Macintosh-Tools
/
dingusppc
mirror of https://github.com/dingusdev/dingusppc.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

ppcfpopcodes: Fix multiply add opcodes. 

Use std::fma for all of them for max accuracy.
For single precision opcodes, convert only the result, not the operands.
This commit is contained in:
joevt 2024-02-05 06:22:17 -08:00 committed by dingusdev
parent 3be22dac99
commit 9199b1e520
1 changed files with 7 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
cpu/ppc/ppcfpopcodes.cpp
View File

@ -319,8 +319,7 @@ void dppc_interpreter::ppc_fmsub() {
ppc_confirm_inf_nan<SUB>(reg_a, reg_b, rc_flag);
}
double ppc_dblresult64_d = (val_reg_a * val_reg_c);
ppc_dblresult64_d -= val_reg_b;
double ppc_dblresult64_d = std::fma(val_reg_a, val_reg_c, -val_reg_b);
ppc_store_dfpresult_flt(reg_d);
fpresult_update(ppc_dblresult64_d);
@ -338,8 +337,7 @@ void dppc_interpreter::ppc_fnmadd() {
ppc_confirm_inf_nan<ADD>(reg_a, reg_b, rc_flag);
}
double ppc_dblresult64_d = std::fma(val_reg_a, val_reg_c, val_reg_b);
ppc_dblresult64_d = -(ppc_dblresult64_d);
double ppc_dblresult64_d = -std::fma(val_reg_a, val_reg_c, val_reg_b);
ppc_store_dfpresult_flt(reg_d);
fpresult_update(ppc_dblresult64_d);
@ -357,9 +355,7 @@ void dppc_interpreter::ppc_fnmsub() {
ppc_confirm_inf_nan<SUB>(reg_a, reg_b, rc_flag);
}
double ppc_dblresult64_d = (val_reg_a * val_reg_c);
ppc_dblresult64_d -= val_reg_b;
ppc_dblresult64_d = -(ppc_dblresult64_d);
double ppc_dblresult64_d = std::fma(-val_reg_a, val_reg_c, val_reg_b);
ppc_store_dfpresult_flt(reg_d);
fpresult_update(ppc_dblresult64_d);
@ -439,8 +435,7 @@ void dppc_interpreter::ppc_fmadds() {
ppc_confirm_inf_nan<ADD>(reg_a, reg_b, rc_flag);
}
double ppc_dblresult64_d = static_cast<double>(
std::fma((float)val_reg_a, (float)val_reg_c, (float)val_reg_b));
double ppc_dblresult64_d = (float)std::fma(val_reg_a, val_reg_c, val_reg_b);
ppc_store_sfpresult_flt(reg_d);
fpresult_update(ppc_dblresult64_d);
@ -458,9 +453,7 @@ void dppc_interpreter::ppc_fmsubs() {
ppc_confirm_inf_nan<ADD>(reg_a, reg_b, rc_flag);
}
float intermediate = float(val_reg_a * val_reg_c);
intermediate -= (float)val_reg_b;
double ppc_dblresult64_d = static_cast<double>(intermediate);
double ppc_dblresult64_d = (float)std::fma(val_reg_a, val_reg_c, -val_reg_b);
ppc_store_sfpresult_flt(reg_d);
fpresult_update(ppc_dblresult64_d);
@ -478,10 +471,7 @@ void dppc_interpreter::ppc_fnmadds() {
ppc_confirm_inf_nan<ADD>(reg_a, reg_b, rc_flag);
}
float intermediate = (float)val_reg_a * (float)val_reg_c;
intermediate += (float)val_reg_b;
intermediate = -intermediate;
double ppc_dblresult64_d = static_cast<double>(intermediate);
double ppc_dblresult64_d = -(float)std::fma(val_reg_a, val_reg_c, val_reg_b);
ppc_store_sfpresult_flt(reg_d);
fpresult_update(ppc_dblresult64_d);
@ -500,10 +490,7 @@ void dppc_interpreter::ppc_fnmsubs() {
ppc_confirm_inf_nan<SUB>(reg_a, reg_b, rc_flag);
}
float intermediate = (float)val_reg_a * (float)val_reg_c;
intermediate -= (float)val_reg_b;
intermediate = -intermediate;
double ppc_dblresult64_d = static_cast<double>(intermediate);
double ppc_dblresult64_d = (float)std::fma(-val_reg_a, val_reg_c, val_reg_b);
ppc_store_sfpresult_flt(reg_d);
fpresult_update(ppc_dblresult64_d);