suyu/dynarmic
28
1
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity

A64: Implement SQSHRUN, SQRSHRUN (vector)

Browse source
This commit is contained in:
MerryMage 2018-07-24 17:20:49 +01:00
parent c5722ec963
commit 712c6c1d7e
2 changed files with 45 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
src/frontend/A64/decoder/a64.inc
View file

@ -811,8 +811,8 @@ INST(SRI_2, "SRI", "0Q101
INST(SLI_2, "SLI", "0Q1011110IIIIiii010101nnnnnddddd")
//INST(SQSHLU_2, "SQSHLU", "0Q1011110IIIIiii011001nnnnnddddd")
//INST(UQSHL_imm_2, "UQSHL (immediate)", "0Q1011110IIIIiii011101nnnnnddddd")
//INST(SQSHRUN_2, "SQSHRUN, SQSHRUN2", "0Q1011110IIIIiii100001nnnnnddddd")
//INST(SQRSHRUN_2, "SQRSHRUN, SQRSHRUN2", "0Q1011110IIIIiii100011nnnnnddddd")
INST(SQSHRUN_2, "SQSHRUN, SQSHRUN2", "0Q1011110IIIIiii100001nnnnnddddd")
INST(SQRSHRUN_2, "SQRSHRUN, SQRSHRUN2", "0Q1011110IIIIiii100011nnnnnddddd")
//INST(UQSHRN_2, "UQSHRN, UQSHRN2", "0Q1011110IIIIiii100101nnnnnddddd")
//INST(UQRSHRN_2, "UQRSHRN, UQRSHRN2", "0Q1011110IIIIiii100111nnnnnddddd")
INST(USHLL, "USHLL, USHLL2", "0Q1011110IIIIiii101001nnnnnddddd")

56
src/frontend/A64/translate/impl/simd_shift_by_immediate.cpp
View file

@ -24,6 +24,17 @@ enum class Signedness {
Unsigned
};
enum class Narrowing {
Truncation,
SaturateToUnsigned,
};
IR::U128 PerformRoundingCorrection(TranslatorVisitor& v, size_t esize, u64 round_value, IR::U128 original, IR::U128 shifted) {
const IR::U128 round_const = v.ir.VectorBroadcast(esize, v.I(esize, round_value));
const IR::U128 round_correction = v.ir.VectorEqual(esize, v.ir.VectorAnd(original, round_const), round_const);
return v.ir.VectorSub(esize, shifted, round_correction);
}
bool ShiftRight(TranslatorVisitor& v, bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd,
Rounding rounding, Accumulating accumulating, Signedness signedness) {
if (immh == 0b0000) {
@ -38,7 +49,6 @@ bool ShiftRight(TranslatorVisitor& v, bool Q, Imm<4> immh, Imm<3> immb, Vec Vn,
const size_t datasize = Q ? 128 : 64;
const u8 shift_amount = static_cast<u8>(2 * esize) - concatenate(immh, immb).ZeroExtend<u8>();
const u64 round_value = 1ULL << (shift_amount - 1);
const IR::U128 operand = v.V(datasize, Vn);
@ -50,9 +60,8 @@ bool ShiftRight(TranslatorVisitor& v, bool Q, Imm<4> immh, Imm<3> immb, Vec Vn,
}();
if (rounding == Rounding::Round) {
const IR::U128 round_const = v.ir.VectorBroadcast(esize, v.I(esize, round_value));
const IR::U128 round_correction = v.ir.VectorEqual(esize, v.ir.VectorAnd(operand, round_const), round_const);
result = v.ir.VectorSub(esize, result, round_correction);
const u64 round_value = 1ULL << (shift_amount - 1);
result = PerformRoundingCorrection(v, esize, round_value, operand, result);
}
if (accumulating == Accumulating::Accumulate) {
@ -65,7 +74,7 @@ bool ShiftRight(TranslatorVisitor& v, bool Q, Imm<4> immh, Imm<3> immb, Vec Vn,
}
bool ShiftRightNarrowing(TranslatorVisitor& v, bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd,
Rounding rounding) {
Rounding rounding, Narrowing narrowing, Signedness signedness) {
if (immh == 0b0000) {
return v.DecodeError();
}
@ -79,19 +88,32 @@ bool ShiftRightNarrowing(TranslatorVisitor& v, bool Q, Imm<4> immh, Imm<3> immb,
const size_t part = Q ? 1 : 0;
const u8 shift_amount = static_cast<u8>(source_esize - concatenate(immh, immb).ZeroExtend());
const u64 round_value = 1ULL << (shift_amount - 1);
const IR::U128 operand = v.V(128, Vn);
IR::U128 wide_result = v.ir.VectorLogicalShiftRight(source_esize, operand, shift_amount);
IR::U128 wide_result = [&] {
if (signedness == Signedness::Signed) {
return v.ir.VectorArithmeticShiftRight(source_esize, operand, shift_amount);
}
return v.ir.VectorLogicalShiftRight(source_esize, operand, shift_amount);
}();
if (rounding == Rounding::Round) {
const IR::U128 round_const = v.ir.VectorBroadcast(esize, v.I(esize, round_value));
const IR::U128 round_correction = v.ir.VectorEqual(esize, v.ir.VectorAnd(operand, round_const), round_const);
wide_result = v.ir.VectorSub(esize, wide_result, round_correction);
const u64 round_value = 1ULL << (shift_amount - 1);
wide_result = PerformRoundingCorrection(v, source_esize, round_value, operand, wide_result);
}
const IR::U128 result = v.ir.VectorNarrow(source_esize, wide_result);
const IR::U128 result = [&] {
switch (narrowing) {
case Narrowing::Truncation:
return v.ir.VectorNarrow(source_esize, wide_result);
case Narrowing::SaturateToUnsigned:
ASSERT(signedness == Signedness::Signed);
return v.ir.VectorSignedSaturatedNarrowToUnsigned(source_esize, wide_result);
}
UNREACHABLE();
return IR::U128{};
}();
v.Vpart(64, Vd, part, result);
return true;
@ -163,11 +185,19 @@ bool TranslatorVisitor::SHL_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd)
}
bool TranslatorVisitor::SHRN(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) {
return ShiftRightNarrowing(*this, Q, immh, immb, Vn, Vd, Rounding::None);
return ShiftRightNarrowing(*this, Q, immh, immb, Vn, Vd, Rounding::None, Narrowing::Truncation, Signedness::Unsigned);
}
bool TranslatorVisitor::RSHRN(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) {
return ShiftRightNarrowing(*this, Q, immh, immb, Vn, Vd, Rounding::Round);
return ShiftRightNarrowing(*this, Q, immh, immb, Vn, Vd, Rounding::Round, Narrowing::Truncation, Signedness::Unsigned);
}
bool TranslatorVisitor::SQSHRUN_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) {
return ShiftRightNarrowing(*this, Q, immh, immb, Vn, Vd, Rounding::None, Narrowing::SaturateToUnsigned, Signedness::Signed);
}
bool TranslatorVisitor::SQRSHRUN_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) {
return ShiftRightNarrowing(*this, Q, immh, immb, Vn, Vd, Rounding::Round, Narrowing::SaturateToUnsigned, Signedness::Signed);
}
bool TranslatorVisitor::SSHLL(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) {