mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

A64: Implement FMAXNM and FMINNM vector variants.

Browse source 
Currently we can implement these in terms of the scalar IR variants.
This commit is contained in:
Lioncash 2018-08-16 11:00:59 -04:00 committed by MerryMage
parent 8b65ea68c0
commit 7eb6be7a6a
2 changed files with 36 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

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

@ -724,7 +724,7 @@ INST(SMAXP, "SMAXP", "0Q001
INST(SMINP, "SMINP", "0Q001110zz1mmmmm101011nnnnnddddd") INST(SMINP, "SMINP", "0Q001110zz1mmmmm101011nnnnnddddd")
//INST(SQDMULH_vec_2, "SQDMULH (vector)", "0Q001110zz1mmmmm101101nnnnnddddd") //INST(SQDMULH_vec_2, "SQDMULH (vector)", "0Q001110zz1mmmmm101101nnnnnddddd")
INST(ADDP_vec, "ADDP (vector)", "0Q001110zz1mmmmm101111nnnnnddddd") INST(ADDP_vec, "ADDP (vector)", "0Q001110zz1mmmmm101111nnnnnddddd")
//INST(FMAXNM_2, "FMAXNM (vector)", "0Q0011100z1mmmmm110001nnnnnddddd") INST(FMAXNM_2, "FMAXNM (vector)", "0Q0011100z1mmmmm110001nnnnnddddd")
INST(FMLA_vec_2, "FMLA (vector)", "0Q0011100z1mmmmm110011nnnnnddddd") INST(FMLA_vec_2, "FMLA (vector)", "0Q0011100z1mmmmm110011nnnnnddddd")
INST(FADD_2, "FADD (vector)", "0Q0011100z1mmmmm110101nnnnnddddd") INST(FADD_2, "FADD (vector)", "0Q0011100z1mmmmm110101nnnnnddddd")
INST(FMAX_2, "FMAX (vector)", "0Q0011100z1mmmmm111101nnnnnddddd") INST(FMAX_2, "FMAX (vector)", "0Q0011100z1mmmmm111101nnnnnddddd")
@ -734,7 +734,7 @@ INST(FCMEQ_reg_4, "FCMEQ (register)", "0Q001
INST(FRECPS_4, "FRECPS", "0Q0011100z1mmmmm111111nnnnnddddd") INST(FRECPS_4, "FRECPS", "0Q0011100z1mmmmm111111nnnnnddddd")
INST(AND_asimd, "AND (vector)", "0Q001110001mmmmm000111nnnnnddddd") INST(AND_asimd, "AND (vector)", "0Q001110001mmmmm000111nnnnnddddd")
INST(BIC_asimd_reg, "BIC (vector, register)", "0Q001110011mmmmm000111nnnnnddddd") INST(BIC_asimd_reg, "BIC (vector, register)", "0Q001110011mmmmm000111nnnnnddddd")
//INST(FMINNM_2, "FMINNM (vector)", "0Q0011101z1mmmmm110001nnnnnddddd") INST(FMINNM_2, "FMINNM (vector)", "0Q0011101z1mmmmm110001nnnnnddddd")
INST(FMLS_vec_2, "FMLS (vector)", "0Q0011101z1mmmmm110011nnnnnddddd") INST(FMLS_vec_2, "FMLS (vector)", "0Q0011101z1mmmmm110011nnnnnddddd")
INST(FSUB_2, "FSUB (vector)", "0Q0011101z1mmmmm110101nnnnnddddd") INST(FSUB_2, "FSUB (vector)", "0Q0011101z1mmmmm110101nnnnnddddd")
//INST(FMLSL_vec_1, "FMLSL, FMLSL2 (vector)", "0Q0011101z1mmmmm111011nnnnnddddd") //INST(FMLSL_vec_1, "FMLSL, FMLSL2 (vector)", "0Q0011101z1mmmmm111011nnnnnddddd")

34
src/frontend/A64/translate/impl/simd_three_same.cpp
View file

@ -208,6 +208,32 @@ bool FPMinMaxOperation(TranslatorVisitor& v, bool Q, bool sz, Vec Vm, Vec Vn, Ve
return true; return true;
} }
bool FPMinMaxNumericOperation(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 IR::U128 operand1 = v.V(datasize, Vn);
const IR::U128 operand2 = v.V(datasize, Vm);
IR::U128 result = v.ir.ZeroVector();
for (size_t i = 0; i < elements; i++) {
const IR::UAny elem1 = v.ir.VectorGetElement(esize, operand1, i);
const IR::UAny elem2 = v.ir.VectorGetElement(esize, operand2, i);
const IR::UAny result_elem = (v.ir.*fn)(elem1, elem2, true);
result = v.ir.VectorSetElement(esize, result, i, result_elem);
}
v.V(datasize, Vd, result);
return true;
}
bool PairedMinMaxOperation(TranslatorVisitor& v, bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd, bool PairedMinMaxOperation(TranslatorVisitor& v, bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd,
MinMaxOperation operation, Signedness sign) { MinMaxOperation operation, Signedness sign) {
if (size == 0b11) { if (size == 0b11) {
@ -930,6 +956,10 @@ bool TranslatorVisitor::FMAX_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
return FPMinMaxOperation(*this, Q, sz, Vm, Vn, Vd, MinMaxOperation::Max); return FPMinMaxOperation(*this, Q, sz, Vm, Vn, Vd, MinMaxOperation::Max);
} }
bool TranslatorVisitor::FMAXNM_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
return FPMinMaxNumericOperation(*this, Q, sz, Vm, Vn, Vd, &IREmitter::FPMaxNumeric);
}
bool TranslatorVisitor::FMAXNMP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { 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); return FPPairedMinMax(*this, Q, sz, Vm, Vn, Vd, &IREmitter::FPMaxNumeric);
} }
@ -942,6 +972,10 @@ bool TranslatorVisitor::FMIN_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
return FPMinMaxOperation(*this, Q, sz, Vm, Vn, Vd, MinMaxOperation::Min); return FPMinMaxOperation(*this, Q, sz, Vm, Vn, Vd, MinMaxOperation::Min);
} }
bool TranslatorVisitor::FMINNM_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
return FPMinMaxNumericOperation(*this, Q, sz, Vm, Vn, Vd, &IREmitter::FPMinNumeric);
}
bool TranslatorVisitor::FMINNMP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { 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); return FPPairedMinMax(*this, Q, sz, Vm, Vn, Vd, &IREmitter::FPMinNumeric);
} }