Emulator-Archive/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2024-11-23 13:30:02 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Merge pull request #18249 from unknownbrackets/arm64jit-vcrsp

Browse Source 
arm64jit: Avoid fused multiplies in vcrsp.t
This commit is contained in:
Henrik Rydgård 2023-09-27 08:49:01 +02:00 committed by GitHub
commit d6a8bfdf3e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 15 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
Core/MIPS/ARM64/Arm64CompVFPU.cpp
View File

@ -1504,7 +1504,7 @@ namespace MIPSComp {
void Arm64Jit::Comp_VCrossQuat(MIPSOpcode op) {
// This op does not support prefixes anyway.
CONDITIONAL_DISABLE(VFPU_VEC);
if (js.HasUnknownPrefix())
if (!js.HasNoPrefix())
DISABLE;
VectorSize sz = GetVecSize(op);
@ -1521,20 +1521,26 @@ namespace MIPSComp {
if (sz == V_Triple) {
MIPSReg temp3 = fpr.GetTempV();
MIPSReg temp4 = fpr.GetTempV();
fpr.MapRegV(temp3, MAP_DIRTY | MAP_NOINIT);
fpr.MapRegV(temp4, MAP_DIRTY | MAP_NOINIT);
// Cross product vcrsp.t
// Compute X
fp.FMUL(S0, fpr.V(sregs[1]), fpr.V(tregs[2]));
fp.FMSUB(S0, fpr.V(sregs[2]), fpr.V(tregs[1]), S0);
// Note: using FMSUB here causes accuracy issues, see #18203.
// Compute X: s[1] * t[2] - s[2] * t[1]
fp.FMUL(fpr.V(temp3), fpr.V(sregs[1]), fpr.V(tregs[2]));
fp.FMUL(fpr.V(temp4), fpr.V(sregs[2]), fpr.V(tregs[1]));
fp.FSUB(S0, fpr.V(temp3), fpr.V(temp4));
// Compute Y
fp.FMUL(S1, fpr.V(sregs[2]), fpr.V(tregs[0]));
fp.FMSUB(S1, fpr.V(sregs[0]), fpr.V(tregs[2]), S1);
// Compute Y: s[2] * t[0] - s[0] * t[2]
fp.FMUL(fpr.V(temp3), fpr.V(sregs[2]), fpr.V(tregs[0]));
fp.FMUL(fpr.V(temp4), fpr.V(sregs[0]), fpr.V(tregs[2]));
fp.FSUB(S1, fpr.V(temp3), fpr.V(temp4));
// Compute Z
// Compute Z: s[0] * t[1] - s[1] * t[0]
fp.FMUL(fpr.V(temp3), fpr.V(sregs[0]), fpr.V(tregs[1]));
fp.FMSUB(fpr.V(temp3), fpr.V(sregs[1]), fpr.V(tregs[0]), fpr.V(temp3));
fp.FMUL(fpr.V(temp4), fpr.V(sregs[1]), fpr.V(tregs[0]));
fp.FSUB(fpr.V(temp3), fpr.V(temp3), fpr.V(temp4));
fpr.MapRegsAndSpillLockV(dregs, sz, MAP_NOINIT);
fp.FMOV(fpr.V(dregs[0]), S0);