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

A64: Implement half-precision vector variants of FMLA/FMLS

Browse source
This commit is contained in:
Lioncash 2019-04-13 01:48:22 -04:00 committed by MerryMage
parent ec6b3ae084
commit d6606deda2
2 changed files with 32 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

@ -583,11 +583,11 @@ INST(INS_elt, "INS (element)", "01101
//INST(FCMGT_reg_3, "FCMGT (register)", "0Q101110110mmmmm001001nnnnnddddd") //INST(FCMGT_reg_3, "FCMGT (register)", "0Q101110110mmmmm001001nnnnnddddd")
//INST(FACGT_3, "FACGT", "0Q101110110mmmmm001011nnnnnddddd") //INST(FACGT_3, "FACGT", "0Q101110110mmmmm001011nnnnnddddd")
//INST(FMAXNM_1, "FMAXNM (vector)", "0Q001110010mmmmm000001nnnnnddddd") //INST(FMAXNM_1, "FMAXNM (vector)", "0Q001110010mmmmm000001nnnnnddddd")
//INST(FMLA_vec_1, "FMLA (vector)", "0Q001110010mmmmm000011nnnnnddddd") INST(FMLA_vec_1, "FMLA (vector)", "0Q001110010mmmmm000011nnnnnddddd")
//INST(FADD_1, "FADD (vector)", "0Q001110010mmmmm000101nnnnnddddd") //INST(FADD_1, "FADD (vector)", "0Q001110010mmmmm000101nnnnnddddd")
//INST(FMAX_1, "FMAX (vector)", "0Q001110010mmmmm001101nnnnnddddd") //INST(FMAX_1, "FMAX (vector)", "0Q001110010mmmmm001101nnnnnddddd")
//INST(FMINNM_1, "FMINNM (vector)", "0Q001110110mmmmm000001nnnnnddddd") //INST(FMINNM_1, "FMINNM (vector)", "0Q001110110mmmmm000001nnnnnddddd")
//INST(FMLS_vec_1, "FMLS (vector)", "0Q001110110mmmmm000011nnnnnddddd") INST(FMLS_vec_1, "FMLS (vector)", "0Q001110110mmmmm000011nnnnnddddd")
//INST(FSUB_1, "FSUB (vector)", "0Q001110110mmmmm000101nnnnnddddd") //INST(FSUB_1, "FSUB (vector)", "0Q001110110mmmmm000101nnnnnddddd")
//INST(FMIN_1, "FMIN (vector)", "0Q001110110mmmmm001101nnnnnddddd") //INST(FMIN_1, "FMIN (vector)", "0Q001110110mmmmm001101nnnnnddddd")
//INST(FMAXNMP_vec_1, "FMAXNMP (vector)", "0Q101110010mmmmm000001nnnnnddddd") //INST(FMAXNMP_vec_1, "FMAXNMP (vector)", "0Q101110010mmmmm000001nnnnnddddd")

30
src/frontend/A64/translate/impl/simd_three_same.cpp
View file

@ -680,10 +680,24 @@ bool TranslatorVisitor::FADD_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
return true; return true;
} }
bool TranslatorVisitor::FMLA_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) {
const size_t datasize = Q ? 128 : 64;
const size_t esize = 16;
const IR::U128 operand1 = V(datasize, Vn);
const IR::U128 operand2 = V(datasize, Vm);
const IR::U128 operand3 = V(datasize, Vd);
const IR::U128 result = ir.FPVectorMulAdd(esize, operand3, operand1, operand2);
V(datasize, Vd, result);
return true;
}
bool TranslatorVisitor::FMLA_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { bool TranslatorVisitor::FMLA_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
if (sz && !Q) { if (sz && !Q) {
return ReservedValue(); return ReservedValue();
} }
const size_t esize = sz ? 64 : 32; const size_t esize = sz ? 64 : 32;
const size_t datasize = Q ? 128 : 64; const size_t datasize = Q ? 128 : 64;
@ -691,6 +705,20 @@ bool TranslatorVisitor::FMLA_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
const IR::U128 operand2 = V(datasize, Vm); const IR::U128 operand2 = V(datasize, Vm);
const IR::U128 operand3 = V(datasize, Vd); const IR::U128 operand3 = V(datasize, Vd);
const IR::U128 result = ir.FPVectorMulAdd(esize, operand3, operand1, operand2); const IR::U128 result = ir.FPVectorMulAdd(esize, operand3, operand1, operand2);
V(datasize, Vd, result);
return true;
}
bool TranslatorVisitor::FMLS_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) {
const size_t datasize = Q ? 128 : 64;
const size_t esize = 16;
const IR::U128 operand1 = V(datasize, Vn);
const IR::U128 operand2 = V(datasize, Vm);
const IR::U128 operand3 = V(datasize, Vd);
const IR::U128 result = ir.FPVectorMulAdd(esize, operand3, ir.FPVectorNeg(esize, operand1), operand2);
V(datasize, Vd, result); V(datasize, Vd, result);
return true; return true;
} }
@ -699,6 +727,7 @@ bool TranslatorVisitor::FMLS_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
if (sz && !Q) { if (sz && !Q) {
return ReservedValue(); return ReservedValue();
} }
const size_t esize = sz ? 64 : 32; const size_t esize = sz ? 64 : 32;
const size_t datasize = Q ? 128 : 64; const size_t datasize = Q ? 128 : 64;
@ -706,6 +735,7 @@ bool TranslatorVisitor::FMLS_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {
const IR::U128 operand2 = V(datasize, Vm); const IR::U128 operand2 = V(datasize, Vm);
const IR::U128 operand3 = V(datasize, Vd); const IR::U128 operand3 = V(datasize, Vd);
const IR::U128 result = ir.FPVectorMulAdd(esize, operand3, ir.FPVectorNeg(esize, operand1), operand2); const IR::U128 result = ir.FPVectorMulAdd(esize, operand3, ir.FPVectorNeg(esize, operand1), operand2);
V(datasize, Vd, result); V(datasize, Vd, result);
return true; return true;
} }