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

Core: Cleanup VFPU float bit handling.

Browse source 
Just to use a common union.
This commit is contained in:
Unknown W. Brackets 2021-04-24 15:49:22 -07:00
parent 86585e9551
commit e9076c90bb
1 changed files with 36 additions and 46 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

82
Core/MIPS/MIPSVFPUUtils.cpp
View file

@ -29,6 +29,11 @@
#define V(i) (currentMIPS->v[voffset[i]]) #define V(i) (currentMIPS->v[voffset[i]])
#define VI(i) (currentMIPS->vi[voffset[i]]) #define VI(i) (currentMIPS->vi[voffset[i]])
union float2int {
uint32_t i;
float f;
};
void GetVectorRegs(u8 regs[4], VectorSize N, int vectorReg) { void GetVectorRegs(u8 regs[4], VectorSize N, int vectorReg) {
int mtx = (vectorReg >> 2) & 7; int mtx = (vectorReg >> 2) & 7;
int col = vectorReg & 3; int col = vectorReg & 3;
@ -610,10 +615,7 @@ bool GetVFPUCtrlMask(int reg, u32 *mask) {
float Float16ToFloat32(unsigned short l) float Float16ToFloat32(unsigned short l)
{ {
union float2int { float2int f2i;
unsigned int i;
float f;
} float2int;
unsigned short float16 = l; unsigned short float16 = l;
unsigned int sign = (float16 >> VFPU_SH_FLOAT16_SIGN) & VFPU_MASK_FLOAT16_SIGN; unsigned int sign = (float16 >> VFPU_SH_FLOAT16_SIGN) & VFPU_MASK_FLOAT16_SIGN;
@ -623,10 +625,10 @@ float Float16ToFloat32(unsigned short l)
float f; float f;
if (exponent == VFPU_FLOAT16_EXP_MAX) if (exponent == VFPU_FLOAT16_EXP_MAX)
{ {
float2int.i = sign << 31; f2i.i = sign << 31;
float2int.i |= 255 << 23; f2i.i |= 255 << 23;
float2int.i |= fraction; f2i.i |= fraction;
f = float2int.f; f = f2i.f;
} }
else if (exponent == 0 && fraction == 0) else if (exponent == 0 && fraction == 0)
{ {
@ -647,10 +649,10 @@ float Float16ToFloat32(unsigned short l)
} }
/* Convert to 32-bit single-precision IEEE754. */ /* Convert to 32-bit single-precision IEEE754. */
float2int.i = sign << 31; f2i.i = sign << 31;
float2int.i |= (exponent + 112) << 23; f2i.i |= (exponent + 112) << 23;
float2int.i |= fraction << 13; f2i.i |= fraction << 13;
f=float2int.f; f=f2i.f;
} }
return f; return f;
} }
@ -674,10 +676,6 @@ static int32_t get_sign(uint32_t x) {
float vfpu_dot(float a[4], float b[4]) { float vfpu_dot(float a[4], float b[4]) {
static const int EXTRA_BITS = 2; static const int EXTRA_BITS = 2;
union float2int {
uint32_t i;
float f;
};
float2int result; float2int result;
float2int src[2]; float2int src[2];
@ -791,31 +789,27 @@ float vfpu_dot(float a[4], float b[4]) {
// TODO: This is still not completely accurate compared to the PSP's vsqrt. // TODO: This is still not completely accurate compared to the PSP's vsqrt.
float vfpu_sqrt(float a) { float vfpu_sqrt(float a) {
union float2int {
uint32_t u;
float f;
};
float2int val; float2int val;
val.f = a; val.f = a;
if ((val.u & 0xff800000) == 0x7f800000) { if ((val.i & 0xff800000) == 0x7f800000) {
if ((val.u & 0x007fffff) != 0) { if ((val.i & 0x007fffff) != 0) {
val.u = 0x7f800001; val.i = 0x7f800001;
} }
return val.f; return val.f;
} }
if ((val.u & 0x7f800000) == 0) { if ((val.i & 0x7f800000) == 0) {
// Kill any sign. // Kill any sign.
val.u = 0; val.i = 0;
return val.f; return val.f;
} }
if (val.u & 0x80000000) { if (val.i & 0x80000000) {
val.u = 0x7f800001; val.i = 0x7f800001;
return val.f; return val.f;
} }
int k = get_exp(val.u); int k = get_exp(val.i);
uint32_t sp = get_mant(val.u); uint32_t sp = get_mant(val.i);
int less_bits = k & 1; int less_bits = k & 1;
k >>= 1; k >>= 1;
@ -826,9 +820,9 @@ float vfpu_sqrt(float a) {
z = (z >> 1) + (uint32_t)(halfsp / z); z = (z >> 1) + (uint32_t)(halfsp / z);
} }
val.u = ((k + 127) << 23) | ((z << less_bits) & 0x007FFFFF); val.i = ((k + 127) << 23) | ((z << less_bits) & 0x007FFFFF);
// The lower two bits never end up set on the PSP, it seems like. // The lower two bits never end up set on the PSP, it seems like.
val.u &= 0xFFFFFFFC; val.i &= 0xFFFFFFFC;
return val.f; return val.f;
} }
@ -842,31 +836,27 @@ static inline uint32_t mant_mul(uint32_t a, uint32_t b) {
} }
float vfpu_rsqrt(float a) { float vfpu_rsqrt(float a) {
union float2int {
uint32_t u;
float f;
};
float2int val; float2int val;
val.f = a; val.f = a;
if (val.u == 0x7f800000) { if (val.i == 0x7f800000) {
return 0.0f; return 0.0f;
} }
if ((val.u & 0x7fffffff) > 0x7f800000) { if ((val.i & 0x7fffffff) > 0x7f800000) {
val.u = (val.u & 0x80000000) | 0x7f800001; val.i = (val.i & 0x80000000) | 0x7f800001;
return val.f; return val.f;
} }
if ((val.u & 0x7f800000) == 0) { if ((val.i & 0x7f800000) == 0) {
val.u = (val.u & 0x80000000) | 0x7f800000; val.i = (val.i & 0x80000000) | 0x7f800000;
return val.f; return val.f;
} }
if (val.u & 0x80000000) { if (val.i & 0x80000000) {
val.u = 0xff800001; val.i = 0xff800001;
return val.f; return val.f;
} }
int k = get_exp(val.u); int k = get_exp(val.i);
uint32_t sp = get_mant(val.u); uint32_t sp = get_mant(val.i);
int less_bits = k & 1; int less_bits = k & 1;
k = -(k >> 1); k = -(k >> 1);
@ -889,8 +879,8 @@ float vfpu_rsqrt(float a) {
z >>= less_bits; z >>= less_bits;
val.u = ((k + 127) << 23) | (z & 0x007FFFFF); val.i = ((k + 127) << 23) | (z & 0x007FFFFF);
val.u &= 0xFFFFFFFC; val.i &= 0xFFFFFFFC;
return val.f; return val.f;
} }