ShuriZma
/
rpcs3
mirror of https://github.com/RPCS3/rpcs3.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #969 from achurch/spu-interpreter-fixes 

Fix SPU interpreter to match real hardware
This commit is contained in:
B1ackDaemon 2015-01-18 20:54:12 +02:00
parent 9709f7f821 e64f6a1db3
commit d9dd3a7eda
5 changed files with 726 additions and 195 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,
};

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

@ -1,4 +1,5 @@
#pragma once #pragma once
#include "Emu/Cell/Common.h"
#include "Emu/Cell/PPCThread.h" #include "Emu/Cell/PPCThread.h"
#include "Emu/Memory/vm.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, 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_SIGN = 0x8000000000000000ULL;
static const u64 DOUBLE_EXP = 0x7FF0000000000000ULL; static const u64 DOUBLE_EXP = 0x7FF0000000000000ULL;
static const u64 DOUBLE_FRAC = 0x000FFFFFFFFFFFFFULL; static const u64 DOUBLE_FRAC = 0x000FFFFFFFFFFFFFULL;

755
rpcs3/Emu/Cell/SPUInterpreter.h
View File

@ -1,5 +1,26 @@
#pragma once #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 UNIMPLEMENTED() UNK(__FUNCTION__)
#define MEM_AND_REG_HASH() \ #define MEM_AND_REG_HASH() \
@ -11,6 +32,54 @@
#define LOG5_OPCODE(...) /// #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 class SPUInterpreter : public SPUOpcodes
{ {
private: private:
@ -47,7 +116,7 @@ private:
} }
void MFSPR(u32 rt, u32 sa) void MFSPR(u32 rt, u32 sa)
{ {
UNIMPLEMENTED(); // not used CPU.GPR[rt].clear(); // All SPRs read as zero.
} }
void RDCH(u32 rt, u32 ra) void RDCH(u32 rt, u32 ra)
{ {
@ -243,7 +312,7 @@ private:
} }
void MTSPR(u32 rt, u32 sa) void MTSPR(u32 rt, u32 sa)
{ {
UNIMPLEMENTED(); // not used // SPR writes are ignored.
} }
void WRCH(u32 ra, u32 rt) void WRCH(u32 ra, u32 rt)
{ {
@ -431,16 +500,41 @@ private:
} }
void FREST(u32 rt, u32 ra) 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++) 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) void FRSQEST(u32 rt, u32 ra)
{ {
//const __u32x4 FloatAbsMask = {0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff}; SetHostRoundingMode(FPSCR_RN_ZERO);
//CPU.GPR[rt]._m128 = _mm_rsqrt_ps(_mm_and_ps(CPU.GPR[ra]._m128, FloatAbsMask.m128));
for (int i = 0; i < 4; i++) 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) void LQX(u32 rt, u32 ra, u32 rb)
{ {
@ -828,37 +922,186 @@ private:
} }
void FCGT(u32 rt, u32 ra, u32 rb) 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; for (int i = 0; i < 4; i++)
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; const u32 a = CPU.GPR[ra]._u32[i];
CPU.GPR[rt]._u32[3] = CPU.GPR[ra]._f[3] > CPU.GPR[rb]._f[3] ? 0xffffffff : 0; 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) void DFCGT(u32 rt, u32 ra, u32 rb)
{ {
UNIMPLEMENTED(); // cannot be used 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++) for (int w = 0; w < 4; w++)
{ {
CPU.GPR[rt]._f[w] = CPU.GPR[ra]._f[w] + CPU.GPR[rb]._f[w]; const float a = CPU.GPR[ra]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f; const float b = sub ? -CPU.GPR[rb]._f[w] : CPU.GPR[rb]._f[w];
} float result;
} if (isdenormal(a))
void FS(u32 rt, u32 ra, u32 rb)
{ {
for (int w = 0; w < 4; w++) CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (b == 0.0f || isdenormal(b))
result = +0.0f;
else
result = b;
}
else if (isdenormal(b))
{ {
CPU.GPR[rt]._f[w] = CPU.GPR[ra]._f[w] - CPU.GPR[rb]._f[w]; CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f; 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) void FM(u32 rt, u32 ra, u32 rb)
{ {
SetHostRoundingMode(FPSCR_RN_ZERO);
for (int w = 0; w < 4; w++) for (int w = 0; w < 4; w++)
{ {
CPU.GPR[rt]._f[w] = CPU.GPR[ra]._f[w] * CPU.GPR[rb]._f[w]; const float a = CPU.GPR[ra]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f; 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) void CLGTH(u32 rt, u32 ra, u32 rb)
@ -873,30 +1116,84 @@ private:
} }
void FCMGT(u32 rt, u32 ra, u32 rb) 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; for (int i = 0; i < 4; i++)
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; const u32 a = CPU.GPR[ra]._u32[i];
CPU.GPR[rt]._u32[3] = fabs(CPU.GPR[ra]._f[3]) > fabs(CPU.GPR[rb]._f[3]) ? 0xffffffff : 0; 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) void DFCMGT(u32 rt, u32 ra, u32 rb)
{ {
UNIMPLEMENTED(); // cannot be used 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]; for (int i = 0; i < 2; i++)
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]; double a = CPU.GPR[ra]._d[i];
CPU.GPR[rt]._d[1] = CPU.GPR[ra]._d[1] - CPU.GPR[rb]._d[1]; double b = CPU.GPR[rb]._d[i];
} if (isdenormal(a))
void DFM(u32 rt, u32 ra, u32 rb)
{ {
CPU.GPR[rt]._d[0] = CPU.GPR[ra]._d[0] * CPU.GPR[rb]._d[0]; CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
CPU.GPR[rt]._d[1] = CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1]; 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) void CLGTB(u32 rt, u32 ra, u32 rb)
{ {
for (int b = 0; b < 16; b++) for (int b = 0; b < 16; b++)
@ -910,26 +1207,64 @@ private:
CPU.Stop(); 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]; for (int i = 0; i < 2; i++)
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]; double a = CPU.GPR[ra]._d[i];
CPU.GPR[rt]._d[1] = CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1] - CPU.GPR[rt]._d[1]; double b = CPU.GPR[rb]._d[i];
} double c = CPU.GPR[rt]._d[i];
void DFNMS(u32 rt, u32 ra, u32 rb) if (isdenormal(a))
{ {
CPU.GPR[rt]._d[0] -= CPU.GPR[ra]._d[0] * CPU.GPR[rb]._d[0]; CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
CPU.GPR[rt]._d[1] -= CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1]; a = copysign(0.0, a);
} }
void DFNMA(u32 rt, u32 ra, u32 rb) if (isdenormal(b))
{ {
CPU.GPR[rt]._d[0] = -(CPU.GPR[ra]._d[0] * CPU.GPR[rb]._d[0] + CPU.GPR[rt]._d[0]); CPU.FPSCR.setDoublePrecisionExceptionFlags(i, FPSCR_DDENORM);
CPU.GPR[rt]._d[1] = -(CPU.GPR[ra]._d[1] * CPU.GPR[rb]._d[1] + CPU.GPR[rt]._d[1]); 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) void CEQ(u32 rt, u32 ra, u32 rb)
{ {
for (int w = 0; w < 4; w++) for (int w = 0; w < 4; w++)
@ -981,29 +1316,67 @@ private:
void FSCRRD(u32 rt) void FSCRRD(u32 rt)
{ {
// TODO (rarely used) CPU.GPR[rt]._u32[3] = CPU.FPSCR._u32[3];
CPU.GPR[rt].clear(); 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) void FESD(u32 rt, u32 ra)
{ {
CPU.GPR[rt]._d[0] = (double)CPU.GPR[ra]._f[1]; for (int i = 0; i < 2; i++)
CPU.GPR[rt]._d[1] = (double)CPU.GPR[ra]._f[3]; {
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) void FRDS(u32 rt, u32 ra)
{ {
CPU.GPR[rt]._f[1] = (float)CPU.GPR[ra]._d[0]; for (int i = 0; i < 2; i++)
CPU.GPR[rt]._u32[0] = 0x00000000; {
CPU.GPR[rt]._f[3] = (float)CPU.GPR[ra]._d[1]; SetHostRoundingMode(CPU.FPSCR.checkSliceRounding(i));
CPU.GPR[rt]._u32[2] = 0x00000000; 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) void FSCRWR(u32 rt, u32 ra)
{ {
// TODO (rarely used) CPU.FPSCR._u32[3] = CPU.GPR[ra]._u32[3] & 0x00000F07;
if (CPU.GPR[ra]._u64[0] || CPU.GPR[ra]._u64[1]) CPU.FPSCR._u32[2] = CPU.GPR[ra]._u32[2] & 0x00003F07;
{ CPU.FPSCR._u32[1] = CPU.GPR[ra]._u32[1] & 0x00003F07;
LOG_ERROR(SPU, "FSCRWR(%d,%d): value = %s", rt, ra, CPU.GPR[ra].to_hex().c_str()); CPU.FPSCR._u32[0] = CPU.GPR[ra]._u32[0] & 0x00000F07;
UNIMPLEMENTED();
}
} }
void DFTSV(u32 rt, u32 ra, s32 i7) void DFTSV(u32 rt, u32 ra, s32 i7)
{ {
@ -1011,10 +1384,17 @@ private:
} }
void FCEQ(u32 rt, u32 ra, u32 rb) 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; for (int i = 0; i < 4; i++)
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; const u32 a = CPU.GPR[ra]._u32[i];
CPU.GPR[rt]._u32[3] = CPU.GPR[ra]._f[3] == CPU.GPR[rb]._f[3] ? 0xffffffff : 0; 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) void DFCEQ(u32 rt, u32 ra, u32 rb)
{ {
@ -1047,10 +1427,17 @@ private:
} }
void FCMEQ(u32 rt, u32 ra, u32 rb) 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; for (int i = 0; i < 4; i++)
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; const u32 a = CPU.GPR[ra]._u32[i];
CPU.GPR[rt]._u32[3] = fabs(CPU.GPR[ra]._f[3]) == fabs(CPU.GPR[rb]._f[3]) ? 0xffffffff : 0; 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) void DFCMEQ(u32 rt, u32 ra, u32 rb)
{ {
@ -1084,53 +1471,54 @@ private:
//0 - 9 //0 - 9
void CFLTS(u32 rt, u32 ra, s32 i8) 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++) for (int i = 0; i < 4; i++)
{ {
u32 exp = ((CPU.GPR[ra]._u32[i] >> 23) & 0xff) + scale; const float a = CPU.GPR[ra]._f[i];
float scaled;
if (exp > 255) if ((fexpf(a)-127) + scale >= 32)
exp = 255; scaled = copysignf(4294967296.0f, a);
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;
else 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) 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++) for (int i = 0; i < 4; i++)
{ {
u32 exp = ((CPU.GPR[ra]._u32[i] >> 23) & 0xff) + scale; const float a = CPU.GPR[ra]._f[i];
float scaled;
if (exp > 255) if ((fexpf(a)-127) + scale >= 32)
exp = 255; scaled = copysignf(4294967296.0f, a);
if (CPU.GPR[ra]._u32[i] & 0x80000000) //if negative, result = 0
CPU.GPR[rt]._u32[i] = 0;
else else
{ scaled = ldexpf(a, scale);
CPU.GPR[rt]._u32[i] = (CPU.GPR[ra]._u32[i] & 0x807fffff) | (exp << 23); u32 result;
if (scaled >= 4294967296.0f)
if (CPU.GPR[rt]._f[i] > 0xffffffff) //if big, result = max result = 0xFFFFFFFF;
CPU.GPR[rt]._u32[i] = 0xffffffff; else if (scaled < 0.0f)
result = 0;
else else
CPU.GPR[rt]._u32[i] = (u32)floor(CPU.GPR[rt]._f[i]); result = (u32)scaled;
} CPU.GPR[rt]._u32[i] = result;
} }
} }
void CSFLT(u32 rt, u32 ra, s32 i8) 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++) 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; u32 exp = ((CPU.GPR[rt]._u32[i] >> 23) & 0xff) - scale;
@ -1138,14 +1526,21 @@ private:
exp = 0; exp = 0;
CPU.GPR[rt]._u32[i] = (CPU.GPR[rt]._u32[i] & 0x807fffff) | (exp << 23); 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) 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++) 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; u32 exp = ((CPU.GPR[rt]._u32[i] >> 23) & 0xff) - scale;
@ -1153,6 +1548,11 @@ private:
exp = 0; exp = 0;
CPU.GPR[rt]._u32[i] = (CPU.GPR[rt]._u32[i] & 0x807fffff) | (exp << 23); 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++) 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]; 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++) 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]; float a = CPU.GPR[ra]._f[w];
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f; 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;
} }
} }
void FMA(u32 rt, u32 ra, u32 rb, u32 rc) else
{ {
for (int w = 0; w < 4; w++) result = fmaf(new_a, new_b, ldexpf_extended(c, -2));
}
if (fabsf(result) >= ldexpf(1.0f, 127))
{ {
CPU.GPR[rt]._f[w] = CPU.GPR[rc]._f[w] + CPU.GPR[ra]._f[w] * CPU.GPR[rb]._f[w]; CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SOVF);
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f; result = extended(sign, 0x7FFFFF);
}
else
{
result = ldexpf_extended(result, 2);
} }
} }
void FMS(u32 rt, u32 ra, u32 rb, u32 rc) }
else if (isextended(c))
{ {
for (int w = 0; w < 4; w++) CPU.FPSCR.setSinglePrecisionExceptionFlags(w, FPSCR_SDIFF);
if (a == 0.0f || b == 0.0f)
{ {
CPU.GPR[rt]._f[w] = CPU.GPR[ra]._f[w] * CPU.GPR[rb]._f[w] - CPU.GPR[rc]._f[w]; result = c;
//if (CPU.GPR[rt]._f[w] == -0.0f) CPU.GPR[rt]._f[w] = 0.0f; }
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;
} }
} }

8
rpcs3/Emu/Cell/SPUThread.cpp
View File

@ -558,6 +558,8 @@ u32 SPUThread::GetChannelCount(u32 ch)
switch (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_WrOutMbox: res = SPU.Out_MBox.GetFreeCount(); break;
case SPU_WrOutIntrMbox: res = SPU.Out_IntrMBox.GetFreeCount(); break; case SPU_WrOutIntrMbox: res = SPU.Out_IntrMBox.GetFreeCount(); break;
case SPU_RdInMbox: res = SPU.In_MBox.GetCount(); break; case SPU_RdInMbox: res = SPU.In_MBox.GetCount(); break;
@ -589,6 +591,9 @@ void SPUThread::WriteChannel(u32 ch, const u128& r)
switch (ch) switch (ch)
{ {
case SPU_WrSRR0:
SRR0 = v & 0x3FFFC; //LSLR & ~3
break;
case SPU_WrOutIntrMbox: case SPU_WrOutIntrMbox:
{ {
if (!group) // if RawSPU if (!group) // if RawSPU
@ -910,6 +915,9 @@ void SPUThread::ReadChannel(u128& r, u32 ch)
switch (ch) switch (ch)
{ {
case SPU_RdSRR0:
v = SRR0;
break;
case SPU_RdInMbox: case SPU_RdInMbox:
{ {
while (!SPU.In_MBox.Pop(v) && !Emu.IsStopped()) while (!SPU.In_MBox.Pop(v) && !Emu.IsStopped())

95
rpcs3/Emu/Cell/SPUThread.h
View File

@ -1,4 +1,5 @@
#pragma once #pragma once
#include "Emu/Cell/Common.h"
#include "Emu/Memory/atomic_type.h" #include "Emu/Memory/atomic_type.h"
#include "PPCThread.h" #include "PPCThread.h"
#include "Emu/SysCalls/lv2/sleep_queue_type.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; 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 //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: public:
u64 low; u32 _u32[4];
u64 hi;
FPSCR() {} SPU_FPSCR() {}
std::string ToString() const 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() void Reset()
{ {
memset(this, 0, sizeof(*this)); memset(this, 0, sizeof(*this));
} }
//slice -> 0 - 1 (4 slices total, only two have rounding) //slice -> 0 - 1 (double-precision slice index)
//0 -> round even //NOTE: slices follow u128 indexing, i.e. slice 0 is RIGHT end of register!
//1 -> round towards zero (truncate) //roundTo -> FPSCR_RN_*
//2 -> round towards positive inf
//3 -> round towards neg inf
void setSliceRounding(u8 slice, u8 roundTo) void setSliceRounding(u8 slice, u8 roundTo)
{ {
u64 mask = roundTo; int shift = 8 + 2*slice;
switch(slice) //rounding is located in the left end of the FPSCR
{ this->_u32[3] = (this->_u32[3] & ~(3 << shift)) | (roundTo << shift);
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;
} }
//Slice 0 or 1 //Slice 0 or 1
u8 checkSliceRounding(u8 slice) const u8 checkSliceRounding(u8 slice) const
@ -213,10 +215,10 @@ public:
switch(slice) switch(slice)
{ {
case 0: case 0:
return this->low >> 20 & 0x3; return this->_u32[3] >> 8 & 0x3;
case 1: case 1:
return this->low >> 22 & 0x3; return this->_u32[3] >> 10 & 0x3;
default: default:
throw fmt::Format("Unexpected slice value in FPSCR::checkSliceRounding(): %d", slice); 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) //slice -> slice number (0-3)
//exception: 1 -> Overflow 2 -> Underflow 4-> Diff (could be IE^3 non compliant) //exception: FPSCR_S* bitmask
void setSinglePrecisionExceptionFlags(u8 slice, u8 exception) void setSinglePrecisionExceptionFlags(u8 slice, u32 exceptions)
{ {
u64 mask = exception; _u32[slice] |= exceptions;
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;
} }
//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 union SPU_SNRConfig_hdr
@ -277,7 +273,8 @@ class SPUThread : public PPCThread
{ {
public: public:
u128 GPR[128]; // General-Purpose Registers 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) SPU_SNRConfig_hdr cfg; // Signal Notification Registers Configuration (OR-mode enabled: 0x1 for SNR1, 0x2 for SNR2)
u64 R_ADDR; // reservation address u64 R_ADDR; // reservation address