[llvm-branch-commits] [llvm] b943932 - Revert "[AMDGPU] Reapply "Support the RF and m inline asm constraints (#19512…"
via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Tue Nov 3 12:02:59 PST 2026
Author: Jun Wang
Date: 2026-11-03T12:02:53-08:00
New Revision: b943932fc28f11957cf4f7da7db232e58ae7d3af
URL: https://github.com/llvm/llvm-project/commit/b943932fc28f11957cf4f7da7db232e58ae7d3af
DIFF: https://github.com/llvm/llvm-project/commit/b943932fc28f11957cf4f7da7db232e58ae7d3af.diff
LOG: Revert "[AMDGPU] Reapply "Support the RF and m inline asm constraints (#19512…"
This reverts commit 0f678c0a737189b7151363c5189ebdbffc96074e.
Added:
Modified:
llvm/include/llvm/IR/InlineAsm.h
llvm/lib/CodeGen/MIRParser/MIParser.cpp
llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.h
llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h
llvm/lib/Target/AMDGPU/SIISelLowering.cpp
llvm/lib/Target/AMDGPU/SIISelLowering.h
llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
llvm/lib/Target/AMDGPU/SIInstrInfo.h
llvm/test/CodeGen/AMDGPU/hazard-inlineasm.mir
llvm/test/CodeGen/AMDGPU/hazard.mir
llvm/test/CodeGen/AMDGPU/vgpr-lowering-gfx1250.mir
Removed:
llvm/test/CodeGen/AMDGPU/inline-mem-constraints.ll
################################################################################
diff --git a/llvm/include/llvm/IR/InlineAsm.h b/llvm/include/llvm/IR/InlineAsm.h
index 63490729d46ed0..f4324066bdf416 100644
--- a/llvm/include/llvm/IR/InlineAsm.h
+++ b/llvm/include/llvm/IR/InlineAsm.h
@@ -273,10 +273,7 @@ class InlineAsm final : public Value {
ZS,
ZT,
- // AMDGPU-specific constraints
- RF,
-
- Max = RF,
+ Max = ZT,
};
// This class is intentionally packed into a 32b value as it is used as a
@@ -540,8 +537,6 @@ class InlineAsm final : public Value {
return "ZS";
case ConstraintCode::ZT:
return "ZT";
- case ConstraintCode::RF:
- return "RF";
default:
llvm_unreachable("Unknown memory constraint");
}
diff --git a/llvm/lib/CodeGen/MIRParser/MIParser.cpp b/llvm/lib/CodeGen/MIRParser/MIParser.cpp
index 83ec4e08633ad2..337f8992e20e52 100644
--- a/llvm/lib/CodeGen/MIRParser/MIParser.cpp
+++ b/llvm/lib/CodeGen/MIRParser/MIParser.cpp
@@ -1919,7 +1919,6 @@ bool MIParser::parseSymbolicInlineAsmOperand(unsigned OpIdx,
.Case("ZR", InlineAsm::ConstraintCode::ZR)
.Case("ZS", InlineAsm::ConstraintCode::ZS)
.Case("ZT", InlineAsm::ConstraintCode::ZT)
- .Case("RF", InlineAsm::ConstraintCode::RF)
.Default(InlineAsm::ConstraintCode::Unknown);
if (CC == InlineAsm::ConstraintCode::Unknown)
return error("unknown memory constraint '" + ConstraintStr + "'");
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 94af58a812ded9..2e35f199c7fbae 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -10767,14 +10767,8 @@ static bool prepareDAGLevelOperands(ConstraintDecisionInfo &Info,
assert((OpInfo.isIndirect ||
OpInfo.ConstraintType != TargetLowering::C_Memory) &&
"Operand must be indirect to be a mem!");
-
assert(InOperandVal.getValueType() ==
- TLI.getPointerTy(
- DAG.getDataLayout(),
- OpInfo.CallOperandVal
- ? cast<PointerType>(OpInfo.CallOperandVal->getType())
- ->getAddressSpace()
- : 0) &&
+ TLI.getPointerTy(DAG.getDataLayout()) &&
"Memory operands expect pointer values");
const InlineAsm::ConstraintCode ConstraintID =
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
index efcff9e7c70255..7a00daad056ee6 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -1959,16 +1959,6 @@ bool AMDGPUAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
return true;
}
-bool AMDGPUAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI,
- unsigned OpNo,
- const char *ExtraCode,
- raw_ostream &OS) {
- const MachineOperand &MO = MI->getOperand(OpNo);
- AMDGPUInstPrinter::printRegOperand(MO.getReg(), OS,
- *MF->getSubtarget().getRegisterInfo());
- return false;
-}
-
void AMDGPUAsmPrinter::getAnalysisUsage(AnalysisUsage &AU) const {
AU.addRequired<AMDGPUResourceUsageAnalysisWrapperPass>();
AU.addPreserved<AMDGPUResourceUsageAnalysisWrapperPass>();
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.h b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.h
index 598510a3eddaf2..4394cde308665d 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.h
@@ -156,9 +156,6 @@ class AMDGPUAsmPrinter final : public AsmPrinter {
bool PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
const char *ExtraCode, raw_ostream &O) override;
- bool PrintAsmMemoryOperand(const MachineInstr *MI, unsigned OpNum,
- const char *ExtraCode, raw_ostream &O) override;
-
protected:
void getAnalysisUsage(AnalysisUsage &AU) const override;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index 60eebe1846f4e7..0944b845550f7e 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -4934,33 +4934,6 @@ void AMDGPUDAGToDAGISel::PostprocessISelDAG() {
} while (IsModified);
}
-bool AMDGPUDAGToDAGISel::SelectInlineAsmMemoryOperand(
- const SDValue &Op, InlineAsm::ConstraintCode ConstraintID,
- std::vector<SDValue> &OutOps) {
- switch (ConstraintID) {
- case InlineAsm::ConstraintCode::m:
- OutOps.push_back(Op);
- return false;
- case InlineAsm::ConstraintCode::RF: {
- // flat_load/flat_store require the address in a VGPR. Always force a copy
- // to the appropriate VGPR class.
- const SIRegisterInfo *TRI = Subtarget->getRegisterInfo();
- MVT VT = Op.getSimpleValueType();
- const TargetRegisterClass *VRC =
- TRI->getVGPRClassForBitWidth(VT.getSizeInBits());
- assert(VRC && "Expected valid VGPR class for RF constraint operand");
- SDLoc DL(Op);
- SDValue RCVal = CurDAG->getTargetConstant(VRC->getID(), DL, MVT::i32);
- SDNode *Copy =
- CurDAG->getMachineNode(AMDGPU::COPY_TO_REGCLASS, DL, VT, {Op, RCVal});
- OutOps.push_back(SDValue(Copy, 0));
- return false;
- }
- default:
- return true;
- }
-}
-
AMDGPUDAGToDAGISelLegacy::AMDGPUDAGToDAGISelLegacy(TargetMachine &TM,
CodeGenOptLevel OptLevel)
: SelectionDAGISelLegacy(
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h
index 4d17be8758c4cb..ca587544b0ea66 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h
@@ -71,9 +71,6 @@ class AMDGPUDAGToDAGISel : public SelectionDAGISel {
void PreprocessISelDAG() override;
void Select(SDNode *N) override;
void PostprocessISelDAG() override;
- bool SelectInlineAsmMemoryOperand(const SDValue &Op,
- InlineAsm::ConstraintCode ConstraintID,
- std::vector<SDValue> &OutOps) override;
protected:
void SelectBuildVector(SDNode *N, unsigned RegClassID);
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 70ed6832a0b29b..99dcdf2424d2fc 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -20301,14 +20301,10 @@ SITargetLowering::getConstraintType(StringRef Constraint) const {
case 'v':
case 'a':
return C_RegisterClass;
- case 'm':
- return C_Memory;
}
} else if (Constraint.size() == 2) {
if (Constraint == "VA")
return C_RegisterClass;
- if (Constraint == "RF")
- return C_Memory;
}
if (isImmConstraint(Constraint)) {
return C_Other;
@@ -20316,14 +20312,6 @@ SITargetLowering::getConstraintType(StringRef Constraint) const {
return TargetLowering::getConstraintType(Constraint);
}
-InlineAsm::ConstraintCode
-SITargetLowering::getInlineAsmMemConstraint(StringRef ConstraintCode) const {
- if (ConstraintCode == "RF")
- return InlineAsm::ConstraintCode::RF;
-
- return TargetLowering::getInlineAsmMemConstraint(ConstraintCode);
-}
-
static uint64_t clearUnusedBits(uint64_t Val, unsigned Size) {
if (!AMDGPU::isInlinableIntLiteral(Val)) {
Val = Val & maskTrailingOnes<uint64_t>(Size);
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.h b/llvm/lib/Target/AMDGPU/SIISelLowering.h
index 1cee4d5b90649b..8ac70ab0c01d60 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.h
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.h
@@ -554,8 +554,6 @@ class SITargetLowering final : public AMDGPUTargetLowering {
getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
StringRef Constraint, MVT VT) const override;
ConstraintType getConstraintType(StringRef Constraint) const override;
- InlineAsm::ConstraintCode
- getInlineAsmMemConstraint(StringRef ConstraintCode) const override;
void LowerAsmOperandForConstraint(SDValue Op, StringRef Constraint,
std::vector<SDValue> &Ops,
SelectionDAG &DAG) const override;
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index b5b17c611b446f..7895c1c8a6a8fe 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -5505,7 +5505,7 @@ bool SIInstrInfo::verifyInstruction(const MachineInstr &MI,
if (!MRI.isSSA() && MI.isCopy())
return verifyCopy(MI, MRI, ErrInfo);
- if (SIInstrInfo::isGenericOpcode(Opcode) && !MI.isInlineAsm())
+ if (SIInstrInfo::isGenericOpcode(Opcode))
return true;
int Src0Idx = AMDGPU::getNamedOperandIdx(Opcode, AMDGPU::OpName::src0);
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.h b/llvm/lib/Target/AMDGPU/SIInstrInfo.h
index f130452d01d4c5..04210db5e6b913 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.h
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.h
@@ -283,14 +283,7 @@ class SIInstrInfo final : public AMDGPUGenInstrInfo {
// constraints should use 64-bit pointers.
const TargetRegisterClass *getInlineAsmMemoryOperandRegClass(
InlineAsm::ConstraintCode C) const override {
- switch (C) {
- case InlineAsm::ConstraintCode::RF:
- case InlineAsm::ConstraintCode::m:
- // flat_load/flat_store require a 64-bit VGPR pair for the address.
- return &AMDGPU::VReg_64RegClass;
- default:
- return &AMDGPU::VGPR_32RegClass;
- }
+ return &AMDGPU::VGPR_32RegClass;
}
const GCNSubtarget &getSubtarget() const {
diff --git a/llvm/test/CodeGen/AMDGPU/hazard-inlineasm.mir b/llvm/test/CodeGen/AMDGPU/hazard-inlineasm.mir
index 211eb1a6e6365d..502f735c3f24ba 100644
--- a/llvm/test/CodeGen/AMDGPU/hazard-inlineasm.mir
+++ b/llvm/test/CodeGen/AMDGPU/hazard-inlineasm.mir
@@ -18,7 +18,7 @@ tracksRegLiveness: false
body: |
bb.0:
FLAT_STORE_DWORDX4 $vgpr49_vgpr50, $vgpr26_vgpr27_vgpr28_vgpr29, 0, 0, implicit $exec, implicit $flat_scr
- INLINEASM &"v_mad_u64_u32 $0, $1, $2, $3, $4", 0, 2752522, def $vgpr26_vgpr27, 4718602, def dead $sgpr14_sgpr15, 1900553, $sgpr12, 1376265, killed $vgpr51, 2752521, $vgpr46_vgpr47
+ INLINEASM &"v_mad_u64_u32 $0, $1, $2, $3, $4", 0, 2621450, def $vgpr26_vgpr27, 2818058, def dead $sgpr14_sgpr15, 589833, $sgpr12, 327689, killed $vgpr51, 2621449, $vgpr46_vgpr47
S_ENDPGM 0
...
diff --git a/llvm/test/CodeGen/AMDGPU/hazard.mir b/llvm/test/CodeGen/AMDGPU/hazard.mir
index 02621c02e547f4..7647b2fda4976d 100644
--- a/llvm/test/CodeGen/AMDGPU/hazard.mir
+++ b/llvm/test/CodeGen/AMDGPU/hazard.mir
@@ -55,7 +55,7 @@ body: |
liveins: $sgpr7, $vgpr4
$m0 = S_MOV_B32 killed $sgpr7
- INLINEASM &"; no-op", 1, 10, def $vgpr5
+ INLINEASM &"; no-op", 1, 327690, def $vgpr5
$vgpr0 = V_INTERP_P1_F32 killed $vgpr4, 0, 0, implicit $mode, implicit $m0, implicit $exec
SI_RETURN_TO_EPILOG killed $vgpr5, killed $vgpr0
...
diff --git a/llvm/test/CodeGen/AMDGPU/inline-mem-constraints.ll b/llvm/test/CodeGen/AMDGPU/inline-mem-constraints.ll
deleted file mode 100644
index d6c89814033545..00000000000000
--- a/llvm/test/CodeGen/AMDGPU/inline-mem-constraints.ll
+++ /dev/null
@@ -1,334 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
-
-; RUN: llc -mtriple=amdgpu8.02 < %s | FileCheck --check-prefix=VI %s
-
-; NOTE: addrspace(2) (region/GDS), addrspace(3) (local/LDS), and addrspace(5)
-; (private/scratch) use 32-bit pointers in AMDGPU, which are incompatible with
-; flat_load/flat_store (which require a 64-bit address operand). Tests for
-; those address spaces are therefore absent.
-
-define void @flat_load_RF(ptr %in) {
-; VI-LABEL: flat_load_RF:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v2, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %addr = load ptr, ptr %in
- %a = tail call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*^RF"(ptr elementtype(i32) %addr)
- ret void
-}
-
-define void @flat_load_RF_as1(ptr %in) {
-; VI-LABEL: flat_load_RF_as1:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v2, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %addr = load ptr addrspace(1), ptr %in
- %a = tail call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*^RF"(ptr addrspace(1) elementtype(i32) %addr)
- ret void
-}
-
-define void @flat_load_RF_as4(ptr %in) {
-; VI-LABEL: flat_load_RF_as4:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v2, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %addr = load ptr addrspace(4), ptr %in
- %a = tail call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*^RF"(ptr addrspace(4) elementtype(i32) %addr)
- ret void
-}
-
-
-define void @flat_load_m_constraint(ptr %in) {
-; VI-LABEL: flat_load_m_constraint:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v2, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %addr = load ptr, ptr %in
- %a = tail call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*m"(ptr elementtype(i32) %addr)
- ret void
-}
-
-define void @flat_load_m_constraint_as1(ptr %in) {
-; VI-LABEL: flat_load_m_constraint_as1:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v2, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %addr = load ptr addrspace(1), ptr %in
- %a = tail call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*m"(ptr addrspace(1) elementtype(i32) %addr)
- ret void
-}
-
-define void @flat_load_m_constraint_as4(ptr %in) {
-; VI-LABEL: flat_load_m_constraint_as4:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v2, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %addr = load ptr addrspace(4), ptr %in
- %a = tail call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*m"(ptr addrspace(4) elementtype(i32) %addr)
- ret void
-}
-
-
-define void @flat_load_with_offset_RF(ptr %in) {
-; VI-LABEL: flat_load_with_offset_RF:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v2, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1] offset:4095
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %addr = load ptr, ptr %in
- %a = tail call i32 asm sideeffect "flat_load_dword $0, $1 offset:4095", "=v,*^RF"(ptr elementtype(i32) %addr)
- ret void
-}
-
-define void @flat_load_with_offset_m_constraint(ptr %in) {
-; VI-LABEL: flat_load_with_offset_m_constraint:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v2, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1] offset:4095
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %addr = load ptr, ptr %in
- %a = tail call i32 asm sideeffect "flat_load_dword $0, $1 offset:4095", "=v,*m"(ptr elementtype(i32) %addr)
- ret void
-}
-
-
-
-define void @flat_store_RF(ptr %in, ptr %in2) {
-; VI-LABEL: flat_store_RF:
-; VI: ; %bb.0:
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v4, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[4:5]
-; VI-NEXT: flat_load_dword v2, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_store_dword v[0:1], v2
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
- %addr = load ptr, ptr %in
- %data = load i32, ptr %in2
- call void asm "flat_store_dword $0, $1", "=*^RF,v"(ptr elementtype(i32) %addr, i32 %data)
- ret void
-}
-
-define void @flat_store_RF_as1(ptr %in, ptr %in2) {
-; VI-LABEL: flat_store_RF_as1:
-; VI: ; %bb.0:
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v4, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[4:5]
-; VI-NEXT: flat_load_dword v2, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_store_dword v[0:1], v2
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
- %addr = load ptr addrspace(1), ptr %in
- %data = load i32, ptr %in2
- call void asm "flat_store_dword $0, $1", "=*^RF,v"(ptr addrspace(1) elementtype(i32) %addr, i32 %data)
- ret void
-}
-
-define void @flat_store_RF_as4(ptr %in, ptr %in2) {
-; VI-LABEL: flat_store_RF_as4:
-; VI: ; %bb.0:
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v4, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[4:5]
-; VI-NEXT: flat_load_dword v2, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_store_dword v[0:1], v2
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
- %addr = load ptr addrspace(4), ptr %in
- %data = load i32, ptr %in2
- call void asm "flat_store_dword $0, $1", "=*^RF,v"(ptr addrspace(4) elementtype(i32) %addr, i32 %data)
- ret void
-}
-
-define void @flat_store_m_constraint(ptr %in, ptr %in2) {
-; VI-LABEL: flat_store_m_constraint:
-; VI: ; %bb.0:
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v4, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[4:5]
-; VI-NEXT: flat_load_dword v2, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_store_dword v[0:1], v2
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
- %addr = load ptr, ptr %in
- %data = load i32, ptr %in2
- call void asm "flat_store_dword $0, $1", "=*m,v"(ptr elementtype(i32) %addr, i32 %data)
- ret void
-}
-
-define void @flat_store_m_constraint_as1(ptr %in, ptr %in2) {
-; VI-LABEL: flat_store_m_constraint_as1:
-; VI: ; %bb.0:
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v4, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[4:5]
-; VI-NEXT: flat_load_dword v2, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_store_dword v[0:1], v2
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
- %addr = load ptr addrspace(1), ptr %in
- %data = load i32, ptr %in2
- call void asm "flat_store_dword $0, $1", "=*m,v"(ptr addrspace(1) elementtype(i32) %addr, i32 %data)
- ret void
-}
-
-define void @flat_store_m_constraint_as4(ptr %in, ptr %in2) {
-; VI-LABEL: flat_store_m_constraint_as4:
-; VI: ; %bb.0:
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_add_u32_e32 v4, vcc, 4, v0
-; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v1, vcc
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: flat_load_dword v1, v[4:5]
-; VI-NEXT: flat_load_dword v2, v[2:3]
-; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_store_dword v[0:1], v2
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
- %addr = load ptr addrspace(4), ptr %in
- %data = load i32, ptr %in2
- call void asm "flat_store_dword $0, $1", "=*m,v"(ptr addrspace(4) elementtype(i32) %addr, i32 %data)
- ret void
-}
-
-
-; Kernel args are loaded from the argument buffer into SGPRs.
-; The RF constraint must insert a SGPR->VGPR copy for the address.
-define amdgpu_kernel void @flat_load_RF_sgpr_addr(ptr %in) {
-; VI-LABEL: flat_load_RF_sgpr_addr:
-; VI: ; %bb.0:
-; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
-; VI-NEXT: s_waitcnt lgkmcnt(0)
-; VI-NEXT: v_mov_b32_e32 v0, s0
-; VI-NEXT: v_mov_b32_e32 v1, s1
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_endpgm
- %a = call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*^RF"(ptr elementtype(i32) %in)
- ret void
-}
-
-; Arguments passed as inreg may be in SGPR, in which case the RF constraint
-; must insert a SGPR->VGPR copy for the address.
-define void @flat_load_RF_inreg(ptr inreg %in) {
-; VI-LABEL: flat_load_RF_inreg:
-; VI: ; %bb.0: ; %entry
-; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT: v_mov_b32_e32 v0, s16
-; VI-NEXT: v_mov_b32_e32 v1, s17
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_setpc_b64 s[30:31]
-entry:
- %a = call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*^RF"(ptr elementtype(i32) %in)
- ret void
-}
-
-define amdgpu_kernel void @flat_load_RF_sgpr_addr_as1(ptr addrspace(1) %in) {
-; VI-LABEL: flat_load_RF_sgpr_addr_as1:
-; VI: ; %bb.0:
-; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
-; VI-NEXT: s_waitcnt lgkmcnt(0)
-; VI-NEXT: v_mov_b32_e32 v0, s0
-; VI-NEXT: v_mov_b32_e32 v1, s1
-; VI-NEXT: ;;#ASMSTART
-; VI-NEXT: flat_load_dword v0, v[0:1]
-; VI-NEXT: ;;#ASMEND
-; VI-NEXT: s_endpgm
- %a = call i32 asm sideeffect "flat_load_dword $0, $1", "=v,*^RF"(ptr addrspace(1) elementtype(i32) %in)
- ret void
-}
diff --git a/llvm/test/CodeGen/AMDGPU/vgpr-lowering-gfx1250.mir b/llvm/test/CodeGen/AMDGPU/vgpr-lowering-gfx1250.mir
index 051b7b8d9b83ac..89c3084160e75a 100644
--- a/llvm/test/CodeGen/AMDGPU/vgpr-lowering-gfx1250.mir
+++ b/llvm/test/CodeGen/AMDGPU/vgpr-lowering-gfx1250.mir
@@ -730,11 +730,11 @@ body: |
; GCN-NOT: s_set_vgpr_msb
; GCN: v_mov_b32_e32 v0 /*v256*/, v1
$vgpr256 = V_MOV_B32_e32 undef $vgpr1, implicit $exec
- INLINEASM &"; def v0", 1, 10, def $vgpr0
- INLINEASM &"; use v0", 1, 10, $vgpr0
- INLINEASM &"; use v1", 1, 10, undef $vgpr1
+ INLINEASM &"; def v0", 1, 327690, def $vgpr0
+ INLINEASM &"; use v0", 1, 327690, $vgpr0
+ INLINEASM &"; use v1", 1, 327690, undef $vgpr1
$vgpr256 = V_MOV_B32_e32 undef $vgpr1, implicit $exec
- INLINEASM &"; no vgprs, mode preserved", 1, 10, undef $sgpr0
+ INLINEASM &"; no vgprs, mode preserved", 1, 327690, undef $sgpr0
$vgpr256 = V_MOV_B32_e32 undef $vgpr1, implicit $exec
; ASM: NumVgprs: 257
More information about the llvm-branch-commits
mailing list