preserve-emulation/ppsspp
1
0 Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2026-10-28 12:48:38 +01:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #7258 from unknownbrackets/jit-minor

Browse Source 
x86jit: Fix vavg, improve vfad
This commit is contained in:
Henrik Rydgård committed 2015-01-03 19:44:22 +01:00
2 parents d545908b6c e6a50a1371
commit 5fffe53671
2 files changed
+18 -11
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

Core/HLE/sceKernelThread.cpp
+1 -1
View File
@@ -1982,7 +1982,7 @@ void ThreadContext::reset()
r[0] = 0;
for (int i = 0; i<128; i++)
{
v[i] = 0.0f;
vi[i] = 0x7f800001;
}
for (int i = 0; i<15; i++)
{
Core/MIPS/x86/CompVFPU.cpp
+17 -10
View File
@@ -3125,8 +3125,12 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
MOVAPS(fpr.VSX(dregs), R(XMM0));
break;
case V_Quad:
XORPS(XMM1, R(XMM1));
MOVAPS(XMM0, fpr.VS(sregs));
DPPS(XMM0, M(&oneOneOneOne), 0xF1);
// In every other case, +0.0 is selected by the mask and added.
// But, here we need to manually add it to the result.
ADDPS(XMM0, R(XMM1));
MOVAPS(fpr.VSX(dregs), R(XMM0));
break;
default:
@@ -3135,15 +3139,17 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
} else {
switch (sz) {
case V_Pair:
XORPS(XMM1, R(XMM1));
MOVAPS(XMM0, fpr.VS(sregs));
MOVAPS(XMM1, R(XMM0));
ADDPS(XMM1, R(XMM0));
SHUFPS(XMM1, R(XMM1), _MM_SHUFFLE(3, 2, 1, 1));
ADDPS(XMM0, R(XMM1));
MOVAPS(fpr.VSX(dregs), R(XMM0));
break;
case V_Triple:
XORPS(XMM1, R(XMM1));
MOVAPS(XMM0, fpr.VS(sregs));
MOVAPS(XMM1, R(XMM0));
ADDPS(XMM1, R(XMM0));
SHUFPS(XMM1, R(XMM1), _MM_SHUFFLE(3, 2, 1, 1));
ADDPS(XMM0, R(XMM1));
SHUFPS(XMM1, R(XMM1), _MM_SHUFFLE(3, 2, 1, 2));
@@ -3151,7 +3157,10 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
MOVAPS(fpr.VSX(dregs), R(XMM0));
break;
case V_Quad:
XORPS(XMM1, R(XMM1));
MOVAPS(XMM0, fpr.VS(sregs));
// This flips the sign of any -0.000.
ADDPS(XMM0, R(XMM1));
MOVHLPS(XMM1, XMM0);
ADDPS(XMM0, R(XMM1));
MOVAPS(XMM1, R(XMM0));
@@ -3164,7 +3173,7 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
}
}
if (((op >> 16) & 31) == 7) { // vavg
MULSS(fpr.VSX(dregs), M(&vavg_table[n]));
MULSS(fpr.VSX(dregs), M(&vavg_table[n - 1]));
}
ApplyPrefixD(dregs, V_Single);
fpr.ReleaseSpillLocks();
@@ -3176,17 +3185,15 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
fpr.SimpleRegsV(dregs, V_Single, MAP_NOINIT | MAP_DIRTY);
X64Reg reg = XMM0;
if (IsOverlapSafeAllowS(dregs[0], 0, n, sregs)) {
if (IsOverlapSafe(dregs[0], 0, n, sregs)) {
fpr.MapRegV(dregs[0], dregs[0] == sregs[0] ? MAP_DIRTY : MAP_NOINIT);
fpr.SpillLockV(dregs[0]);
reg = fpr.VX(dregs[0]);
}
// We use a temp reg in case of overlap.
if (!fpr.V(sregs[0]).IsSimpleReg(reg)) {
MOVSS(reg, fpr.V(sregs[0]));
}
for (int i = 1; i < n; ++i) {
// We have to start zt +0.000 in case any values are -0.000.
XORPS(reg, R(reg));
for (int i = 0; i < n; ++i) {
ADDSS(reg, fpr.V(sregs[i]));
}
@@ -3194,7 +3201,7 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
case 6: // vfad
break;
case 7: // vavg
MULSS(reg, M(&vavg_table[n]));
MULSS(reg, M(&vavg_table[n - 1]));
break;
}