[llvm] [AMDGPU] Make direct f16 <-> i32 conversion illegal (PR #205565)
Lukas Sommer via llvm-commits
llvm-commits at lists.llvm.org
Wed Jun 24 06:47:55 PDT 2026
https://github.com/sommerlukas created https://github.com/llvm/llvm-project/pull/205565
Neither SALU nor VALU support direct conversion from f16 to/from i32.
Previously, this was still legal and handled by instruction selection patterns, forming chains f16 -> f32 -> i32 and i32 -> f32 -> f16 for the two cases, respectively.
This change marks the conversion illegal and creates the same chains as the pattern during (operation) legalization.
This had the added benefit that a combination of FNEG and FPTOSI/UI can now merge the float negation into the source modifier of the f16-to-f32 conversion, as demonstrated by the GlobalISel tests.
This fixes https://github.com/llvm/llvm-project/issues/177342.
>From e1bb5d73f76223b4a08b872581ba9dbfb5de50e0 Mon Sep 17 00:00:00 2001
From: Lukas Sommer <lukas.sommer at amd.com>
Date: Wed, 24 Jun 2026 08:22:26 -0500
Subject: [PATCH] [AMDGPU] Make direct f16 <-> i32 conversion illegal
Neither SALU nor VALU support direct conversion from f16 to/from i32.
Previously, this was still legal and handled by instruction selection
patterns, forming chains f16 -> f32 -> i32 and i32 -> f32 -> f16 for the
two cases, respectively.
This change marks the conversion illegal and creates the same chains as
the pattern during (operation) legalization.
This had the added benefit that a combination of FNEG and FPTOSI/UI can
now merge the float negation into the source modifier of the f16-to-f32
conversion, as demonstrated by the GlobalISel tests.
Signed-off-by: Lukas Sommer <lukas.sommer at amd.com>
---
llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp | 61 +++----
llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h | 1 +
.../lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp | 10 +-
llvm/lib/Target/AMDGPU/SIISelLowering.cpp | 8 +-
llvm/lib/Target/AMDGPU/SOPInstructions.td | 12 --
.../AMDGPU/GlobalISel/inst-select-fptosi.mir | 162 +++++++++---------
.../AMDGPU/GlobalISel/inst-select-fptoui.mir | 162 +++++++++---------
.../AMDGPU/GlobalISel/inst-select-sitofp.mir | 46 ++---
.../AMDGPU/GlobalISel/inst-select-uitofp.mir | 46 ++---
.../AMDGPU/GlobalISel/legalize-fpowi.mir | 9 +-
.../AMDGPU/GlobalISel/legalize-fptosi.mir | 36 ++--
.../AMDGPU/GlobalISel/legalize-fptoui.mir | 36 ++--
.../AMDGPU/GlobalISel/legalize-sitofp.mir | 15 +-
.../AMDGPU/GlobalISel/legalize-uitofp.mir | 15 +-
14 files changed, 304 insertions(+), 315 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index 278c2dc94c56b..9151d68521543 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -3460,6 +3460,16 @@ SDValue AMDGPUTargetLowering::LowerCTLS(SDValue Op, SelectionDAG &DAG) const {
DAG.getAllOnesConstant(SL, MVT::i32));
}
+SDValue AMDGPUTargetLowering::LowerINT_TO_FP16(SDValue Op, SelectionDAG &DAG,
+ EVT FP16Ty) const {
+ assert(FP16Ty == MVT::f16 || FP16Ty == MVT::bf16);
+ SDLoc SL(Op);
+ SDValue Src = Op.getOperand(0);
+ SDValue ToF32 = DAG.getNode(Op.getOpcode(), SL, MVT::f32, Src);
+ SDValue FPRoundFlag = DAG.getIntPtrConstant(0, SL, /*isTarget=*/true);
+ return DAG.getNode(ISD::FP_ROUND, SL, FP16Ty, ToF32, FPRoundFlag);
+}
+
SDValue AMDGPUTargetLowering::LowerINT_TO_FP32(SDValue Op, SelectionDAG &DAG,
bool Signed) const {
// The regular method converting a 64-bit integer to float roughly consists of
@@ -3625,28 +3635,16 @@ SDValue AMDGPUTargetLowering::LowerUINT_TO_FP(SDValue Op,
return DAG.getNode(ISD::UINT_TO_FP, DL, DestVT, Ext);
}
- if (DestVT == MVT::bf16) {
- SDLoc SL(Op);
- SDValue ToF32 = DAG.getNode(ISD::UINT_TO_FP, SL, MVT::f32, Src);
- SDValue FPRoundFlag = DAG.getIntPtrConstant(0, SL, /*isTarget=*/true);
- return DAG.getNode(ISD::FP_ROUND, SL, MVT::bf16, ToF32, FPRoundFlag);
- }
+ if (DestVT == MVT::bf16)
+ return LowerINT_TO_FP16(Op, DAG, MVT::bf16);
+
+ if (DestVT == MVT::f16 && isTypeLegal(MVT::f16) &&
+ (SrcVT == MVT::i32 || SrcVT == MVT::i64))
+ return LowerINT_TO_FP16(Op, DAG, MVT::f16);
if (SrcVT != MVT::i64)
return Op;
- if (DestVT == MVT::f16 && isTypeLegal(MVT::f16)) {
- SDLoc DL(Op);
-
- SDValue IntToFp32 = DAG.getNode(Op.getOpcode(), DL, MVT::f32, Src);
- SDValue FPRoundFlag =
- DAG.getIntPtrConstant(0, SDLoc(Op), /*isTarget=*/true);
- SDValue FPRound =
- DAG.getNode(ISD::FP_ROUND, DL, MVT::f16, IntToFp32, FPRoundFlag);
-
- return FPRound;
- }
-
if (DestVT == MVT::f32)
return LowerINT_TO_FP32(Op, DAG, false);
@@ -3671,31 +3669,18 @@ SDValue AMDGPUTargetLowering::LowerSINT_TO_FP(SDValue Op,
return DAG.getNode(ISD::SINT_TO_FP, DL, DestVT, Ext);
}
- if (DestVT == MVT::bf16) {
- SDLoc SL(Op);
- SDValue ToF32 = DAG.getNode(ISD::SINT_TO_FP, SL, MVT::f32, Src);
- SDValue FPRoundFlag = DAG.getIntPtrConstant(0, SL, /*isTarget=*/true);
- return DAG.getNode(ISD::FP_ROUND, SL, MVT::bf16, ToF32, FPRoundFlag);
- }
+ if (DestVT == MVT::bf16)
+ return LowerINT_TO_FP16(Op, DAG, MVT::bf16);
+
+ if (DestVT == MVT::f16 && isTypeLegal(MVT::f16) &&
+ (SrcVT == MVT::i32 || SrcVT == MVT::i64))
+ return LowerINT_TO_FP16(Op, DAG, MVT::f16);
if (SrcVT != MVT::i64)
return Op;
// TODO: Factor out code common with LowerUINT_TO_FP.
- if (DestVT == MVT::f16 && isTypeLegal(MVT::f16)) {
- SDLoc DL(Op);
- SDValue Src = Op.getOperand(0);
-
- SDValue IntToFp32 = DAG.getNode(Op.getOpcode(), DL, MVT::f32, Src);
- SDValue FPRoundFlag =
- DAG.getIntPtrConstant(0, SDLoc(Op), /*isTarget=*/true);
- SDValue FPRound =
- DAG.getNode(ISD::FP_ROUND, DL, MVT::f16, IntToFp32, FPRoundFlag);
-
- return FPRound;
- }
-
if (DestVT == MVT::f32)
return LowerINT_TO_FP32(Op, DAG, true);
@@ -3894,7 +3879,7 @@ SDValue AMDGPUTargetLowering::LowerFP_TO_INT(const SDValue Op,
if (SrcVT == MVT::f16 && DestVT == MVT::i16)
return Op;
- if (SrcVT == MVT::bf16) {
+ if (SrcVT == MVT::bf16 || (SrcVT == MVT::f16 && DestVT == MVT::i32)) {
SDLoc DL(Op);
SDValue PromotedSrc = DAG.getNode(ISD::FP_EXTEND, DL, MVT::f32, Src);
return DAG.getNode(Op.getOpcode(), DL, DestVT, PromotedSrc);
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h
index bd429074f4a96..a0a451e05630c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h
@@ -92,6 +92,7 @@ class AMDGPUTargetLowering : public TargetLowering {
SDValue LowerCTLZ_CTTZ(SDValue Op, SelectionDAG &DAG) const;
+ SDValue LowerINT_TO_FP16(SDValue Op, SelectionDAG &DAG, EVT FP16Ty) const;
SDValue LowerINT_TO_FP32(SDValue Op, SelectionDAG &DAG, bool Signed) const;
SDValue LowerINT_TO_FP64(SDValue Op, SelectionDAG &DAG, bool Signed) const;
SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 65ae77cec0846..0ced081d6ed66 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -1213,7 +1213,8 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
// TODO: Split s1->s64 during regbankselect for VALU.
auto &IToFP = getActionDefinitionsBuilder({G_SITOFP, G_UITOFP})
- .legalFor({{S32, S32}, {S64, S32}, {S16, S32}})
+ .legalFor({{S32, S32}, {S64, S32}})
+ .widenScalarFor({{S16, S32}}, changeTo(0, S32))
.lowerIf(typeIs(1, S1))
.customFor({{S32, S64}, {S64, S64}});
if (ST.has16BitInsts())
@@ -1224,9 +1225,10 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
.widenScalarToNextPow2(1);
auto &FPToI = getActionDefinitionsBuilder({G_FPTOSI, G_FPTOUI})
- .legalFor({{S32, S32}, {S32, S64}, {S32, S16}})
- .customFor({{S64, S32}, {S64, S64}})
- .narrowScalarFor({{S64, S16}}, changeTo(0, S32));
+ .legalFor({{S32, S32}, {S32, S64}})
+ .customFor({{S64, S32}, {S64, S64}})
+ .widenScalarFor({{S32, S16}}, changeTo(1, S32))
+ .narrowScalarFor({{S64, S16}}, changeTo(0, S32));
if (ST.has16BitInsts())
FPToI.legalFor({{S16, S16}});
else
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index e22acbe288d0d..14989e513e0c4 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -607,6 +607,7 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM,
AddPromotedToType(ISD::FP_TO_FP16, MVT::i16, MVT::i32);
setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT}, MVT::i16, Custom);
+ setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT}, MVT::i32, Custom);
setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP}, MVT::i16, Custom);
setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP}, MVT::i1, Custom);
@@ -637,13 +638,6 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM,
if (Subtarget->hasBF16TransInsts())
setOperationAction({ISD::FCOS, ISD::FSIN, ISD::FDIV}, MVT::bf16, Custom);
- setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT, ISD::FP_TO_SINT_SAT,
- ISD::FP_TO_UINT_SAT},
- MVT::f16, Promote);
- setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT, ISD::FP_TO_SINT_SAT,
- ISD::FP_TO_UINT_SAT},
- MVT::bf16, Promote);
-
// F16 - VOP2 Actions.
setOperationAction({ISD::BR_CC, ISD::SELECT_CC}, {MVT::f16, MVT::bf16},
Expand);
diff --git a/llvm/lib/Target/AMDGPU/SOPInstructions.td b/llvm/lib/Target/AMDGPU/SOPInstructions.td
index 01dd444c477e2..573651c2c04f8 100644
--- a/llvm/lib/Target/AMDGPU/SOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SOPInstructions.td
@@ -477,18 +477,6 @@ let SubtargetPredicate = HasSALUFloatInsts, AddedComplexity = 9 in {
(S_CVT_I32_F32 $src0)>;
def : GCNPat<(i16 (UniformUnaryFrag<fp_to_uint> f32:$src0)),
(S_CVT_U32_F32 $src0)>;
- // f16 -> i32 : form chain f16 -> f32 -> i32
- def : GCNPat<(i32 (UniformUnaryFrag<fp_to_sint> f16:$src0)),
- (S_CVT_I32_F32 (S_CVT_F32_F16 $src0))>;
- def : GCNPat<(i32 (UniformUnaryFrag<fp_to_uint> f16:$src0)),
- (S_CVT_U32_F32 (S_CVT_F32_F16 $src0))>;
-
- // i32 -> f16 : form chain i32 -> f32 -> f16
- def : GCNPat<(f16 (UniformUnaryFrag<sint_to_fp> i32:$src0)),
- (S_CVT_F16_F32 (S_CVT_F32_I32 $src0))>;
- def : GCNPat<(f16 (UniformUnaryFrag<uint_to_fp> i32:$src0)),
- (S_CVT_F16_F32 (S_CVT_F32_U32 $src0))>;
-
// f32 -> i32 saturated conversions
def : GCNPat<(i32 (UniformBinFrag<fp_to_sint_sat> f32:$src0, i32)),
(S_CVT_I32_F32 (f32 $src0))>;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptosi.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptosi.mir
index 03cb907f82a16..6d84952fb884f 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptosi.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptosi.mir
@@ -125,16 +125,16 @@ body: |
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; VI-LABEL: name: fptosi_s16_to_s32_vv
; VI: liveins: $vgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptosi_s16_to_s32_vv
; GFX11-TRUE16: liveins: $vgpr0
@@ -142,20 +142,21 @@ body: |
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; GFX11-FAKE16-LABEL: name: fptosi_s16_to_s32_vv
; GFX11-FAKE16: liveins: $vgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
- %2:vgpr(s32) = G_FPTOSI %1
- $vgpr0 = COPY %2
+ %2:vgpr(s32) = G_FPEXT %1
+ %3:vgpr(s32) = G_FPTOSI %2
+ $vgpr0 = COPY %3
...
---
@@ -173,36 +174,37 @@ body: |
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; VI-LABEL: name: fptosi_s16_to_s32_vs
; VI: liveins: $sgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptosi_s16_to_s32_vs
; GFX11-TRUE16: liveins: $sgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[COPY]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; GFX11-FAKE16-LABEL: name: fptosi_s16_to_s32_vs
; GFX11-FAKE16: liveins: $sgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
%0:sgpr(s32) = COPY $sgpr0
%1:sgpr(s16) = G_TRUNC %0
- %2:vgpr(s32) = G_FPTOSI %1
- $vgpr0 = COPY %2
+ %2:vgpr(s32) = G_FPEXT %1
+ %3:vgpr(s32) = G_FPTOSI %2
+ $vgpr0 = COPY %3
...
---
@@ -219,46 +221,40 @@ body: |
; GCN: liveins: $vgpr0
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GCN-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; GCN-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; VI-LABEL: name: fptosi_s16_to_s32_fneg_vv
; VI: liveins: $vgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; VI-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; VI-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptosi_s16_to_s32_fneg_vv
; GFX11-TRUE16: liveins: $vgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
- ; GFX11-TRUE16-NEXT: [[V_XOR_B16_t16_e64_:%[0-9]+]]:vgpr_16 = V_XOR_B16_t16_e64 0, 32768, 0, [[COPY1]], 0, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[V_XOR_B16_t16_e64_]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 1, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
;
; GFX11-FAKE16-LABEL: name: fptosi_s16_to_s32_fneg_vv
; GFX11-FAKE16: liveins: $vgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-FAKE16-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; GFX11-FAKE16-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_I32_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
%2:vgpr(s16) = G_FNEG %1
- %3:vgpr(s32) = G_FPTOSI %2
- $vgpr0 = COPY %3
+ %3:vgpr(s32) = G_FPEXT %2
+ %4:vgpr(s32) = G_FPTOSI %3
+ $vgpr0 = COPY %4
...
---
@@ -276,16 +272,16 @@ body: |
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
;
; VI-LABEL: name: fptosi_s16_to_s1_vv
; VI: liveins: $vgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptosi_s16_to_s1_vv
; GFX11-TRUE16: liveins: $vgpr0
@@ -293,8 +289,8 @@ body: |
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_I32_F32_e32_]].lo16
+ ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_I32_F32_e64_]].lo16
; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[COPY2]]
;
; GFX11-FAKE16-LABEL: name: fptosi_s16_to_s1_vv
@@ -302,13 +298,14 @@ body: |
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
- %2:vgpr(s32) = G_FPTOSI %1
- %3:vgpr(s1) = G_TRUNC %2
- S_ENDPGM 0, implicit %3
+ %2:vgpr(s32) = G_FPEXT %1
+ %3:vgpr(s32) = G_FPTOSI %2
+ %4:vgpr(s1) = G_TRUNC %3
+ S_ENDPGM 0, implicit %4
...
---
@@ -326,24 +323,24 @@ body: |
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
;
; VI-LABEL: name: fptosi_s16_to_s1_vs
; VI: liveins: $sgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptosi_s16_to_s1_vs
; GFX11-TRUE16: liveins: $sgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[COPY]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_I32_F32_e32_]].lo16
+ ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_I32_F32_e64_]].lo16
; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[COPY1]]
;
; GFX11-FAKE16-LABEL: name: fptosi_s16_to_s1_vs
@@ -351,13 +348,14 @@ body: |
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
%0:sgpr(s32) = COPY $sgpr0
%1:sgpr(s16) = G_TRUNC %0
- %2:vgpr(s32) = G_FPTOSI %1
- %3:vgpr(s1) = G_TRUNC %2
- S_ENDPGM 0, implicit %3
+ %2:vgpr(s32) = G_FPEXT %1
+ %3:vgpr(s32) = G_FPTOSI %2
+ %4:vgpr(s1) = G_TRUNC %3
+ S_ENDPGM 0, implicit %4
...
---
@@ -374,46 +372,40 @@ body: |
; GCN: liveins: $vgpr0
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GCN-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; GCN-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
;
; VI-LABEL: name: fptosi_s16_to_s1_fneg_vv
; VI: liveins: $vgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; VI-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; VI-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptosi_s16_to_s1_fneg_vv
; GFX11-TRUE16: liveins: $vgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
- ; GFX11-TRUE16-NEXT: [[V_XOR_B16_t16_e64_:%[0-9]+]]:vgpr_16 = V_XOR_B16_t16_e64 0, 32768, 0, [[COPY1]], 0, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[V_XOR_B16_t16_e64_]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_I32_F32_e32_]].lo16
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 1, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_I32_F32_e64_]].lo16
; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[COPY2]]
;
; GFX11-FAKE16-LABEL: name: fptosi_s16_to_s1_fneg_vv
; GFX11-FAKE16: liveins: $vgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-FAKE16-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; GFX11-FAKE16-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_I32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_I32_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
%2:vgpr(s16) = G_FNEG %1
- %3:vgpr(s32) = G_FPTOSI %2
- %4:vgpr(s1) = G_TRUNC %3
- S_ENDPGM 0, implicit %4
+ %3:vgpr(s32) = G_FPEXT %2
+ %4:vgpr(s32) = G_FPTOSI %3
+ %5:vgpr(s1) = G_TRUNC %4
+ S_ENDPGM 0, implicit %5
...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir
index 521a0e8a2a796..93c271e59e5f9 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir
@@ -75,16 +75,16 @@ body: |
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; VI-LABEL: name: fptoui_s16_to_s32_vv
; VI: liveins: $vgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptoui_s16_to_s32_vv
; GFX11-TRUE16: liveins: $vgpr0
@@ -92,20 +92,21 @@ body: |
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; GFX11-FAKE16-LABEL: name: fptoui_s16_to_s32_vv
; GFX11-FAKE16: liveins: $vgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
- %2:vgpr(s32) = G_FPTOUI %1
- $vgpr0 = COPY %2
+ %2:vgpr(s32) = G_FPEXT %1
+ %3:vgpr(s32) = G_FPTOUI %2
+ $vgpr0 = COPY %3
...
---
@@ -123,36 +124,37 @@ body: |
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; VI-LABEL: name: fptoui_s16_to_s32_vs
; VI: liveins: $sgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptoui_s16_to_s32_vs
; GFX11-TRUE16: liveins: $sgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[COPY]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; GFX11-FAKE16-LABEL: name: fptoui_s16_to_s32_vs
; GFX11-FAKE16: liveins: $sgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
%0:sgpr(s32) = COPY $sgpr0
%1:sgpr(s16) = G_TRUNC %0
- %2:vgpr(s32) = G_FPTOUI %1
- $vgpr0 = COPY %2
+ %2:vgpr(s32) = G_FPEXT %1
+ %3:vgpr(s32) = G_FPTOUI %2
+ $vgpr0 = COPY %3
...
---
@@ -169,46 +171,40 @@ body: |
; GCN: liveins: $vgpr0
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GCN-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; GCN-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; VI-LABEL: name: fptoui_s16_to_s32_fneg_vv
; VI: liveins: $vgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; VI-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; VI-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptoui_s16_to_s32_fneg_vv
; GFX11-TRUE16: liveins: $vgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
- ; GFX11-TRUE16-NEXT: [[V_XOR_B16_t16_e64_:%[0-9]+]]:vgpr_16 = V_XOR_B16_t16_e64 0, 32768, 0, [[COPY1]], 0, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[V_XOR_B16_t16_e64_]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 1, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
;
; GFX11-FAKE16-LABEL: name: fptoui_s16_to_s32_fneg_vv
; GFX11-FAKE16: liveins: $vgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-FAKE16-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; GFX11-FAKE16-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_U32_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
%2:vgpr(s16) = G_FNEG %1
- %3:vgpr(s32) = G_FPTOUI %2
- $vgpr0 = COPY %3
+ %3:vgpr(s32) = G_FPEXT %2
+ %4:vgpr(s32) = G_FPTOUI %3
+ $vgpr0 = COPY %4
...
---
@@ -226,16 +222,16 @@ body: |
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
;
; VI-LABEL: name: fptoui_s16_to_s1_vv
; VI: liveins: $vgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptoui_s16_to_s1_vv
; GFX11-TRUE16: liveins: $vgpr0
@@ -243,8 +239,8 @@ body: |
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_U32_F32_e32_]].lo16
+ ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_U32_F32_e64_]].lo16
; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[COPY2]]
;
; GFX11-FAKE16-LABEL: name: fptoui_s16_to_s1_vv
@@ -252,13 +248,14 @@ body: |
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
- %2:vgpr(s32) = G_FPTOUI %1
- %3:vgpr(s1) = G_TRUNC %2
- S_ENDPGM 0, implicit %3
+ %2:vgpr(s32) = G_FPEXT %1
+ %3:vgpr(s32) = G_FPTOUI %2
+ %4:vgpr(s1) = G_TRUNC %3
+ S_ENDPGM 0, implicit %4
...
---
@@ -276,24 +273,24 @@ body: |
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
;
; VI-LABEL: name: fptoui_s16_to_s1_vs
; VI: liveins: $sgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptoui_s16_to_s1_vs
; GFX11-TRUE16: liveins: $sgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[COPY]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_U32_F32_e32_]].lo16
+ ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_U32_F32_e64_]].lo16
; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[COPY1]]
;
; GFX11-FAKE16-LABEL: name: fptoui_s16_to_s1_vs
@@ -301,13 +298,14 @@ body: |
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
%0:sgpr(s32) = COPY $sgpr0
%1:sgpr(s16) = G_TRUNC %0
- %2:vgpr(s32) = G_FPTOUI %1
- %3:vgpr(s1) = G_TRUNC %2
- S_ENDPGM 0, implicit %3
+ %2:vgpr(s32) = G_FPEXT %1
+ %3:vgpr(s32) = G_FPTOUI %2
+ %4:vgpr(s1) = G_TRUNC %3
+ S_ENDPGM 0, implicit %4
...
---
@@ -324,46 +322,40 @@ body: |
; GCN: liveins: $vgpr0
; GCN-NEXT: {{ $}}
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GCN-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; GCN-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GCN-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; GCN-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
;
; VI-LABEL: name: fptoui_s16_to_s1_fneg_vv
; VI: liveins: $vgpr0
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; VI-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; VI-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; VI-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_e64_]], implicit $mode, implicit $exec
- ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; VI-NEXT: [[V_CVT_F32_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; VI-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: fptoui_s16_to_s1_fneg_vv
; GFX11-TRUE16: liveins: $vgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
- ; GFX11-TRUE16-NEXT: [[V_XOR_B16_t16_e64_:%[0-9]+]]:vgpr_16 = V_XOR_B16_t16_e64 0, 32768, 0, [[COPY1]], 0, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 0, [[V_XOR_B16_t16_e64_]], 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_t16_e64_]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_U32_F32_e32_]].lo16
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F32_F16_t16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_t16_e64 1, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_t16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[V_CVT_U32_F32_e64_]].lo16
; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[COPY2]]
;
; GFX11-FAKE16-LABEL: name: fptoui_s16_to_s1_fneg_vv
; GFX11-FAKE16: liveins: $vgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-FAKE16-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 32768
- ; GFX11-FAKE16-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 0, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e32 [[V_CVT_F32_F16_fake16_e64_]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e32_]]
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F32_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F32_F16_fake16_e64 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_U32_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, [[V_CVT_F32_F16_fake16_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: S_ENDPGM 0, implicit [[V_CVT_U32_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
%2:vgpr(s16) = G_FNEG %1
- %3:vgpr(s32) = G_FPTOUI %2
- %4:vgpr(s1) = G_TRUNC %3
- S_ENDPGM 0, implicit %4
+ %3:vgpr(s32) = G_FPEXT %2
+ %4:vgpr(s32) = G_FPTOUI %3
+ %5:vgpr(s1) = G_TRUNC %4
+ S_ENDPGM 0, implicit %5
...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-sitofp.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-sitofp.mir
index 3888ce87b46fd..feac719a4cf1a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-sitofp.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-sitofp.mir
@@ -74,24 +74,24 @@ body: |
; WAVE64: liveins: $vgpr0
; WAVE64-NEXT: {{ $}}
; WAVE64-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; WAVE64-NEXT: [[V_CVT_F32_I32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e32 [[COPY]], implicit $mode, implicit $exec
- ; WAVE64-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_I32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE64-NEXT: [[V_CVT_F32_I32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE64-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_I32_e64_]], 0, 0, implicit $mode, implicit $exec
; WAVE64-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_e64_]]
;
; WAVE32-LABEL: name: sitofp_s32_to_s16_vv
; WAVE32: liveins: $vgpr0
; WAVE32-NEXT: {{ $}}
; WAVE32-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; WAVE32-NEXT: [[V_CVT_F32_I32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e32 [[COPY]], implicit $mode, implicit $exec
- ; WAVE32-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_I32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE32-NEXT: [[V_CVT_F32_I32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE32-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_I32_e64_]], 0, 0, implicit $mode, implicit $exec
; WAVE32-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: sitofp_s32_to_s16_vv
; GFX11-TRUE16: liveins: $vgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-TRUE16-NEXT: [[V_CVT_F32_I32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e32 [[COPY]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[V_CVT_F32_I32_e32_]], 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F32_I32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[V_CVT_F32_I32_e64_]], 0, 0, 0, implicit $mode, implicit $exec
; GFX11-TRUE16-NEXT: [[DEF:%[0-9]+]]:vgpr_16 = IMPLICIT_DEF
; GFX11-TRUE16-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vgpr_32 = REG_SEQUENCE [[V_CVT_F16_F32_t16_e64_]], %subreg.lo16, [[DEF]], %subreg.hi16
; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[REG_SEQUENCE]]
@@ -100,13 +100,14 @@ body: |
; GFX11-FAKE16: liveins: $vgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-FAKE16-NEXT: [[V_CVT_F32_I32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e32 [[COPY]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_F16_F32_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_fake16_e64 0, [[V_CVT_F32_I32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F32_I32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F16_F32_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_fake16_e64 0, [[V_CVT_F32_I32_e64_]], 0, 0, implicit $mode, implicit $exec
; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_fake16_e64_]]
%0:vgpr(s32) = COPY $vgpr0
- %1:vgpr(s16) = G_SITOFP %0
- %2:vgpr(s32) = G_ANYEXT %1
- $vgpr0 = COPY %2
+ %1:vgpr(s32) = G_SITOFP %0
+ %2:vgpr(s16) = G_FPTRUNC %1
+ %3:vgpr(s32) = G_ANYEXT %2
+ $vgpr0 = COPY %3
...
---
@@ -123,24 +124,24 @@ body: |
; WAVE64: liveins: $sgpr0
; WAVE64-NEXT: {{ $}}
; WAVE64-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
- ; WAVE64-NEXT: [[V_CVT_F32_I32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e32 [[COPY]], implicit $mode, implicit $exec
- ; WAVE64-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_I32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE64-NEXT: [[V_CVT_F32_I32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE64-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_I32_e64_]], 0, 0, implicit $mode, implicit $exec
; WAVE64-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_e64_]]
;
; WAVE32-LABEL: name: sitofp_s32_to_s16_vs
; WAVE32: liveins: $sgpr0
; WAVE32-NEXT: {{ $}}
; WAVE32-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
- ; WAVE32-NEXT: [[V_CVT_F32_I32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e32 [[COPY]], implicit $mode, implicit $exec
- ; WAVE32-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_I32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE32-NEXT: [[V_CVT_F32_I32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE32-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_I32_e64_]], 0, 0, implicit $mode, implicit $exec
; WAVE32-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: sitofp_s32_to_s16_vs
; GFX11-TRUE16: liveins: $sgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
- ; GFX11-TRUE16-NEXT: [[V_CVT_F32_I32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e32 [[COPY]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[V_CVT_F32_I32_e32_]], 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F32_I32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[V_CVT_F32_I32_e64_]], 0, 0, 0, implicit $mode, implicit $exec
; GFX11-TRUE16-NEXT: [[DEF:%[0-9]+]]:vgpr_16 = IMPLICIT_DEF
; GFX11-TRUE16-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vgpr_32 = REG_SEQUENCE [[V_CVT_F16_F32_t16_e64_]], %subreg.lo16, [[DEF]], %subreg.hi16
; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[REG_SEQUENCE]]
@@ -149,11 +150,12 @@ body: |
; GFX11-FAKE16: liveins: $sgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
- ; GFX11-FAKE16-NEXT: [[V_CVT_F32_I32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e32 [[COPY]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_F16_F32_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_fake16_e64 0, [[V_CVT_F32_I32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F32_I32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_I32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F16_F32_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_fake16_e64 0, [[V_CVT_F32_I32_e64_]], 0, 0, implicit $mode, implicit $exec
; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_fake16_e64_]]
%0:sgpr(s32) = COPY $sgpr0
- %1:vgpr(s16) = G_SITOFP %0
- %2:vgpr(s32) = G_ANYEXT %1
- $vgpr0 = COPY %2
+ %1:vgpr(s32) = G_SITOFP %0
+ %2:vgpr(s16) = G_FPTRUNC %1
+ %3:vgpr(s32) = G_ANYEXT %2
+ $vgpr0 = COPY %3
...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-uitofp.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-uitofp.mir
index 35d622dc57d18..cd427e65b52c0 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-uitofp.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-uitofp.mir
@@ -88,24 +88,24 @@ body: |
; WAVE64: liveins: $vgpr0
; WAVE64-NEXT: {{ $}}
; WAVE64-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; WAVE64-NEXT: [[V_CVT_F32_U32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e32 [[COPY]], implicit $mode, implicit $exec
- ; WAVE64-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_U32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE64-NEXT: [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE64-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_U32_e64_]], 0, 0, implicit $mode, implicit $exec
; WAVE64-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_e64_]]
;
; WAVE32-LABEL: name: uitofp_s32_to_s16_vv
; WAVE32: liveins: $vgpr0
; WAVE32-NEXT: {{ $}}
; WAVE32-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; WAVE32-NEXT: [[V_CVT_F32_U32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e32 [[COPY]], implicit $mode, implicit $exec
- ; WAVE32-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_U32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE32-NEXT: [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE32-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_U32_e64_]], 0, 0, implicit $mode, implicit $exec
; WAVE32-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: uitofp_s32_to_s16_vv
; GFX11-TRUE16: liveins: $vgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-TRUE16-NEXT: [[V_CVT_F32_U32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e32 [[COPY]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[V_CVT_F32_U32_e32_]], 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[V_CVT_F32_U32_e64_]], 0, 0, 0, implicit $mode, implicit $exec
; GFX11-TRUE16-NEXT: [[DEF:%[0-9]+]]:vgpr_16 = IMPLICIT_DEF
; GFX11-TRUE16-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vgpr_32 = REG_SEQUENCE [[V_CVT_F16_F32_t16_e64_]], %subreg.lo16, [[DEF]], %subreg.hi16
; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[REG_SEQUENCE]]
@@ -114,13 +114,14 @@ body: |
; GFX11-FAKE16: liveins: $vgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-FAKE16-NEXT: [[V_CVT_F32_U32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e32 [[COPY]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_F16_F32_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_fake16_e64 0, [[V_CVT_F32_U32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F16_F32_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_fake16_e64 0, [[V_CVT_F32_U32_e64_]], 0, 0, implicit $mode, implicit $exec
; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_fake16_e64_]]
%0:vgpr(s32) = COPY $vgpr0
- %1:vgpr(s16) = G_UITOFP %0
- %2:vgpr(s32) = G_ANYEXT %1
- $vgpr0 = COPY %2
+ %1:vgpr(s32) = G_UITOFP %0
+ %2:vgpr(s16) = G_FPTRUNC %1
+ %3:vgpr(s32) = G_ANYEXT %2
+ $vgpr0 = COPY %3
...
---
@@ -137,24 +138,24 @@ body: |
; WAVE64: liveins: $sgpr0
; WAVE64-NEXT: {{ $}}
; WAVE64-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
- ; WAVE64-NEXT: [[V_CVT_F32_U32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e32 [[COPY]], implicit $mode, implicit $exec
- ; WAVE64-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_U32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE64-NEXT: [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE64-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_U32_e64_]], 0, 0, implicit $mode, implicit $exec
; WAVE64-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_e64_]]
;
; WAVE32-LABEL: name: uitofp_s32_to_s16_vs
; WAVE32: liveins: $sgpr0
; WAVE32-NEXT: {{ $}}
; WAVE32-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
- ; WAVE32-NEXT: [[V_CVT_F32_U32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e32 [[COPY]], implicit $mode, implicit $exec
- ; WAVE32-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_U32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE32-NEXT: [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; WAVE32-NEXT: [[V_CVT_F16_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_e64 0, [[V_CVT_F32_U32_e64_]], 0, 0, implicit $mode, implicit $exec
; WAVE32-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_e64_]]
;
; GFX11-TRUE16-LABEL: name: uitofp_s32_to_s16_vs
; GFX11-TRUE16: liveins: $sgpr0
; GFX11-TRUE16-NEXT: {{ $}}
; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
- ; GFX11-TRUE16-NEXT: [[V_CVT_F32_U32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e32 [[COPY]], implicit $mode, implicit $exec
- ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[V_CVT_F32_U32_e32_]], 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[V_CVT_F32_U32_e64_]], 0, 0, 0, implicit $mode, implicit $exec
; GFX11-TRUE16-NEXT: [[DEF:%[0-9]+]]:vgpr_16 = IMPLICIT_DEF
; GFX11-TRUE16-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vgpr_32 = REG_SEQUENCE [[V_CVT_F16_F32_t16_e64_]], %subreg.lo16, [[DEF]], %subreg.hi16
; GFX11-TRUE16-NEXT: $vgpr0 = COPY [[REG_SEQUENCE]]
@@ -163,11 +164,12 @@ body: |
; GFX11-FAKE16: liveins: $sgpr0
; GFX11-FAKE16-NEXT: {{ $}}
; GFX11-FAKE16-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
- ; GFX11-FAKE16-NEXT: [[V_CVT_F32_U32_e32_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e32 [[COPY]], implicit $mode, implicit $exec
- ; GFX11-FAKE16-NEXT: [[V_CVT_F16_F32_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_fake16_e64 0, [[V_CVT_F32_U32_e32_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-FAKE16-NEXT: [[V_CVT_F16_F32_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_F16_F32_fake16_e64 0, [[V_CVT_F32_U32_e64_]], 0, 0, implicit $mode, implicit $exec
; GFX11-FAKE16-NEXT: $vgpr0 = COPY [[V_CVT_F16_F32_fake16_e64_]]
%0:sgpr(s32) = COPY $sgpr0
- %1:vgpr(s16) = G_UITOFP %0
- %2:vgpr(s32) = G_ANYEXT %1
- $vgpr0 = COPY %2
+ %1:vgpr(s32) = G_UITOFP %0
+ %2:vgpr(s16) = G_FPTRUNC %1
+ %3:vgpr(s32) = G_ANYEXT %2
+ $vgpr0 = COPY %3
...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fpowi.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fpowi.mir
index 3b0b2cf363d9e..249350b41d4a1 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fpowi.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fpowi.mir
@@ -40,13 +40,14 @@ body: |
; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; GFX9-NEXT: [[SITOFP:%[0-9]+]]:_(s16) = G_SITOFP [[COPY1]](s32)
+ ; GFX9-NEXT: [[SITOFP:%[0-9]+]]:_(s32) = G_SITOFP [[COPY1]](s32)
+ ; GFX9-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[SITOFP]](s32)
; GFX9-NEXT: [[FLOG2_:%[0-9]+]]:_(s16) = nnan G_FLOG2 [[TRUNC]]
; GFX9-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = nnan G_FPEXT [[FLOG2_]](s16)
- ; GFX9-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = nnan G_FPEXT [[SITOFP]](s16)
+ ; GFX9-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = nnan G_FPEXT [[FPTRUNC]](s16)
; GFX9-NEXT: [[INT:%[0-9]+]]:_(s32) = nnan G_INTRINSIC intrinsic(@llvm.amdgcn.fmul.legacy), [[FPEXT]](s32), [[FPEXT1]](s32)
- ; GFX9-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[INT]](s32)
- ; GFX9-NEXT: [[FEXP2_:%[0-9]+]]:_(s16) = nnan G_FEXP2 [[FPTRUNC]]
+ ; GFX9-NEXT: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[INT]](s32)
+ ; GFX9-NEXT: [[FEXP2_:%[0-9]+]]:_(s16) = nnan G_FEXP2 [[FPTRUNC1]]
; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FEXP2_]](s16)
; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
%0:_(s32) = COPY $vgpr0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
index 653d38cd389eb..c9412367d9679 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
@@ -566,7 +566,8 @@ body: |
; SI-NEXT: {{ $}}
; SI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; SI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; SI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; SI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; SI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; SI-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI]](s32)
; SI-NEXT: $vgpr0_vgpr1 = COPY [[SEXT]](s64)
; VI-LABEL: name: test_fptosi_s16_to_s64
@@ -574,7 +575,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI]](s32)
; VI-NEXT: $vgpr0_vgpr1 = COPY [[SEXT]](s64)
%0:_(s32) = COPY $vgpr0
@@ -598,9 +600,11 @@ body: |
; SI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
- ; SI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; SI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; SI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; SI-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI]](s32)
- ; SI-NEXT: [[FPTOSI1:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC1]](s16)
+ ; SI-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
+ ; SI-NEXT: [[FPTOSI1:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT1]](s32)
; SI-NEXT: [[SEXT1:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI1]](s32)
; SI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[SEXT]](s64), [[SEXT1]](s64)
; SI-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
@@ -613,9 +617,11 @@ body: |
; VI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI]](s32)
- ; VI-NEXT: [[FPTOSI1:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC1]](s16)
+ ; VI-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
+ ; VI-NEXT: [[FPTOSI1:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT1]](s32)
; VI-NEXT: [[SEXT1:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI1]](s32)
; VI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[SEXT]](s64), [[SEXT1]](s64)
; VI-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
@@ -643,7 +649,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: [[TRUNC1:%[0-9]+]]:_(s1) = G_TRUNC [[FPTOSI]](s32)
; VI-NEXT: S_ENDPGM 0, implicit [[TRUNC1]](s1)
%0:_(s32) = COPY $vgpr0
@@ -671,7 +678,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOSI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
@@ -699,7 +707,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOSI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
@@ -785,7 +794,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOSI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
@@ -813,7 +823,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOSI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
@@ -841,7 +852,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOSI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
index f57148cb0aff3..999d64c036de4 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
@@ -514,7 +514,8 @@ body: |
; SI-NEXT: {{ $}}
; SI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; SI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; SI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; SI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; SI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; SI-NEXT: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI]](s32)
; SI-NEXT: $vgpr0_vgpr1 = COPY [[ZEXT]](s64)
; VI-LABEL: name: test_fptoui_s16_to_s64
@@ -522,7 +523,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; VI-NEXT: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI]](s32)
; VI-NEXT: $vgpr0_vgpr1 = COPY [[ZEXT]](s64)
%0:_(s32) = COPY $vgpr0
@@ -546,9 +548,11 @@ body: |
; SI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
- ; SI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; SI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; SI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; SI-NEXT: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI]](s32)
- ; SI-NEXT: [[FPTOUI1:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC1]](s16)
+ ; SI-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
+ ; SI-NEXT: [[FPTOUI1:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT1]](s32)
; SI-NEXT: [[ZEXT1:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI1]](s32)
; SI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[ZEXT]](s64), [[ZEXT1]](s64)
; SI-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
@@ -561,9 +565,11 @@ body: |
; VI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
- ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; VI-NEXT: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI]](s32)
- ; VI-NEXT: [[FPTOUI1:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC1]](s16)
+ ; VI-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
+ ; VI-NEXT: [[FPTOUI1:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT1]](s32)
; VI-NEXT: [[ZEXT1:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI1]](s32)
; VI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[ZEXT]](s64), [[ZEXT1]](s64)
; VI-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
@@ -591,7 +597,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[FPEXT]](s32)
; VI-NEXT: [[TRUNC1:%[0-9]+]]:_(s1) = G_TRUNC [[FPTOSI]](s32)
; VI-NEXT: S_ENDPGM 0, implicit [[TRUNC1]](s1)
%0:_(s32) = COPY $vgpr0
@@ -619,7 +626,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOUI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
@@ -647,7 +655,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOUI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
@@ -715,7 +724,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOUI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
@@ -743,7 +753,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOUI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
@@ -771,7 +782,8 @@ body: |
; VI-NEXT: {{ $}}
; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+ ; VI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+ ; VI-NEXT: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[FPEXT]](s32)
; VI-NEXT: $vgpr0 = COPY [[FPTOUI]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir
index 238b7f6c09f1b..af4b4d946749b 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir
@@ -205,8 +205,9 @@ body: |
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX6-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 16
- ; GFX6-NEXT: [[SITOFP:%[0-9]+]]:_(s16) = G_SITOFP [[SEXT_INREG]](s32)
- ; GFX6-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SITOFP]](s16)
+ ; GFX6-NEXT: [[SITOFP:%[0-9]+]]:_(s32) = G_SITOFP [[SEXT_INREG]](s32)
+ ; GFX6-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[SITOFP]](s32)
+ ; GFX6-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
; GFX6-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX8-LABEL: name: test_sitofp_s16_to_s16
; GFX8: liveins: $vgpr0
@@ -286,16 +287,18 @@ body: |
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX6-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 8
- ; GFX6-NEXT: [[SITOFP:%[0-9]+]]:_(s16) = G_SITOFP [[SEXT_INREG]](s32)
- ; GFX6-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SITOFP]](s16)
+ ; GFX6-NEXT: [[SITOFP:%[0-9]+]]:_(s32) = G_SITOFP [[SEXT_INREG]](s32)
+ ; GFX6-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[SITOFP]](s32)
+ ; GFX6-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
; GFX6-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX8-LABEL: name: test_sitofp_s8_to_s16
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX8-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 8
- ; GFX8-NEXT: [[SITOFP:%[0-9]+]]:_(s16) = G_SITOFP [[SEXT_INREG]](s32)
- ; GFX8-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SITOFP]](s16)
+ ; GFX8-NEXT: [[SITOFP:%[0-9]+]]:_(s32) = G_SITOFP [[SEXT_INREG]](s32)
+ ; GFX8-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[SITOFP]](s32)
+ ; GFX8-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
; GFX8-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s8) = G_TRUNC %0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-uitofp.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-uitofp.mir
index e75495f53e410..347fe28641a5a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-uitofp.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-uitofp.mir
@@ -167,8 +167,9 @@ body: |
; GFX6-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX6-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; GFX6-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C]]
- ; GFX6-NEXT: [[UITOFP:%[0-9]+]]:_(s16) = G_UITOFP [[AND]](s32)
- ; GFX6-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UITOFP]](s16)
+ ; GFX6-NEXT: [[UITOFP:%[0-9]+]]:_(s32) = G_UITOFP [[AND]](s32)
+ ; GFX6-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[UITOFP]](s32)
+ ; GFX6-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
; GFX6-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX8-LABEL: name: test_uitofp_s16_to_s16
; GFX8: liveins: $vgpr0
@@ -253,8 +254,9 @@ body: |
; GFX6-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX6-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
; GFX6-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C]]
- ; GFX6-NEXT: [[UITOFP:%[0-9]+]]:_(s16) = G_UITOFP [[AND]](s32)
- ; GFX6-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UITOFP]](s16)
+ ; GFX6-NEXT: [[UITOFP:%[0-9]+]]:_(s32) = G_UITOFP [[AND]](s32)
+ ; GFX6-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[UITOFP]](s32)
+ ; GFX6-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
; GFX6-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX8-LABEL: name: test_uitofp_s8_to_s16
; GFX8: liveins: $vgpr0
@@ -262,8 +264,9 @@ body: |
; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
; GFX8-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C]]
- ; GFX8-NEXT: [[UITOFP:%[0-9]+]]:_(s16) = G_UITOFP [[AND]](s32)
- ; GFX8-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UITOFP]](s16)
+ ; GFX8-NEXT: [[UITOFP:%[0-9]+]]:_(s32) = G_UITOFP [[AND]](s32)
+ ; GFX8-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[UITOFP]](s32)
+ ; GFX8-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
; GFX8-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s8) = G_TRUNC %0
More information about the llvm-commits
mailing list