/// Extract all of the characters from the number \p Num one by one and
/// insert them into the string builder \p SB.
static void DecodeFixedWidth(APInt &Num, std::string &SB) {
uint64_t CL = Huffman::CharsetLength;
// NL is the number of characters that we can hold in a 64bit number.
// Each letter takes Log2(CL) bits. Doing this computation in floating-
// point arithmetic could give a slightly better (more optimistic) result,
// but the computation may not be constant at compile time.
uint64_t NumLetters = 64 / Log2_64_Ceil(CL);
assert(Num.getBitWidth() > 8 &&
"Not enough bits for arithmetic on this alphabet");
// Try to decode eight numbers at once. It is much faster to work with
// local 64bit numbers than working with APInt. In this loop we try to
// extract NL characters at once and process them using a local 64-bit
// number.
// Calculate CharsetLength**NumLetters (CL to the power of NL), which is the
// highest numeric value that can hold NumLetters characters in a 64bit
// number. Notice: this loop is optimized away and CLX is computed to a
// constant integer at compile time.
uint64_t CLX = 1;
for (unsigned i = 0; i < NumLetters; i++) { CLX *= CL; }
while (Num.ugt(CLX)) {
unsigned BW = Num.getBitWidth();
APInt C = APInt(BW, CLX);
APInt Quotient(1, 0), Remainder(1, 0);
APInt::udivrem(Num, C, Quotient, Remainder);
// Try to reduce the bitwidth of the API after the division. This can
// accelerate the division operation in future iterations because the
// number becomes smaller (fewer bits) with each iteration. However,
// We can't reduce the number to something too small because we still
// need to be able to perform the "mod charset_length" operation.
Num = Quotient.zextOrTrunc(std::max(Quotient.getActiveBits(), 64u));
uint64_t Tail = Remainder.getZExtValue();
for (unsigned i = 0; i < NumLetters; i++) {
SB += Huffman::Charset[Tail % CL];
Tail = Tail / CL;
}
}
// Pop characters out of the APInt one by one.
while (Num.getBoolValue()) {
unsigned BW = Num.getBitWidth();
APInt C = APInt(BW, CL);
APInt Quotient(1, 0), Remainder(1, 0);
APInt::udivrem(Num, C, Quotient, Remainder);
Num = Quotient;
SB += Huffman::Charset[Remainder.getZExtValue()];
}
}
bool X86MCInstrAnalysis::clearsSuperRegisters(const MCRegisterInfo &MRI,
const MCInst &Inst,
APInt &Mask) const {
const MCInstrDesc &Desc = Info->get(Inst.getOpcode());
unsigned NumDefs = Desc.getNumDefs();
unsigned NumImplicitDefs = Desc.getNumImplicitDefs();
assert(Mask.getBitWidth() == NumDefs + NumImplicitDefs &&
"Unexpected number of bits in the mask!");
bool HasVEX = (Desc.TSFlags & X86II::EncodingMask) == X86II::VEX;
bool HasEVEX = (Desc.TSFlags & X86II::EncodingMask) == X86II::EVEX;
bool HasXOP = (Desc.TSFlags & X86II::EncodingMask) == X86II::XOP;
const MCRegisterClass &GR32RC = MRI.getRegClass(X86::GR32RegClassID);
const MCRegisterClass &VR128XRC = MRI.getRegClass(X86::VR128XRegClassID);
const MCRegisterClass &VR256XRC = MRI.getRegClass(X86::VR256XRegClassID);
auto ClearsSuperReg = [=](unsigned RegID) {
// On X86-64, a general purpose integer register is viewed as a 64-bit
// register internal to the processor.
// An update to the lower 32 bits of a 64 bit integer register is
// architecturally defined to zero extend the upper 32 bits.
if (GR32RC.contains(RegID))
return true;
// Early exit if this instruction has no vex/evex/xop prefix.
if (!HasEVEX && !HasVEX && !HasXOP)
return false;
// All VEX and EVEX encoded instructions are defined to zero the high bits
// of the destination register up to VLMAX (i.e. the maximum vector register
// width pertaining to the instruction).
// We assume the same behavior for XOP instructions too.
return VR128XRC.contains(RegID) || VR256XRC.contains(RegID);
};
Mask.clearAllBits();
for (unsigned I = 0, E = NumDefs; I < E; ++I) {
const MCOperand &Op = Inst.getOperand(I);
if (ClearsSuperReg(Op.getReg()))
Mask.setBit(I);
}
for (unsigned I = 0, E = NumImplicitDefs; I < E; ++I) {
const MCPhysReg Reg = Desc.getImplicitDefs()[I];
if (ClearsSuperReg(Reg))
Mask.setBit(NumDefs + I);
}
return Mask.getBoolValue();
}
/// Extract all of the characters from the number \p Num one by one and
/// insert them into the string builder \p SB.
static void DecodeFixedWidth(APInt &Num, std::string &SB) {
uint64_t CL = Huffman::CharsetLength;
assert(Num.getBitWidth() > 8 &&
"Not enough bits for arithmetic on this alphabet");
// Try to decode eight numbers at once. It is much faster to work with
// local 64bit numbers than working with APInt. In this loop we try to
// extract 8 characters at one and process them using a local 64bit number.
// In this code we assume a worse case scenario where our alphabet is a full
// 8-bit ascii. It is possible to improve this code by packing one or two
// more characters into the 64bit local variable.
uint64_t CL8 = CL * CL * CL * CL * CL * CL * CL * CL;
while (Num.ugt(CL8)) {
unsigned BW = Num.getBitWidth();
APInt C = APInt(BW, CL8);
APInt Quotient(1, 0), Remainder(1, 0);
APInt::udivrem(Num, C, Quotient, Remainder);
// Try to reduce the bitwidth of the API after the division. This can
// accelerate the division operation in future iterations because the
// number becomes smaller (fewer bits) with each iteration. However,
// We can't reduce the number to something too small because we still
// need to be able to perform the "mod charset_length" operation.
Num = Quotient.zextOrTrunc(std::max(Quotient.getActiveBits(), 64u));
uint64_t Tail = Remainder.getZExtValue();
for (int i=0; i < 8; i++) {
SB += Huffman::Charset[Tail % CL];
Tail = Tail / CL;
}
}
// Pop characters out of the APInt one by one.
while (Num.getBoolValue()) {
unsigned BW = Num.getBitWidth();
APInt C = APInt(BW, CL);
APInt Quotient(1, 0), Remainder(1, 0);
APInt::udivrem(Num, C, Quotient, Remainder);
Num = Quotient;
SB += Huffman::Charset[Remainder.getZExtValue()];
}
}
