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

A64: Implement FRSQRTS' half-precision scalar variant

Browse source 
With the necessary machinery in place, we can now handle the
half-precision variant.
This commit is contained in:
Lioncash 2019-04-14 21:04:38 -04:00 committed by MerryMage
parent 96356fac93
commit f9b2862217
2 changed files with 12 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/frontend/A64/decoder/a64.inc
View file

@ -386,7 +386,7 @@ INST(FMULX_vec_2, "FMULX", "01011
INST(FCMEQ_reg_2, "FCMEQ (register)", "010111100z1mmmmm111001nnnnnddddd")
INST(FRECPS_1, "FRECPS", "01011110010mmmmm001111nnnnnddddd")
INST(FRECPS_2, "FRECPS", "010111100z1mmmmm111111nnnnnddddd")
//INST(FRSQRTS_1, "FRSQRTS", "01011110110mmmmm001111nnnnnddddd")
INST(FRSQRTS_1, "FRSQRTS", "01011110110mmmmm001111nnnnnddddd")
INST(FRSQRTS_2, "FRSQRTS", "010111101z1mmmmm111111nnnnnddddd")
//INST(FCMGE_reg_1, "FCMGE (register)", "01111110010mmmmm001001nnnnnddddd")
INST(FCMGE_reg_2, "FCMGE (register)", "011111100z1mmmmm111001nnnnnddddd")

11
src/frontend/A64/translate/impl/simd_scalar_three_same.cpp
View file

@ -316,6 +316,17 @@ bool TranslatorVisitor::FRECPS_2(bool sz, Vec Vm, Vec Vn, Vec Vd) {
return true;
}
bool TranslatorVisitor::FRSQRTS_1(Vec Vm, Vec Vn, Vec Vd) {
const size_t esize = 16;
const IR::U16 operand1 = V_scalar(esize, Vn);
const IR::U16 operand2 = V_scalar(esize, Vm);
const IR::U16 result = ir.FPRSqrtStepFused(operand1, operand2);
V_scalar(esize, Vd, result);
return true;
}
bool TranslatorVisitor::FRSQRTS_2(bool sz, Vec Vm, Vec Vn, Vec Vd) {
const size_t esize = sz ? 64 : 32;