+
; GFX6: [[AND1:%[0-9]+]]:_(s64) = G_AND [[AND]], [[C5]]
567+
; GFX6: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 40
568+
; GFX6: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[AND]], [[C6]](s32)
569+
; GFX6: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 23
570+
; GFX6: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[SELECT]], [[C7]](s32)
571+
; GFX6: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[LSHR]](s64)
572+
; GFX6: [[OR:%[0-9]+]]:_(s32) = G_OR [[SHL1]], [[TRUNC]]
573+
; GFX6: [[C8:%[0-9]+]]:_(s64) = G_CONSTANT i64 549755813888
574+
; GFX6: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[AND1]](s64), [[C8]]
575+
; GFX6: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[AND1]](s64), [[C8]]
576+
; GFX6: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
577+
; GFX6: [[AND2:%[0-9]+]]:_(s32) = G_AND [[OR]], [[C9]]
578+
; GFX6: [[SELECT1:%[0-9]+]]:_(s32) = G_SELECT [[ICMP2]](s1), [[AND2]], [[C1]]
579+
; GFX6: [[SELECT2:%[0-9]+]]:_(s32) = G_SELECT [[ICMP1]](s1), [[C9]], [[SELECT1]]
580+
; GFX6: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[OR]], [[SELECT2]]
581+
; GFX6: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[ADD]]
582+
; GFX6: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[ASHR]](s64), [[C2]]
583+
; GFX6: [[SELECT3:%[0-9]+]]:_(s32) = G_SELECT [[ICMP3]](s1), [[FNEG]], [[ADD]]
584+
; GFX6: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[SELECT3]](s32)
585+
; GFX6: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
586+
; GFX6: $vgpr0 = COPY [[ANYEXT]](s32)
587+
; GFX8-LABEL: name: test_sitofp_s64_to_s16
588+
; GFX8: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
589+
; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 63
590+
; GFX8: [[ASHR:%[0-9]+]]:_(s64) = G_ASHR [[COPY]], [[C]](s32)
591+
; GFX8: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
592+
; GFX8: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ASHR]](s64)
593+
; GFX8: [[UADDO:%[0-9]+]]:_(s32), [[UADDO1:%[0-9]+]]:_(s1) = G_UADDO [[UV]], [[UV2]]
594+
; GFX8: [[UADDE:%[0-9]+]]:_(s32), [[UADDE1:%[0-9]+]]:_(s1) = G_UADDE [[UV1]], [[UV3]], [[UADDO1]]
595+
; GFX8: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[UADDO]](s32), [[UADDE]](s32)
596+
; GFX8: [[XOR:%[0-9]+]]:_(s64) = G_XOR [[MV]], [[ASHR]]
597+
; GFX8: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
598+
; GFX8: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
599+
; GFX8: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[XOR]](s64)
600+
; GFX8: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 190
601+
; GFX8: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[C3]], [[CTLZ_ZERO_UNDEF]]
602+
; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[XOR]](s64), [[C2]]
603+
; GFX8: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[ICMP]](s1), [[SUB]], [[C1]]
604+
; GFX8: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 9223372036854775807
605+
; GFX8: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[XOR]], [[CTLZ_ZERO_UNDEF]](s32)
606+
; GFX8: [[AND:%[0-9]+]]:_(s64) = G_AND [[SHL]], [[C4]]
607+
; GFX8: [[C5:%[0-9]+]]:_(s64) = G_CONSTANT i64 1099511627775
608+
; GFX8: [[AND1:%[0-9]+]]:_(s64) = G_AND [[AND]], [[C5]]
609+
; GFX8: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 40
610+
; GFX8: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[AND]], [[C6]](s32)
611+
; GFX8: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 23
612+
; GFX8: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[SELECT]], [[C7]](s32)
613+
; GFX8: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[LSHR]](s64)
614+
; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[SHL1]], [[TRUNC]]
615+
; GFX8: [[C8:%[0-9]+]]:_(s64) = G_CONSTANT i64 549755813888
616+
; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[AND1]](s64), [[C8]]
617+
; GFX8: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[AND1]](s64), [[C8]]
618+
; GFX8: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
619+
; GFX8: [[AND2:%[0-9]+]]:_(s32) = G_AND [[OR]], [[C9]]
620+
; GFX8: [[SELECT1:%[0-9]+]]:_(s32) = G_SELECT [[ICMP2]](s1), [[AND2]], [[C1]]
621+
; GFX8: [[SELECT2:%[0-9]+]]:_(s32) = G_SELECT [[ICMP1]](s1), [[C9]], [[SELECT1]]
622+
; GFX8: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[OR]], [[SELECT2]]
623+
; GFX8: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[ADD]]
624+
; GFX8: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[ASHR]](s64), [[C2]]
625+
; GFX8: [[SELECT3:%[0-9]+]]:_(s32) = G_SELECT [[ICMP3]](s1), [[FNEG]], [[ADD]]
626+
; GFX8: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[SELECT3]](s32)
627+
; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
628+
; GFX8: $vgpr0 = COPY [[ANYEXT]](s32)
629+
%0:_(s64) = COPY $vgpr0_vgpr1
630+
%1:_(s16) = G_SITOFP %0
631+
%2:_(s32) = G_ANYEXT %1
632+
$vgpr0 = COPY %2
633+
...
634+
635+
---
636+
name: test_sitofp_v2s64_to_v2s16
637+
body: |
638+
bb.0:
639+
liveins: $vgpr0_vgpr1_vgpr2_vgpr3
640+
641+
; GFX6-LABEL: name: test_sitofp_v2s64_to_v2s16
642+
; GFX6: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
643+
; GFX6: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
644+
; GFX6: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 63
645+
; GFX6: [[ASHR:%[0-9]+]]:_(s64) = G_ASHR [[UV]], [[C]](s32)
646+
; GFX6: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[UV]](s64)
647+
; GFX6: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ASHR]](s64)
648+
; GFX6: [[UADDO:%[0-9]+]]:_(s32), [[UADDO1:%[0-9]+]]:_(s1) = G_UADDO [[UV2]], [[UV4]]
649+
; GFX6: [[UADDE:%[0-9]+]]:_(s32), [[UADDE1:%[0-9]+]]:_(s1) = G_UADDE [[UV3]], [[UV5]], [[UADDO1]]
650+
; GFX6: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[UADDO]](s32), [[UADDE]](s32)
651+
; GFX6: [[XOR:%[0-9]+]]:_(s64) = G_XOR [[MV]], [[ASHR]]
652+
; GFX6: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
653+
; GFX6: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
654+
; GFX6: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[XOR]](s64)
655+
; GFX6: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 190
656+
; GFX6: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[C3]], [[CTLZ_ZERO_UNDEF]]
657+
; GFX6: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[XOR]](s64), [[C2]]
658+
; GFX6: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[ICMP]](s1), [[SUB]], [[C1]]
659+
; GFX6: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 9223372036854775807
660+
; GFX6: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[XOR]], [[CTLZ_ZERO_UNDEF]](s32)
661+
; GFX6: [[AND:%[0-9]+]]:_(s64) = G_AND [[SHL]], [[C4]]
662+
; GFX6: [[C5:%[0-9]+]]:_(s64) = G_CONSTANT i64 1099511627775
663+
; GFX6: [[AND1:%[0-9]+]]:_(s64) = G_AND [[AND]], [[C5]]
664+
; GFX6: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 40
665+
; GFX6: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[AND]], [[C6]](s32)
666+
; GFX6: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 23
667+
; GFX6: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[SELECT]], [[C7]](s32)
668+
; GFX6: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[LSHR]](s64)
669+
; GFX6: [[OR:%[0-9]+]]:_(s32) = G_OR [[SHL1]], [[TRUNC]]
670+
; GFX6: [[C8:%[0-9]+]]:_(s64) = G_CONSTANT i64 549755813888
671+
; GFX6: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[AND1]](s64), [[C8]]
672+
; GFX6: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[AND1]](s64), [[C8]]
673+
; GFX6: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
674+
; GFX6: [[AND2:%[0-9]+]]:_(s32) = G_AND [[OR]], [[C9]]
675+
; GFX6: [[SELECT1:%[0-9]+]]:_(s32) = G_SELECT [[ICMP2]](s1), [[AND2]], [[C1]]
676+
; GFX6: [[SELECT2:%[0-9]+]]:_(s32) = G_SELECT [[ICMP1]](s1), [[C9]], [[SELECT1]]
677+
; GFX6: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[OR]], [[SELECT2]]
678+
; GFX6: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[ADD]]
679+
; GFX6: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[ASHR]](s64), [[C2]]
680+
; GFX6: [[SELECT3:%[0-9]+]]:_(s32) = G_SELECT [[ICMP3]](s1), [[FNEG]], [[ADD]]
681+
; GFX6: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[SELECT3]](s32)
682+
; GFX6: [[ASHR1:%[0-9]+]]:_(s64) = G_ASHR [[UV1]], [[C]](s32)
683+
; GFX6: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[UV1]](s64)
684+
; GFX6: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ASHR1]](s64)
685+
; GFX6: [[UADDO2:%[0-9]+]]:_(s32), [[UADDO3:%[0-9]+]]:_(s1) = G_UADDO [[UV6]], [[UV8]]
686+
; GFX6: [[UADDE2:%[0-9]+]]:_(s32), [[UADDE3:%[0-9]+]]:_(s1) = G_UADDE [[UV7]], [[UV9]], [[UADDO3]]
687+
; GFX6: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[UADDO2]](s32), [[UADDE2]](s32)
688+
; GFX6: [[XOR1:%[0-9]+]]:_(s64) = G_XOR [[MV1]], [[ASHR1]]
689+
; GFX6: [[CTLZ_ZERO_UNDEF1:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[XOR1]](s64)
690+
; GFX6: [[SUB1:%[0-9]+]]:_(s32) = G_SUB [[C3]], [[CTLZ_ZERO_UNDEF1]]
691+
; GFX6: [[ICMP4:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[XOR1]](s64), [[C2]]
692+
; GFX6: [[SELECT4:%[0-9]+]]:_(s32) = G_SELECT [[ICMP4]](s1), [[SUB1]], [[C1]]
693+
; GFX6: [[SHL2:%[0-9]+]]:_(s64) = G_SHL [[XOR1]], [[CTLZ_ZERO_UNDEF1]](s32)
694+
; GFX6: [[AND3:%[0-9]+]]:_(s64) = G_AND [[SHL2]], [[C4]]
695+
; GFX6: [[AND4:%[0-9]+]]:_(s64) = G_AND [[AND3]], [[C5]]
696+
; GFX6: [[LSHR1:%[0-9]+]]:_(s64) = G_LSHR [[AND3]], [[C6]](s32)
697+
; GFX6: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[SELECT4]], [[C7]](s32)
698+
; GFX6: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[LSHR1]](s64)
699+
; GFX6: [[OR1:%[0-9]+]]:_(s32) = G_OR [[SHL3]], [[TRUNC1]]
700+
; GFX6: [[ICMP5:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[AND4]](s64), [[C8]]
701+
; GFX6: [[ICMP6:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[AND4]](s64), [[C8]]
702+
; GFX6: [[AND5:%[0-9]+]]:_(s32) = G_AND [[OR1]], [[C9]]
703+
; GFX6: [[SELECT5:%[0-9]+]]:_(s32) = G_SELECT [[ICMP6]](s1), [[AND5]], [[C1]]
704+
; GFX6: [[SELECT6:%[0-9]+]]:_(s32) = G_SELECT [[ICMP5]](s1), [[C9]], [[SELECT5]]
705+
; GFX6: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[OR1]], [[SELECT6]]
706+
; GFX6: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[ADD1]]
707+
; GFX6: [[ICMP7:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[ASHR1]](s64), [[C2]]
708+
; GFX6: [[SELECT7:%[0-9]+]]:_(s32) = G_SELECT [[ICMP7]](s1), [[FNEG1]], [[ADD1]]
709+
; GFX6: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[SELECT7]](s32)
710+
; GFX6: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
711+
; GFX6: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
712+
; GFX6: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
713+
; GFX6: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
714+
; GFX6: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
715+
; GFX6: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
716+
; GFX6: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
717+
; GFX8-LABEL: name: test_sitofp_v2s64_to_v2s16
718+
; GFX8: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
719+
; GFX8: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
720+
; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 63
721+
; GFX8: [[ASHR:%[0-9]+]]:_(s64) = G_ASHR [[UV]], [[C]](s32)
722+
; GFX8: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[UV]](s64)
723+
; GFX8: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ASHR]](s64)
724+
; GFX8: [[UADDO:%[0-9]+]]:_(s32), [[UADDO1:%[0-9]+]]:_(s1) = G_UADDO [[UV2]], [[UV4]]
725+
; GFX8: [[UADDE:%[0-9]+]]:_(s32), [[UADDE1:%[0-9]+]]:_(s1) = G_UADDE [[UV3]], [[UV5]], [[UADDO1]]
726+
; GFX8: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[UADDO]](s32), [[UADDE]](s32)
727+
; GFX8: [[XOR:%[0-9]+]]:_(s64) = G_XOR [[MV]], [[ASHR]]
728+
; GFX8: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
729+
; GFX8: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
730+
; GFX8: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[XOR]](s64)
731+
; GFX8: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 190
732+
; GFX8: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[C3]], [[CTLZ_ZERO_UNDEF]]
733+
; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[XOR]](s64), [[C2]]
734+
; GFX8: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[ICMP]](s1), [[SUB]], [[C1]]
735+
; GFX8: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 9223372036854775807
736+
; GFX8: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[XOR]], [[CTLZ_ZERO_UNDEF]](s32)
737+
; GFX8: [[AND:%[0-9]+]]:_(s64) = G_AND [[SHL]], [[C4]]
738+
; GFX8: [[C5:%[0-9]+]]:_(s64) = G_CONSTANT i64 1099511627775
739+
; GFX8: [[AND1:%[0-9]+]]:_(s64) = G_AND [[AND]], [[C5]]
740+
; GFX8: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 40
741+
; GFX8: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[AND]], [[C6]](s32)
742+
; GFX8: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 23
743+
; GFX8: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[SELECT]], [[C7]](s32)
744+
; GFX8: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[LSHR]](s64)
745+
; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[SHL1]], [[TRUNC]]
746+
; GFX8: [[C8:%[0-9]+]]:_(s64) = G_CONSTANT i64 549755813888
747+
; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[AND1]](s64), [[C8]]
748+
; GFX8: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[AND1]](s64), [[C8]]
749+
; GFX8: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
750+
; GFX8: [[AND2:%[0-9]+]]:_(s32) = G_AND [[OR]], [[C9]]
751+
; GFX8: [[SELECT1:%[0-9]+]]:_(s32) = G_SELECT [[ICMP2]](s1), [[AND2]], [[C1]]
752+
; GFX8: [[SELECT2:%[0-9]+]]:_(s32) = G_SELECT [[ICMP1]](s1), [[C9]], [[SELECT1]]
753+
; GFX8: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[OR]], [[SELECT2]]
754+
; GFX8: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[ADD]]
755+
; GFX8: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[ASHR]](s64), [[C2]]
756+
; GFX8: [[SELECT3:%[0-9]+]]:_(s32) = G_SELECT [[ICMP3]](s1), [[FNEG]], [[ADD]]
757+
; GFX8: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[SELECT3]](s32)
758+
; GFX8: [[ASHR1:%[0-9]+]]:_(s64) = G_ASHR [[UV1]], [[C]](s32)
759+
; GFX8: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[UV1]](s64)
760+
; GFX8: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ASHR1]](s64)
761+
; GFX8: [[UADDO2:%[0-9]+]]:_(s32), [[UADDO3:%[0-9]+]]:_(s1) = G_UADDO [[UV6]], [[UV8]]
762+
; GFX8: [[UADDE2:%[0-9]+]]:_(s32), [[UADDE3:%[0-9]+]]:_(s1) = G_UADDE [[UV7]], [[UV9]], [[UADDO3]]
763+
; GFX8: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[UADDO2]](s32), [[UADDE2]](s32)
764+
; GFX8: [[XOR1:%[0-9]+]]:_(s64) = G_XOR [[MV1]], [[ASHR1]]
765+
; GFX8: [[CTLZ_ZERO_UNDEF1:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[XOR1]](s64)
766+
; GFX8: [[SUB1:%[0-9]+]]:_(s32) = G_SUB [[C3]], [[CTLZ_ZERO_UNDEF1]]
767+
; GFX8: [[ICMP4:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[XOR1]](s64), [[C2]]
768+
; GFX8: [[SELECT4:%[0-9]+]]:_(s32) = G_SELECT [[ICMP4]](s1), [[SUB1]], [[C1]]
769+
; GFX8: [[SHL2:%[0-9]+]]:_(s64) = G_SHL [[XOR1]], [[CTLZ_ZERO_UNDEF1]](s32)
770+
; GFX8: [[AND3:%[0-9]+]]:_(s64) = G_AND [[SHL2]], [[C4]]
771+
; GFX8: [[AND4:%[0-9]+]]:_(s64) = G_AND [[AND3]], [[C5]]
772+
; GFX8: [[LSHR1:%[0-9]+]]:_(s64) = G_LSHR [[AND3]], [[C6]](s32)
773+
; GFX8: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[SELECT4]], [[C7]](s32)
774+
; GFX8: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[LSHR1]](s64)
775+
; GFX8: [[OR1:%[0-9]+]]:_(s32) = G_OR [[SHL3]], [[TRUNC1]]
776+
; GFX8: [[ICMP5:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[AND4]](s64), [[C8]]
777+
; GFX8: [[ICMP6:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[AND4]](s64), [[C8]]
778+
; GFX8: [[AND5:%[0-9]+]]:_(s32) = G_AND [[OR1]], [[C9]]
779+
; GFX8: [[SELECT5:%[0-9]+]]:_(s32) = G_SELECT [[ICMP6]](s1), [[AND5]], [[C1]]
780+
; GFX8: [[SELECT6:%[0-9]+]]:_(s32) = G_SELECT [[ICMP5]](s1), [[C9]], [[SELECT5]]
781+
; GFX8: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[OR1]], [[SELECT6]]
782+
; GFX8: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[ADD1]]
783+
; GFX8: [[ICMP7:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[ASHR1]](s64), [[C2]]
784+
; GFX8: [[SELECT7:%[0-9]+]]:_(s32) = G_SELECT [[ICMP7]](s1), [[FNEG1]], [[ADD1]]
785+
; GFX8: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[SELECT7]](s32)
786+
; GFX8: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
787+
; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
788+
; GFX8: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
789+
; GFX8: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
790+
; GFX8: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
791+
; GFX8: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
792+
; GFX8: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
793+
%0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
794+
%1:_(<2 x s16>) = G_SITOFP %0
795+
$vgpr0 = COPY %1
796+
...

0 commit comments

Comments
 (0)