A64: Implement FMAXP, FMAXNMP, FMINP, and FMINNMP's vector variants
We can just implement these in terms of scalars for the time being.
This commit is contained in:
parent
ec76f95f5a
commit
8b65ea68c0
2 changed files with 52 additions and 4 deletions
|
@ -764,22 +764,22 @@ INST(PMUL, "PMUL", "0Q101
|
|||
INST(UMAXP, "UMAXP", "0Q101110zz1mmmmm101001nnnnnddddd")
|
||||
INST(UMINP, "UMINP", "0Q101110zz1mmmmm101011nnnnnddddd")
|
||||
//INST(SQRDMULH_vec_2, "SQRDMULH (vector)", "0Q101110zz1mmmmm101101nnnnnddddd")
|
||||
//INST(FMAXNMP_vec_2, "FMAXNMP (vector)", "0Q1011100z1mmmmm110001nnnnnddddd")
|
||||
INST(FMAXNMP_vec_2, "FMAXNMP (vector)", "0Q1011100z1mmmmm110001nnnnnddddd")
|
||||
//INST(FMLAL_vec_2, "FMLAL, FMLAL2 (vector)", "0Q1011100z1mmmmm110011nnnnnddddd")
|
||||
INST(FADDP_vec_2, "FADDP (vector)", "0Q1011100z1mmmmm110101nnnnnddddd")
|
||||
INST(FMUL_vec_2, "FMUL (vector)", "0Q1011100z1mmmmm110111nnnnnddddd")
|
||||
INST(FCMGE_reg_4, "FCMGE (register)", "0Q1011100z1mmmmm111001nnnnnddddd")
|
||||
INST(FACGE_4, "FACGE", "0Q1011100z1mmmmm111011nnnnnddddd")
|
||||
//INST(FMAXP_vec_2, "FMAXP (vector)", "0Q1011100z1mmmmm111101nnnnnddddd")
|
||||
INST(FMAXP_vec_2, "FMAXP (vector)", "0Q1011100z1mmmmm111101nnnnnddddd")
|
||||
INST(FDIV_2, "FDIV (vector)", "0Q1011100z1mmmmm111111nnnnnddddd")
|
||||
INST(EOR_asimd, "EOR (vector)", "0Q101110001mmmmm000111nnnnnddddd")
|
||||
INST(BSL, "BSL", "0Q101110011mmmmm000111nnnnnddddd")
|
||||
//INST(FMINNMP_vec_2, "FMINNMP (vector)", "0Q1011101z1mmmmm110001nnnnnddddd")
|
||||
INST(FMINNMP_vec_2, "FMINNMP (vector)", "0Q1011101z1mmmmm110001nnnnnddddd")
|
||||
//INST(FMLSL_vec_2, "FMLSL, FMLSL2 (vector)", "0Q1011101z1mmmmm110011nnnnnddddd")
|
||||
INST(FABD_4, "FABD", "0Q1011101z1mmmmm110101nnnnnddddd")
|
||||
INST(FCMGT_reg_4, "FCMGT (register)", "0Q1011101z1mmmmm111001nnnnnddddd")
|
||||
INST(FACGT_4, "FACGT", "0Q1011101z1mmmmm111011nnnnnddddd")
|
||||
//INST(FMINP_vec_2, "FMINP (vector)", "0Q1011101z1mmmmm111101nnnnnddddd")
|
||||
INST(FMINP_vec_2, "FMINP (vector)", "0Q1011101z1mmmmm111101nnnnnddddd")
|
||||
INST(BIT, "BIT", "0Q101110101mmmmm000111nnnnnddddd")
|
||||
INST(BIF, "BIF", "0Q101110111mmmmm000111nnnnnddddd")
|
||||
|
||||
|
|
|
@ -247,6 +247,38 @@ bool PairedMinMaxOperation(TranslatorVisitor& v, bool Q, Imm<2> size, Vec Vm, Ve
|
|||
return true;
|
||||
}
|
||||
|
||||
bool FPPairedMinMax(TranslatorVisitor& v, bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd,
|
||||
IR::U32U64 (IREmitter::* fn)(const IR::U32U64&, const IR::U32U64&, bool)) {
|
||||
if (sz && !Q) {
|
||||
return v.ReservedValue();
|
||||
}
|
||||
|
||||
const size_t esize = sz ? 64 : 32;
|
||||
const size_t datasize = Q ? 128 : 64;
|
||||
const size_t elements = datasize / esize;
|
||||
const size_t boundary = elements / 2;
|
||||
|
||||
const IR::U128 operand1 = v.V(datasize, Vn);
|
||||
const IR::U128 operand2 = v.V(datasize, Vm);
|
||||
IR::U128 result = v.ir.ZeroVector();
|
||||
|
||||
const auto operation = [&](IR::U128 operand, size_t result_start_index) {
|
||||
for (size_t i = 0; i < elements; i += 2, result_start_index++) {
|
||||
const IR::UAny elem1 = v.ir.VectorGetElement(esize, operand, i);
|
||||
const IR::UAny elem2 = v.ir.VectorGetElement(esize, operand, i + 1);
|
||||
const IR::UAny result_elem = (v.ir.*fn)(elem1, elem2, true);
|
||||
|
||||
result = v.ir.VectorSetElement(esize, result, result_start_index, result_elem);
|
||||
}
|
||||
};
|
||||
|
||||
operation(operand1, 0);
|
||||
operation(operand2, boundary);
|
||||
|
||||
v.V(datasize, Vd, result);
|
||||
return true;
|
||||
}
|
||||
|
||||
bool SaturatingArithmeticOperation(TranslatorVisitor& v, bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd,
|
||||
Operation op, Signedness sign) {
|
||||
if (size == 0b11 && !Q) {
|
||||
|
@ -898,10 +930,26 @@ bool TranslatorVisitor::FMAX_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
|
|||
return FPMinMaxOperation(*this, Q, sz, Vm, Vn, Vd, MinMaxOperation::Max);
|
||||
}
|
||||
|
||||
bool TranslatorVisitor::FMAXNMP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
|
||||
return FPPairedMinMax(*this, Q, sz, Vm, Vn, Vd, &IREmitter::FPMaxNumeric);
|
||||
}
|
||||
|
||||
bool TranslatorVisitor::FMAXP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
|
||||
return FPPairedMinMax(*this, Q, sz, Vm, Vn, Vd, &IREmitter::FPMax);
|
||||
}
|
||||
|
||||
bool TranslatorVisitor::FMIN_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
|
||||
return FPMinMaxOperation(*this, Q, sz, Vm, Vn, Vd, MinMaxOperation::Min);
|
||||
}
|
||||
|
||||
bool TranslatorVisitor::FMINNMP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
|
||||
return FPPairedMinMax(*this, Q, sz, Vm, Vn, Vd, &IREmitter::FPMinNumeric);
|
||||
}
|
||||
|
||||
bool TranslatorVisitor::FMINP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
|
||||
return FPPairedMinMax(*this, Q, sz, Vm, Vn, Vd, &IREmitter::FPMin);
|
||||
}
|
||||
|
||||
bool TranslatorVisitor::FADDP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
|
||||
if (sz && !Q) {
|
||||
return ReservedValue();
|
||||
|
|
Loading…
Reference in a new issue