AMDGPU: Use RegClassByHwMode to manage operand VGPR operand constraints #158272
+2,770
−2,606
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
arsenm
added
backend:AMDGPU
backend:SystemZ
llvm:codegen
llvm:globalisel
llvm:mc
Graphite App
This stack of pull requests is managed by Graphite. Learn more about stacking. |
|
@llvm/pr-subscribers-tablegen @llvm/pr-subscribers-llvm-globalisel Author: Matt Arsenault (arsenm) ChangesThis removes special case processing in TargetInstrInfo::getRegClass to This regresses assembler diagnostics, which currently work by hackily Patch is 919.75 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/158272.diff 30 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index ffbda14dcd849..de7d4fac58296 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -2735,6 +2735,9 @@ def HasGetWaveIdInst : Predicate<"Subtarget->hasGetWaveIdInst()">,
def HasMAIInsts : Predicate<"Subtarget->hasMAIInsts()">,
AssemblerPredicate<(all_of FeatureMAIInsts)>;
+def NotHasMAIInsts : Predicate<"!Subtarget->hasMAIInsts()">,
+ AssemblerPredicate<(all_of (not FeatureMAIInsts))>;
+
def HasSMemRealTime : Predicate<"Subtarget->hasSMemRealTime()">,
AssemblerPredicate<(all_of FeatureSMemRealTime)>;
@@ -2909,6 +2912,23 @@ def HasLdsBarrierArriveAtomic : Predicate<"Subtarget->hasLdsBarrierArriveAtomic(
def HasSetPrioIncWgInst : Predicate<"Subtarget->hasSetPrioIncWgInst()">,
AssemblerPredicate<(all_of FeatureSetPrioIncWgInst)>;
+def NeedsAlignedVGPRs : Predicate<"Subtarget->needsAlignedVGPRs()">,
+ AssemblerPredicate<(all_of FeatureRequiresAlignedVGPRs)>;
+
+def HasAVAlign2AndAVLoadStore : Predicate<"Subtarget->needsAlignedVGPRs() && Subtarget->hasMAIInsts()">;
+def HasVGPRAlign2NoAGPR : Predicate<"Subtarget->needsAlignedVGPRs() && !Subtarget->hasMAIInsts()">;
+
+//===----------------------------------------------------------------------===//
+// HwModes
+//===----------------------------------------------------------------------===//
+
+// gfx90a-gfx950. Has AGPRs, and also the align2 VGPR/AGPR requirement
+def AVAlign2LoadStoreMode : HwMode<[HasMAIInsts, NeedsAlignedVGPRs]>;
+
+// gfx1250, has alignment requirement but no AGPRs.
+def AlignedVGPRNoAGPRMode : HwMode<[NotHasMAIInsts, NeedsAlignedVGPRs]>;
+
+
// Include AMDGPU TD files
include "SISchedule.td"
include "GCNProcessors.td"
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index c2fca79979e1b..cf0cb69d529e1 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -393,12 +393,13 @@ const TargetRegisterClass *AMDGPUDAGToDAGISel::getOperandRegClass(SDNode *N,
switch (N->getMachineOpcode()) {
default: {
- const MCInstrDesc &Desc =
- Subtarget->getInstrInfo()->get(N->getMachineOpcode());
+ const SIInstrInfo *TII = Subtarget->getInstrInfo();
+ const MCInstrDesc &Desc = TII->get(N->getMachineOpcode());
unsigned OpIdx = Desc.getNumDefs() + OpNo;
if (OpIdx >= Desc.getNumOperands())
return nullptr;
- int RegClass = Desc.operands()[OpIdx].RegClass;
+
+ int16_t RegClass = TII->getOpRegClassID(Desc.operands()[OpIdx]);
if (RegClass == -1)
return nullptr;
@@ -4338,7 +4339,8 @@ bool AMDGPUDAGToDAGISel::isVGPRImm(const SDNode * N) const {
if (!RC || SIRI->isSGPRClass(RC))
return false;
- if (RC != &AMDGPU::VS_32RegClass && RC != &AMDGPU::VS_64RegClass) {
+ if (RC != &AMDGPU::VS_32RegClass && RC != &AMDGPU::VS_64RegClass &&
+ RC != &AMDGPU::VS_64_Align2RegClass) {
AllUsesAcceptSReg = false;
SDNode *User = U->getUser();
if (User->isMachineOpcode()) {
@@ -4352,7 +4354,8 @@ bool AMDGPUDAGToDAGISel::isVGPRImm(const SDNode * N) const {
const TargetRegisterClass *CommutedRC =
getOperandRegClass(U->getUser(), CommutedOpNo);
if (CommutedRC == &AMDGPU::VS_32RegClass ||
- CommutedRC == &AMDGPU::VS_64RegClass)
+ CommutedRC == &AMDGPU::VS_64RegClass ||
+ CommutedRC == &AMDGPU::VS_64_Align2RegClass)
AllUsesAcceptSReg = true;
}
}
diff --git a/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp b/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
index e420f2ad676f9..7f22e84670fc0 100644
--- a/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
+++ b/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
@@ -1385,6 +1385,7 @@ class AMDGPUAsmParser : public MCTargetAsmParser {
bool ForcedDPP = false;
bool ForcedSDWA = false;
KernelScopeInfo KernelScope;
+ const unsigned HwMode;
/// @name Auto-generated Match Functions
/// {
@@ -1394,6 +1395,13 @@ class AMDGPUAsmParser : public MCTargetAsmParser {
/// }
+ /// Get size of register operand
+ unsigned getRegOperandSize(const MCInstrDesc &Desc, unsigned OpNo) const {
+ assert(OpNo < Desc.NumOperands);
+ int16_t RCID = MII.getOpRegClassID(Desc.operands()[OpNo], HwMode);
+ return getRegBitWidth(RCID) / 8;
+ }
+
private:
void createConstantSymbol(StringRef Id, int64_t Val);
@@ -1480,9 +1488,9 @@ class AMDGPUAsmParser : public MCTargetAsmParser {
using OptionalImmIndexMap = std::map<AMDGPUOperand::ImmTy, unsigned>;
AMDGPUAsmParser(const MCSubtargetInfo &STI, MCAsmParser &_Parser,
- const MCInstrInfo &MII,
- const MCTargetOptions &Options)
- : MCTargetAsmParser(Options, STI, MII), Parser(_Parser) {
+ const MCInstrInfo &MII, const MCTargetOptions &Options)
+ : MCTargetAsmParser(Options, STI, MII), Parser(_Parser),
+ HwMode(STI.getHwMode()) {
MCAsmParserExtension::Initialize(Parser);
if (getFeatureBits().none()) {
@@ -4197,7 +4205,7 @@ bool AMDGPUAsmParser::validateMIMGDataSize(const MCInst &Inst,
if ((DMaskIdx == -1 || TFEIdx == -1) && isGFX10_AEncoding()) // intersect_ray
return true;
- unsigned VDataSize = AMDGPU::getRegOperandSize(getMRI(), Desc, VDataIdx);
+ unsigned VDataSize = getRegOperandSize(Desc, VDataIdx);
unsigned TFESize = (TFEIdx != -1 && Inst.getOperand(TFEIdx).getImm()) ? 1 : 0;
unsigned DMask = Inst.getOperand(DMaskIdx).getImm() & 0xf;
if (DMask == 0)
@@ -4262,8 +4270,7 @@ bool AMDGPUAsmParser::validateMIMGAddrSize(const MCInst &Inst,
const AMDGPU::MIMGDimInfo *DimInfo = AMDGPU::getMIMGDimInfoByEncoding(Dim);
bool IsNSA = SrsrcIdx - VAddr0Idx > 1;
unsigned ActualAddrSize =
- IsNSA ? SrsrcIdx - VAddr0Idx
- : AMDGPU::getRegOperandSize(getMRI(), Desc, VAddr0Idx) / 4;
+ IsNSA ? SrsrcIdx - VAddr0Idx : getRegOperandSize(Desc, VAddr0Idx) / 4;
unsigned ExpectedAddrSize =
AMDGPU::getAddrSizeMIMGOp(BaseOpcode, DimInfo, IsA16, hasG16());
@@ -4273,8 +4280,7 @@ bool AMDGPUAsmParser::validateMIMGAddrSize(const MCInst &Inst,
ExpectedAddrSize >
getNSAMaxSize(Desc.TSFlags & SIInstrFlags::VSAMPLE)) {
int VAddrLastIdx = SrsrcIdx - 1;
- unsigned VAddrLastSize =
- AMDGPU::getRegOperandSize(getMRI(), Desc, VAddrLastIdx) / 4;
+ unsigned VAddrLastSize = getRegOperandSize(Desc, VAddrLastIdx) / 4;
ActualAddrSize = VAddrLastIdx - VAddr0Idx + VAddrLastSize;
}
@@ -4526,7 +4532,8 @@ bool AMDGPUAsmParser::validateMFMA(const MCInst &Inst,
return true;
const MCRegisterInfo *TRI = getContext().getRegisterInfo();
- if (TRI->getRegClass(Desc.operands()[0].RegClass).getSizeInBits() <= 128)
+ if (TRI->getRegClass(MII.getOpRegClassID(Desc.operands()[0], HwMode))
+ .getSizeInBits() <= 128)
return true;
if (TRI->regsOverlap(Src2Reg, DstReg)) {
@@ -5071,7 +5078,7 @@ bool AMDGPUAsmParser::validateDPP(const MCInst &Inst,
unsigned DppCtrl = Inst.getOperand(DppCtrlIdx).getImm();
if (!AMDGPU::isLegalDPALU_DPPControl(getSTI(), DppCtrl) &&
- AMDGPU::isDPALU_DPP(MII.get(Opc), getSTI())) {
+ AMDGPU::isDPALU_DPP(MII.get(Opc), MII, getSTI())) {
// DP ALU DPP is supported for row_newbcast only on GFX9* and row_share
// only on GFX12.
SMLoc S = getImmLoc(AMDGPUOperand::ImmTyDppCtrl, Operands);
@@ -5589,7 +5596,8 @@ bool AMDGPUAsmParser::validateWMMA(const MCInst &Inst,
unsigned Fmt = Inst.getOperand(FmtIdx).getImm();
int SrcIdx = AMDGPU::getNamedOperandIdx(Opc, SrcOp);
unsigned RegSize =
- TRI->getRegClass(Desc.operands()[SrcIdx].RegClass).getSizeInBits();
+ TRI->getRegClass(MII.getOpRegClassID(Desc.operands()[SrcIdx], HwMode))
+ .getSizeInBits();
if (RegSize == AMDGPU::wmmaScaleF8F6F4FormatToNumRegs(Fmt) * 32)
return true;
diff --git a/llvm/lib/Target/AMDGPU/BUFInstructions.td b/llvm/lib/Target/AMDGPU/BUFInstructions.td
index 09a66d785d5cf..b97b7385dc1ff 100644
--- a/llvm/lib/Target/AMDGPU/BUFInstructions.td
+++ b/llvm/lib/Target/AMDGPU/BUFInstructions.td
@@ -417,10 +417,10 @@ class getBUFVDataRegisterOperandForOp<RegisterOperand Op, bit isTFE> {
}
class getMUBUFInsDA<list<RegisterOperand> vdataList,
- list<RegisterClass> vaddrList, bit isTFE, bit hasRestrictedSOffset> {
+ list<RegisterClassLike> vaddrList, bit isTFE, bit hasRestrictedSOffset> {
RegisterOperand vdataClass = !if(!empty(vdataList), ?, !head(vdataList));
- RegisterClass vaddrClass = !if(!empty(vaddrList), ?, !head(vaddrList));
- RegisterOperand vdata_op = getBUFVDataRegisterOperandForOp<vdataClass, isTFE>.ret;
+ RegisterClassLike vaddrClass = !if(!empty(vaddrList), ?, !head(vaddrList));
+ RegisterOperand vdata_op = getBUFVDataRegisterOperand<!cast<SIRegisterClassLike>(vdataClass.RegClass).Size, isTFE>.ret;
dag SOffset = !if(hasRestrictedSOffset, (ins SReg_32:$soffset), (ins SCSrc_b32:$soffset));
dag NonVaddrInputs = !con((ins SReg_128_XNULL:$srsrc), SOffset, (ins Offset:$offset, CPol_0:$cpol, i1imm_0:$swz));
@@ -453,8 +453,8 @@ class getMUBUFIns<int addrKind, list<RegisterOperand> vdataList, bit isTFE, bit
!if(!eq(addrKind, BUFAddrKind.Offset), getMUBUFInsDA<vdataList, [], isTFE, hasRestrictedSOffset>.ret,
!if(!eq(addrKind, BUFAddrKind.OffEn), getMUBUFInsDA<vdataList, [VGPR_32], isTFE, hasRestrictedSOffset>.ret,
!if(!eq(addrKind, BUFAddrKind.IdxEn), getMUBUFInsDA<vdataList, [VGPR_32], isTFE, hasRestrictedSOffset>.ret,
- !if(!eq(addrKind, BUFAddrKind.BothEn), getMUBUFInsDA<vdataList, [VReg_64], isTFE, hasRestrictedSOffset>.ret,
- !if(!eq(addrKind, BUFAddrKind.Addr64), getMUBUFInsDA<vdataList, [VReg_64], isTFE, hasRestrictedSOffset>.ret,
+ !if(!eq(addrKind, BUFAddrKind.BothEn), getMUBUFInsDA<vdataList, [VReg_64_AlignTarget], isTFE, hasRestrictedSOffset>.ret,
+ !if(!eq(addrKind, BUFAddrKind.Addr64), getMUBUFInsDA<vdataList, [VReg_64_AlignTarget], isTFE, hasRestrictedSOffset>.ret,
(ins))))));
}
@@ -677,8 +677,8 @@ class MUBUF_Pseudo_Store_Lds<string opName>
}
class getMUBUFAtomicInsDA<RegisterOperand vdata_op, bit vdata_in, bit hasRestrictedSOffset,
- list<RegisterClass> vaddrList=[]> {
- RegisterClass vaddrClass = !if(!empty(vaddrList), ?, !head(vaddrList));
+ list<RegisterClassLike> vaddrList=[]> {
+ RegisterClassLike vaddrClass = !if(!empty(vaddrList), ?, !head(vaddrList));
dag VData = !if(vdata_in, (ins vdata_op:$vdata_in), (ins vdata_op:$vdata));
dag Data = !if(!empty(vaddrList), VData, !con(VData, (ins vaddrClass:$vaddr)));
@@ -702,9 +702,9 @@ class getMUBUFAtomicIns<int addrKind,
!if(!eq(addrKind, BUFAddrKind.IdxEn),
getMUBUFAtomicInsDA<vdataClass, vdata_in, hasRestrictedSOffset, [VGPR_32]>.ret,
!if(!eq(addrKind, BUFAddrKind.BothEn),
- getMUBUFAtomicInsDA<vdataClass, vdata_in, hasRestrictedSOffset, [VReg_64]>.ret,
+ getMUBUFAtomicInsDA<vdataClass, vdata_in, hasRestrictedSOffset, [VReg_64_AlignTarget]>.ret,
!if(!eq(addrKind, BUFAddrKind.Addr64),
- getMUBUFAtomicInsDA<vdataClass, vdata_in, hasRestrictedSOffset, [VReg_64]>.ret,
+ getMUBUFAtomicInsDA<vdataClass, vdata_in, hasRestrictedSOffset, [VReg_64_AlignTarget]>.ret,
(ins))))));
}
@@ -1568,11 +1568,12 @@ multiclass BufferAtomicCmpSwapPat_Common<ValueType vt, ValueType data_vt, string
# !if(!eq(RtnMode, "ret"), "", "_noret")
# "_" # vt);
defvar InstSuffix = !if(!eq(RtnMode, "ret"), "_RTN", "");
- defvar data_vt_RC = getVregSrcForVT<data_vt>.ret.RegClass;
+ defvar data_op = getVregSrcForVT<data_vt>.ret;
+ defvar data_vt_RC = getVregClassForVT<data_vt>.ret;
let AddedComplexity = !if(!eq(RtnMode, "ret"), 0, 1) in {
defvar OffsetResDag = (!cast<MUBUF_Pseudo>(Inst # "_OFFSET" # InstSuffix)
- data_vt_RC:$vdata_in, SReg_128:$srsrc, SCSrc_b32:$soffset,
+ data_op:$vdata_in, SReg_128:$srsrc, SCSrc_b32:$soffset,
Offset:$offset);
def : GCNPat<
(vt (Op (MUBUFOffset v4i32:$srsrc, i32:$soffset, i32:$offset), data_vt:$vdata_in)),
@@ -1583,7 +1584,7 @@ multiclass BufferAtomicCmpSwapPat_Common<ValueType vt, ValueType data_vt, string
>;
defvar Addr64ResDag = (!cast<MUBUF_Pseudo>(Inst # "_ADDR64" # InstSuffix)
- data_vt_RC:$vdata_in, VReg_64:$vaddr, SReg_128:$srsrc,
+ data_op:$vdata_in, VReg_64:$vaddr, SReg_128:$srsrc,
SCSrc_b32:$soffset, Offset:$offset);
def : GCNPat<
(vt (Op (MUBUFAddr64 v4i32:$srsrc, i64:$vaddr, i32:$soffset, i32:$offset),
@@ -1832,7 +1833,7 @@ multiclass SIBufferAtomicCmpSwapPat_Common<ValueType vt, ValueType data_vt, stri
(extract_cpol_set_glc $auxiliary),
(extract_cpol $auxiliary));
defvar SrcRC = getVregSrcForVT<vt>.ret;
- defvar DataRC = getVregSrcForVT<data_vt>.ret.RegClass;
+ defvar DataRC = getVregClassForVT<data_vt>.ret;
defvar SubLo = !if(!eq(vt, i32), sub0, sub0_sub1);
defvar SubHi = !if(!eq(vt, i32), sub1, sub2_sub3);
@@ -2088,7 +2089,7 @@ defm : MUBUFStore_PatternOffset <"BUFFER_STORE_SHORT", i16, store_global>;
multiclass MUBUFScratchStorePat_Common <string Instr,
ValueType vt, PatFrag st,
- RegisterClass rc = VGPR_32> {
+ RegisterClassLike rc = VGPR_32> {
def : GCNPat <
(st vt:$value, (MUBUFScratchOffen v4i32:$srsrc, i32:$vaddr,
i32:$soffset, i32:$offset)),
@@ -2104,7 +2105,7 @@ multiclass MUBUFScratchStorePat_Common <string Instr,
multiclass MUBUFScratchStorePat <string Instr,
ValueType vt, PatFrag st,
- RegisterClass rc = VGPR_32> {
+ RegisterClassLike rc = VGPR_32> {
let SubtargetPredicate = HasUnrestrictedSOffset in {
defm : MUBUFScratchStorePat_Common<Instr, vt, st, rc>;
}
diff --git a/llvm/lib/Target/AMDGPU/DSInstructions.td b/llvm/lib/Target/AMDGPU/DSInstructions.td
index f2e432fa8d7f5..d9825460cf389 100644
--- a/llvm/lib/Target/AMDGPU/DSInstructions.td
+++ b/llvm/lib/Target/AMDGPU/DSInstructions.td
@@ -904,7 +904,7 @@ let SubtargetPredicate = isGFX1250Plus in {
let WaveSizePredicate = isWave32, mayStore = 0 in {
let OtherPredicates = [HasTransposeLoadF4F6Insts] in {
defm DS_LOAD_TR4_B64 : DS_1A_RET_NoM0<"ds_load_tr4_b64", VGPROp_64>;
-defm DS_LOAD_TR6_B96 : DS_1A_RET_NoM0<"ds_load_tr6_b96", VGPROp_96>;
+defm DS_LOAD_TR6_B96 : DS_1A_RET_NoM0<"ds_load_tr6_b96", VGPROp_96_Align1>;
} // End OtherPredicates = [HasTransposeLoadF4F6Insts]
defm DS_LOAD_TR8_B64 : DS_1A_RET_NoM0<"ds_load_tr8_b64", VGPROp_64>;
defm DS_LOAD_TR16_B128 : DS_1A_RET_NoM0<"ds_load_tr16_b128", VGPROp_128>;
@@ -934,7 +934,7 @@ let WaveSizePredicate = isWave64, SubtargetPredicate = HasGFX950Insts, mayStore
defm DS_READ_B64_TR_B4 : DS_1A_RET_NoM0<"ds_read_b64_tr_b4", AVLdSt_64>;
defm DS_READ_B64_TR_B8 : DS_1A_RET_NoM0<"ds_read_b64_tr_b8", AVLdSt_64>;
defm DS_READ_B64_TR_B16 : DS_1A_RET_NoM0<"ds_read_b64_tr_b16", AVLdSt_64>;
- defm DS_READ_B96_TR_B6 : DS_1A_RET_NoM0<"ds_read_b96_tr_b6", AVLdSt_96>;
+ defm DS_READ_B96_TR_B6 : DS_1A_RET_NoM0<"ds_read_b96_tr_b6", AVLdSt_96_Align1>;
}
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
index d3db1b7394675..8887299ba476c 100644
--- a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
+++ b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
@@ -56,7 +56,9 @@ static int64_t getInlineImmVal64(unsigned Imm);
AMDGPUDisassembler::AMDGPUDisassembler(const MCSubtargetInfo &STI,
MCContext &Ctx, MCInstrInfo const *MCII)
: MCDisassembler(STI, Ctx), MCII(MCII), MRI(*Ctx.getRegisterInfo()),
- MAI(*Ctx.getAsmInfo()), TargetMaxInstBytes(MAI.getMaxInstLength(&STI)),
+ MAI(*Ctx.getAsmInfo()),
+ HwModeRegClass(STI.getHwMode(MCSubtargetInfo::HwMode_RegClass)),
+ TargetMaxInstBytes(MAI.getMaxInstLength(&STI)),
CodeObjectVersion(AMDGPU::getDefaultAMDHSACodeObjectVersion()) {
// ToDo: AMDGPUDisassembler supports only VI ISA.
if (!STI.hasFeature(AMDGPU::FeatureGCN3Encoding) && !isGFX10Plus())
@@ -824,7 +826,8 @@ DecodeStatus AMDGPUDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
}
}
- if (MCII->get(MI.getOpcode()).TSFlags & SIInstrFlags::MIMG) {
+ const MCInstrDesc &Desc = MCII->get(MI.getOpcode());
+ if (Desc.TSFlags & SIInstrFlags::MIMG) {
int VAddr0Idx =
AMDGPU::getNamedOperandIdx(MI.getOpcode(), AMDGPU::OpName::vaddr0);
int RsrcIdx =
@@ -837,7 +840,7 @@ DecodeStatus AMDGPUDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
for (unsigned i = 0; i < NSAArgs; ++i) {
const unsigned VAddrIdx = VAddr0Idx + 1 + i;
auto VAddrRCID =
- MCII->get(MI.getOpcode()).operands()[VAddrIdx].RegClass;
+ MCII->getOpRegClassID(Desc.operands()[VAddrIdx], HwModeRegClass);
MI.insert(MI.begin() + VAddrIdx, createRegOperand(VAddrRCID, Bytes[i]));
}
Bytes = Bytes.slice(4 * NSAWords);
@@ -1310,7 +1313,8 @@ void AMDGPUDisassembler::convertMIMGInst(MCInst &MI) const {
// Widen the register to the correct number of enabled channels.
MCRegister NewVdata;
if (DstSize != Info->VDataDwords) {
- auto DataRCID = MCII->get(NewOpcode).operands()[VDataIdx].RegClass;
+ auto DataRCID = MCII->getOpRegClassID(
+ MCII->get(NewOpcode).operands()[VDataIdx], HwModeRegClass);
// Get first subregister of VData
MCRegister Vdata0 = MI.getOperand(VDataIdx).getReg();
@@ -1337,7 +1341,9 @@ void AMDGPUDisassembler::convertMIMGInst(MCInst &MI) const {
MCRegister VAddrSubSA = MRI.getSubReg(VAddrSA, AMDGPU::sub0);
VAddrSA = VAddrSubSA ? VAddrSubSA : VAddrSA;
- auto AddrRCID = MCII->get(NewOpcode).operands()[VAddrSAIdx].RegClass;
+ auto AddrRCID = MCII->getOpRegClassID(
+ MCII->get(NewOpcode).operands()[VAddrSAIdx], HwModeRegClass);
+
const MCRegisterClass &NewRC = MRI.getRegClass(AddrRCID);
NewVAddrSA = MRI.getMatchingSuperReg(VAddrSA, AMDGPU::sub0, &NewRC);
NewVAddrSA = CheckVGPROverflow(NewVAddrSA, NewRC, MRI);
diff --git a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h
index c1131c2936fc7..6139ce105437b 100644
--- a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h
+++ b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h
@@ -41,6 +41,7 @@ class AMDGPUDisassembler : public MCDisassembler {
std::unique_ptr<MCInstrInfo const> const MCII;
const MCRegisterInfo &MRI;
const MCAsmInfo &MAI;
+ const unsigned HwModeRegClass;
const unsigned TargetMaxInstBytes;
mutable ArrayRef<uint8_t> Bytes;
mutable uint32_t Literal;
diff --git a/llvm/lib/Target/AMDGPU/FLATInstructions.td b/llvm/lib/Target/AMDGPU/FLATInstructions.td
index a1306565bbe29..aaf173dcb8fae 100644
--- a/llvm/lib/Target/AMDGPU/FLATInstructions.td
+++ b/llvm/lib/Target/AMDGPU/FLATInstructions.td
@@ -235,7 +235,7 @@ class FLAT_Load_Pseudo<
let InOperandList = !con(
!if(EnableSaddr,
(ins SReg_64_XEXEC_XNULL:$saddr, VGPR_32:$vaddr),
- (ins VReg_64:$vaddr)),
+ (ins VReg_64_AlignTarget:$vaddr)),
(ins flat_offset:$offset),
// FIXME: Operands with default values do not work with following
// non-optional operands.
@@ -274,7 +274,7 @@ class FLAT_Store_Pseudo <string opName, RegisterOperand vdataClass,
!con(
!if(EnableSaddr,
(ins VGPR_32:$vaddr, vdataClass:$vdata, SReg_64_XEXEC_XNULL:$saddr),
- (ins VReg_64:$vaddr, vdataClass:$vdata)),
+ (ins VReg_64_AlignTarget:$vaddr, vdataClass:$vdata)),
(ins flat_offset:$offset, CPol_0:$cpol)),
" $vaddr, $vdata"#!if(HasSaddr, !if(EnableSaddr, ", $saddr", ", off"), "")#"$offset$cpol"> {
let mayLoad = 0;
@@ -388,7 +388,7 @@ class FLAT_Global_Load_LDS_Pseudo <string opName, bit EnableSaddr = 0, bit IsAsy
(outs ...
[truncated]
|
This was referenced Sep 12, 2025
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
arsenm
force-pushed
the
users/arsenm/amdgpu/use-regclassbyhwmode-av-align-registers
branch
from
0a70e35 to
77b6693
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/use-regclassbyhwmode-av-align-registers
branch
from
77b6693 to
45d3ca9
Compare
arsenm
force-pushed
the
users/arsenm/codegen/targetinstrinfo-add-regclass-by-hwmode
branch
from
ba508eb to
b0954df
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/use-regclassbyhwmode-av-align-registers
branch
from
45d3ca9 to
67fb474
Compare
arsenm
force-pushed
the
users/arsenm/codegen/targetinstrinfo-add-regclass-by-hwmode
branch
from
b0954df to
480926a
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/use-regclassbyhwmode-av-align-registers
branch
from
67fb474 to
c6ac1f0
Compare
arsenm
force-pushed
the
users/arsenm/codegen/targetinstrinfo-add-regclass-by-hwmode
branch
from
480926a to
bdfdc33
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/use-regclassbyhwmode-av-align-registers
branch
2 times, most recently
from
19fd65e to
a9b4b37
Compare
arsenm
force-pushed
the
users/arsenm/codegen/targetinstrinfo-add-regclass-by-hwmode
branch
from
da6fc27 to
8bfcdcc
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/use-regclassbyhwmode-av-align-registers
branch
from
a9b4b37 to
2166f0b
Compare
cdevadas
reviewed
Sep 19, 2025
| def AVAlign2LoadStoreMode : HwMode<[HasMAIInsts, NeedsAlignedVGPRs]>; | ||
|
|
||
| // gfx1250, has alignment requirement but no AGPRs. | ||
| def AlignedVGPRNoAGPRMode : HwMode<[NotHasMAIInsts, NeedsAlignedVGPRs]>; |
There was a problem hiding this comment.
What will be the Mode for gfx908 that has AGPRs but no strict VGPR align requirement?
There was a problem hiding this comment.
DefaultMode. That's mostly the reason for having separate AV and AV_LdSt cases
Base automatically changed from
users/arsenm/codegen/targetinstrinfo-add-regclass-by-hwmode
to
main
September 19, 2025 11:08
arsenm
force-pushed
the
users/arsenm/amdgpu/use-regclassbyhwmode-av-align-registers
branch
from
2166f0b to
5b4d86d
Compare
This was referenced Sep 19, 2025
|
ping |
1 similar comment
|
ping |
This removes special case processing in TargetInstrInfo::getRegClass to fixup register operands which depending on the subtarget support AGPRs, or require even aligned registers. This regresses assembler diagnostics, which currently work by hackily accepting invalid cases and then post-rejecting a validly parsed instruction. On the plus side this now emits a comment when disassembling unaligned registers for targets with the alignment requirement.
arsenm
force-pushed
the
users/arsenm/amdgpu/use-regclassbyhwmode-av-align-registers
branch
from
5b4d86d to
c1ac2e0
Compare
|
ping |
|
ping. I'm submitting this today before vacation if there's no comments |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This removes special case processing in TargetInstrInfo::getRegClass to
fixup register operands which depending on the subtarget support AGPRs,
or require even aligned registers.
This regresses assembler diagnostics, which currently work by hackily
accepting invalid cases and then post-rejecting a validly parsed instruction.
On the plus side this now emits a comment when disassembling unaligned
registers for targets with the alignment requirement.