[llvm] [LLVM][TableGen] Paramaterize NumToSkip in DecoderEmitter. (PR #135882)
Rahul Joshi via llvm-commits
llvm-commits at lists.llvm.org
Tue Apr 15 16:44:36 PDT 2025
https://github.com/jurahul created https://github.com/llvm/llvm-project/pull/135882
- Add command line option `num-to-skip-size` to parameterize the size of `NumToSkip` bytes in the decoder table. Default value will be 2, and targets that need larger size can use 3.
- Keep all existing targets, except AArch64, to use size 2, and change AArch64 to use size 3 since it run into the "disassembler decoding table too large" error with size 2.
- Following is a rough reduction in size for the decoder tables by switching to size 2.
```
Target Old Size New Size % Reduction
================================================
AArch64 153254 153254 0.00
AMDGPU 471566 412805 12.46
ARC 5724 5061 11.58
ARM 84936 73831 13.07
AVR 1497 1306 12.76
BPF 2172 1927 11.28
CSKY 10064 8692 13.63
Hexagon 47967 41965 12.51
Lanai 1108 982 11.37
LoongArch 24446 21621 11.56
MSP430 4200 3716 11.52
Mips 36330 31415 13.53
PPC 31897 28098 11.91
RISCV 37979 32790 13.66
Sparc 8331 7252 12.95
SystemZ 36722 32248 12.18
VE 48296 42873 11.23
XCore 2590 2316 10.58
Xtensa 3827 3316 13.35
```
>From 17e9bb3bbe74e9e88937cf591708ee13e2e56b40 Mon Sep 17 00:00:00 2001
From: Rahul Joshi <rjoshi at nvidia.com>
Date: Tue, 15 Apr 2025 15:45:24 -0700
Subject: [PATCH] [LLVM][TableGen] Paramaterize NumToSkip in DecoderEmitter.
- Add command line option `num-to-skip-size` to parameterize the
size of `NumToSkip` bytes in the decoder table. Default value
will be 2, and targets that need larger size can use 3.
- Keep all existing targets, except AArch64, to use size 2, and
change AArch64 to use size 3 since it run into the
"disassembler decoding table too large" error with size 2.
- Following is a rough reduction in size for the decoder tables
by switching to size 2.
Target Old Size New Size % Reduction
================================================
AArch64 153254 153254 0.00
AMDGPU 471566 412805 12.46
ARC 5724 5061 11.58
ARM 84936 73831 13.07
AVR 1497 1306 12.76
BPF 2172 1927 11.28
CSKY 10064 8692 13.63
Hexagon 47967 41965 12.51
Lanai 1108 982 11.37
LoongArch 24446 21621 11.56
MSP430 4200 3716 11.52
Mips 36330 31415 13.53
PPC 31897 28098 11.91
RISCV 37979 32790 13.66
Sparc 8331 7252 12.95
SystemZ 36722 32248 12.18
VE 48296 42873 11.23
XCore 2590 2316 10.58
Xtensa 3827 3316 13.35
---
llvm/lib/Target/AArch64/CMakeLists.txt | 2 +-
llvm/utils/TableGen/DecoderEmitter.cpp | 105 ++++++++++++++-----------
2 files changed, 62 insertions(+), 45 deletions(-)
diff --git a/llvm/lib/Target/AArch64/CMakeLists.txt b/llvm/lib/Target/AArch64/CMakeLists.txt
index 2300e479bc110..ba1d1605ec104 100644
--- a/llvm/lib/Target/AArch64/CMakeLists.txt
+++ b/llvm/lib/Target/AArch64/CMakeLists.txt
@@ -7,7 +7,7 @@ tablegen(LLVM AArch64GenAsmWriter.inc -gen-asm-writer)
tablegen(LLVM AArch64GenAsmWriter1.inc -gen-asm-writer -asmwriternum=1)
tablegen(LLVM AArch64GenCallingConv.inc -gen-callingconv)
tablegen(LLVM AArch64GenDAGISel.inc -gen-dag-isel)
-tablegen(LLVM AArch64GenDisassemblerTables.inc -gen-disassembler)
+tablegen(LLVM AArch64GenDisassemblerTables.inc -gen-disassembler --num-to-skip-size=3)
tablegen(LLVM AArch64GenFastISel.inc -gen-fast-isel)
tablegen(LLVM AArch64GenGlobalISel.inc -gen-global-isel)
tablegen(LLVM AArch64GenO0PreLegalizeGICombiner.inc -gen-global-isel-combiner
diff --git a/llvm/utils/TableGen/DecoderEmitter.cpp b/llvm/utils/TableGen/DecoderEmitter.cpp
index 9c6015cc24576..3e3f6ed7c565b 100644
--- a/llvm/utils/TableGen/DecoderEmitter.cpp
+++ b/llvm/utils/TableGen/DecoderEmitter.cpp
@@ -32,8 +32,10 @@
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/FormatVariadic.h"
#include "llvm/Support/FormattedStream.h"
#include "llvm/Support/LEB128.h"
+#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/TableGen/Error.h"
#include "llvm/TableGen/Record.h"
@@ -76,6 +78,12 @@ static cl::opt<SuppressLevel> DecoderEmitterSuppressDuplicates(
"significantly reducing Table Duplications")),
cl::init(SUPPRESSION_DISABLE), cl::cat(DisassemblerEmitterCat));
+static cl::opt<uint32_t> NumToSkipSizeInBytes(
+ "num-to-skip-size",
+ cl::desc(
+ "number of bytes to use for num-to-skip entries in the decoder table"),
+ cl::init(2), cl::cat(DisassemblerEmitterCat));
+
STATISTIC(NumEncodings, "Number of encodings considered");
STATISTIC(NumEncodingsLackingDisasm,
"Number of encodings without disassembler info");
@@ -130,9 +138,20 @@ struct DecoderTable : public std::vector<uint8_t> {
// in the table for patching.
size_t insertNumToSkip() {
size_t Size = size();
- insert(end(), 3, 0);
+ insert(end(), NumToSkipSizeInBytes, 0);
return Size;
}
+
+ void patchNumToSkip(size_t FixupIdx, uint32_t Value) {
+ if (!isUIntN(8 * NumToSkipSizeInBytes, Value))
+ PrintFatalError(
+ "disassembler decoding table too large, try --num-to-skip-size=3");
+
+ (*this)[FixupIdx] = static_cast<uint8_t>(Value);
+ (*this)[FixupIdx + 1] = static_cast<uint8_t>(Value >> 8);
+ if (NumToSkipSizeInBytes == 3)
+ (*this)[FixupIdx + 2] = static_cast<uint8_t>(Value >> 16);
+ }
};
struct DecoderTableInfo {
DecoderTable Table;
@@ -692,16 +711,11 @@ static void resolveTableFixups(DecoderTable &Table, const FixupList &Fixups,
// current location.
for (uint32_t FixupIdx : reverse(Fixups)) {
// Calculate the distance from the byte following the fixup entry byte
- // to the destination. The Target is calculated from after the 24-bit
- // NumToSkip entry itself, so subtract three from the displacement here
- // to account for that.
- uint32_t Delta = DestIdx - FixupIdx - 3;
- // Our NumToSkip entries are 24-bits. Make sure our table isn't too
- // big.
- assert(isUInt<24>(Delta));
- Table[FixupIdx] = (uint8_t)Delta;
- Table[FixupIdx + 1] = (uint8_t)(Delta >> 8);
- Table[FixupIdx + 2] = (uint8_t)(Delta >> 16);
+ // to the destination. The Target is calculated from after the
+ // `NumToSkipSizeInBytes`-byte NumToSkip entry itself, so subtract
+ // `NumToSkipSizeInBytes` from the displacement here to account for that.
+ uint32_t Delta = DestIdx - FixupIdx - NumToSkipSizeInBytes;
+ Table.patchNumToSkip(FixupIdx, Delta);
}
}
@@ -760,13 +774,11 @@ void Filter::emitTableEntry(DecoderTableInfo &TableInfo) const {
// Now that we've emitted the body of the handler, update the NumToSkip
// of the filter itself to be able to skip forward when false. Subtract
- // three as to account for the width of the NumToSkip field itself.
+ // `NumToSkipSizeInBytes` as to account for the width of the NumToSkip
+ // field itself.
if (PrevFilter) {
- uint32_t NumToSkip = Table.size() - PrevFilter - 3;
- assert(isUInt<24>(NumToSkip) && "disassembler decoding table too large!");
- Table[PrevFilter] = (uint8_t)NumToSkip;
- Table[PrevFilter + 1] = (uint8_t)(NumToSkip >> 8);
- Table[PrevFilter + 2] = (uint8_t)(NumToSkip >> 16);
+ uint32_t NumToSkip = Table.size() - PrevFilter - NumToSkipSizeInBytes;
+ Table.patchNumToSkip(PrevFilter, NumToSkip);
}
}
@@ -814,7 +826,8 @@ void DecoderEmitter::emitTable(formatted_raw_ostream &OS, DecoderTable &Table,
OS << (unsigned)*I++ << ", ";
};
- // Emit 24-bit numtoskip value to OS, returning the NumToSkip value.
+ // Emit `NumToSkipSizeInBytes`-byte numtoskip value to OS, returning the
+ // NumToSkip value.
auto emitNumToSkip = [](DecoderTable::const_iterator &I,
formatted_raw_ostream &OS) {
uint8_t Byte = *I++;
@@ -823,9 +836,11 @@ void DecoderEmitter::emitTable(formatted_raw_ostream &OS, DecoderTable &Table,
Byte = *I++;
OS << (unsigned)Byte << ", ";
NumToSkip |= Byte << 8;
- Byte = *I++;
- OS << (unsigned)(Byte) << ", ";
- NumToSkip |= Byte << 16;
+ if (NumToSkipSizeInBytes == 3) {
+ Byte = *I++;
+ OS << (unsigned)(Byte) << ", ";
+ NumToSkip |= Byte << 16;
+ }
return NumToSkip;
};
@@ -867,7 +882,7 @@ void DecoderEmitter::emitTable(formatted_raw_ostream &OS, DecoderTable &Table,
// The filter value is ULEB128 encoded.
emitULEB128(I, OS);
- // 24-bit numtoskip value.
+ // numtoskip value.
uint32_t NumToSkip = emitNumToSkip(I, OS);
OS << "// Skip to: " << ((I - Table.begin()) + NumToSkip) << "\n";
break;
@@ -883,7 +898,7 @@ void DecoderEmitter::emitTable(formatted_raw_ostream &OS, DecoderTable &Table,
// ULEB128 encoded field value.
emitULEB128(I, OS);
- // 24-bit numtoskip value.
+ // numtoskip value.
uint32_t NumToSkip = emitNumToSkip(I, OS);
OS << "// Skip to: " << ((I - Table.begin()) + NumToSkip) << "\n";
break;
@@ -893,7 +908,7 @@ void DecoderEmitter::emitTable(formatted_raw_ostream &OS, DecoderTable &Table,
OS << Indent << "MCD::OPC_CheckPredicate, ";
emitULEB128(I, OS);
- // 24-bit numtoskip value.
+ // numtoskip value.
uint32_t NumToSkip = emitNumToSkip(I, OS);
OS << "// Skip to: " << ((I - Table.begin()) + NumToSkip) << "\n";
break;
@@ -925,7 +940,7 @@ void DecoderEmitter::emitTable(formatted_raw_ostream &OS, DecoderTable &Table,
// Fallthrough for OPC_TryDecode.
- // 24-bit numtoskip value.
+ // numtoskip value.
uint32_t NumToSkip = emitNumToSkip(I, OS);
OS << "// Opcode: " << NumberedEncodings[EncodingID]
@@ -1411,9 +1426,9 @@ void FilterChooser::emitSingletonTableEntry(DecoderTableInfo &TableInfo,
TableInfo.Table.push_back(NumBits);
TableInfo.Table.insertULEB128(Ilnd.FieldVal);
- // The fixup is always 24-bits, so go ahead and allocate the space
- // in the table so all our relative position calculations work OK even
- // before we fully resolve the real value here.
+ // Allocate space in the table for fixup (NumToSkipSizeInBytes) so all
+ // our relative position calculations work OK even before we fully
+ // resolve the real value here.
// Push location for NumToSkip backpatching.
TableInfo.FixupStack.back().push_back(TableInfo.Table.insertNumToSkip());
@@ -2157,7 +2172,18 @@ insertBits(InsnType &field, uint64_t bits, unsigned startBit, unsigned numBits)
// decodeInstruction().
static void emitDecodeInstruction(formatted_raw_ostream &OS,
bool IsVarLenInst) {
+ OS << formatv("\nconstexpr unsigned NumToSkipSizeInBytes = {};\n",
+ NumToSkipSizeInBytes);
+
OS << R"(
+inline unsigned decodeNumToSkip(const uint8_t *&Ptr) {
+ unsigned NumToSkip = *Ptr++;
+ NumToSkip |= (*Ptr++) << 8;
+ if constexpr (NumToSkipSizeInBytes == 3)
+ NumToSkip |= (*Ptr++) << 16;
+ return NumToSkip;
+}
+
template <typename InsnType>
static DecodeStatus decodeInstruction(const uint8_t DecodeTable[], MCInst &MI,
InsnType insn, uint64_t Address,
@@ -2195,10 +2221,7 @@ static DecodeStatus decodeInstruction(const uint8_t DecodeTable[], MCInst &MI,
// Decode the field value.
uint64_t Val = decodeULEB128AndIncUnsafe(++Ptr);
bool Failed = Val != CurFieldValue;
- // NumToSkip is a plain 24-bit integer.
- unsigned NumToSkip = *Ptr++;
- NumToSkip |= (*Ptr++) << 8;
- NumToSkip |= (*Ptr++) << 16;
+ unsigned NumToSkip = decodeNumToSkip(Ptr);
// Perform the filter operation.
if (Failed)
@@ -2222,10 +2245,7 @@ static DecodeStatus decodeInstruction(const uint8_t DecodeTable[], MCInst &MI,
uint64_t ExpectedValue = decodeULEB128(++Ptr, &PtrLen);
Ptr += PtrLen;
bool Failed = ExpectedValue != FieldValue;
- // NumToSkip is a plain 24-bit integer.
- unsigned NumToSkip = *Ptr++;
- NumToSkip |= (*Ptr++) << 8;
- NumToSkip |= (*Ptr++) << 16;
+ unsigned NumToSkip = decodeNumToSkip(Ptr);
// If the actual and expected values don't match, skip.
if (Failed)
@@ -2240,10 +2260,7 @@ static DecodeStatus decodeInstruction(const uint8_t DecodeTable[], MCInst &MI,
case MCD::OPC_CheckPredicate: {
// Decode the Predicate Index value.
unsigned PIdx = decodeULEB128AndIncUnsafe(++Ptr);
- // NumToSkip is a plain 24-bit integer.
- unsigned NumToSkip = *Ptr++;
- NumToSkip |= (*Ptr++) << 8;
- NumToSkip |= (*Ptr++) << 16;
+ unsigned NumToSkip = decodeNumToSkip(Ptr);
// Check the predicate.
bool Failed = !checkDecoderPredicate(PIdx, Bits);
if (Failed)
@@ -2278,10 +2295,7 @@ static DecodeStatus decodeInstruction(const uint8_t DecodeTable[], MCInst &MI,
// Decode the Opcode value.
unsigned Opc = decodeULEB128AndIncUnsafe(++Ptr);
unsigned DecodeIdx = decodeULEB128AndIncUnsafe(Ptr);
- // NumToSkip is a plain 24-bit integer.
- unsigned NumToSkip = *Ptr++;
- NumToSkip |= (*Ptr++) << 8;
- NumToSkip |= (*Ptr++) << 16;
+ unsigned NumToSkip = decodeNumToSkip(Ptr);
// Perform the decode operation.
MCInst TmpMI;
@@ -2406,6 +2420,9 @@ handleHwModesUnrelatedEncodings(const CodeGenInstruction *Instr,
// Emits disassembler code for instruction decoding.
void DecoderEmitter::run(raw_ostream &o) {
+ if (NumToSkipSizeInBytes != 2 && NumToSkipSizeInBytes != 3)
+ PrintFatalError("Invalid value for num-to-skip-size, must be 2 or 3");
+
formatted_raw_ostream OS(o);
OS << R"(
#include "llvm/MC/MCInst.h"
More information about the llvm-commits
mailing list