mirror/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-04-16 08:42:32 +00:00
Code Issues Packages Projects Releases Wiki Activity

Handle NaNs correctly in vector FP instructions.

Browse Source
This commit is contained in:
Andrew Church 2015-01-18 07:05:18 +09:00
parent 2e2dbf6c33
commit 10c706d9a5
1 changed files with 148 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

140
rpcs3/Emu/Cell/PPUInterpreter.h
View File

@ -247,7 +247,16 @@ private:
{
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = CPU.VPR[va]._f[w] + CPU.VPR[vb]._f[w];
const float a = CPU.VPR[va]._f[w];
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(a))
CPU.VPR[vd]._f[w] = SilenceNaN(a);
else if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else if (std::isinf(a) && std::isinf(b) && a != b)
CPU.VPR[vd]._f[w] = (float)FPR_NAN;
else
CPU.VPR[vd]._f[w] = a + b;
}
}
void VADDSBS(u32 vd, u32 va, u32 vb) //nf
@ -759,8 +768,14 @@ private:
for (uint w = 0; w < 4; w++)
{
double result = (double)CPU.VPR[vb]._f[w] * nScale;
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
{
CPU.VPR[vd]._s32[w] = 0;
}
else
{
double result = (double)b * nScale;
if (result > 0x7fffffff)
{
CPU.VPR[vd]._s32[w] = (int)0x7fffffff;
@ -775,15 +790,22 @@ private:
CPU.VPR[vd]._s32[w] = (int)result;
}
}
}
void VCTUXS(u32 vd, u32 uimm5, u32 vb)
{
u32 nScale = 1 << uimm5;
for (uint w = 0; w < 4; w++)
{
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
{
CPU.VPR[vd]._s32[w] = 0;
}
else
{
// C rounding = Round towards 0
double result = (double)CPU.VPR[vb]._f[w] * nScale;
double result = (double)b * nScale;
if (result > 0xffffffffu)
{
CPU.VPR[vd]._u32[w] = 0xffffffffu;
@ -798,6 +820,7 @@ private:
CPU.VPR[vd]._u32[w] = (u32)result;
}
}
}
void VEXPTEFP(u32 vd, u32 vb)
{
// vd = 2^x
@ -805,7 +828,11 @@ private:
// and between different executions on the same implementation.
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = powf(2.0f, CPU.VPR[vb]._f[w]);
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = powf(2.0f, b);
}
}
void VLOGEFP(u32 vd, u32 vb)
@ -814,21 +841,48 @@ private:
// and between different executions on the same implementation.
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = log2f(CPU.VPR[vb]._f[w]);
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = log2f(b);
}
}
void VMADDFP(u32 vd, u32 va, u32 vc, u32 vb)
{
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = CPU.VPR[va]._f[w] * CPU.VPR[vc]._f[w] + CPU.VPR[vb]._f[w];
const float a = CPU.VPR[va]._f[w];
const float b = CPU.VPR[vb]._f[w];
const float c = CPU.VPR[vc]._f[w];
if (std::isnan(a))
CPU.VPR[vd]._f[w] = SilenceNaN(a);
else if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else if (std::isnan(c))
CPU.VPR[vd]._f[w] = SilenceNaN(c);
else if ((std::isinf(a) && c == 0) || (a == 0 && std::isinf(c)))
CPU.VPR[vd]._f[w] = (float)FPR_NAN;
else
{
CPU.VPR[vd]._f[w] = a * c + b;
if (std::isnan(CPU.VPR[vd]._f[w]))
CPU.VPR[vd]._f[w] = (float)FPR_NAN;
}
}
}
void VMAXFP(u32 vd, u32 va, u32 vb)
{
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = std::max(CPU.VPR[va]._f[w], CPU.VPR[vb]._f[w]);
const float a = CPU.VPR[va]._f[w];
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(a))
CPU.VPR[vd]._f[w] = SilenceNaN(a);
else if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = std::max(a, b);
}
}
void VMAXSB(u32 vd, u32 va, u32 vb) //nf
@ -915,7 +969,14 @@ private:
{
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = std::min(CPU.VPR[va]._f[w], CPU.VPR[vb]._f[w]);
const float a = CPU.VPR[va]._f[w];
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(a))
CPU.VPR[vd]._f[w] = SilenceNaN(a);
else if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = std::min(a, b);
}
}
void VMINSB(u32 vd, u32 va, u32 vb) //nf
@ -1202,7 +1263,23 @@ private:
{
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = -(CPU.VPR[va]._f[w] * CPU.VPR[vc]._f[w] - CPU.VPR[vb]._f[w]);
const float a = CPU.VPR[va]._f[w];
const float b = CPU.VPR[vb]._f[w];
const float c = CPU.VPR[vc]._f[w];
if (std::isnan(a))
CPU.VPR[vd]._f[w] = SilenceNaN(a);
else if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else if (std::isnan(c))
CPU.VPR[vd]._f[w] = SilenceNaN(c);
else if ((std::isinf(a) && c == 0) || (a == 0 && std::isinf(c)))
CPU.VPR[vd]._f[w] = (float)FPR_NAN;
else
{
CPU.VPR[vd]._f[w] = -(a * c - b);
if (std::isnan(CPU.VPR[vd]._f[w]))
CPU.VPR[vd]._f[w] = (float)FPR_NAN;
}
}
}
void VNOR(u32 vd, u32 va, u32 vb)
@ -1477,13 +1554,21 @@ private:
{
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = 1.0f / CPU.VPR[vb]._f[w];
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = 1.0f / b;
}
}
void VRFIM(u32 vd, u32 vb)
{
for (uint w = 0; w < 4; w++)
{
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = floorf(CPU.VPR[vb]._f[w]);
}
}
@ -1491,6 +1576,10 @@ private:
{
for (uint w = 0; w < 4; w++)
{
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = nearbyintf(CPU.VPR[vb]._f[w]);
}
}
@ -1498,6 +1587,10 @@ private:
{
for (uint w = 0; w < 4; w++)
{
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = ceilf(CPU.VPR[vb]._f[w]);
}
}
@ -1505,6 +1598,10 @@ private:
{
for (uint w = 0; w < 4; w++)
{
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else
CPU.VPR[vd]._f[w] = truncf(CPU.VPR[vb]._f[w]);
}
}
@ -1536,7 +1633,13 @@ private:
for (uint w = 0; w < 4; w++)
{
//TODO: accurate div
CPU.VPR[vd]._f[w] = 1.0f / sqrtf(CPU.VPR[vb]._f[w]);
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else if (b < 0)
CPU.VPR[vd]._f[w] = (float)FPR_NAN;
else
CPU.VPR[vd]._f[w] = 1.0f / sqrtf(b);
}
}
void VSEL(u32 vd, u32 va, u32 vb, u32 vc)
@ -1729,7 +1832,16 @@ private:
{
for (uint w = 0; w < 4; w++)
{
CPU.VPR[vd]._f[w] = CPU.VPR[va]._f[w] - CPU.VPR[vb]._f[w];
const float a = CPU.VPR[va]._f[w];
const float b = CPU.VPR[vb]._f[w];
if (std::isnan(a))
CPU.VPR[vd]._f[w] = SilenceNaN(a);
else if (std::isnan(b))
CPU.VPR[vd]._f[w] = SilenceNaN(b);
else if (std::isinf(a) && std::isinf(b) && a == b)
CPU.VPR[vd]._f[w] = (float)FPR_NAN;
else
CPU.VPR[vd]._f[w] = a - b;
}
}
void VSUBSBS(u32 vd, u32 va, u32 vb) //nf