Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Get rid of sincostemp global. Solution not tested on linux yet.

Browse source
This commit is contained in:
Henrik Rydgård 2017-07-05 16:44:22 +02:00
parent cfa7c61651
commit 077fafba64
2 changed files with 39 additions and 24 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
Core/MIPS/MIPS.h
View file

@ -237,6 +237,8 @@ public:
u32 MEMORY_ALIGNED16(vcmpResult[4]); u32 MEMORY_ALIGNED16(vcmpResult[4]);
float sincostemp[2];
static const u32 FCR0_VALUE = 0x00003351; static const u32 FCR0_VALUE = 0x00003351;
#if defined(PPSSPP_ARCH_X86) || defined(PPSSPP_ARCH_AMD64) #if defined(PPSSPP_ARCH_X86) || defined(PPSSPP_ARCH_AMD64)

61
Core/MIPS/x86/CompVFPU.cpp
View file

@ -2120,8 +2120,6 @@ void Jit::Comp_Vbfy(MIPSOpcode op) {
fpr.ReleaseSpillLocks(); fpr.ReleaseSpillLocks();
} }
static float sincostemp[2];
union u32float { union u32float {
u32 u; u32 u;
float f; float f;
@ -2142,29 +2140,29 @@ typedef float SinCosArg;
typedef u32float SinCosArg; typedef u32float SinCosArg;
#endif #endif
void SinCos(SinCosArg angle) { void SinCos(SinCosArg angle, float *output) {
vfpu_sincos(angle, sincostemp[0], sincostemp[1]); vfpu_sincos(angle, output[0], output[1]);
} }
void SinOnly(SinCosArg angle) { void SinOnly(SinCosArg angle, float *output) {
sincostemp[0] = vfpu_sin(angle); output[0] = vfpu_sin(angle);
} }
void NegSinOnly(SinCosArg angle) { void NegSinOnly(SinCosArg angle, float *output) {
sincostemp[0] = -vfpu_sin(angle); output[0] = -vfpu_sin(angle);
} }
void CosOnly(SinCosArg angle) { void CosOnly(SinCosArg angle, float *output) {
sincostemp[1] = vfpu_cos(angle); output[1] = vfpu_cos(angle);
} }
void ASinScaled(SinCosArg angle) { void ASinScaled(SinCosArg angle, float *output) {
sincostemp[0] = vfpu_asin(angle); output[0] = vfpu_asin(angle);
} }
void SinCosNegSin(SinCosArg angle) { void SinCosNegSin(SinCosArg angle, float *output) {
vfpu_sincos(angle, sincostemp[0], sincostemp[1]); vfpu_sincos(angle, output[0], output[1]);
sincostemp[0] = -sincostemp[0]; output[0] = -output[0];
} }
void Jit::Comp_VV2Op(MIPSOpcode op) { void Jit::Comp_VV2Op(MIPSOpcode op) {
@ -2173,9 +2171,15 @@ void Jit::Comp_VV2Op(MIPSOpcode op) {
if (js.HasUnknownPrefix()) if (js.HasUnknownPrefix())
DISABLE; DISABLE;
auto trigCallHelper = [this](void (*sinCosFunc)(SinCosArg), u8 sreg) { auto trigCallHelper = [this](void (*sinCosFunc)(SinCosArg, float *output), u8 sreg) {
#ifdef _M_X64 #ifdef _M_X64
MOVSS(XMM0, fpr.V(sreg)); MOVSS(XMM0, fpr.V(sreg));
// TODO: This reg might be different on Linux...
#ifdef _WIN32
LEA(64, RDX, MIPSSTATE_VAR(sincostemp));
#else
LEA(64, RSI, MIPSSTATE_VAR(sincostemp));
#endif
ABI_CallFunction(thunks.ProtectFunction((const void *)sinCosFunc, 0)); ABI_CallFunction(thunks.ProtectFunction((const void *)sinCosFunc, 0));
#else #else
// Sigh, passing floats with cdecl isn't pretty, ends up on the stack. // Sigh, passing floats with cdecl isn't pretty, ends up on the stack.
@ -2184,7 +2188,7 @@ void Jit::Comp_VV2Op(MIPSOpcode op) {
} else { } else {
MOV(32, R(EAX), fpr.V(sreg)); MOV(32, R(EAX), fpr.V(sreg));
} }
CallProtectedFunction((const void *)sinCosFunc, R(EAX)); CallProtectedFunction((const void *)sinCosFunc, R(EAX), Imm32((uint32_t)(uintptr_t)mips_->sincostemp));
#endif #endif
}; };
@ -2350,12 +2354,14 @@ void Jit::Comp_VV2Op(MIPSOpcode op) {
DIVSS(tempxregs[i], R(XMM0)); DIVSS(tempxregs[i], R(XMM0));
break; break;
case 18: // d[i] = sinf((float)M_PI_2 * s[i]); break; //vsin case 18: // d[i] = sinf((float)M_PI_2 * s[i]); break; //vsin
LEA(64, RDX, MIPSSTATE_VAR(sincostemp));
trigCallHelper(&SinOnly, sregs[i]); trigCallHelper(&SinOnly, sregs[i]);
MOVSS(tempxregs[i], M(&sincostemp[0])); MOVSS(tempxregs[i], MIPSSTATE_VAR(sincostemp[0]));
break; break;
case 19: // d[i] = cosf((float)M_PI_2 * s[i]); break; //vcos case 19: // d[i] = cosf((float)M_PI_2 * s[i]); break; //vcos
LEA(64, RDX, MIPSSTATE_VAR(sincostemp));
trigCallHelper(&CosOnly, sregs[i]); trigCallHelper(&CosOnly, sregs[i]);
MOVSS(tempxregs[i], M(&sincostemp[1])); MOVSS(tempxregs[i], MIPSSTATE_VAR(sincostemp[1]));
break; break;
case 20: // d[i] = powf(2.0f, s[i]); break; //vexp2 case 20: // d[i] = powf(2.0f, s[i]); break; //vexp2
DISABLE; DISABLE;
@ -2369,8 +2375,9 @@ void Jit::Comp_VV2Op(MIPSOpcode op) {
ANDPS(tempxregs[i], MatR(TEMPREG)); ANDPS(tempxregs[i], MatR(TEMPREG));
break; break;
case 23: // d[i] = asinf(s[i]) / M_PI_2; break; //vasin case 23: // d[i] = asinf(s[i]) / M_PI_2; break; //vasin
LEA(64, RDX, MIPSSTATE_VAR(sincostemp));
trigCallHelper(&ASinScaled, sregs[i]); trigCallHelper(&ASinScaled, sregs[i]);
MOVSS(tempxregs[i], M(&sincostemp[0])); MOVSS(tempxregs[i], MIPSSTATE_VAR(sincostemp[0]));
break; break;
case 24: // d[i] = -1.0f / s[i]; break; // vnrcp case 24: // d[i] = -1.0f / s[i]; break; // vnrcp
// Rare so let's not bother checking for RipAccessible. // Rare so let's not bother checking for RipAccessible.
@ -2380,8 +2387,9 @@ void Jit::Comp_VV2Op(MIPSOpcode op) {
MOVSS(tempxregs[i], R(XMM0)); MOVSS(tempxregs[i], R(XMM0));
break; break;
case 26: // d[i] = -sinf((float)M_PI_2 * s[i]); break; // vnsin case 26: // d[i] = -sinf((float)M_PI_2 * s[i]); break; // vnsin
LEA(64, RDX, MIPSSTATE_VAR(sincostemp));
trigCallHelper(&NegSinOnly, sregs[i]); trigCallHelper(&NegSinOnly, sregs[i]);
MOVSS(tempxregs[i], M(&sincostemp[0])); MOVSS(tempxregs[i], MIPSSTATE_VAR(sincostemp[0]));
break; break;
case 28: // d[i] = 1.0f / expf(s[i] * (float)M_LOG2E); break; // vrexp2 case 28: // d[i] = 1.0f / expf(s[i] * (float)M_LOG2E); break; // vrexp2
DISABLE; DISABLE;
@ -3514,15 +3522,20 @@ void Jit::Comp_VRot(MIPSOpcode op) {
bool negSin1 = (imm & 0x10) ? true : false; bool negSin1 = (imm & 0x10) ? true : false;
#ifdef _M_X64 #ifdef _M_X64
#ifdef _WIN32
LEA(64, RDX, MIPSSTATE_VAR(sincostemp));
#else
LEA(64, RSI, MIPSSTATE_VAR(sincostemp));
#endif
MOVSS(XMM0, fpr.V(sreg)); MOVSS(XMM0, fpr.V(sreg));
ABI_CallFunction(negSin1 ? (const void *)&SinCosNegSin : (const void *)&SinCos); ABI_CallFunction(negSin1 ? (const void *)&SinCosNegSin : (const void *)&SinCos);
#else #else
// Sigh, passing floats with cdecl isn't pretty, ends up on the stack. // Sigh, passing floats with cdecl isn't pretty, ends up on the stack.
ABI_CallFunctionA(negSin1 ? (const void *)&SinCosNegSin : (const void *)&SinCos, fpr.V(sreg)); ABI_CallFunctionAC(negSin1 ? (const void *)&SinCosNegSin : (const void *)&SinCos, fpr.V(sreg), (uintptr_t)mips_->sincostemp);
#endif #endif
MOVSS(XMM0, M(&sincostemp[0])); MOVSS(XMM0, MIPSSTATE_VAR(sincostemp[0]));
MOVSS(XMM1, M(&sincostemp[1])); MOVSS(XMM1, MIPSSTATE_VAR(sincostemp[1]));
CompVrotShuffle(dregs, imm, n, false); CompVrotShuffle(dregs, imm, n, false);
if (vd2 != -1) { if (vd2 != -1) {