mirror/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-01-30 12:32:43 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' of https://github.com/DHrpcs3/rpcs3

Browse Source
This commit is contained in:
Nekotekina 2015-01-19 00:29:41 +03:00
commit 87f1a9d9dc
9 changed files with 805 additions and 205 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
rpcs3/Emu/Cell/Common.h Normal file
View File

@ -0,0 +1,10 @@
#pragma once
// Floating-point rounding mode (for both PPU and SPU)
enum FPSCR_RN
{
FPSCR_RN_NEAR = 0,
FPSCR_RN_ZERO = 1,
FPSCR_RN_PINF = 2,
FPSCR_RN_MINF = 3,
};

79
rpcs3/Emu/Cell/PPUInterpreter.h
View File

@ -17,6 +17,8 @@
#define _rotl64(x,r) (((u64)(x) << (r)) | ((u64)(x) >> (64 - (r))))
#endif
#include <fenv.h>
#if 0//def _DEBUG
#define HLE_CALL_DEBUG
#endif
@ -60,6 +62,26 @@ static double SilenceNaN(double x)
return (double&)bits;
}
static void SetHostRoundingMode(u32 rn)
{
switch (rn)
{
case FPSCR_RN_NEAR:
fesetround(FE_TONEAREST);
break;
case FPSCR_RN_ZERO:
fesetround(FE_TOWARDZERO);
break;
case FPSCR_RN_PINF:
fesetround(FE_UPWARD);
break;
case FPSCR_RN_MINF:
fesetround(FE_DOWNWARD);
break;
}
}
namespace ppu_recompiler_llvm {
class Compiler;
}
@ -79,6 +101,13 @@ public:
}
private:
void CheckHostFPExceptions()
{
CPU.SetFPSCR_FI(fetestexcept(FE_INEXACT) != 0);
if (fetestexcept(FE_UNDERFLOW)) CPU.SetFPSCRException(FPSCR_UX);
if (fetestexcept(FE_OVERFLOW)) CPU.SetFPSCRException(FPSCR_OX);
}
void Exit() {}
void SysCall()
@ -245,6 +274,7 @@ private:
}
void VADDFP(u32 vd, u32 va, u32 vb)
{
SetHostRoundingMode(FPSCR_RN_NEAR);
for (uint w = 0; w < 4; w++)
{
const float a = CheckVSCR_NJ(CPU.VPR[va]._f[w]);
@ -441,6 +471,7 @@ private:
}
void VCFSX(u32 vd, u32 uimm5, u32 vb)
{
SetHostRoundingMode(FPSCR_RN_NEAR);
u32 scale = 1 << uimm5;
for (uint w = 0; w < 4; w++)
@ -450,6 +481,7 @@ private:
}
void VCFUX(u32 vd, u32 uimm5, u32 vb)
{
SetHostRoundingMode(FPSCR_RN_NEAR);
u32 scale = 1 << uimm5;
for (uint w = 0; w < 4; w++)
@ -786,8 +818,8 @@ private:
CPU.VPR[vd]._s32[w] = (int)0x80000000;
CPU.VSCR.SAT = 1;
}
else // C rounding = Round towards 0
CPU.VPR[vd]._s32[w] = (int)result;
else
CPU.VPR[vd]._s32[w] = (int)trunc(result);
}
}
}
@ -804,7 +836,6 @@ private:
}
else
{
// C rounding = Round towards 0
double result = (double)b * nScale;
if (result > 0xffffffffu)
{
@ -817,7 +848,7 @@ private:
CPU.VSCR.SAT = 1;
}
else
CPU.VPR[vd]._u32[w] = (u32)result;
CPU.VPR[vd]._u32[w] = (u32)trunc(result);
}
}
}
@ -826,6 +857,7 @@ private:
// vd = 2^x
// ISA : Note that the value placed into the element of vD may vary between implementations
// and between different executions on the same implementation.
SetHostRoundingMode(FPSCR_RN_NEAR);
for (uint w = 0; w < 4; w++)
{
const float b = CheckVSCR_NJ(CPU.VPR[vb]._f[w]);
@ -839,6 +871,7 @@ private:
{
// ISA : Note that the value placed into the element of vD may vary between implementations
// and between different executions on the same implementation.
SetHostRoundingMode(FPSCR_RN_NEAR);
for (uint w = 0; w < 4; w++)
{
const float b = CheckVSCR_NJ(CPU.VPR[vb]._f[w]);
@ -850,6 +883,7 @@ private:
}
void VMADDFP(u32 vd, u32 va, u32 vc, u32 vb)
{
SetHostRoundingMode(FPSCR_RN_NEAR);
for (uint w = 0; w < 4; w++)
{
const float a = CheckVSCR_NJ(CPU.VPR[va]._f[w]);
@ -1275,6 +1309,7 @@ private:
}
void VNMSUBFP(u32 vd, u32 va, u32 vc, u32 vb)
{
SetHostRoundingMode(FPSCR_RN_NEAR);
for (uint w = 0; w < 4; w++)
{
const float a = CheckVSCR_NJ(CPU.VPR[va]._f[w]);
@ -1568,6 +1603,7 @@ private:
}
void VREFP(u32 vd, u32 vb)
{
SetHostRoundingMode(FPSCR_RN_NEAR);
for (uint w = 0; w < 4; w++)
{
const float b = CheckVSCR_NJ(CPU.VPR[vb]._f[w]);
@ -1596,7 +1632,10 @@ private:
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
{
SetHostRoundingMode(FPSCR_RN_NEAR);
CPU.VPR[vd]._f[w] = nearbyintf(CPU.VPR[vb]._f[w]);
}
}
}
void VRFIP(u32 vd, u32 vb)
@ -1646,6 +1685,7 @@ private:
}
void VRSQRTEFP(u32 vd, u32 vb)
{
SetHostRoundingMode(FPSCR_RN_NEAR);
for (uint w = 0; w < 4; w++)
{
//TODO: accurate div
@ -1846,6 +1886,7 @@ private:
}
void VSUBFP(u32 vd, u32 va, u32 vb)
{
SetHostRoundingMode(FPSCR_RN_NEAR);
for (uint w = 0; w < 4; w++)
{
const float a = CheckVSCR_NJ(CPU.VPR[va]._f[w]);
@ -3663,6 +3704,7 @@ private:
void FSQRTS(u32 frd, u32 frb, bool rc) {FSQRT(frd, frb, rc, true);}
void FRES(u32 frd, u32 frb, bool rc)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double b = CPU.FPR[frb];
if(FPRdouble::IsSNaN(b))
{
@ -3693,7 +3735,9 @@ private:
}
else
{
feclearexcept(FE_ALL_EXCEPT);
CPU.FPR[frd] = static_cast<float>(1.0 / b);
CheckHostFPExceptions();
}
CPU.FPSCR.FPRF = CPU.FPR[frd].GetType();
if(rc) CPU.UpdateCR1();
@ -3797,6 +3841,7 @@ private:
}
void FRSP(u32 frd, u32 frb, bool rc)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double b = CPU.FPR[frb];
if (FPRdouble::IsSNaN(b))
{
@ -3814,6 +3859,7 @@ private:
{
if (((u64&)b0 & DOUBLE_EXP) < 0x3800000000000000ULL) (u64&)b0 &= DOUBLE_SIGN;
}
feclearexcept(FE_ALL_EXCEPT);
const double r = static_cast<float>(b0);
if (FPRdouble::IsNaN(r))
{
@ -3823,7 +3869,7 @@ private:
else
{
CPU.FPSCR.FR = fabs(r) > fabs(b);
CPU.SetFPSCR_FI(b != r);
CheckHostFPExceptions();
}
u32 type = PPCdouble(r).GetType();
if (type == FPR_PN && r < ldexp(1.0, -126)) type = FPR_PD;
@ -3869,6 +3915,7 @@ private:
switch(rn)
{
case FPSCR_RN_NEAR:
SetHostRoundingMode(FPSCR_RN_NEAR);
i = (s32)nearbyint(b);
break;
case FPSCR_RN_ZERO:
@ -3902,6 +3949,7 @@ private:
void FDIV(u32 frd, u32 fra, u32 frb, bool rc) {FDIV(frd, fra, frb, rc, false);}
void FDIV(u32 frd, u32 fra, u32 frb, bool rc, bool single)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double a = CPU.FPR[fra];
const double b = CPU.FPR[frb];
if(FPRdouble::IsSNaN(a) || FPRdouble::IsSNaN(b))
@ -3960,9 +4008,11 @@ private:
return;
}
}
feclearexcept(FE_ALL_EXCEPT);
const double res = a / b;
if(single) CPU.FPR[frd] = (float)res;
else CPU.FPR[frd] = res;
CheckHostFPExceptions();
}
CPU.FPSCR.FPRF = CPU.FPR[frd].GetType();
@ -3971,6 +4021,7 @@ private:
void FSUB(u32 frd, u32 fra, u32 frb, bool rc) {FSUB(frd, fra, frb, rc, false);}
void FSUB(u32 frd, u32 fra, u32 frb, bool rc, bool single)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double a = CPU.FPR[fra];
const double b = CPU.FPR[frb];
if(FPRdouble::IsSNaN(a) || FPRdouble::IsSNaN(b))
@ -4006,9 +4057,11 @@ private:
}
else
{
feclearexcept(FE_ALL_EXCEPT);
const double res = a - b;
if(single) CPU.FPR[frd] = (float)res;
else CPU.FPR[frd] = res;
CheckHostFPExceptions();
}
CPU.FPSCR.FPRF = CPU.FPR[frd].GetType();
if(rc) CPU.UpdateCR1();
@ -4016,6 +4069,7 @@ private:
void FADD(u32 frd, u32 fra, u32 frb, bool rc) {FADD(frd, fra, frb, rc, false);}
void FADD(u32 frd, u32 fra, u32 frb, bool rc, bool single)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double a = CPU.FPR[fra];
const double b = CPU.FPR[frb];
if(FPRdouble::IsSNaN(a) || FPRdouble::IsSNaN(b))
@ -4051,9 +4105,11 @@ private:
}
else
{
feclearexcept(FE_ALL_EXCEPT);
const double res = a + b;
if(single) CPU.FPR[frd] = (float)res;
else CPU.FPR[frd] = res;
CheckHostFPExceptions();
}
CPU.FPSCR.FPRF = CPU.FPR[frd].GetType();
if(rc) CPU.UpdateCR1();
@ -4061,6 +4117,7 @@ private:
void FSQRT(u32 frd, u32 frb, bool rc) {FSQRT(frd, frb, rc, false);}
void FSQRT(u32 frd, u32 frb, bool rc, bool single)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double b = CPU.FPR[frb];
if(FPRdouble::IsSNaN(b))
{
@ -4091,9 +4148,11 @@ private:
}
else
{
feclearexcept(FE_ALL_EXCEPT);
const double res = sqrt(b);
if(single) CPU.FPR[frd] = (float)res;
else CPU.FPR[frd] = res;
CheckHostFPExceptions();
}
CPU.FPSCR.FPRF = CPU.FPR[frd].GetType();
if(rc) CPU.UpdateCR1();
@ -4106,6 +4165,7 @@ private:
void FMUL(u32 frd, u32 fra, u32 frc, bool rc) {FMUL(frd, fra, frc, rc, false);}
void FMUL(u32 frd, u32 fra, u32 frc, bool rc, bool single)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double a = CPU.FPR[fra];
const double c = CPU.FPR[frc];
if(FPRdouble::IsSNaN(a) || FPRdouble::IsSNaN(c))
@ -4141,15 +4201,18 @@ private:
}
else
{
feclearexcept(FE_ALL_EXCEPT);
const double res = a * c;
if(single) CPU.FPR[frd] = (float)res;
else CPU.FPR[frd] = res;
CheckHostFPExceptions();
}
CPU.FPSCR.FPRF = CPU.FPR[frd].GetType();
if(rc) CPU.UpdateCR1();
}
void FRSQRTE(u32 frd, u32 frb, bool rc)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double b = CPU.FPR[frb];
if(FPRdouble::IsSNaN(b))
{
@ -4192,7 +4255,9 @@ private:
}
else
{
feclearexcept(FE_ALL_EXCEPT);
CPU.FPR[frd] = 1.0 / sqrt(b);
CheckHostFPExceptions();
}
CPU.FPSCR.FPRF = CPU.FPR[frd].GetType();
if(rc) CPU.UpdateCR1();
@ -4201,6 +4266,7 @@ private:
void FMADD(u32 frd, u32 fra, u32 frc, u32 frb, bool rc) {FMADD(frd, fra, frc, frb, rc, false, false, false);}
void FMADD(u32 frd, u32 fra, u32 frc, u32 frb, bool rc, bool neg, bool sub, bool single)
{
SetHostRoundingMode(CPU.FPSCR.RN);
const double a = CPU.FPR[fra];
const double b = CPU.FPR[frb];
const double c = CPU.FPR[frc];
@ -4256,8 +4322,10 @@ private:
}
else
{
feclearexcept(FE_ALL_EXCEPT);
if(single) CPU.FPR[frd] = (float)(neg ? -res : res);
else CPU.FPR[frd] = (neg ? -res : res);
CheckHostFPExceptions();
}
}
CPU.FPSCR.FPRF = CPU.FPR[frd].GetType();
@ -4344,6 +4412,7 @@ private:
switch(rn)
{
case FPSCR_RN_NEAR:
SetHostRoundingMode(FPSCR_RN_NEAR);
i = (s64)nearbyint(b);
break;
case FPSCR_RN_ZERO:

9
rpcs3/Emu/Cell/PPUThread.h
View File

@ -1,4 +1,5 @@
#pragma once
#include "Emu/Cell/Common.h"
#include "Emu/Cell/PPCThread.h"
#include "Emu/Memory/vm.h"
@ -57,14 +58,6 @@ enum FPSCR_EXP
FPSCR_VX_ALL = FPSCR_VXSNAN | FPSCR_VXISI | FPSCR_VXIDI | FPSCR_VXZDZ | FPSCR_VXIMZ | FPSCR_VXVC | FPSCR_VXSOFT | FPSCR_VXSQRT | FPSCR_VXCVI,
};
enum FPSCR_RN
{
FPSCR_RN_NEAR = 0,
FPSCR_RN_ZERO = 1,
FPSCR_RN_PINF = 2,
FPSCR_RN_MINF = 3,
};
static const u64 DOUBLE_SIGN = 0x8000000000000000ULL;
static const u64 DOUBLE_EXP = 0x7FF0000000000000ULL;
static const u64 DOUBLE_FRAC = 0x000FFFFFFFFFFFFFULL;

789
rpcs3/Emu/Cell/SPUInterpreter.h
View File

@ -1,5 +1,26 @@
#pragma once
#include <fenv.h>
static void SetHostRoundingMode(u32 rn)
{
switch (rn)
{
case FPSCR_RN_NEAR:
fesetround(FE_TONEAREST);
break;
case FPSCR_RN_ZERO:
fesetround(FE_TOWARDZERO);
break;
case FPSCR_RN_PINF:
fesetround(FE_UPWARD);
break;
case FPSCR_RN_MINF:
fesetround(FE_DOWNWARD);
break;
}
}
#define UNIMPLEMENTED() UNK(__FUNCTION__)
#define MEM_AND_REG_HASH() \
@ -11,6 +32,54 @@
#define LOG5_OPCODE(...) ///
// Floating-point utility constants and functions
static const u32 FLOAT_MAX_NORMAL_I = 0x7F7FFFFF;
static const float& FLOAT_MAX_NORMAL = (float&)FLOAT_MAX_NORMAL_I;
static const u32 FLOAT_NAN_I = 0x7FC00000;
static const float& FLOAT_NAN = (float&)FLOAT_NAN_I;
static const u64 DOUBLE_NAN_I = 0x7FF8000000000000ULL;
static const double& DOUBLE_NAN = (double&)DOUBLE_NAN_I;
static inline bool issnan(double x) {return isnan(x) && ((s64&)x)<<12 > 0;}
static inline bool issnan(float x) {return isnan(x) && ((s32&)x)<<9 > 0;}
static inline int fexpf(float x) {return ((u32&)x >> 23) & 0xFF;}
static inline bool isextended(float x) {return fexpf(x) == 255;}
static inline float extended(bool sign, u32 mantissa) // returns -1^sign * 2^127 * (1.mantissa)
{
u32 bits = sign<<31 | 0x7F800000 | mantissa;
return (float&)bits;
}
static inline float ldexpf_extended(float x, int exp) // ldexpf() for extended values, assumes result is in range
{
u32 bits = (u32&)x;
if (bits << 1 != 0) bits += exp * 0x00800000;
return (float&)bits;
}
static inline bool isdenormal(float x)
{
const int fpc = _fpclass(x);
#ifdef __GNUG__
return fpc == FP_SUBNORMAL;
#else
return (fpc & (_FPCLASS_PD | _FPCLASS_ND)) != 0;
#endif
}
static inline bool isdenormal(double x)
{
const int fpc = _fpclass(x);
#ifdef __GNUG__
return fpc == FP_SUBNORMAL;
#else
return (fpc & (_FPCLASS_PD | _FPCLASS_ND)) != 0;
#endif
}
static double SilenceNaN(double x)
{
u64 bits = (u64&)x;
bits |= 0x0008000000000000ULL;
return (double&)bits;
}
class SPUInterpreter : public SPUOpcodes
{
private:
@ -47,7 +116,7 @@ private:
}
void MFSPR(u32 rt, u32 sa)
{
UNIMPLEMENTED(); // not used
CPU.GPR[rt].clear(); // All SPRs read as zero.
}
void RDCH(u32 rt, u32 ra)
{
@ -243,7 +312,7 @@ private:
}
void MTSPR(u32 rt, u32 sa)
{
UNIMPLEMENTED(); // not used
// SPR writes are ignored.
}
void WRCH(u32 ra, u32 rt)
{
@ -431,16 +500,41 @@ private:
}
void FREST(u32 rt, u32 ra)
{
//CPU.GPR[rt]._m128 = _mm_rcp_ps(CPU.GPR[ra]._m128);
SetHostRoundingMode(FPSCR_RN_ZERO);
for (int i = 0; i < 4; i++)
CPU.GPR[rt]._f[i] = 1 / CPU.GPR[ra]._f[i];
{
const float a = CPU.GPR[ra]._f[i];
float result;
if (fexpf(a) == 0)
{
CPU.FPSCR.setDivideByZeroFlag(i);
result = extended(std::signbit(a), 0x7FFFFF);
}
else if (isextended(a))
result = 0.0f;
else
result = 1 / a;
CPU.GPR[rt]._f[i] = result;
}
}
void FRSQEST(u32 rt, u32 ra)
{
//const __u32x4 FloatAbsMask = {0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff};
//CPU.GPR[rt]._m128 = _mm_rsqrt_ps(_mm_and_ps(CPU.GPR[ra]._m128, FloatAbsMask.m128));
SetHostRoundingMode(FPSCR_RN_ZERO);
for (int i = 0; i < 4; i++)
CPU.GPR[rt]._f[i] = 1 / sqrt(abs(CPU.GPR[ra]._f[i]));
{
const float a = CPU.GPR[ra]._f[i];
float result;
if (fexpf(a) == 0)
{
CPU.FPSCR.setDivideByZeroFlag(i);
result = extended(0, 0x7FFFFF);
}
else if (isextended(a))
result = 0.5f / sqrtf(fabsf(ldexpf_extended(a, -2)));
else
result = 1 / sqrtf(fabsf(a));
CPU.GPR[rt]._f[i] = result;
}
}
void LQX(u32 rt, u32 ra, u32 rb)
{
@ -828,37 +922,186 @@ private:
}
void FCGT(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._u32[0] = CPU.GPR[ra]._f[0] > CPU.GPR[rb]._f[0] ? 0xffffffff : 0;
CPU.GPR[rt]._u32[1] = CPU.GPR[ra]._f[1] > CPU.GPR[rb]._f[1] ? 0xffffffff : 0;
CPU.GPR[rt]._u32[2] = CPU.GPR[ra]._f[2] > CPU.GPR[rb]._f[2] ? 0xffffffff : 0;
CPU.GPR[rt]._u32[3] = CPU.GPR[ra]._f[3] > CPU.GPR[rb]._f[3] ? 0xffffffff : 0;
for (int i = 0; i < 4; i++)
{
const u32 a = CPU.GPR[ra]._u32[i];
const u32 b = CPU.GPR[rb]._u32[i];
const u32 abs_a = a & 0x7FFFFFFF;
const u32 abs_b = b & 0x7FFFFFFF;
const bool a_zero = (abs_a < 0x00800000);
const bool b_zero = (abs_b < 0x00800000);
bool pass;
if (a_zero)
pass = b >= 0x80800000;
else if (b_zero)
pass = (s32)a >= 0x00800000;
else if (a >= 0x80000000)
pass = (b >= 0x80000000 && a < b);
else
pass = (b >= 0x80000000 || a > b);
CPU.GPR[rt]._u32[i] = pass ? 0xFFFFFFFF : 0;
}
}
void DFCGT(u32 rt, u32 ra, u32 rb)
{
UNIMPLEMENTED(); // cannot be used
}
void FA(u32 rt, u32 ra, u32 rb)
void FA_FS(u32 rt, u32 ra, u32 rb, bool sub)
{
SetHostRoundingMode(FPSCR_RN_ZERO);
for (int w = 0; w < 4; w++)
{
CPU.GPR[rt]._f[w] = CPU.GPR[ra]._f[w] + CPU.GPR[rb]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f;
}
}
void FS(u32 rt, u32 ra, u32 rb)
{
for (int w = 0; w < 4; w++)
{
CPU.GPR[rt]._f[w] = CPU.GPR[ra]._f[w] - CPU.GPR[rb]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f;
const float a = CPU.GPR[ra]._f[w];
const float b = sub ? -CPU.GPR[rb]._f[w] : CPU.GPR[rb]._f[w];
float result;
if (isdenormal(a))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (b == 0.0f || isdenormal(b))
result = +0.0f;
else
result = b;
}
else if (isdenormal(b))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (a == 0.0f)
result = +0.0f;
else
result = a;
}
else if (isextended(a) || isextended(b))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (isextended(a) && fexpf(b) < 255-32)
{
if (std::signbit(a) != std::signbit(b))
{
const u32 bits = (u32&)a - 1;
result = (float&)bits;
}
else
result = a;
}
else if (isextended(b) && fexpf(a) < 255-32)
{
if (std::signbit(a) != std::signbit(b))
{
const u32 bits = (u32&)b - 1;
result = (float&)bits;
}
else
result = b;
}
else
{
feclearexcept(FE_ALL_EXCEPT);
result = ldexpf_extended(a, -1) + ldexpf_extended(b, -1);
result = ldexpf_extended(result, 1);
if (fetestexcept(FE_OVERFLOW))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
result = extended(std::signbit(result), 0x7FFFFF);
}
}
}
else
{
result = a + b;
if (result == copysignf(FLOAT_MAX_NORMAL, result))
{
result = ldexpf_extended(ldexpf(a,-1) + ldexpf(b,-1), 1);
if (isextended(result))
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
}
else if (isdenormal(result))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SUNF | FPSCR_SDIFF);
result = +0.0f;
}
else if (result == 0.0f)
{
if (fabsf(a) != fabsf(b))
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SUNF | FPSCR_SDIFF);
result = +0.0f;
}
}
CPU.GPR[rt]._f[w] = result;
}
}
void FA(u32 rt, u32 ra, u32 rb) {FA_FS(rt, ra, rb, false);}
void FS(u32 rt, u32 ra, u32 rb) {FA_FS(rt, ra, rb, true);}
void FM(u32 rt, u32 ra, u32 rb)
{
SetHostRoundingMode(FPSCR_RN_ZERO);
for (int w = 0; w < 4; w++)
{
CPU.GPR[rt]._f[w] = CPU.GPR[ra]._f[w] * CPU.GPR[rb]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f;
const float a = CPU.GPR[ra]._f[w];
const float b = CPU.GPR[rb]._f[w];
float result;
if (isdenormal(a) || isdenormal(b))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
result = +0.0f;
}
else if (isextended(a) || isextended(b))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
const bool sign = std::signbit(a) ^ std::signbit(b);
if (a == 0.0f || b == 0.0f)
{
result = +0.0f;
}
else if ((fexpf(a)-127) + (fexpf(b)-127) >= 129)
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
result = extended(sign, 0x7FFFFF);
}
else
{
if (isextended(a))
result = ldexpf_extended(a, -1) * b;
else
result = a * ldexpf_extended(b, -1);
if (result == copysignf(FLOAT_MAX_NORMAL, result))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
result = extended(sign, 0x7FFFFF);
}
else
result = ldexpf_extended(result, 1);
}
}
else
{
result = a * b;
if (result == copysignf(FLOAT_MAX_NORMAL, result))
{
feclearexcept(FE_ALL_EXCEPT);
if (fexpf(a) > fexpf(b))
result = ldexpf(a, -1) * b;
else
result = a * ldexpf(b, -1);
result = ldexpf_extended(result, 1);
if (isextended(result))
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (fetestexcept(FE_OVERFLOW))
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
}
else if (isdenormal(result))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SUNF | FPSCR_SDIFF);
result = +0.0f;
}
else if (result == 0.0f)
{
if (a != 0.0f & b != 0.0f)
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SUNF | FPSCR_SDIFF);
result = +0.0f;
}
}
CPU.GPR[rt]._f[w] = result;
}
}
void CLGTH(u32 rt, u32 ra, u32 rb)
@ -873,30 +1116,84 @@ private:
}
void FCMGT(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._u32[0] = fabs(CPU.GPR[ra]._f[0]) > fabs(CPU.GPR[rb]._f[0]) ? 0xffffffff : 0;
CPU.GPR[rt]._u32[1] = fabs(CPU.GPR[ra]._f[1]) > fabs(CPU.GPR[rb]._f[1]) ? 0xffffffff : 0;
CPU.GPR[rt]._u32[2] = fabs(CPU.GPR[ra]._f[2]) > fabs(CPU.GPR[rb]._f[2]) ? 0xffffffff : 0;
CPU.GPR[rt]._u32[3] = fabs(CPU.GPR[ra]._f[3]) > fabs(CPU.GPR[rb]._f[3]) ? 0xffffffff : 0;
for (int i = 0; i < 4; i++)
{
const u32 a = CPU.GPR[ra]._u32[i];
const u32 b = CPU.GPR[rb]._u32[i];
const u32 abs_a = a & 0x7FFFFFFF;
const u32 abs_b = b & 0x7FFFFFFF;
const bool a_zero = (abs_a < 0x00800000);
const bool b_zero = (abs_b < 0x00800000);
bool pass;
if (a_zero)
pass = false;
else if (b_zero)
pass = !a_zero;
else
pass = abs_a > abs_b;
CPU.GPR[rt]._u32[i] = pass ? 0xFFFFFFFF : 0;
}
}
void DFCMGT(u32 rt, u32 ra, u32 rb)
{
UNIMPLEMENTED(); // cannot be used
}
void DFA(u32 rt, u32 ra, u32 rb)
enum DoubleOp {DFASM_A, DFASM_S, DFASM_M};
void DFASM(u32 rt, u32 ra, u32 rb, DoubleOp op)
{
CPU.GPR[rt]._d[0] = CPU.GPR[ra]._d[0] + CPU.GPR[rb]._d[0];
CPU.GPR[rt]._d[1] = CPU.GPR[ra]._d[1] + CPU.GPR[rb]._d[1];
}
void DFS(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._d[0] = CPU.GPR[ra]._d[0] - CPU.GPR[rb]._d[0];
CPU.GPR[rt]._d[1] = CPU.GPR[ra]._d[1] - CPU.GPR[rb]._d[1];
}
void DFM(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._d[0] = CPU.GPR[ra]._d[0] * CPU.GPR[rb]._d[0];
CPU.GPR[rt]._d[1] = CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1];
for (int i = 0; i < 2; i++)
{
double a = CPU.GPR[ra]._d[i];
double b = CPU.GPR[rb]._d[i];
if (isdenormal(a))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
a = copysign(0.0, a);
}
if (isdenormal(b))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
b = copysign(0.0, b);
}
double result;
if (isnan(a) || isnan(b))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DNAN);
if (issnan(a) || issnan(b))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINV);
result = DOUBLE_NAN;
}
else
{
SetHostRoundingMode(CPU.FPSCR.checkSliceRounding(i));
feclearexcept(FE_ALL_EXCEPT);
switch (op)
{
case DFASM_A: result = a + b; break;
case DFASM_S: result = a - b; break;
case DFASM_M: result = a * b; break;
}
if (fetestexcept(FE_INVALID))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINV);
result = DOUBLE_NAN;
}
else
{
if (fetestexcept(FE_OVERFLOW))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DOVF);
if (fetestexcept(FE_UNDERFLOW))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DUNF);
if (fetestexcept(FE_INEXACT))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINX);
}
}
CPU.GPR[rt]._d[i] = result;
}
}
void DFA(u32 rt, u32 ra, u32 rb) {DFASM(rt, ra, rb, DFASM_A);}
void DFS(u32 rt, u32 ra, u32 rb) {DFASM(rt, ra, rb, DFASM_S);}
void DFM(u32 rt, u32 ra, u32 rb) {DFASM(rt, ra, rb, DFASM_M);}
void CLGTB(u32 rt, u32 ra, u32 rb)
{
for (int b = 0; b < 16; b++)
@ -910,26 +1207,64 @@ private:
CPU.Stop();
}
}
void DFMA(u32 rt, u32 ra, u32 rb)
void DFMA(u32 rt, u32 ra, u32 rb, bool neg, bool sub)
{
CPU.GPR[rt]._d[0] += CPU.GPR[ra]._d[0] * CPU.GPR[rb]._d[0];
CPU.GPR[rt]._d[1] += CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1];
}
void DFMS(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._d[0] = CPU.GPR[ra]._d[0] * CPU.GPR[rb]._d[0] - CPU.GPR[rt]._d[0];
CPU.GPR[rt]._d[1] = CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1] - CPU.GPR[rt]._d[1];
}
void DFNMS(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._d[0] -= CPU.GPR[ra]._d[0] * CPU.GPR[rb]._d[0];
CPU.GPR[rt]._d[1] -= CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1];
}
void DFNMA(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._d[0] = -(CPU.GPR[ra]._d[0] * CPU.GPR[rb]._d[0] + CPU.GPR[rt]._d[0]);
CPU.GPR[rt]._d[1] = -(CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1] + CPU.GPR[rt]._d[1]);
for (int i = 0; i < 2; i++)
{
double a = CPU.GPR[ra]._d[i];
double b = CPU.GPR[rb]._d[i];
double c = CPU.GPR[rt]._d[i];
if (isdenormal(a))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
a = copysign(0.0, a);
}
if (isdenormal(b))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
b = copysign(0.0, b);
}
if (isdenormal(c))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
c = copysign(0.0, c);
}
double result;
if (isnan(a) || isnan(b) || isnan(c))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DNAN);
if (issnan(a) || issnan(b) || issnan(c) || (isinf(a) && b == 0.0f) || (a == 0.0f && isinf(b)))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINV);
result = DOUBLE_NAN;
}
else
{
SetHostRoundingMode(CPU.FPSCR.checkSliceRounding(i));
feclearexcept(FE_ALL_EXCEPT);
result = fma(a, b, sub ? -c : c);
if (fetestexcept(FE_INVALID))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINV);
result = DOUBLE_NAN;
}
else
{
if (fetestexcept(FE_OVERFLOW))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DOVF);
if (fetestexcept(FE_UNDERFLOW))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DUNF);
if (fetestexcept(FE_INEXACT))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINX);
if (neg) result = -result;
}
}
CPU.GPR[rt]._d[i] = result;
}
}
void DFMA(u32 rt, u32 ra, u32 rb) {DFMA(rt, ra, rb, false, false);}
void DFMS(u32 rt, u32 ra, u32 rb) {DFMA(rt, ra, rb, false, true);}
void DFNMS(u32 rt, u32 ra, u32 rb) {DFMA(rt, ra, rb, true, true);}
void DFNMA(u32 rt, u32 ra, u32 rb) {DFMA(rt, ra, rb, true, false);}
void CEQ(u32 rt, u32 ra, u32 rb)
{
for (int w = 0; w < 4; w++)
@ -981,29 +1316,67 @@ private:
void FSCRRD(u32 rt)
{
// TODO (rarely used)
CPU.GPR[rt].clear();
CPU.GPR[rt]._u32[3] = CPU.FPSCR._u32[3];
CPU.GPR[rt]._u32[2] = CPU.FPSCR._u32[2];
CPU.GPR[rt]._u32[1] = CPU.FPSCR._u32[1];
CPU.GPR[rt]._u32[0] = CPU.FPSCR._u32[0];
}
void FESD(u32 rt, u32 ra)
{
CPU.GPR[rt]._d[0] = (double)CPU.GPR[ra]._f[1];
CPU.GPR[rt]._d[1] = (double)CPU.GPR[ra]._f[3];
for (int i = 0; i < 2; i++)
{
const float a = CPU.GPR[ra]._f[i*2+1];
if (isnan(a))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DNAN);
if (issnan(a))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINV);
CPU.GPR[rt]._d[i] = DOUBLE_NAN;
}
else if (isdenormal(a))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
CPU.GPR[rt]._d[i] = 0.0;
}
else
{
CPU.GPR[rt]._d[i] = (double)a;
}
}
}
void FRDS(u32 rt, u32 ra)
{
CPU.GPR[rt]._f[1] = (float)CPU.GPR[ra]._d[0];
CPU.GPR[rt]._u32[0] = 0x00000000;
CPU.GPR[rt]._f[3] = (float)CPU.GPR[ra]._d[1];
CPU.GPR[rt]._u32[2] = 0x00000000;
for (int i = 0; i < 2; i++)
{
SetHostRoundingMode(CPU.FPSCR.checkSliceRounding(i));
const double a = CPU.GPR[ra]._d[i];
if (isnan(a))
{
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DNAN);
if (issnan(a))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINV);
CPU.GPR[rt]._f[i*2+1] = FLOAT_NAN;
}
else
{
feclearexcept(FE_ALL_EXCEPT);
CPU.GPR[rt]._f[i*2+1] = (float)a;
if (fetestexcept(FE_OVERFLOW))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DOVF);
if (fetestexcept(FE_UNDERFLOW))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DUNF);
if (fetestexcept(FE_INEXACT))
CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DINX);
}
CPU.GPR[rt]._u32[i*2] = 0;
}
}
void FSCRWR(u32 rt, u32 ra)
{
// TODO (rarely used)
if (CPU.GPR[ra]._u64[0] || CPU.GPR[ra]._u64[1])
{
LOG_ERROR(SPU, "FSCRWR(%d,%d): value = %s", rt, ra, CPU.GPR[ra].to_hex().c_str());
UNIMPLEMENTED();
}
CPU.FPSCR._u32[3] = CPU.GPR[ra]._u32[3] & 0x00000F07;
CPU.FPSCR._u32[2] = CPU.GPR[ra]._u32[2] & 0x00003F07;
CPU.FPSCR._u32[1] = CPU.GPR[ra]._u32[1] & 0x00003F07;
CPU.FPSCR._u32[0] = CPU.GPR[ra]._u32[0] & 0x00000F07;
}
void DFTSV(u32 rt, u32 ra, s32 i7)
{
@ -1011,10 +1384,17 @@ private:
}
void FCEQ(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._u32[0] = CPU.GPR[ra]._f[0] == CPU.GPR[rb]._f[0] ? 0xffffffff : 0;
CPU.GPR[rt]._u32[1] = CPU.GPR[ra]._f[1] == CPU.GPR[rb]._f[1] ? 0xffffffff : 0;
CPU.GPR[rt]._u32[2] = CPU.GPR[ra]._f[2] == CPU.GPR[rb]._f[2] ? 0xffffffff : 0;
CPU.GPR[rt]._u32[3] = CPU.GPR[ra]._f[3] == CPU.GPR[rb]._f[3] ? 0xffffffff : 0;
for (int i = 0; i < 4; i++)
{
const u32 a = CPU.GPR[ra]._u32[i];
const u32 b = CPU.GPR[rb]._u32[i];
const u32 abs_a = a & 0x7FFFFFFF;
const u32 abs_b = b & 0x7FFFFFFF;
const bool a_zero = (abs_a < 0x00800000);
const bool b_zero = (abs_b < 0x00800000);
const bool pass = a == b || (a_zero && b_zero);
CPU.GPR[rt]._u32[i] = pass ? 0xFFFFFFFF : 0;
}
}
void DFCEQ(u32 rt, u32 ra, u32 rb)
{
@ -1047,10 +1427,17 @@ private:
}
void FCMEQ(u32 rt, u32 ra, u32 rb)
{
CPU.GPR[rt]._u32[0] = fabs(CPU.GPR[ra]._f[0]) == fabs(CPU.GPR[rb]._f[0]) ? 0xffffffff : 0;
CPU.GPR[rt]._u32[1] = fabs(CPU.GPR[ra]._f[1]) == fabs(CPU.GPR[rb]._f[1]) ? 0xffffffff : 0;
CPU.GPR[rt]._u32[2] = fabs(CPU.GPR[ra]._f[2]) == fabs(CPU.GPR[rb]._f[2]) ? 0xffffffff : 0;
CPU.GPR[rt]._u32[3] = fabs(CPU.GPR[ra]._f[3]) == fabs(CPU.GPR[rb]._f[3]) ? 0xffffffff : 0;
for (int i = 0; i < 4; i++)
{
const u32 a = CPU.GPR[ra]._u32[i];
const u32 b = CPU.GPR[rb]._u32[i];
const u32 abs_a = a & 0x7FFFFFFF;
const u32 abs_b = b & 0x7FFFFFFF;
const bool a_zero = (abs_a < 0x00800000);
const bool b_zero = (abs_b < 0x00800000);
const bool pass = abs_a == abs_b || (a_zero && b_zero);
CPU.GPR[rt]._u32[i] = pass ? 0xFFFFFFFF : 0;
}
}
void DFCMEQ(u32 rt, u32 ra, u32 rb)
{
@ -1084,53 +1471,54 @@ private:
//0 - 9
void CFLTS(u32 rt, u32 ra, s32 i8)
{
const u32 scale = 173 - (i8 & 0xff); //unsigned immediate
const int scale = 173 - (i8 & 0xff); //unsigned immediate
for (int i = 0; i < 4; i++)
{
u32 exp = ((CPU.GPR[ra]._u32[i] >> 23) & 0xff) + scale;
if (exp > 255)
exp = 255;
CPU.GPR[rt]._u32[i] = (CPU.GPR[ra]._u32[i] & 0x807fffff) | (exp << 23);
if (CPU.GPR[rt]._f[i] > 0x7fffffff)
CPU.GPR[rt]._u32[i] = 0x7fffffff;
else if (CPU.GPR[rt]._f[i] < -pow(2, 31))
CPU.GPR[rt]._u32[i] = 0x80000000;
const float a = CPU.GPR[ra]._f[i];
float scaled;
if ((fexpf(a)-127) + scale >= 32)
scaled = copysignf(4294967296.0f, a);
else
CPU.GPR[rt]._s32[i] = (s32)CPU.GPR[rt]._f[i]; //trunc
scaled = ldexpf(a, scale);
s32 result;
if (scaled >= 2147483648.0f)
result = 0x7FFFFFFF;
else if (scaled < -2147483648.0f)
result = 0x80000000;
else
result = (s32)scaled;
CPU.GPR[rt]._s32[i] = result;
}
}
void CFLTU(u32 rt, u32 ra, s32 i8)
{
const u32 scale = 173 - (i8 & 0xff); //unsigned immediate
const int scale = 173 - (i8 & 0xff); //unsigned immediate
for (int i = 0; i < 4; i++)
{
u32 exp = ((CPU.GPR[ra]._u32[i] >> 23) & 0xff) + scale;
if (exp > 255)
exp = 255;
if (CPU.GPR[ra]._u32[i] & 0x80000000) //if negative, result = 0
CPU.GPR[rt]._u32[i] = 0;
const float a = CPU.GPR[ra]._f[i];
float scaled;
if ((fexpf(a)-127) + scale >= 32)
scaled = copysignf(4294967296.0f, a);
else
{
CPU.GPR[rt]._u32[i] = (CPU.GPR[ra]._u32[i] & 0x807fffff) | (exp << 23);
if (CPU.GPR[rt]._f[i] > 0xffffffff) //if big, result = max
CPU.GPR[rt]._u32[i] = 0xffffffff;
else
CPU.GPR[rt]._u32[i] = (u32)floor(CPU.GPR[rt]._f[i]);
}
scaled = ldexpf(a, scale);
u32 result;
if (scaled >= 4294967296.0f)
result = 0xFFFFFFFF;
else if (scaled < 0.0f)
result = 0;
else
result = (u32)scaled;
CPU.GPR[rt]._u32[i] = result;
}
}
void CSFLT(u32 rt, u32 ra, s32 i8)
{
const u32 scale = 155 - (i8 & 0xff); //unsigned immediate
SetHostRoundingMode(FPSCR_RN_ZERO);
const int scale = 155 - (i8 & 0xff); //unsigned immediate
for (int i = 0; i < 4; i++)
{
CPU.GPR[rt]._f[i] = (float)CPU.GPR[ra]._s32[i];
const s32 a = CPU.GPR[ra]._s32[i];
CPU.GPR[rt]._f[i] = (float)a;
u32 exp = ((CPU.GPR[rt]._u32[i] >> 23) & 0xff) - scale;
@ -1138,14 +1526,21 @@ private:
exp = 0;
CPU.GPR[rt]._u32[i] = (CPU.GPR[rt]._u32[i] & 0x807fffff) | (exp << 23);
if (isdenormal(CPU.GPR[rt]._f[i]) || (CPU.GPR[rt]._f[i] == 0.0f && a != 0))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(i, FPSCR_SUNF | FPSCR_SDIFF);
CPU.GPR[rt]._f[i] = 0.0f;
}
}
}
void CUFLT(u32 rt, u32 ra, s32 i8)
{
const u32 scale = 155 - (i8 & 0xff); //unsigned immediate
SetHostRoundingMode(FPSCR_RN_ZERO);
const int scale = 155 - (i8 & 0xff); //unsigned immediate
for (int i = 0; i < 4; i++)
{
CPU.GPR[rt]._f[i] = (float)CPU.GPR[ra]._u32[i];
const u32 a = CPU.GPR[ra]._u32[i];
CPU.GPR[rt]._f[i] = (float)a;
u32 exp = ((CPU.GPR[rt]._u32[i] >> 23) & 0xff) - scale;
@ -1153,6 +1548,11 @@ private:
exp = 0;
CPU.GPR[rt]._u32[i] = (CPU.GPR[rt]._u32[i] & 0x807fffff) | (exp << 23);
if (isdenormal(CPU.GPR[rt]._f[i]) || (CPU.GPR[rt]._f[i] == 0.0f && a != 0))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(i, FPSCR_SUNF | FPSCR_SDIFF);
CPU.GPR[rt]._f[i] = 0.0f;
}
}
}
@ -1524,28 +1924,151 @@ private:
for (int w = 0; w < 4; w++)
CPU.GPR[rt]._s32[w] = CPU.GPR[ra]._s16[w*2] * CPU.GPR[rb]._s16[w*2] + CPU.GPR[rc]._s32[w];
}
void FNMS(u32 rt, u32 ra, u32 rb, u32 rc)
void FNMS(u32 rt, u32 ra, u32 rb, u32 rc) {FMA(rt, ra, rb, rc, true, true);}
void FMA(u32 rt, u32 ra, u32 rb, u32 rc) {FMA(rt, ra, rb, rc, false, false);}
void FMS(u32 rt, u32 ra, u32 rb, u32 rc) {FMA(rt, ra, rb, rc, false, true);}
void FMA(u32 rt, u32 ra, u32 rb, u32 rc, bool neg, bool sub)
{
SetHostRoundingMode(FPSCR_RN_ZERO);
for (int w = 0; w < 4; w++)
{
CPU.GPR[rt]._f[w] = CPU.GPR[rc]._f[w] - CPU.GPR[ra]._f[w] * CPU.GPR[rb]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f;
}
}
void FMA(u32 rt, u32 ra, u32 rb, u32 rc)
{
for (int w = 0; w < 4; w++)
{
CPU.GPR[rt]._f[w] = CPU.GPR[rc]._f[w] + CPU.GPR[ra]._f[w] * CPU.GPR[rb]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f;
}
}
void FMS(u32 rt, u32 ra, u32 rb, u32 rc)
{
for (int w = 0; w < 4; w++)
{
CPU.GPR[rt]._f[w] = CPU.GPR[ra]._f[w] * CPU.GPR[rb]._f[w] - CPU.GPR[rc]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f;
float a = CPU.GPR[ra]._f[w];
float b = neg ? -CPU.GPR[rb]._f[w] : CPU.GPR[rb]._f[w];
float c = (neg != sub) ? -CPU.GPR[rc]._f[w] : CPU.GPR[rc]._f[w];
if (isdenormal(a))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
a = 0.0f;
}
if (isdenormal(b))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
b = 0.0f;
}
if (isdenormal(c))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
c = 0.0f;
}
const bool sign = std::signbit(a) ^ std::signbit(b);
float result;
if (isextended(a) || isextended(b))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (a == 0.0f || b == 0.0f)
{
result = c;
}
else if ((fexpf(a)-127) + (fexpf(b)-127) >= 130)
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
result = extended(sign, 0x7FFFFF);
}
else
{
float new_a, new_b;
if (isextended(a))
{
new_a = ldexpf_extended(a, -2);
new_b = b;
}
else
{
new_a = a;
new_b = ldexpf_extended(b, -2);
}
if (fexpf(c) < 3)
{
result = new_a * new_b;
if (c != 0.0f && std::signbit(c) != sign)
{
u32 bits = (u32&)result - 1;
result = (float&)bits;
}
}
else
{
result = fmaf(new_a, new_b, ldexpf_extended(c, -2));
}
if (fabsf(result) >= ldexpf(1.0f, 127))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
result = extended(sign, 0x7FFFFF);
}
else
{
result = ldexpf_extended(result, 2);
}
}
}
else if (isextended(c))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (a == 0.0f || b == 0.0f)
{
result = c;
}
else if ((fexpf(a)-127) + (fexpf(b)-127) < 96)
{
result = c;
if (sign != std::signbit(c))
{
u32 bits = (u32&)result - 1;
result = (float&)bits;
}
}
else
{
result = fmaf(ldexpf(a,-1), ldexpf(b,-1), ldexpf_extended(c,-2));
if (fabsf(result) >= ldexpf(1.0f, 127))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
result = extended(sign, 0x7FFFFF);
}
else
{
result = ldexpf_extended(result, 2);
}
}
}
else
{
feclearexcept(FE_ALL_EXCEPT);
result = fmaf(a, b, c);
if (fetestexcept(FE_OVERFLOW))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (fexpf(a) > fexpf(b))
result = fmaf(ldexpf(a,-2), b, ldexpf(c,-2));
else
result = fmaf(a, ldexpf(b,-2), ldexpf(c,-2));
if (fabsf(result) >= ldexpf(1.0f, 127))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
result = extended(sign, 0x7FFFFF);
}
else
{
result = ldexpf_extended(result, 2);
}
}
else if (fetestexcept(FE_UNDERFLOW))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SUNF | FPSCR_SDIFF);
}
}
if (isdenormal(result))
{
CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SUNF | FPSCR_SDIFF);
result = 0.0f;
}
else if (result == 0.0f)
{
result = +0.0f;
}
CPU.GPR[rt]._f[w] = result;
}
}

10
rpcs3/Emu/Cell/SPUThread.cpp
View File

@ -558,6 +558,8 @@ u32 SPUThread::GetChannelCount(u32 ch)
switch (ch)
{
case SPU_WrSRR0: res = 1; break;
case SPU_RdSRR0: res = 1; break;
case SPU_WrOutMbox: res = SPU.Out_MBox.GetFreeCount(); break;
case SPU_WrOutIntrMbox: res = SPU.Out_IntrMBox.GetFreeCount(); break;
case SPU_RdInMbox: res = SPU.In_MBox.GetCount(); break;
@ -589,6 +591,9 @@ void SPUThread::WriteChannel(u32 ch, const u128& r)
switch (ch)
{
case SPU_WrSRR0:
SRR0 = v & 0x3FFFC; //LSLR & ~3
break;
case SPU_WrOutIntrMbox:
{
if (!group) // if RawSPU
@ -910,6 +915,9 @@ void SPUThread::ReadChannel(u128& r, u32 ch)
switch (ch)
{
case SPU_RdSRR0:
v = SRR0;
break;
case SPU_RdInMbox:
{
while (!SPU.In_MBox.Pop(v) && !Emu.IsStopped())
@ -1223,4 +1231,4 @@ spu_thread::spu_thread(u32 entry, const std::string& name, u32 stack_size, u32 p
thread->SetPrio(prio ? prio : Emu.GetInfo().GetProcParam().primary_prio);
argc = 0;
}
}

103
rpcs3/Emu/Cell/SPUThread.h
View File

@ -1,4 +1,5 @@
#pragma once
#include "Emu/Cell/Common.h"
#include "Emu/Memory/atomic_type.h"
#include "PPCThread.h"
#include "Emu/SysCalls/lv2/sleep_queue_type.h"
@ -167,45 +168,46 @@ struct g_imm_table_struct
extern const g_imm_table_struct g_imm_table;
//Floating point status and control register. Unsure if this is one of the GPRs or SPRs
enum FPSCR_EX
{
//Single-precision exceptions
FPSCR_SOVF = 1 << 2, //Overflow
FPSCR_SUNF = 1 << 1, //Underflow
FPSCR_SDIFF = 1 << 0, //Different (could be IEEE non-compliant)
//Double-precision exceptions
FPSCR_DOVF = 1 << 13, //Overflow
FPSCR_DUNF = 1 << 12, //Underflow
FPSCR_DINX = 1 << 11, //Inexact
FPSCR_DINV = 1 << 10, //Invalid operation
FPSCR_DNAN = 1 << 9, //NaN
FPSCR_DDENORM = 1 << 8, //Denormal
};
//Is 128 bits, but bits 0-19, 24-28, 32-49, 56-60, 64-81, 88-92, 96-115, 120-124 are unused
class FPSCR
class SPU_FPSCR
{
public:
u64 low;
u64 hi;
u32 _u32[4];
FPSCR() {}
SPU_FPSCR() {}
std::string ToString() const
{
return "FPSCR writer not yet implemented"; //fmt::Format("%08x%08x%08x%08x", _u32[3], _u32[2], _u32[1], _u32[0]);
return fmt::Format("%08x%08x%08x%08x", _u32[3], _u32[2], _u32[1], _u32[0]);
}
void Reset()
{
memset(this, 0, sizeof(*this));
}
//slice -> 0 - 1 (4 slices total, only two have rounding)
//0 -> round even
//1 -> round towards zero (truncate)
//2 -> round towards positive inf
//3 -> round towards neg inf
//slice -> 0 - 1 (double-precision slice index)
//NOTE: slices follow u128 indexing, i.e. slice 0 is RIGHT end of register!
//roundTo -> FPSCR_RN_*
void setSliceRounding(u8 slice, u8 roundTo)
{
u64 mask = roundTo;
switch(slice)
{
case 0:
mask = mask << 20;
break;
case 1:
mask = mask << 22;
break;
}
//rounding is located in the low end of the FPSCR
this->low = this->low & mask;
int shift = 8 + 2*slice;
//rounding is located in the left end of the FPSCR
this->_u32[3] = (this->_u32[3] & ~(3 << shift)) | (roundTo << shift);
}
//Slice 0 or 1
u8 checkSliceRounding(u8 slice) const
@ -213,10 +215,10 @@ public:
switch(slice)
{
case 0:
return this->low >> 20 & 0x3;
return this->_u32[3] >> 8 & 0x3;
case 1:
return this->low >> 22 & 0x3;
return this->_u32[3] >> 10 & 0x3;
default:
throw fmt::Format("Unexpected slice value in FPSCR::checkSliceRounding(): %d", slice);
@ -224,34 +226,28 @@ public:
}
}
//Single Precision Exception Flags (all 3 slices)
//Single-precision exception flags (all 4 slices)
//slice -> slice number (0-3)
//exception: 1 -> Overflow 2 -> Underflow 4-> Diff (could be IE^3 non compliant)
void setSinglePrecisionExceptionFlags(u8 slice, u8 exception)
//exception: FPSCR_S* bitmask
void setSinglePrecisionExceptionFlags(u8 slice, u32 exceptions)
{
u64 mask = exception;
switch(slice)
{
case 0:
mask = mask << 29;
this->low = this->low & mask;
break;
case 1:
mask = mask << 61;
this->low = this->low & mask;
break;
case 2:
mask = mask << 29;
this->hi = this->hi & mask;
break;
case 3:
mask = mask << 61;
this->hi = this->hi & mask;
break;
}
_u32[slice] |= exceptions;
}
//Single-precision divide-by-zero flags (all 4 slices)
//slice -> slice number (0-3)
void setDivideByZeroFlag(u8 slice)
{
_u32[0] |= 1 << (8 + slice);
}
//Double-precision exception flags
//slice -> slice number (0-1)
//exception: FPSCR_D* bitmask
void setDoublePrecisionExceptionFlags(u8 slice, u32 exceptions)
{
_u32[1+slice] |= exceptions;
}
};
union SPU_SNRConfig_hdr
@ -277,7 +273,8 @@ class SPUThread : public PPCThread
{
public:
u128 GPR[128]; // General-Purpose Registers
//FPSCR FPSCR;
SPU_FPSCR FPSCR;
u32 SRR0;
SPU_SNRConfig_hdr cfg; // Signal Notification Registers Configuration (OR-mode enabled: 0x1 for SNR1, 0x2 for SNR2)
u64 R_ADDR; // reservation address
@ -630,4 +627,4 @@ public:
return *this;
}
};
};

2
rpcs3/Emu/RSX/GL/GLFragmentProgram.cpp
View File

@ -350,7 +350,7 @@ std::string GLFragmentDecompilerThread::BuildCode()
p += param.Format();
}
return std::string("#version 330\n"
return std::string("#version 420\n"
"\n"
+ p + "\n"
"void main()\n{\n" + main + "}\n");

6
rpcs3/Emu/RSX/GL/GLGSRender.cpp
View File

@ -734,7 +734,7 @@ void DrawCursorObj::Draw()
void DrawCursorObj::InitializeShaders()
{
m_vp.shader =
"#version 330\n"
"#version 420\n"
"\n"
"uniform vec4 in_pos;\n"
"uniform vec2 in_tc;\n"
@ -747,10 +747,10 @@ void DrawCursorObj::InitializeShaders()
"}\n";
m_fp.shader =
"#version 330\n"
"#version 420\n"
"\n"
"in vec2 tc;\n"
"uniform sampler2D tex0;\n"
"layout (binding = 0) uniform sampler2D tex0;\n"
"layout (location = 0) out vec4 res;\n"
"\n"
"void main()\n"

2
rpcs3/Emu/RSX/GL/GLVertexProgram.cpp
View File

@ -498,7 +498,7 @@ std::string GLVertexDecompilerThread::BuildCode()
}
static const std::string& prot =
"#version 330\n"
"#version 420\n"
"\n"
"uniform mat4 scaleOffsetMat = mat4(1.0);\n"
"%s\n"