[llvm] r372290 - AMDGPU/GlobalISel: RegBankSelect llvm.amdgcn.raw.buffer.{load|store}
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Wed Sep 18 19:25:09 PDT 2019
Author: arsenm
Date: Wed Sep 18 19:25:09 2019
New Revision: 372290
URL: http://llvm.org/viewvc/llvm-project?rev=372290&view=rev
Log:
AMDGPU/GlobalISel: RegBankSelect llvm.amdgcn.raw.buffer.{load|store}
Added:
llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.load.ll
llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.store.ll
Modified:
llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.h
Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp?rev=372290&r1=372289&r2=372290&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp Wed Sep 18 19:25:09 2019
@@ -1452,6 +1452,12 @@ void AMDGPURegisterBankInfo::applyMappin
constrainOpWithReadfirstlane(MI, MRI, 2); // M0
return;
}
+ case Intrinsic::amdgcn_raw_buffer_load:
+ case Intrinsic::amdgcn_raw_buffer_store: {
+ applyDefaultMapping(OpdMapper);
+ executeInWaterfallLoop(MI, MRI, {2, 4});
+ return;
+ }
default: {
if (const AMDGPU::RsrcIntrinsic *RSrcIntrin =
AMDGPU::lookupRsrcIntrinsic(IntrID)) {
@@ -1656,11 +1662,31 @@ AMDGPURegisterBankInfo::getRegBankID(Reg
return Bank ? Bank->getID() : Default;
}
+
static unsigned regBankUnion(unsigned RB0, unsigned RB1) {
return (RB0 == AMDGPU::SGPRRegBankID && RB1 == AMDGPU::SGPRRegBankID) ?
AMDGPU::SGPRRegBankID : AMDGPU::VGPRRegBankID;
}
+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
/// AMDGPURegisterBankInfo::getInstrAlternativeMappings() is not called
@@ -2362,7 +2388,6 @@ AMDGPURegisterBankInfo::getInstrMapping(
OpdsMapping[1] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
break;
}
-
case Intrinsic::amdgcn_else: {
unsigned WaveSize = getSizeInBits(MI.getOperand(1).getReg(), MRI, *TRI);
OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::VCCRegBankID, 1);
@@ -2374,6 +2399,22 @@ AMDGPURegisterBankInfo::getInstrMapping(
OpdsMapping[1] = AMDGPU::getValueMapping(AMDGPU::VCCRegBankID, 1);
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:
if (const AMDGPU::RsrcIntrinsic *RSrcIntrin =
AMDGPU::lookupRsrcIntrinsic(IntrID)) {
Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.h?rev=372290&r1=372289&r2=372290&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.h (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPURegisterBankInfo.h Wed Sep 18 19:25:09 2019
@@ -62,6 +62,16 @@ class AMDGPURegisterBankInfo : public AM
const TargetRegisterInfo &TRI,
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
/// Regs. This appropriately sets the regbank of the new registers.
void split64BitValueForMapping(MachineIRBuilder &B,
Added: llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.load.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.load.ll?rev=372290&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.load.ll (added)
+++ llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.load.ll Wed Sep 18 19:25:09 2019
@@ -0,0 +1,173 @@
+; 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)
Added: llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.store.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.store.ll?rev=372290&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.store.ll (added)
+++ llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.store.ll Wed Sep 18 19:25:09 2019
@@ -0,0 +1,168 @@
+; 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)
More information about the llvm-commits
mailing list