This is an archive of the discontinued LLVM Phabricator instance.

Differential D65027

AMDGPU/GlobalISel: RegBankSelect llvm.amdgcn.raw.buffer.{load|store}
ClosedPublic

Authored by arsenm on Jul 19 2019, 4:04 PM.

Details

Reviewers
tstellar
nhaehnle

Diff Detail

Event Timeline

arsenm created this revision.Jul 19 2019, 4:04 PM
Herald added subscribers: Petar.Avramovic, t-tye, tpr and 6 others. · View Herald TranscriptJul 19 2019, 4:04 PM
arsenm updated this revision to Diff 210929.Jul 19 2019, 4:16 PM
arsenm retitled this revision from AMDGPU/GlobalISel: RegBankSelect llvm.amdgcn.raw.buffer.store to AMDGPU/GlobalISel: RegBankSelect llvm.amdgcn.raw.buffer.{load|store}.

Also handle the load

nhaehnle accepted this revision.Jul 24 2019, 1:45 AM

LGTM

This revision is now accepted and ready to land.Jul 24 2019, 1:45 AM
arsenm closed this revision.Sep 18 2019, 7:23 PM

r372290

Revision Contents

PathSize
lib/
Target/
AMDGPU/
10 lines
41 lines
test/
CodeGen/
AMDGPU/
GlobalISel/
173 lines
168 lines

Diff 210929

lib/Target/AMDGPU/AMDGPURegisterBankInfo.h

Show First 20 Lines • Show All 56 Lines • ▼ Show 20 Linesclass AMDGPURegisterBankInfo : public AMDGPUGenRegisterBankInfo {
const RegisterBankInfo::InstructionMapping & const RegisterBankInfo::InstructionMapping &
getInstrMappingForLoad(const MachineInstr &MI) const; getInstrMappingForLoad(const MachineInstr &MI) const;
unsigned getRegBankID(Register Reg, const MachineRegisterInfo &MRI, unsigned getRegBankID(Register Reg, const MachineRegisterInfo &MRI,
const TargetRegisterInfo &TRI, const TargetRegisterInfo &TRI,
unsigned Default = AMDGPU::VGPRRegBankID) const; unsigned Default = AMDGPU::VGPRRegBankID) const;
// Return a value mapping for an operand that is required to be an SGPR.
const ValueMapping *getSGPROpMapping(Register Reg,
const MachineRegisterInfo &MRI,
const TargetRegisterInfo &TRI) const;
// Return a value mapping for an operand that is required to be a VGPR.
const ValueMapping *getVGPROpMapping(Register Reg,
const MachineRegisterInfo &MRI,
const TargetRegisterInfo &TRI) const;
/// Split 64-bit value \p Reg into two 32-bit halves and populate them into \p /// Split 64-bit value \p Reg into two 32-bit halves and populate them into \p
/// Regs. This appropriately sets the regbank of the new registers. /// Regs. This appropriately sets the regbank of the new registers.
void split64BitValueForMapping(MachineIRBuilder &B, void split64BitValueForMapping(MachineIRBuilder &B,
SmallVector<Register, 2> &Regs, SmallVector<Register, 2> &Regs,
LLT HalfTy, LLT HalfTy,
Register Reg) const; Register Reg) const;
template <unsigned NumOps> template <unsigned NumOps>
▲ Show 20 Lines • Show All 49 Lines • Show Last 20 Lines

lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

Show First 20 Lines • Show All 1,379 Lines • ▼ Show 20 Lines case Intrinsic::amdgcn_ds_ordered_swap: {
return; return;
} }
case Intrinsic::amdgcn_s_sendmsg: case Intrinsic::amdgcn_s_sendmsg:
case Intrinsic::amdgcn_s_sendmsghalt: { case Intrinsic::amdgcn_s_sendmsghalt: {
// FIXME: Should this use a waterfall loop? // FIXME: Should this use a waterfall loop?
constrainOpWithReadfirstlane(MI, MRI, 2); // M0 constrainOpWithReadfirstlane(MI, MRI, 2); // M0
return; return;
} }
case Intrinsic::amdgcn_raw_buffer_load:
case Intrinsic::amdgcn_raw_buffer_store: {
applyDefaultMapping(OpdMapper);
executeInWaterfallLoop(MI, MRI, {2, 4});
return;
}
default: { default: {
if (const AMDGPU::RsrcIntrinsic *RSrcIntrin = if (const AMDGPU::RsrcIntrinsic *RSrcIntrin =
AMDGPU::lookupRsrcIntrinsic(IntrID)) { AMDGPU::lookupRsrcIntrinsic(IntrID)) {
// Non-images can have complications from operands that allow both SGPR // Non-images can have complications from operands that allow both SGPR
// and VGPR. For now it's too complicated to figure out the final opcode // and VGPR. For now it's too complicated to figure out the final opcode
// to derive the register bank from the MCInstrDesc. // to derive the register bank from the MCInstrDesc.
if (RSrcIntrin->IsImage) { if (RSrcIntrin->IsImage) {
applyMappingImage(MI, OpdMapper, MRI, RSrcIntrin->RsrcArg); applyMappingImage(MI, OpdMapper, MRI, RSrcIntrin->RsrcArg);
▲ Show 20 Lines • Show All 183 Lines • ▼ Show 20 LinesAMDGPURegisterBankInfo::getRegBankID(Register Reg,
const MachineRegisterInfo &MRI, const MachineRegisterInfo &MRI,
const TargetRegisterInfo &TRI, const TargetRegisterInfo &TRI,
unsigned Default) const { unsigned Default) const {
const RegisterBank *Bank = getRegBank(Reg, MRI, TRI); const RegisterBank *Bank = getRegBank(Reg, MRI, TRI);
return Bank ? Bank->getID() : Default; return Bank ? Bank->getID() : Default;
} }
const RegisterBankInfo::ValueMapping *
AMDGPURegisterBankInfo::getSGPROpMapping(Register Reg,
const MachineRegisterInfo &MRI,
const TargetRegisterInfo &TRI) const {
// Lie and claim anything is legal, even though this needs to be an SGPR
// applyMapping will have to deal with it as a waterfall loop.
unsigned Bank = getRegBankID(Reg, MRI, TRI, AMDGPU::SGPRRegBankID);
unsigned Size = getSizeInBits(Reg, MRI, TRI);
return AMDGPU::getValueMapping(Bank, Size);
}
const RegisterBankInfo::ValueMapping *
AMDGPURegisterBankInfo::getVGPROpMapping(Register Reg,
const MachineRegisterInfo &MRI,
const TargetRegisterInfo &TRI) const {
unsigned Size = getSizeInBits(Reg, MRI, TRI);
return AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size);
}
/// ///
/// This function must return a legal mapping, because /// This function must return a legal mapping, because
/// AMDGPURegisterBankInfo::getInstrAlternativeMappings() is not called /// AMDGPURegisterBankInfo::getInstrAlternativeMappings() is not called
/// in RegBankSelect::Mode::Fast. Any mapping that would cause a /// in RegBankSelect::Mode::Fast. Any mapping that would cause a
/// VGPR to SGPR generated is illegal. /// VGPR to SGPR generated is illegal.
/// ///
// Operands that must be SGPRs must accept potentially divergent VGPRs as // Operands that must be SGPRs must accept potentially divergent VGPRs as
// legal. These will be dealt with in applyMappingImpl. // legal. These will be dealt with in applyMappingImpl.
▲ Show 20 Lines • Show All 687 Lines • ▼ Show 20 Lines case Intrinsic::amdgcn_s_sendmsghalt: {
OpdsMapping[2] = AMDGPU::getValueMapping(Bank, 32); OpdsMapping[2] = AMDGPU::getValueMapping(Bank, 32);
break; break;
} }
case Intrinsic::amdgcn_end_cf: { case Intrinsic::amdgcn_end_cf: {
unsigned Size = getSizeInBits(MI.getOperand(1).getReg(), MRI, *TRI); unsigned Size = getSizeInBits(MI.getOperand(1).getReg(), MRI, *TRI);
OpdsMapping[1] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size); OpdsMapping[1] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
break; break;
} }
case Intrinsic::amdgcn_raw_buffer_load: {
// FIXME: Should make intrinsic ID the last operand of the instruction,
// then this would be the same as store
OpdsMapping[0] = getVGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);
OpdsMapping[2] = getSGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);
OpdsMapping[3] = getVGPROpMapping(MI.getOperand(3).getReg(), MRI, *TRI);
OpdsMapping[4] = getSGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI);
break;
}
case Intrinsic::amdgcn_raw_buffer_store: {
OpdsMapping[1] = getVGPROpMapping(MI.getOperand(1).getReg(), MRI, *TRI);
OpdsMapping[2] = getSGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);
OpdsMapping[3] = getVGPROpMapping(MI.getOperand(3).getReg(), MRI, *TRI);
OpdsMapping[4] = getSGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI);
break;
}
default: default:
if (const AMDGPU::RsrcIntrinsic *RSrcIntrin = if (const AMDGPU::RsrcIntrinsic *RSrcIntrin =
AMDGPU::lookupRsrcIntrinsic(IntrID)) { AMDGPU::lookupRsrcIntrinsic(IntrID)) {
// Non-images can have complications from operands that allow both SGPR // Non-images can have complications from operands that allow both SGPR
// and VGPR. For now it's too complicated to figure out the final opcode // and VGPR. For now it's too complicated to figure out the final opcode
// to derive the register bank from the MCInstrDesc. // to derive the register bank from the MCInstrDesc.
if (RSrcIntrin->IsImage) if (RSrcIntrin->IsImage)
return getImageMapping(MRI, MI, RSrcIntrin->RsrcArg); return getImageMapping(MRI, MI, RSrcIntrin->RsrcArg);
▲ Show 20 Lines • Show All 77 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.load.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck %s
; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck %s
; Natural mapping
define amdgpu_ps float @raw_buffer_load__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
; CHECK-LABEL: name: raw_buffer_load__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
; CHECK: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
; CHECK: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
; CHECK: [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[INT:%[0-9]+]]:vgpr(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.load), [[BUILD_VECTOR]](<4 x s32>), [[COPY4]](s32), [[COPY5]](s32), 0 :: (dereferenceable load 4 from custom TargetCustom7, align 1, addrspace 4)
; CHECK: $vgpr0 = COPY [[INT]](s32)
; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0
%val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret float %val
}
; Copies for VGPR arguments
define amdgpu_ps float @raw_buffer_load__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 inreg %voffset, i32 inreg %soffset) {
; CHECK-LABEL: name: raw_buffer_load__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7
; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
; CHECK: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
; CHECK: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
; CHECK: [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
; CHECK: [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[COPY6:%[0-9]+]]:vgpr(s32) = COPY [[COPY4]](s32)
; CHECK: [[INT:%[0-9]+]]:vgpr(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.load), [[BUILD_VECTOR]](<4 x s32>), [[COPY6]](s32), [[COPY5]](s32), 0 :: (dereferenceable load 4 from custom TargetCustom7, align 1, addrspace 4)
; CHECK: $vgpr0 = COPY [[INT]](s32)
; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0
%val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret float %val
}
; Waterfall for rsrc
define amdgpu_ps float @raw_buffer_load__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 inreg %soffset) {
; CHECK-LABEL: name: raw_buffer_load__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: successors: %bb.2(0x80000000)
; CHECK: liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
; CHECK: [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
; CHECK: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
; CHECK: [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
; CHECK: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
; CHECK: bb.2:
; CHECK: successors: %bb.3(0x40000000), %bb.2(0x40000000)
; CHECK: [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF1]], %bb.1, %15, %bb.2
; CHECK: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.1, %9(s32), %bb.2
; CHECK: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
; CHECK: [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
; CHECK: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
; CHECK: [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
; CHECK: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
; CHECK: [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
; CHECK: [[INT:%[0-9]+]]:vgpr(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.load), [[BUILD_VECTOR1]](<4 x s32>), [[COPY4]](s32), [[COPY5]](s32), 0 :: (dereferenceable load 4 from custom TargetCustom7, align 1, addrspace 4)
; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec
; CHECK: bb.3:
; CHECK: successors: %bb.4(0x80000000)
; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
; CHECK: bb.4:
; CHECK: $vgpr0 = COPY [[INT]](s32)
; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0
%val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret float %val
}
; Waterfall for soffset
define amdgpu_ps float @raw_buffer_load__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 %soffset) {
; CHECK-LABEL: name: raw_buffer_load__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: successors: %bb.2(0x80000000)
; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1
; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
; CHECK: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
; CHECK: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
; CHECK: [[COPY5:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr1
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
; CHECK: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
; CHECK: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
; CHECK: bb.2:
; CHECK: successors: %bb.3(0x40000000), %bb.2(0x40000000)
; CHECK: [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF1]], %bb.1, %15, %bb.2
; CHECK: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.1, %9(s32), %bb.2
; CHECK: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY5]](s32), implicit $exec
; CHECK: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY5]](s32), implicit $exec
; CHECK: [[INT:%[0-9]+]]:vgpr(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.load), [[BUILD_VECTOR]](<4 x s32>), [[COPY4]](s32), [[V_READFIRSTLANE_B32_]](s32), 0 :: (dereferenceable load 4 from custom TargetCustom7, align 1, addrspace 4)
; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec
; CHECK: bb.3:
; CHECK: successors: %bb.4(0x80000000)
; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
; CHECK: bb.4:
; CHECK: $vgpr0 = COPY [[INT]](s32)
; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0
%val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret float %val
}
; Waterfall for rsrc and soffset
define amdgpu_ps float @raw_buffer_load__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
; CHECK-LABEL: name: raw_buffer_load__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: successors: %bb.2(0x80000000)
; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
; CHECK: [[COPY5:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr5
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
; CHECK: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
; CHECK: [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
; CHECK: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
; CHECK: bb.2:
; CHECK: successors: %bb.3(0x40000000), %bb.2(0x40000000)
; CHECK: [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF1]], %bb.1, %15, %bb.2
; CHECK: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.1, %9(s32), %bb.2
; CHECK: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
; CHECK: [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
; CHECK: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
; CHECK: [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
; CHECK: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
; CHECK: [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
; CHECK: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY5]](s32), implicit $exec
; CHECK: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]](s32), [[COPY5]](s32), implicit $exec
; CHECK: [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
; CHECK: [[INT:%[0-9]+]]:vgpr(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.load), [[BUILD_VECTOR1]](<4 x s32>), [[COPY4]](s32), [[V_READFIRSTLANE_B32_4]](s32), 0 :: (dereferenceable load 4 from custom TargetCustom7, align 1, addrspace 4)
; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec
; CHECK: bb.3:
; CHECK: successors: %bb.4(0x80000000)
; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
; CHECK: bb.4:
; CHECK: $vgpr0 = COPY [[INT]](s32)
; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0
%val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret float %val
}
declare float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32>, i32, i32, i32 immarg)

test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.store.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck %s
; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck %s
; Natural mapping
define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
; CHECK: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
; CHECK: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
; CHECK: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
; CHECK: [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY4]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY5]](s32), [[COPY6]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
; CHECK: S_ENDPGM 0
call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret void
}
; Copies for VGPR arguments
define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, float inreg %val, i32 inreg %voffset, i32 inreg %soffset) {
; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8
; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
; CHECK: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
; CHECK: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
; CHECK: [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
; CHECK: [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
; CHECK: [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[COPY7:%[0-9]+]]:vgpr(s32) = COPY [[COPY4]](s32)
; CHECK: [[COPY8:%[0-9]+]]:vgpr(s32) = COPY [[COPY5]](s32)
; CHECK: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY7]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY8]](s32), [[COPY6]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
; CHECK: S_ENDPGM 0
call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret void
}
; Waterfall for rsrc
define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: successors: %bb.2(0x80000000)
; CHECK: liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
; CHECK: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
; CHECK: [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
; CHECK: [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
; CHECK: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
; CHECK: bb.2:
; CHECK: successors: %bb.3(0x40000000), %bb.2(0x40000000)
; CHECK: [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF]], %bb.1, %13, %bb.2
; CHECK: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
; CHECK: [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
; CHECK: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
; CHECK: [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
; CHECK: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
; CHECK: [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
; CHECK: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY4]](s32), [[BUILD_VECTOR1]](<4 x s32>), [[COPY5]](s32), [[COPY6]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec
; CHECK: bb.3:
; CHECK: successors: %bb.4(0x80000000)
; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
; CHECK: bb.4:
; CHECK: S_ENDPGM 0
call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret void
}
; Waterfall for soffset
define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 %soffset) {
; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: successors: %bb.2(0x80000000)
; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1, $vgpr2
; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
; CHECK: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
; CHECK: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
; CHECK: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
; CHECK: [[COPY6:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr2
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
; CHECK: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
; CHECK: bb.2:
; CHECK: successors: %bb.3(0x40000000), %bb.2(0x40000000)
; CHECK: [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF]], %bb.1, %13, %bb.2
; CHECK: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY6]](s32), implicit $exec
; CHECK: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY6]](s32), implicit $exec
; CHECK: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY4]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY5]](s32), [[V_READFIRSTLANE_B32_]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec
; CHECK: bb.3:
; CHECK: successors: %bb.4(0x80000000)
; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
; CHECK: bb.4:
; CHECK: S_ENDPGM 0
call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret void
}
; Waterfall for rsrc and soffset
define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
; CHECK: bb.1 (%ir-block.0):
; CHECK: successors: %bb.2(0x80000000)
; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
; CHECK: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
; CHECK: [[COPY6:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr6
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
; CHECK: [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
; CHECK: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
; CHECK: bb.2:
; CHECK: successors: %bb.3(0x40000000), %bb.2(0x40000000)
; CHECK: [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF]], %bb.1, %13, %bb.2
; CHECK: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
; CHECK: [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
; CHECK: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
; CHECK: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
; CHECK: [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
; CHECK: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
; CHECK: [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
; CHECK: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY6]](s32), implicit $exec
; CHECK: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]](s32), [[COPY6]](s32), implicit $exec
; CHECK: [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
; CHECK: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY4]](s32), [[BUILD_VECTOR1]](<4 x s32>), [[COPY5]](s32), [[V_READFIRSTLANE_B32_4]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec
; CHECK: bb.3:
; CHECK: successors: %bb.4(0x80000000)
; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
; CHECK: bb.4:
; CHECK: S_ENDPGM 0
call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret void
}
declare void @llvm.amdgcn.raw.buffer.store.f32(float, <4 x i32>, i32, i32, i32 immarg)