libretro/ppsspp
1
0
Fork 0
mirror of https://github.com/libretro/ppsspp.git synced 2024-12-14 20:48:30 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

SIMD-ify a couple instructions a bit

Browse Source
This commit is contained in:
Henrik Rydgård 2014-11-25 00:52:39 +01:00 committed by Henrik Rydgard
parent ced555ee47
commit 4db6b7f3e2
1 changed files with 22 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
Core/MIPS/x86/CompVFPU.cpp
View File

@ -447,8 +447,20 @@ void Jit::Comp_VVectorInit(MIPSOpcode op) {
if (js.HasUnknownPrefix()) if (js.HasUnknownPrefix())
DISABLE; DISABLE;
switch ((op >> 16) & 0xF) VectorSize sz = GetVecSize(op);
{ int type = (op >> 16) & 0xF;
u8 dregs[4];
GetVectorRegsPrefixD(dregs, sz, _VD);
// vzero only for now
if (type == 6 && fpr.TryMapRegsVS(dregs, sz, MAP_NOINIT | MAP_DIRTY)) {
XORPS(fpr.VSX(dregs[0]), fpr.VS(dregs[0]));
ApplyPrefixD(dregs, sz);
fpr.ReleaseSpillLocks();
return;
}
switch (type) {
case 6: // v=zeros; break; //vzero case 6: // v=zeros; break; //vzero
XORPS(XMM0, R(XMM0)); XORPS(XMM0, R(XMM0));
break; break;
@ -460,10 +472,7 @@ void Jit::Comp_VVectorInit(MIPSOpcode op) {
break; break;
} }
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz); int n = GetNumVectorElements(sz);
u8 dregs[4];
GetVectorRegsPrefixD(dregs, sz, _VD);
fpr.MapRegsV(dregs, sz, MAP_NOINIT | MAP_DIRTY); fpr.MapRegsV(dregs, sz, MAP_NOINIT | MAP_DIRTY);
for (int i = 0; i < n; ++i) for (int i = 0; i < n; ++i)
MOVSS(fpr.VX(dregs[i]), R(XMM0)); MOVSS(fpr.VX(dregs[i]), R(XMM0));
@ -472,8 +481,6 @@ void Jit::Comp_VVectorInit(MIPSOpcode op) {
fpr.ReleaseSpillLocks(); fpr.ReleaseSpillLocks();
} }
void Jit::Comp_VIdt(MIPSOpcode op) { void Jit::Comp_VIdt(MIPSOpcode op) {
CONDITIONAL_DISABLE; CONDITIONAL_DISABLE;
if (js.HasUnknownPrefix()) if (js.HasUnknownPrefix())
@ -1647,6 +1654,14 @@ void Jit::Comp_Vcst(MIPSOpcode op) {
GetVectorRegsPrefixD(dregs, sz, _VD); GetVectorRegsPrefixD(dregs, sz, _VD);
MOVSS(XMM0, M(&cst_constants[conNum])); MOVSS(XMM0, M(&cst_constants[conNum]));
if (fpr.TryMapRegsVS(dregs, sz, MAP_NOINIT | MAP_DIRTY)) {
SHUFPS(XMM0, R(XMM0), _MM_SHUFFLE(0,0,0,0));
MOVAPS(fpr.VS(dregs[0]), XMM0);
fpr.ReleaseSpillLocks();
return;
}
fpr.MapRegsV(dregs, sz, MAP_NOINIT | MAP_DIRTY); fpr.MapRegsV(dregs, sz, MAP_NOINIT | MAP_DIRTY);
for (int i = 0; i < n; i++) { for (int i = 0; i < n; i++) {
MOVSS(fpr.V(dregs[i]), XMM0); MOVSS(fpr.V(dregs[i]), XMM0);
@ -2210,7 +2225,6 @@ void Jit::Comp_VScl(MIPSOpcode op) {
u8 sregs[4], dregs[4], scale; u8 sregs[4], dregs[4], scale;
GetVectorRegsPrefixS(sregs, sz, _VS); GetVectorRegsPrefixS(sregs, sz, _VS);
// TODO: Prefixes seem strange...
GetVectorRegsPrefixT(&scale, V_Single, _VT); GetVectorRegsPrefixT(&scale, V_Single, _VT);
GetVectorRegsPrefixD(dregs, sz, _VD); GetVectorRegsPrefixD(dregs, sz, _VD);