void Jit::Comp_VDot(u32 op)
{
// DISABLE;
CONDITIONAL_DISABLE;
// WARNING: No prefix support!
if (js.MayHavePrefix()) {
Comp_Generic(op);
js.EatPrefix();
return;
}
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
// TODO: Force read one of them into regs? probably not.
u8 sregs[4], tregs[4];
GetVectorRegs(sregs, sz, vs);
GetVectorRegs(tregs, sz, vt);
// TODO: applyprefixST here somehow (shuffle, etc...)
fpr.MapRegsV(sregs, sz, 0);
fpr.MapRegsV(tregs, sz, 0);
VMUL(S0, fpr.V(sregs[0]), fpr.V(tregs[0]));
int n = GetNumVectorElements(sz);
for (int i = 1; i < n; i++) {
// sum += s[i]*t[i];
VMLA(S0, fpr.V(sregs[i]), fpr.V(tregs[i]));
}
fpr.ReleaseSpillLocks();
fpr.MapRegV(vd, MAP_NOINIT | MAP_DIRTY);
// TODO: applyprefixD here somehow (write mask etc..)
VMOV(fpr.V(vd), S0);
fpr.ReleaseSpillLocks();
js.EatPrefix();
}
void Jit::Comp_VVectorInit(u32 op)
{
CONDITIONAL_DISABLE;
// WARNING: No prefix support!
if (js.MayHavePrefix()) {
Comp_Generic(op);
js.EatPrefix();
return;
}
switch ((op >> 16) & 0xF)
{
case 6: // v=zeros; break; //vzero
MOVI2F(S0, 0.0f, R0);
break;
case 7: // v=ones; break; //vone
MOVI2F(S0, 1.0f, R0);
break;
default:
DISABLE;
break;
}
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
u8 dregs[4];
GetVectorRegsPrefixD(dregs, sz, _VD);
fpr.MapRegsV(dregs, sz, MAP_NOINIT | MAP_DIRTY);
for (int i = 0; i < n; ++i)
VMOV(fpr.V(dregs[i]), S0);
ApplyPrefixD(dregs, sz);
fpr.ReleaseSpillLocks();
}
void Jit::Comp_VVectorInit(MIPSOpcode op)
{
CONDITIONAL_DISABLE;
// WARNING: No prefix support!
if (js.HasUnknownPrefix() || disablePrefixes) {
DISABLE;
}
switch ((op >> 16) & 0xF)
{
case 6: // v=zeros; break; //vzero
MOVI2F(S0, 0.0f, R0);
break;
case 7: // v=ones; break; //vone
MOVI2F(S0, 1.0f, R0);
break;
default:
DISABLE;
break;
}
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
u8 dregs[4];
GetVectorRegsPrefixD(dregs, sz, _VD);
fpr.MapRegsAndSpillLockV(dregs, sz, MAP_NOINIT | MAP_DIRTY);
for (int i = 0; i < n; ++i)
VMOV(fpr.V(dregs[i]), S0);
ApplyPrefixD(dregs, sz);
fpr.ReleaseSpillLocksAndDiscardTemps();
}
void Jit::Comp_VV2Op(u32 op) {
CONDITIONAL_DISABLE;
DISABLE;
if (js.HasUnknownPrefix())
DISABLE;
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
u8 sregs[4], dregs[4];
GetVectorRegsPrefixS(sregs, sz, _VS);
GetVectorRegsPrefixD(dregs, sz, _VD);
ARMReg tempxregs[4];
for (int i = 0; i < n; ++i)
{
if (!IsOverlapSafeAllowS(dregs[i], i, n, sregs))
{
int reg = fpr.GetTempV();
fpr.MapRegV(reg, MAP_NOINIT | MAP_DIRTY);
fpr.SpillLockV(reg);
tempxregs[i] = fpr.V(reg);
}
else
{
fpr.MapRegV(dregs[i], (dregs[i] == sregs[i] ? 0 : MAP_NOINIT) | MAP_DIRTY);
fpr.SpillLockV(dregs[i]);
tempxregs[i] = fpr.V(dregs[i]);
}
}
// Warning: sregs[i] and tempxregs[i] may be the same reg.
// Helps for vmov, hurts for vrcp, etc.
for (int i = 0; i < n; ++i)
{
switch ((op >> 16) & 0x1f)
{
case 0: // d[i] = s[i]; break; //vmov
// Probably for swizzle.
VMOV(tempxregs[i], fpr.V(sregs[i]));
break;
case 1: // d[i] = fabsf(s[i]); break; //vabs
//if (!fpr.V(sregs[i]).IsSimpleReg(tempxregs[i]))
VABS(tempxregs[i], fpr.V(sregs[i]));
break;
case 2: // d[i] = -s[i]; break; //vneg
VNEG(tempxregs[i], fpr.V(sregs[i]));
break;
case 4: // if (s[i] < 0) d[i] = 0; else {if(s[i] > 1.0f) d[i] = 1.0f; else d[i] = s[i];} break; // vsat0
DISABLE;
break;
case 5: // if (s[i] < -1.0f) d[i] = -1.0f; else {if(s[i] > 1.0f) d[i] = 1.0f; else d[i] = s[i];} break; // vsat1
DISABLE;
break;
case 16: // d[i] = 1.0f / s[i]; break; //vrcp
MOVI2F(S0, 1.0f, R0);
VDIV(tempxregs[i], S0, fpr.V(sregs[i]));
break;
case 17: // d[i] = 1.0f / sqrtf(s[i]); break; //vrsq
MOVI2F(S0, 1.0f, R0);
VSQRT(S1, fpr.V(sregs[i]));
VDIV(tempxregs[i], S0, S1);
break;
case 18: // d[i] = sinf((float)M_PI_2 * s[i]); break; //vsin
DISABLE;
break;
case 19: // d[i] = cosf((float)M_PI_2 * s[i]); break; //vcos
DISABLE;
break;
case 20: // d[i] = powf(2.0f, s[i]); break; //vexp2
DISABLE;
break;
case 21: // d[i] = logf(s[i])/log(2.0f); break; //vlog2
DISABLE;
break;
case 22: // d[i] = sqrtf(s[i]); break; //vsqrt
VSQRT(tempxregs[i], fpr.V(sregs[i]));
VABS(tempxregs[i], tempxregs[i]);
break;
case 23: // d[i] = asinf(s[i] * (float)M_2_PI); break; //vasin
DISABLE;
break;
case 24: // d[i] = -1.0f / s[i]; break; // vnrcp
MOVI2F(S0, -1.0f, R0);
VDIV(tempxregs[i], S0, fpr.V(sregs[i]));
break;
case 26: // d[i] = -sinf((float)M_PI_2 * s[i]); break; // vnsin
DISABLE;
break;
case 28: // d[i] = 1.0f / expf(s[i] * (float)M_LOG2E); break; // vrexp2
DISABLE;
break;
}
}
fpr.MapRegsV(dregs, sz, MAP_NOINIT | MAP_DIRTY);
for (int i = 0; i < n; ++i)
{
VMOV(fpr.V(dregs[i]), tempxregs[i]);
}
ApplyPrefixD(dregs, sz);
fpr.ReleaseSpillLocks();
}
void Jit::Comp_VecDo3(u32 op)
{
CONDITIONAL_DISABLE;
DISABLE;
// WARNING: No prefix support!
if (js.MayHavePrefix())
{
Comp_Generic(op);
js.EatPrefix();
return;
}
int vd = _VD;
int vs = _VS;
int vt = _VT;
void (ARMXEmitter::*triop)(ARMReg, ARMReg, ARMReg) = NULL;
switch (op >> 26)
{
case 24: //VFPU0
switch ((op >> 23)&7)
{
case 0: // d[i] = s[i] + t[i]; break; //vadd
triop = &ARMXEmitter::VADD;
break;
case 1: // d[i] = s[i] - t[i]; break; //vsub
triop = &ARMXEmitter::VSUB;
break;
case 7: // d[i] = s[i] / t[i]; break; //vdiv
triop = &ARMXEmitter::VDIV;
break;
}
break;
case 25: //VFPU1
switch ((op >> 23)&7)
{
case 0: // d[i] = s[i] * t[i]; break; //vmul
triop = &ARMXEmitter::VMUL;
break;
}
break;
}
if (!triop) {
DISABLE;
}
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
u8 sregs[4], tregs[4], dregs[4];
GetVectorRegsPrefixS(sregs, sz, _VS);
GetVectorRegsPrefixT(tregs, sz, _VT);
GetVectorRegsPrefixD(dregs, sz, _VD);
MIPSReg tempregs[4];
for (int i = 0; i < n; i++) {
if (!IsOverlapSafeAllowS(dregs[i], i, n, sregs, n, tregs)) {
tempregs[i] = fpr.GetTempV();
} else {
fpr.MapRegV(dregs[i], (dregs[i] == sregs[i] || dregs[i] == tregs[i] ? 0 : MAP_NOINIT) | MAP_DIRTY);
tempregs[i] = dregs[i];
}
}
for (int i = 0; i < n; i++) {
fpr.SpillLockV(sregs[i]);
fpr.SpillLockV(tregs[i]);
fpr.MapRegV(sregs[i]);
fpr.MapRegV(tregs[i]);
fpr.MapRegV(tempregs[i]);
(this->*triop)(fpr.V(tempregs[i]), fpr.V(sregs[i]), fpr.V(tregs[i]));
fpr.ReleaseSpillLockV(sregs[i]);
fpr.ReleaseSpillLockV(tregs[i]);
}
fpr.MapRegsV(dregs, sz, MAP_DIRTY);
for (int i = 0; i < n; i++) {
if (dregs[i] != tempregs[i])
VMOV(fpr.V(dregs[i]), fpr.V(tempregs[i]));
}
ApplyPrefixD(dregs, sz);
fpr.ReleaseSpillLocks();
js.EatPrefix();
}