ppsspp/Core/MIPS/MIPSIntVFPU.cpp

1894 lines
42 KiB
C++
Raw Normal View History

2012-11-01 15:19:01 +00:00
// Copyright (c) 2012- PPSSPP Project.
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0 or later versions.
2012-11-01 15:19:01 +00:00
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License 2.0 for more details.
// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
// TODO: Test and maybe fix: https://code.google.com/p/jpcsp/source/detail?r=3082#
2012-11-01 15:19:01 +00:00
#include <cmath>
#include <limits>
2013-09-17 04:48:14 +00:00
#include <algorithm>
2012-11-01 15:19:01 +00:00
#include "math/math_util.h"
#include "Core/Core.h"
#include "Core/Reporting.h"
#include "Core/MemMap.h"
#include "Core/MIPS/MIPS.h"
#include "Core/MIPS/MIPSInt.h"
#include "Core/MIPS/MIPSTables.h"
#include "Core/MIPS/MIPSVFPUUtils.h"
2012-11-01 15:19:01 +00:00
#define R(i) (currentMIPS->r[i])
#define V(i) (currentMIPS->v[voffset[i]])
#define VI(i) (currentMIPS->vi[voffset[i]])
#define FI(i) (currentMIPS->fi[i])
#define FsI(i) (currentMIPS->fs[i])
2012-11-01 15:19:01 +00:00
#define PC (currentMIPS->pc)
#define _RS ((op>>21) & 0x1F)
#define _RT ((op>>16) & 0x1F)
2012-11-01 15:19:01 +00:00
#define _RD ((op>>11) & 0x1F)
#define _FS ((op>>11) & 0x1F)
#define _FT ((op>>16) & 0x1F)
2012-11-01 15:19:01 +00:00
#define _FD ((op>>6 ) & 0x1F)
#define _POS ((op>>6 ) & 0x1F)
#define _SIZE ((op>>11) & 0x1F)
2012-11-01 15:19:01 +00:00
#define HI currentMIPS->hi
#define LO currentMIPS->lo
#ifndef M_LOG2E
2012-11-11 21:38:19 +00:00
#define M_E 2.71828182845904523536f
#define M_LOG2E 1.44269504088896340736f
#define M_LOG10E 0.434294481903251827651f
#define M_LN2 0.693147180559945309417f
#define M_LN10 2.30258509299404568402f
#undef M_PI
2012-11-11 21:38:19 +00:00
#define M_PI 3.14159265358979323846f
#ifndef M_PI_2
2012-11-11 21:38:19 +00:00
#define M_PI_2 1.57079632679489661923f
#endif
2012-11-11 21:38:19 +00:00
#define M_PI_4 0.785398163397448309616f
#define M_1_PI 0.318309886183790671538f
#define M_2_PI 0.636619772367581343076f
#define M_2_SQRTPI 1.12837916709551257390f
#define M_SQRT2 1.41421356237309504880f
#define M_SQRT1_2 0.707106781186547524401f
2012-11-01 15:19:01 +00:00
#endif
union FloatBits {
float f[4];
u32 u[4];
int i[4];
};
2013-02-24 22:39:13 +00:00
// Preserves NaN in first param, takes sign of equal second param.
// Technically, std::max may do this but it's undefined.
inline float nanmax(float f, float cst)
{
return f <= cst ? cst : f;
}
// Preserves NaN in first param, takes sign of equal second param.
inline float nanmin(float f, float cst)
{
return f >= cst ? cst : f;
}
// Preserves NaN in first param, takes sign of equal value in others.
inline float nanclamp(float f, float lower, float upper)
{
return nanmin(nanmax(f, lower), upper);
}
2013-01-29 01:15:23 +00:00
void ApplyPrefixST(float *r, u32 data, VectorSize size)
2012-11-01 15:19:01 +00:00
{
// Possible optimization shortcut:
if (data == 0xe4)
return;
2012-11-01 15:19:01 +00:00
int n = GetNumVectorElements(size);
float origV[4];
static const float constantArray[8] = {0.f, 1.f, 2.f, 0.5f, 3.f, 1.f/3.f, 0.25f, 1.f/6.f};
for (int i = 0; i < n; i++)
{
origV[i] = r[i];
2012-11-01 15:19:01 +00:00
}
for (int i = 0; i < n; i++)
{
int regnum = (data >> (i*2)) & 3;
int abs = (data >> (8+i)) & 1;
int negate = (data >> (16+i)) & 1;
int constants = (data >> (12+i)) & 1;
if (!constants)
{
// Prefix may say "z, z, z, z" but if this is a pair, we force to x.
// TODO: But some ops seem to use const 0 instead?
if (regnum >= n) {
ERROR_LOG_REPORT(CPU, "Invalid VFPU swizzle: %08x: %i / %d at PC = %08x (%s)", data, regnum, n, currentMIPS->pc, MIPSDisasmAt(currentMIPS->pc));
//for (int i = 0; i < 12; i++) {
// ERROR_LOG(CPU, " vfpuCtrl[%i] = %08x", i, currentMIPS->vfpuCtrl[i]);
//}
regnum = 0;
}
r[i] = origV[regnum];
2012-11-01 15:19:01 +00:00
if (abs)
r[i] = fabs(r[i]);
2012-11-01 15:19:01 +00:00
}
else
{
r[i] = constantArray[regnum + (abs<<2)];
2012-11-01 15:19:01 +00:00
}
if (negate)
r[i] = -r[i];
2012-11-01 15:19:01 +00:00
}
}
2012-11-17 14:06:10 +00:00
inline void ApplySwizzleS(float *v, VectorSize size)
2012-11-01 15:19:01 +00:00
{
ApplyPrefixST(v, currentMIPS->vfpuCtrl[VFPU_CTRL_SPREFIX], size);
}
2012-11-17 14:06:10 +00:00
inline void ApplySwizzleT(float *v, VectorSize size)
2012-11-01 15:19:01 +00:00
{
ApplyPrefixST(v, currentMIPS->vfpuCtrl[VFPU_CTRL_TPREFIX], size);
}
2012-11-11 21:38:19 +00:00
void ApplyPrefixD(float *v, VectorSize size, bool onlyWriteMask = false)
2012-11-01 15:19:01 +00:00
{
2012-11-17 14:06:10 +00:00
u32 data = currentMIPS->vfpuCtrl[VFPU_CTRL_DPREFIX];
if (!data || onlyWriteMask)
2012-11-17 14:06:10 +00:00
return;
2012-11-01 15:19:01 +00:00
int n = GetNumVectorElements(size);
for (int i = 0; i < n; i++)
{
int sat = (data >> (i * 2)) & 3;
if (sat == 1)
{
if (v[i] > 1.0f) v[i] = 1.0f;
// This includes -0.0f -> +0.0f.
if (v[i] <= 0.0f) v[i] = 0.0f;
}
else if (sat == 3)
{
if (v[i] > 1.0f) v[i] = 1.0f;
if (v[i] < -1.0f) v[i] = -1.0f;
2012-11-01 15:19:01 +00:00
}
}
}
void EatPrefixes()
{
2012-11-17 14:06:10 +00:00
currentMIPS->vfpuCtrl[VFPU_CTRL_SPREFIX] = 0xe4; // passthru
currentMIPS->vfpuCtrl[VFPU_CTRL_TPREFIX] = 0xe4; // passthru
2012-11-01 15:19:01 +00:00
currentMIPS->vfpuCtrl[VFPU_CTRL_DPREFIX] = 0;
}
namespace MIPSInt
{
void Int_VPFX(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int data = op & 0xFFFFF;
int regnum = (op >> 24) & 3;
currentMIPS->vfpuCtrl[VFPU_CTRL_SPREFIX + regnum] = data;
PC += 4;
}
void Int_SVQ(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int imm = (signed short)(op&0xFFFC);
int rs = _RS;
int vt = (((op >> 16) & 0x1f)) | ((op&1) << 5);
u32 addr = R(rs) + imm;
switch (op >> 26)
{
case 53: //lvl.q/lvr.q
2012-11-18 22:34:37 +00:00
{
if (addr & 0x3)
2012-11-18 22:34:37 +00:00
{
_dbg_assert_msg_(CPU, 0, "Misaligned lvX.q");
2012-11-18 22:34:37 +00:00
}
float d[4];
ReadVector(d, V_Quad, vt);
int offset = (addr >> 2) & 3;
if ((op & 2) == 0)
{
// It's an LVL
for (int i = 0; i < offset + 1; i++)
{
d[3 - i] = Memory::Read_Float(addr - 4 * i);
}
}
else
2012-11-18 22:34:37 +00:00
{
// It's an LVR
for (int i = 0; i < (3 - offset) + 1; i++)
{
d[i] = Memory::Read_Float(addr + 4 * i);
}
2012-11-18 22:34:37 +00:00
}
WriteVector(d, V_Quad, vt);
}
break;
2012-11-01 15:19:01 +00:00
case 54: //lv.q
2012-11-18 22:34:37 +00:00
if (addr & 0xF)
{
_dbg_assert_msg_(CPU, 0, "Misaligned lv.q");
}
2013-07-30 14:40:35 +00:00
#ifndef COMMON_BIG_ENDIAN
2012-11-01 15:19:01 +00:00
WriteVector((const float*)Memory::GetPointer(addr), V_Quad, vt);
2013-07-30 14:40:35 +00:00
#else
float lvqd[4];
lvqd[0] = Memory::Read_Float(addr);
lvqd[1] = Memory::Read_Float(addr + 4);
lvqd[2] = Memory::Read_Float(addr + 8);
lvqd[3] = Memory::Read_Float(addr + 12);
WriteVector(lvqd, V_Quad, vt);
#endif
2012-11-01 15:19:01 +00:00
break;
case 61: // svl.q/svr.q
2012-11-18 22:34:37 +00:00
{
if (addr & 0x3)
2012-11-18 22:34:37 +00:00
{
_dbg_assert_msg_(CPU, 0, "Misaligned svX.q");
2012-11-18 22:34:37 +00:00
}
float d[4];
ReadVector(d, V_Quad, vt);
int offset = (addr >> 2) & 3;
if ((op&2) == 0)
2012-11-18 22:34:37 +00:00
{
// It's an SVL
for (int i = 0; i < offset + 1; i++)
{
Memory::Write_Float(d[3 - i], addr - i * 4);
}
2012-11-18 22:34:37 +00:00
}
else
{
// It's an SVR
for (int i = 0; i < (3 - offset) + 1; i++)
{
Memory::Write_Float(d[i], addr + 4 * i);
}
}
break;
2012-11-18 22:34:37 +00:00
}
2012-11-01 15:19:01 +00:00
case 62: //sv.q
2012-11-18 22:34:37 +00:00
if (addr & 0xF)
{
_dbg_assert_msg_(CPU, 0, "Misaligned sv.q");
}
2013-07-30 14:40:35 +00:00
#ifndef COMMON_BIG_ENDIAN
ReadVector(reinterpret_cast<float *>(Memory::GetPointer(addr)), V_Quad, vt);
2013-07-30 14:40:35 +00:00
#else
float svqd[4];
ReadVector(svqd, V_Quad, vt);
Memory::Write_Float(svqd[0], addr);
Memory::Write_Float(svqd[1], addr + 4);
Memory::Write_Float(svqd[2], addr + 8);
Memory::Write_Float(svqd[3], addr + 12);
#endif
2012-11-01 15:19:01 +00:00
break;
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret VQ instruction that can't be interpreted");
break;
}
PC += 4;
}
void Int_VMatrixInit(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
static const float idt[16] =
{
1,0,0,0,
0,1,0,0,
0,0,1,0,
0,0,0,1,
};
static const float zero[16] =
{
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
};
static const float one[16] =
{
1,1,1,1,
1,1,1,1,
1,1,1,1,
1,1,1,1,
};
int vd = _VD;
MatrixSize sz = GetMtxSize(op);
const float *m;
switch ((op >> 16) & 0xF)
{
2012-11-11 21:38:19 +00:00
case 3: m=idt; break; //identity // vmidt
case 6: m=zero; break; // vmzero
case 7: m=one; break; // vmone
2012-11-01 15:19:01 +00:00
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
2013-08-11 23:28:20 +00:00
PC += 4;
EatPrefixes();
2012-11-01 15:19:01 +00:00
return;
}
WriteMatrix(m, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_VVectorInit(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int vd = _VD;
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
2012-11-11 21:38:19 +00:00
static const float ones[4] = {1,1,1,1};
static const float zeros[4] = {0,0,0,0};
2012-11-01 15:19:01 +00:00
const float *v;
switch ((op >> 16) & 0xF)
{
2012-11-11 21:38:19 +00:00
case 6: v=zeros; break; //vzero
case 7: v=ones; break; //vone
2012-11-01 15:19:01 +00:00
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
2013-08-11 23:28:20 +00:00
PC += 4;
EatPrefixes();
2012-11-01 15:19:01 +00:00
return;
}
float o[4];
for (int i = 0; i < n; i++)
o[i] = v[i];
ApplyPrefixD(o, sz);
WriteVector(o, sz, vd);
EatPrefixes();
PC += 4;
}
void Int_Viim(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int vt = _VT;
s32 imm = (s16)(op&0xFFFF);
u16 uimm16 = (op&0xFFFF);
2012-11-01 15:19:01 +00:00
//V(vt) = (float)imm;
2012-11-11 21:38:19 +00:00
float f[1];
2012-11-01 15:19:01 +00:00
int type = (op >> 23) & 7;
if (type == 6)
2012-11-11 21:38:19 +00:00
f[0] = (float)imm; // viim
2012-11-01 15:19:01 +00:00
else if (type == 7)
2012-11-11 21:38:19 +00:00
f[0] = Float16ToFloat32((u16)uimm16); // vfim
2012-11-01 15:19:01 +00:00
else
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
2012-11-11 21:38:19 +00:00
ApplyPrefixD(f, V_Single);
V(vt) = f[0];
2012-11-01 15:19:01 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vidt(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int vd = _VD;
VectorSize sz = GetVecSize(op);
float f[4];
switch (sz)
{
case V_Pair:
f[0] = (vd&1)==0 ? 1.0f : 0.0f;
f[1] = (vd&1)==1 ? 1.0f : 0.0f;
break;
case V_Quad:
f[0] = (vd&3)==0 ? 1.0f : 0.0f;
f[1] = (vd&3)==1 ? 1.0f : 0.0f;
f[2] = (vd&3)==2 ? 1.0f : 0.0f;
f[3] = (vd&3)==3 ? 1.0f : 0.0f;
break;
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
break;
}
2012-11-11 21:38:19 +00:00
ApplyPrefixD(f, sz);
2012-11-01 15:19:01 +00:00
WriteVector(f, sz, vd);
PC += 4;
EatPrefixes();
}
// The test really needs some work.
void Int_Vmmul(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
float s[16];
float t[16];
float d[16];
int vd = _VD;
int vs = _VS;
int vt = _VT;
MatrixSize sz = GetMtxSize(op);
int n = GetMatrixSide(sz);
ReadMatrix(s, sz, vs);
ReadMatrix(t, sz, vt);
for (int a = 0; a < n; a++)
{
for (int b = 0; b < n; b++)
{
float sum = 0.0f;
for (int c = 0; c < n; c++)
{
sum += s[b*4 + c] * t[a*4 + c];
}
d[a*4 + b] = sum;
2012-11-01 15:19:01 +00:00
}
}
WriteMatrix(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vmscl(MIPSOpcode op)
2012-11-18 22:34:37 +00:00
{
float d[16];
float s[16];
float t[1];
int vd = _VD;
int vs = _VS;
int vt = _VT;
MatrixSize sz = GetMtxSize(op);
int n = GetMatrixSide(sz);
ReadMatrix(s, sz, vs);
ReadVector(t, V_Single, vt);
for (int a = 0; a < n; a++)
{
for (int b = 0; b < n; b++)
{
d[a*4 + b] = s[a*4 + b] * t[0];
}
}
WriteMatrix(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-11-01 15:19:01 +00:00
void Int_Vmmov(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
float s[16];
int vd = _VD;
int vs = _VS;
MatrixSize sz = GetMtxSize(op);
ReadMatrix(s, sz, vs);
// This is just for matrices. No prefixes.
2012-11-18 22:34:37 +00:00
WriteMatrix(s, sz, vd);
2012-11-01 15:19:01 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vflush(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
// DEBUG_LOG(CPU,"vflush");
2012-11-01 15:19:01 +00:00
PC += 4;
}
void Int_VV2Op(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
float s[4], d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
for (int i = 0; i < GetNumVectorElements(sz); i++)
{
switch ((op >> 16) & 0x1f)
{
case 0: d[i] = s[i]; break; //vmov
case 1: d[i] = fabsf(s[i]); break; //vabs
case 2: d[i] = -s[i]; break; //vneg
// vsat0 changes -0.0 to +0.0, both retain NAN.
case 4: if (s[i] <= 0) d[i] = 0; else {if(s[i] > 1.0f) d[i] = 1.0f; else d[i] = s[i];} break; // vsat0
2012-11-01 15:19:01 +00:00
case 5: if (s[i] < -1.0f) d[i] = -1.0f; else {if(s[i] > 1.0f) d[i] = 1.0f; else d[i] = s[i];} break; // vsat1
case 16: d[i] = 1.0f / s[i]; break; //vrcp
case 17: d[i] = 1.0f / sqrtf(s[i]); break; //vrsq
case 18: { d[i] = vfpu_sin(s[i]); } break; //vsin
case 19: { d[i] = vfpu_cos(s[i]); } break; //vcos
case 20: d[i] = powf(2.0f, s[i]); break; //vexp2
case 21: d[i] = logf(s[i])/log(2.0f); break; //vlog2
2013-02-19 05:46:33 +00:00
case 22: d[i] = fabsf(sqrtf(s[i])); break; //vsqrt
2013-03-24 10:34:29 +00:00
case 23: d[i] = asinf(s[i]) / M_PI_2; break; //vasin
case 24: d[i] = -1.0f / s[i]; break; // vnrcp
case 26: { d[i] = -vfpu_sin(s[i]); } break; // vnsin
case 28: d[i] = 1.0f / powf(2.0, s[i]); break; // vrexp2
2012-11-01 15:19:01 +00:00
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret VV2Op instruction that can't be interpreted");
break;
}
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vocp(MIPSOpcode op)
2012-11-15 09:15:40 +00:00
{
float s[4], d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
for (int i = 0; i < GetNumVectorElements(sz); i++)
{
2013-02-24 22:39:13 +00:00
// Always positive NaN.
d[i] = my_isnan(s[i]) ? fabsf(s[i]) : 1.0f - s[i];
2012-11-15 09:15:40 +00:00
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-12-21 03:05:53 +00:00
void Int_Vsocp(MIPSOpcode op)
2012-12-21 03:05:53 +00:00
{
float s[4], d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
int n = GetNumVectorElements(sz);
2012-12-21 03:05:53 +00:00
float x = s[0];
2013-02-24 22:39:13 +00:00
d[0] = nanclamp(1.0f - x, 0.0f, 1.0f);
d[1] = nanclamp(x, 0.0f, 1.0f);
VectorSize outSize = V_Pair;
2012-12-21 09:30:17 +00:00
if (n > 1) {
float y = s[1];
2013-02-24 22:39:13 +00:00
d[2] = nanclamp(1.0f - y, 0.0f, 1.0f);
d[3] = nanclamp(y, 0.0f, 1.0f);
outSize = V_Quad;
2012-12-21 09:30:17 +00:00
}
WriteVector(d, outSize, vd);
2012-12-21 03:05:53 +00:00
PC += 4;
EatPrefixes();
}
2012-11-01 15:19:01 +00:00
void Int_Vsgn(MIPSOpcode op)
2012-11-15 09:15:40 +00:00
{
float s[4], d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
for (int i = 0; i < GetNumVectorElements(sz); i++)
{
// To handle NaNs correctly, we do this with integer hackery
u32 val;
memcpy(&val, &s[i], sizeof(u32));
if (val == 0 || val == 0x80000000)
d[i] = 0.0f;
else if ((val >> 31) == 0)
2012-11-11 21:38:19 +00:00
d[i] = 1.0f;
else
2012-11-15 09:15:40 +00:00
d[i] = -1.0f;
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-11-01 15:19:01 +00:00
2013-08-07 19:30:57 +00:00
inline int round_vfpu_n(double param) {
2013-07-29 20:36:31 +00:00
// return floorf(param);
return (int)round_ieee_754(param);
}
void Int_Vf2i(MIPSOpcode op) {
2012-11-01 15:19:01 +00:00
float s[4];
int d[4];
int vd = _VD;
int vs = _VS;
int imm = (op >> 16) & 0x1f;
float mult = (float)(1UL << imm);
2012-11-01 15:19:01 +00:00
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz); //TODO: and the mask to kill everything but swizzle
2013-08-07 19:30:57 +00:00
for (int i = 0; i < GetNumVectorElements(sz); i++) {
if (my_isnan(s[i])) {
2013-03-24 13:28:32 +00:00
d[i] = 0x7FFFFFFF;
continue;
}
double sv = s[i] * mult; // (float)0x7fffffff == (float)0x80000000
// Cap/floor it to 0x7fffffff / 0x80000000
2013-08-07 19:30:57 +00:00
if (sv > (double)0x7fffffff) {
d[i] = 0x7fffffff;
} else if (sv <= (double)(int)0x80000000) {
d[i] = 0x80000000;
} else {
switch ((op >> 21) & 0x1f)
{
case 16: d[i] = (int)round_vfpu_n(sv); break; //(floor(sv + 0.5f)); break; //n
2013-08-07 19:30:57 +00:00
case 17: d[i] = s[i]>=0 ? (int)floor(sv) : (int)ceil(sv); break; //z
case 18: d[i] = (int)ceil(sv); break; //u
case 19: d[i] = (int)floor(sv); break; //d
default: d[i] = 0x7FFFFFFF; break;
}
2012-11-01 15:19:01 +00:00
}
}
ApplyPrefixD(reinterpret_cast<float *>(d), sz, true);
WriteVector(reinterpret_cast<float *>(d), sz, vd);
2012-11-01 15:19:01 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vi2f(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int s[4];
float d[4];
int vd = _VD;
int vs = _VS;
int imm = (op >> 16) & 0x1f;
float mult = 1.0f/(float)(1UL << imm);
2012-11-01 15:19:01 +00:00
VectorSize sz = GetVecSize(op);
ReadVector(reinterpret_cast<float *>(s), sz, vs);
ApplySwizzleS(reinterpret_cast<float *>(s), sz); //TODO: and the mask to kill everything but swizzle
2012-11-01 15:19:01 +00:00
for (int i = 0; i < GetNumVectorElements(sz); i++)
{
d[i] = (float)s[i] * mult;
}
ApplyPrefixD(d, sz); //TODO: and the mask to kill everything but mask
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-11-06 19:56:19 +00:00
void Int_Vh2f(MIPSOpcode op)
2012-11-11 21:38:19 +00:00
{
u32 s[4];
2012-11-11 21:38:19 +00:00
float d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(reinterpret_cast<float *>(s), sz, vs);
ApplySwizzleS(reinterpret_cast<float *>(s), sz);
2012-11-11 21:38:19 +00:00
VectorSize outsize = V_Pair;
switch (sz) {
case V_Single:
outsize = V_Pair;
d[0] = ExpandHalf(s[0] & 0xFFFF);
d[1] = ExpandHalf(s[0] >> 16);
break;
case V_Pair:
outsize = V_Quad;
d[0] = ExpandHalf(s[0] & 0xFFFF);
d[1] = ExpandHalf(s[0] >> 16);
d[2] = ExpandHalf(s[1] & 0xFFFF);
d[3] = ExpandHalf(s[1] >> 16);
break;
2014-12-20 17:11:45 +00:00
default:
_dbg_assert_msg_(CPU, 0, "Trying to interpret Int_Vh2f instruction that can't be interpreted");
memset(d, 0, sizeof(d));
break;
2012-11-11 21:38:19 +00:00
}
2013-02-23 19:34:32 +00:00
ApplyPrefixD(d, outsize);
WriteVector(d, outsize, vd);
2012-11-11 21:38:19 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vf2h(MIPSOpcode op)
2012-11-06 19:56:19 +00:00
{
float s[4];
u32 d[4];
2012-11-06 19:56:19 +00:00
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
2013-02-23 19:34:32 +00:00
ApplySwizzleS(s, sz);
2012-11-06 19:56:19 +00:00
VectorSize outsize = V_Single;
switch (sz) {
case V_Pair:
outsize = V_Single;
d[0] = ShrinkToHalf(s[0]) | ((u32)ShrinkToHalf(s[1]) << 16);
break;
case V_Quad:
outsize = V_Pair;
d[0] = ShrinkToHalf(s[0]) | ((u32)ShrinkToHalf(s[1]) << 16);
d[1] = ShrinkToHalf(s[2]) | ((u32)ShrinkToHalf(s[3]) << 16);
break;
2014-12-20 17:11:45 +00:00
default:
_dbg_assert_msg_(CPU, 0, "Trying to interpret Int_Vf2h instruction that can't be interpreted");
break;
2012-11-06 19:56:19 +00:00
}
ApplyPrefixD(reinterpret_cast<float *>(d), outsize);
WriteVector(reinterpret_cast<float *>(d), outsize, vd);
2012-11-06 19:56:19 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vx2i(MIPSOpcode op)
2012-11-11 17:44:20 +00:00
{
u32 s[4];
u32 d[4] = {0};
2012-11-11 17:44:20 +00:00
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
VectorSize oz = sz;
ReadVector(reinterpret_cast<float *>(s), sz, vs);
2012-11-11 17:44:20 +00:00
// ForbidVPFXS
switch ((op >> 16) & 3) {
case 0: // vuc2i
// Quad is the only option.
// This operation is weird. This particular way of working matches hw but does not
// seem quite sane.
// I guess it's used for fixed-point math, and fills more bits to facilitate
// conversion between 8-bit and 16-bit values. But then why not do it in vc2i?
2012-11-11 17:44:20 +00:00
{
u32 value = s[0];
u32 value2 = value / 2;
for (int i = 0; i < 4; i++) {
d[i] = (u32)((value & 0xFF) * 0x01010101) >> 1;
2012-11-11 17:44:20 +00:00
value >>= 8;
}
oz = V_Quad;
2012-11-11 17:44:20 +00:00
}
break;
2012-11-11 17:44:20 +00:00
case 1: // vc2i
// Quad is the only option
{
u32 value = s[0];
d[0] = (value & 0xFF) << 24;
d[1] = (value & 0xFF00) << 16;
d[2] = (value & 0xFF0000) << 8;
d[3] = (value & 0xFF000000);
oz = V_Quad;
2012-11-11 17:44:20 +00:00
}
break;
case 2: // vus2i
2012-11-12 13:35:10 +00:00
oz = V_Pair;
switch (sz)
{
case V_Pair:
oz = V_Quad;
// Intentional fallthrough.
case V_Single:
for (int i = 0; i < GetNumVectorElements(sz); i++) {
u32 value = s[i];
d[i * 2] = (value & 0xFFFF) << 15;
d[i * 2 + 1] = (value & 0xFFFF0000) >> 1;
}
break;
default:
ERROR_LOG_REPORT(CPU, "vus2i with more than 2 elements.");
break;
}
2012-11-12 13:35:10 +00:00
break;
2012-11-11 17:44:20 +00:00
case 3: // vs2i
oz = V_Pair;
switch (sz)
{
case V_Pair:
oz = V_Quad;
// Intentional fallthrough.
case V_Single:
for (int i = 0; i < GetNumVectorElements(sz); i++) {
u32 value = s[i];
d[i * 2] = (value & 0xFFFF) << 16;
d[i * 2 + 1] = value & 0xFFFF0000;
}
break;
default:
ERROR_LOG_REPORT(CPU, "vs2i with more than 2 elements.");
break;
}
2012-11-11 17:44:20 +00:00
break;
2012-11-11 17:44:20 +00:00
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
break;
}
ApplyPrefixD(reinterpret_cast<float *>(d),oz, true); // Only write mask
WriteVector(reinterpret_cast<float *>(d),oz,vd);
2012-11-11 17:44:20 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vi2x(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int s[4];
2012-11-18 22:34:37 +00:00
u32 d[2] = {0};
2012-11-01 15:19:01 +00:00
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
VectorSize oz;
ReadVector(reinterpret_cast<float *>(s), sz, vs);
ApplySwizzleS(reinterpret_cast<float *>(s), sz); //TODO: and the mask to kill everything but swizzle
2012-11-01 15:19:01 +00:00
switch ((op >> 16)&3)
{
case 0: //vi2uc
{
for (int i = 0; i < 4; i++)
{
int v = s[i];
if (v < 0) v = 0;
v >>= 23;
d[0] |= ((u32)v & 0xFF) << (i * 8);
}
oz = V_Single;
}
break;
case 1: //vi2c
2012-11-18 22:34:37 +00:00
{
for (int i = 0; i < 4; i++)
{
u32 v = s[i];
d[0] |= (v >> 24) << (i * 8);
2012-11-18 22:34:37 +00:00
}
oz = V_Single;
}
break;
case 2: //vi2us
2012-11-07 18:10:52 +00:00
{
for (int i = 0; i < GetNumVectorElements(sz) / 2; i++) {
int low = s[i * 2];
int high = s[i * 2 + 1];
if (low < 0) low = 0;
if (high < 0) high = 0;
low >>= 15;
high >>= 15;
d[i] = low | (high << 16);
}
switch (sz) {
case V_Quad: oz = V_Pair; break;
case V_Pair: oz = V_Single; break;
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
oz = V_Single;
break;
}
2012-11-07 18:10:52 +00:00
}
break;
2012-11-01 15:19:01 +00:00
case 3: //vi2s
2012-11-07 18:10:52 +00:00
{
for (int i = 0; i < GetNumVectorElements(sz) / 2; i++) {
u32 low = s[i * 2];
u32 high = s[i * 2 + 1];
low >>= 16;
high >>= 16;
d[i] = low | (high << 16);
}
switch (sz) {
case V_Quad: oz = V_Pair; break;
case V_Pair: oz = V_Single; break;
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
oz = V_Single;
break;
}
2012-11-07 18:10:52 +00:00
}
break;
2012-11-01 15:19:01 +00:00
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
oz = V_Single;
2012-11-01 15:19:01 +00:00
break;
}
ApplyPrefixD(reinterpret_cast<float *>(d),oz);
WriteVector(reinterpret_cast<float *>(d),oz,vd);
2012-11-01 15:19:01 +00:00
PC += 4;
EatPrefixes();
}
void Int_ColorConv(MIPSOpcode op)
2012-11-18 22:34:37 +00:00
{
int vd = _VD;
int vs = _VS;
u32 s[4];
VectorSize sz = V_Quad;
ReadVector(reinterpret_cast<float *>(s), sz, vs);
2012-11-18 22:34:37 +00:00
u16 colors[4];
for (int i = 0; i < 4; i++)
{
u32 in = s[i];
u16 col = 0;
switch ((op >> 16) & 3)
{
case 1: // 4444
{
int a = ((in >> 24) & 0xFF) >> 4;
int b = ((in >> 16) & 0xFF) >> 4;
int g = ((in >> 8) & 0xFF) >> 4;
int r = ((in) & 0xFF) >> 4;
col = (a << 12) | (b << 8) | (g << 4 ) | (r);
break;
}
case 2: // 5551
{
int a = ((in >> 24) & 0xFF) >> 7;
int b = ((in >> 16) & 0xFF) >> 3;
int g = ((in >> 8) & 0xFF) >> 3;
int r = ((in) & 0xFF) >> 3;
col = (a << 15) | (b << 10) | (g << 5) | (r);
break;
}
case 3: // 565
{
int b = ((in >> 16) & 0xFF) >> 3;
int g = ((in >> 8) & 0xFF) >> 2;
int r = ((in) & 0xFF) >> 3;
col = (b << 11) | (g << 5) | (r);
break;
}
}
colors[i] = col;
}
u32 ov[2] = {(u32)colors[0] | (colors[1] << 16), (u32)colors[2] | (colors[3] << 16)};
WriteVector((const float *)ov, V_Pair, vd);
PC += 4;
EatPrefixes();
}
2012-11-01 15:19:01 +00:00
void Int_VDot(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
float s[4], t[4];
float d;
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
ReadVector(t, sz, vt);
ApplySwizzleT(t, sz);
float sum = 0.0f;
int n = GetNumVectorElements(sz);
for (int i = 0; i < n; i++)
{
sum += s[i]*t[i];
}
d = sum;
ApplyPrefixD(&d,V_Single);
WriteVector(&d, V_Single, vd);
2012-11-01 15:19:01 +00:00
PC += 4;
EatPrefixes();
}
void Int_VHdp(MIPSOpcode op)
2012-12-18 21:54:08 +00:00
{
float s[4], t[4];
float d;
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
ReadVector(t, sz, vt);
ApplySwizzleT(t, sz);
float sum = 0.0f;
int n = GetNumVectorElements(sz);
for (int i = 0; i < n; i++)
{
sum += (i == n - 1) ? t[i] : s[i]*t[i];
}
d = my_isnan(sum) ? fabsf(sum) : sum;
2012-12-18 21:54:08 +00:00
ApplyPrefixD(&d,V_Single);
WriteVector(&d, V_Single, vd);
2012-12-18 21:54:08 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vbfy(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
float s[4];
float d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
int n = GetNumVectorElements(sz);
if (op & 0x10000)
{
// vbfy2
2012-11-01 15:19:01 +00:00
d[0] = s[0] + s[2];
d[1] = s[1] + s[3];
d[2] = s[0] - s[2];
d[3] = s[1] - s[3];
}
else
{
d[0] = s[0] + s[1];
d[1] = s[0] - s[1];
if (n == 4) {
d[2] = s[2] + s[3];
d[3] = s[2] - s[3];
2012-11-01 15:19:01 +00:00
}
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-12-21 06:30:34 +00:00
void Int_Vsrt1(MIPSOpcode op)
2012-12-21 06:30:34 +00:00
{
float s[4];
float d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
float x = s[0];
float y = s[1];
float z = s[2];
float w = s[3];
d[0] = std::min(x, y);
d[1] = std::max(x, y);
d[2] = std::min(z, w);
d[3] = std::max(z, w);
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vsrt2(MIPSOpcode op)
2012-12-21 06:30:34 +00:00
{
float s[4];
float d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
float x = s[0];
float y = s[1];
float z = s[2];
float w = s[3];
d[0] = std::min(x, w);
d[1] = std::min(y, z);
d[2] = std::max(y, z);
d[3] = std::max(x, w);
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-11-01 15:19:01 +00:00
void Int_Vsrt3(MIPSOpcode op)
2012-12-21 06:30:34 +00:00
{
float s[4];
float d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
float x = s[0];
float y = s[1];
float z = s[2];
float w = s[3];
d[0] = std::max(x, y);
d[1] = std::min(x, y);
d[2] = std::max(z, w);
d[3] = std::min(z, w);
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vsrt4(MIPSOpcode op)
2012-12-21 06:30:34 +00:00
{
float s[4];
float d[4];
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
float x = s[0];
float y = s[1];
float z = s[2];
float w = s[3];
d[0] = std::max(x, w);
d[1] = std::max(y, z);
d[2] = std::min(y, z);
d[3] = std::min(x, w);
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vcrs(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
//half a cross product
float s[4], t[4];
float d[4];
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
if (sz != V_Triple)
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
ReadVector(s, sz, vs);
ReadVector(t, sz, vt);
// no swizzles allowed
d[0] = s[1] * t[2];
d[1] = s[2] * t[0];
d[2] = s[0] * t[1];
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-12-20 13:11:21 +00:00
void Int_Vdet(MIPSOpcode op)
2012-12-20 13:11:21 +00:00
{
float s[4], t[4];
float d[4];
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
if (sz != V_Pair)
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
2012-12-20 13:11:21 +00:00
ReadVector(t, sz, vt);
// TODO: The swizzle on t behaves oddly with constants, but sign changes seem to work.
// Also, seems to round in a non-standard way (sometimes toward zero, not always.)
2012-12-20 13:11:21 +00:00
d[0] = s[0] * t[1] - s[1] * t[0];
ApplyPrefixD(d, sz);
WriteVector(d, V_Single, vd);
2012-12-20 13:11:21 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vfad(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
float s[4];
float d;
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
float sum = 0.0f;
int n = GetNumVectorElements(sz);
for (int i = 0; i < n; i++)
{
sum += s[i];
}
d = sum;
ApplyPrefixD(&d,V_Single);
V(vd) = d;
PC += 4;
EatPrefixes();
}
void Int_Vavg(MIPSOpcode op)
2012-11-18 22:34:37 +00:00
{
float s[4];
float d;
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
float sum = 0.0f;
int n = GetNumVectorElements(sz);
for (int i = 0; i < n; i++)
{
sum += s[i];
}
d = sum / n;
ApplyPrefixD(&d, V_Single);
V(vd) = d;
PC += 4;
EatPrefixes();
}
2012-11-01 15:19:01 +00:00
void Int_VScl(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
float s[4];
float d[4];
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
float scale = V(vt);
if (currentMIPS->vfpuCtrl[VFPU_CTRL_TPREFIX] != 0xE4)
{
// WARN_LOG(CPU, "Broken T prefix used with VScl: %08x / %08x", currentMIPS->vfpuCtrl[VFPU_CTRL_TPREFIX], op);
ApplySwizzleT(&scale, V_Single);
}
2012-11-01 15:19:01 +00:00
int n = GetNumVectorElements(sz);
for (int i = 0; i < n; i++)
{
d[i] = s[i] * scale;
2012-11-01 15:19:01 +00:00
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vrnds(MIPSOpcode op)
2012-11-18 22:34:37 +00:00
{
int vd = _VD;
int seed = VI(vd);
currentMIPS->rng.Init(seed);
2012-11-25 02:25:42 +00:00
PC += 4;
EatPrefixes();
2012-11-18 22:34:37 +00:00
}
void Int_VrndX(MIPSOpcode op)
2012-11-18 22:34:37 +00:00
{
FloatBits d;
2012-11-18 22:34:37 +00:00
int vd = _VD;
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
for (int i = 0; i < n; i++)
{
switch ((op >> 16) & 0x1f)
{
case 1: d.u[i] = currentMIPS->rng.R32(); break; // vrndi
case 2: d.f[i] = 1.0f + ((float)currentMIPS->rng.R32() / 0xFFFFFFFF); break; // vrndf1 TODO: make more accurate
case 3: d.f[i] = 2.0f + 2 * ((float)currentMIPS->rng.R32() / 0xFFFFFFFF); break; // vrndf2 TODO: make more accurate
default: _dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
2012-11-18 22:34:37 +00:00
}
}
ApplyPrefixD(d.f, sz);
WriteVector(d.f, sz, vd);
2012-11-18 22:34:37 +00:00
PC += 4;
EatPrefixes();
}
// Generates one line of a rotation matrix around one of the three axes
void Int_Vrot(MIPSOpcode op) {
2012-11-01 15:19:01 +00:00
int vd = _VD;
int vs = _VS;
int imm = (op >> 16) & 0x1f;
VectorSize sz = GetVecSize(op);
bool negSin = (imm & 0x10) ? true : false;
float sine, cosine;
vfpu_sincos(V(vs), sine, cosine);
2012-11-01 15:19:01 +00:00
if (negSin)
2012-11-18 22:34:37 +00:00
sine = -sine;
2012-11-01 15:19:01 +00:00
float d[4] = {0};
if (((imm >> 2) & 3) == (imm & 3)) {
2012-11-01 15:19:01 +00:00
for (int i = 0; i < 4; i++)
d[i] = sine;
}
d[(imm >> 2) & 3] = sine;
d[imm & 3] = cosine;
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vtfm(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int vd = _VD;
int vs = _VS;
int vt = _VT;
int ins = (op >> 23) & 7;
VectorSize sz = GetVecSize(op);
MatrixSize msz = GetMtxSize(op);
2012-11-18 22:34:37 +00:00
int n = GetNumVectorElements(sz);
bool homogenous = false;
if (n == ins)
{
n++;
sz = (VectorSize)((int)(sz) + 1);
msz = (MatrixSize)((int)(msz) + 1);
homogenous = true;
}
2012-11-01 15:19:01 +00:00
float s[16];
ReadMatrix(s, msz, vs);
float t[4];
ReadVector(t, sz, vt);
float d[4];
2012-11-18 22:34:37 +00:00
2012-11-01 15:19:01 +00:00
if (homogenous)
{
for (int i = 0; i < n; i++)
{
d[i] = 0.0f;
for (int k = 0; k < n; k++)
{
d[i] += (k == n-1) ? s[i*4+k] : (s[i*4+k] * t[k]);
}
}
}
else if (n == ins + 1)
2012-11-01 15:19:01 +00:00
{
for (int i = 0; i < n; i++)
{
d[i] = 0.0f;
for (int k = 0; k < n; k++)
{
d[i] += s[i*4+k] * t[k];
}
}
}
else
{
Reporting::ReportMessage("Trying to interpret instruction that can't be interpreted (BADVTFM)");
2012-11-01 15:19:01 +00:00
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted (BADVTFM)");
memset(d, 0, sizeof(d));
2012-11-01 15:19:01 +00:00
}
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-11-18 22:34:37 +00:00
void Int_SV(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
s32 imm = (signed short)(op&0xFFFC);
int vt = ((op >> 16) & 0x1f) | ((op & 3) << 5);
int rs = _RS;
2012-11-01 15:19:01 +00:00
u32 addr = R(rs) + imm;
switch (op >> 26)
{
case 50: //lv.s
2012-11-18 22:34:37 +00:00
VI(vt) = Memory::Read_U32(addr);
2012-11-01 15:19:01 +00:00
break;
case 58: //sv.s
Memory::Write_U32(VI(vt), addr);
break;
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
break;
}
PC += 4;
}
void Int_Mftv(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
2012-11-22 19:14:24 +00:00
int imm = op & 0xFF;
2012-11-01 15:19:01 +00:00
int rt = _RT;
switch ((op >> 21) & 0x1f)
{
case 3: //mfv / mfvc
// rt = 0, imm = 255 appears to be used as a CPU interlock by some games.
if (rt != 0) {
if (imm < 128) {
R(rt) = VI(imm);
} else if (imm < 128 + VFPU_CTRL_MAX) { //mtvc
R(rt) = currentMIPS->vfpuCtrl[imm - 128];
} else {
//ERROR - maybe need to make this value too an "interlock" value?
_dbg_assert_msg_(CPU,0,"mfv - invalid register");
}
2012-11-22 19:14:24 +00:00
}
2012-11-01 15:19:01 +00:00
break;
2012-11-01 15:19:01 +00:00
case 7: //mtv
2012-11-22 19:14:24 +00:00
if (imm < 128) {
VI(imm) = R(rt);
} else if (imm < 128 + VFPU_CTRL_MAX) { //mtvc
2012-11-22 19:14:24 +00:00
currentMIPS->vfpuCtrl[imm - 128] = R(rt);
} else {
//ERROR
_dbg_assert_msg_(CPU,0,"mtv - invalid register");
}
2012-11-01 15:19:01 +00:00
break;
2012-11-01 15:19:01 +00:00
default:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
break;
}
PC += 4;
}
void Int_Vmfvc(MIPSOpcode op) {
int vs = _VS;
int imm = op & 0xFF;
if (imm >= 128 && imm < 128 + VFPU_CTRL_MAX) {
VI(vs) = currentMIPS->vfpuCtrl[imm - 128];
}
PC += 4;
}
void Int_Vmtvc(MIPSOpcode op) {
int vs = _VS;
int imm = op & 0xFF;
if (imm >= 128 && imm < 128 + VFPU_CTRL_MAX) {
currentMIPS->vfpuCtrl[imm - 128] = VI(vs);
}
PC += 4;
}
void Int_Vcst(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int conNum = (op >> 16) & 0x1f;
int vd = _VD;
VectorSize sz = GetVecSize(op);
float c = cst_constants[conNum];
2012-11-01 15:19:01 +00:00
float temp[4] = {c,c,c,c};
ApplyPrefixD(temp, sz);
WriteVector(temp, sz, vd);
2012-11-01 15:19:01 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vcmp(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int vs = _VS;
2012-11-22 19:14:24 +00:00
int vt = _VT;
int cond = op & 0xf;
2012-11-01 15:19:01 +00:00
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
float s[4];
float t[4];
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
ReadVector(t, sz, vt);
ApplySwizzleT(t, sz);
int cc = 0;
int or_val = 0;
int and_val = 1;
int affected_bits = (1 << 4) | (1 << 5); // 4 and 5
2012-11-01 15:19:01 +00:00
for (int i = 0; i < n; i++)
{
int c;
2012-11-18 22:35:02 +00:00
// These set c to 0 or 1, nothing else.
2012-11-01 15:19:01 +00:00
switch (cond)
{
2012-11-22 19:14:24 +00:00
case VC_FL: c = 0; break;
2012-11-06 18:33:56 +00:00
case VC_EQ: c = s[i] == t[i]; break;
2012-11-01 15:19:01 +00:00
case VC_LT: c = s[i] < t[i]; break;
case VC_LE: c = s[i] <= t[i]; break;
2012-11-22 19:14:24 +00:00
2012-11-01 15:19:01 +00:00
case VC_TR: c = 1; break;
case VC_NE: c = s[i] != t[i]; break;
case VC_GE: c = s[i] >= t[i]; break;
2012-11-22 19:14:24 +00:00
case VC_GT: c = s[i] > t[i]; break;
case VC_EZ: c = s[i] == 0.0f || s[i] == -0.0f; break;
case VC_EN: c = my_isnan(s[i]); break;
case VC_EI: c = my_isinf(s[i]); break;
case VC_ES: c = my_isnanorinf(s[i]); break; // Tekken Dark Resurrection
2012-11-22 19:14:24 +00:00
case VC_NZ: c = s[i] != 0; break;
case VC_NN: c = !my_isnan(s[i]); break;
case VC_NI: c = !my_isinf(s[i]); break;
case VC_NS: c = !(my_isnanorinf(s[i])); break; // How about t[i] ?
2013-02-24 23:07:29 +00:00
2012-11-01 15:19:01 +00:00
default:
2012-11-08 13:23:55 +00:00
_dbg_assert_msg_(CPU,0,"Unsupported vcmp condition code %d", cond);
2012-11-22 19:14:24 +00:00
PC += 4;
2013-02-24 23:07:29 +00:00
EatPrefixes();
2012-11-01 15:19:01 +00:00
return;
}
cc |= (c<<i);
or_val |= c;
and_val &= c;
affected_bits |= 1 << i;
2012-11-01 15:19:01 +00:00
}
// Use masking to only change the affected bits
2012-11-18 22:35:02 +00:00
currentMIPS->vfpuCtrl[VFPU_CTRL_CC] =
(currentMIPS->vfpuCtrl[VFPU_CTRL_CC] & ~affected_bits) |
((cc | (or_val << 4) | (and_val << 5)) & affected_bits);
2012-11-01 15:19:01 +00:00
PC += 4;
EatPrefixes();
}
void Int_Vminmax(MIPSOpcode op) {
int vt = _VT;
int vs = _VS;
int vd = _VD;
int cond = op&15;
VectorSize sz = GetVecSize(op);
int numElements = GetNumVectorElements(sz);
FloatBits s;
FloatBits t;
FloatBits d;
ReadVector(s.f, sz, vs);
ApplySwizzleS(s.f, sz);
ReadVector(t.f, sz, vt);
ApplySwizzleT(t.f, sz);
// If both are zero, take t's sign.
// Otherwise: -NAN < -INF < real < INF < NAN (higher mantissa is farther from 0.)
switch ((op >> 23) & 3) {
case 2: // vmin
for (int i = 0; i < numElements; i++) {
if (my_isnanorinf(s.f[i]) || my_isnanorinf(t.f[i])) {
// If both are negative, we flip the comparison (not two's compliment.)
if (s.i[i] < 0 && t.i[i] < 0) {
// If at least one side is NAN, we take the highest mantissa bits.
d.i[i] = std::max(t.i[i], s.i[i]);
} else {
// Otherwise, we take the lowest value (negative or lowest mantissa.)
d.i[i] = std::min(t.i[i], s.i[i]);
}
} else {
d.f[i] = std::min(t.f[i], s.f[i]);
}
}
break;
case 3: // vmax
for (int i = 0; i < numElements; i++) {
// This is the same logic as vmin, just reversed.
if (my_isnanorinf(s.f[i]) || my_isnanorinf(t.f[i])) {
if (s.i[i] < 0 && t.i[i] < 0) {
d.i[i] = std::min(t.i[i], s.i[i]);
} else {
d.i[i] = std::max(t.i[i], s.i[i]);
}
} else {
d.f[i] = std::max(t.f[i], s.f[i]);
}
}
break;
default:
_dbg_assert_msg_(CPU,0,"unknown min/max op %d", cond);
2013-08-11 23:28:20 +00:00
PC += 4;
EatPrefixes();
return;
}
ApplyPrefixD(d.f, sz);
WriteVector(d.f, sz, vd);
PC += 4;
EatPrefixes();
}
2012-12-20 13:39:10 +00:00
// This doesn't quite pass all the tests :/
void Int_Vscmp(MIPSOpcode op) {
2012-12-20 13:39:10 +00:00
int vt = _VT;
int vs = _VS;
int vd = _VD;
VectorSize sz = GetVecSize(op);
float s[4];
float t[4];
float d[4];
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
ReadVector(t, sz, vt);
ApplySwizzleT(t, sz);
int n = GetNumVectorElements(sz);
for (int i = 0; i < n ; i++) {
float a = s[i] - t[i];
d[i] = (float) ((0.0 < a) - (a < 0.0));
2012-12-20 13:39:10 +00:00
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2013-03-24 13:28:32 +00:00
void Int_Vsge(MIPSOpcode op) {
int vt = _VT;
int vs = _VS;
int vd = _VD;
int cond = op&15;
VectorSize sz = GetVecSize(op);
int numElements = GetNumVectorElements(sz);
float s[4];
float t[4];
float d[4];
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
ReadVector(t, sz, vt);
ApplySwizzleT(t, sz);
2013-01-29 01:15:23 +00:00
for (int i = 0; i < GetNumVectorElements(sz); i++) {
if ( my_isnan(s[i]) || my_isnan(t[i]) )
2013-01-29 01:15:23 +00:00
d[i] = 0.0f;
else
d[i] = s[i] >= t[i] ? 1.0f : 0.0f;
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vslt(MIPSOpcode op) {
int vt = _VT;
int vs = _VS;
int vd = _VD;
int cond = op&15;
VectorSize sz = GetVecSize(op);
int numElements = GetNumVectorElements(sz);
float s[4];
float t[4];
float d[4];
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
ReadVector(t, sz, vt);
ApplySwizzleT(t, sz);
2013-01-29 01:15:23 +00:00
for (int i = 0; i < GetNumVectorElements(sz); i++) {
if ( my_isnan(s[i]) || my_isnan(t[i]) )
2013-01-29 01:15:23 +00:00
d[i] = 0.0f;
else
d[i] = s[i] < t[i] ? 1.0f : 0.0f;
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vcmov(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int vs = _VS;
2013-03-24 13:28:32 +00:00
int vd = _VD;
int tf = (op >> 19) & 1;
int imm3 = (op >> 16) & 7;
2012-11-01 15:19:01 +00:00
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
float s[4];
float d[4];
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
ReadVector(d, sz, vd); //Yes!
2012-11-01 15:19:01 +00:00
int CC = currentMIPS->vfpuCtrl[VFPU_CTRL_CC];
2012-11-11 21:38:19 +00:00
if (imm3 < 6)
2012-11-01 15:19:01 +00:00
{
2012-11-11 21:38:19 +00:00
if (((CC >> imm3) & 1) == !tf)
2012-11-01 15:19:01 +00:00
{
for (int i = 0; i < n; i++)
d[i] = s[i];
}
}
else if (imm3 == 6)
{
for (int i = 0; i < n; i++)
{
if (((CC >> i) & 1) == !tf)
2012-11-01 15:19:01 +00:00
d[i] = s[i];
}
}
2013-03-24 13:28:32 +00:00
else
2012-11-01 15:19:01 +00:00
{
ERROR_LOG_REPORT(CPU, "Bad Imm3 in cmov: %d", imm3);
2012-11-01 15:19:01 +00:00
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_VecDo3(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
float s[4];
float t[4];
float d[4];
ReadVector(s, sz, vs);
ApplySwizzleS(s, sz);
ReadVector(t, sz, vt);
ApplySwizzleT(t, sz);
for (int i = 0; i < GetNumVectorElements(sz); i++)
{
switch(op >> 26)
{
case 24: //VFPU0
switch ((op >> 23)&7)
{
case 0: d[i] = s[i] + t[i]; break; //vadd
case 1: d[i] = s[i] - t[i]; break; //vsub
case 7: d[i] = s[i] / t[i]; break; //vdiv
default: goto bad;
}
break;
case 25: //VFPU1
switch ((op >> 23)&7)
{
case 0: d[i] = s[i] * t[i]; break; //vmul
default: goto bad;
}
break;
default:
bad:
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
break;
}
}
ApplyPrefixD(d, sz);
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
2012-11-11 21:38:19 +00:00
void Int_CrossQuat(MIPSOpcode op)
2012-11-01 15:19:01 +00:00
{
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
float s[4];
float t[4];
float d[4];
ReadVector(s, sz, vs);
ReadVector(t, sz, vt);
switch (sz)
{
case V_Triple: // vcrsp.t
2012-11-01 15:19:01 +00:00
d[0] = s[1]*t[2] - s[2]*t[1];
d[1] = s[2]*t[0] - s[0]*t[2];
d[2] = s[0]*t[1] - s[1]*t[0];
break;
case V_Quad: // vqmul.q
d[0] = s[0]*t[3] + s[1]*t[2] - s[2]*t[1] + s[3]*t[0];
d[1] = -s[0]*t[2] + s[1]*t[3] + s[2]*t[0] + s[3]*t[1];
d[2] = s[0]*t[1] - s[1]*t[0] + s[2]*t[3] + s[3]*t[2];
d[3] = -s[0]*t[0] - s[1]*t[1] - s[2]*t[2] + s[3]*t[3];
break;
2012-11-01 15:19:01 +00:00
default:
Reporting::ReportMessage("CrossQuat instruction with wrong size");
2012-11-01 15:19:01 +00:00
_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
break;
}
WriteVector(d, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vlgb(MIPSOpcode op)
{
// S & D valid
Reporting::ReportMessage("vlgb not implemented");
if (!PSP_CoreParameter().headLess) {
_dbg_assert_msg_(CPU,0,"vlgb not implemented");
}
PC += 4;
EatPrefixes();
}
// There has to be a concise way of expressing this in terms of
// bit manipulation on the raw floats.
void Int_Vwbn(MIPSOpcode op) {
int vd = _VD;
int vs = _VS;
VectorSize sz = GetVecSize(op);
FloatBits d;
FloatBits s;
u8 exp = (u8)((op >> 16) & 0xFF);
ReadVector(s.f, sz, vs);
// TODO: Test swizzle, t?
ApplySwizzleS(s.f, sz);
if (sz != V_Single) {
ERROR_LOG_REPORT(CPU, "vwbn not implemented for size %d", GetNumVectorElements(sz));
}
for (int i = 0; i < GetNumVectorElements(sz); ++i) {
u32 sigbit = s.u[i] & 0x80000000;
u32 prevExp = (s.u[i] & 0x7F800000) >> 23;
2014-02-16 08:00:29 +00:00
u32 mantissa = (s.u[i] & 0x007FFFFF) | 0x00800000;
if (prevExp != 0xFF && prevExp != 0) {
if (exp > prevExp) {
s8 shift = (exp - prevExp) & 0xF;
mantissa = mantissa >> shift;
} else {
s8 shift = (prevExp - exp) & 0xF;
mantissa = mantissa << shift;
}
d.u[i] = sigbit | (mantissa & 0x007FFFFF) | (exp << 23);
} else {
d.u[i] = s.u[i] | (exp << 23);
}
}
ApplyPrefixD(d.f, sz);
WriteVector(d.f, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vsbn(MIPSOpcode op)
{
int vd = _VD;
int vs = _VS;
int vt = _VT;
VectorSize sz = GetVecSize(op);
FloatBits d;
FloatBits s;
u8 exp = (u8)(127 + VI(vt));
ReadVector(s.f, sz, vs);
// TODO: Test swizzle, t?
ApplySwizzleS(s.f, sz);
if (sz != V_Single) {
ERROR_LOG_REPORT(CPU, "vsbn not implemented for size %d", GetNumVectorElements(sz));
}
for (int i = 0; i < GetNumVectorElements(sz); ++i) {
// Simply replace the exponent bits.
u32 prev = s.u[i] & 0x7F800000;
if (prev != 0 && prev != 0x7F800000) {
d.u[i] = (s.u[i] & ~0x7F800000) | (exp << 23);
} else {
d.u[i] = s.u[i];
}
}
ApplyPrefixD(d.f, sz);
WriteVector(d.f, sz, vd);
PC += 4;
EatPrefixes();
}
void Int_Vsbz(MIPSOpcode op)
{
Reporting::ReportMessage("vsbz not implemented");
if (!PSP_CoreParameter().headLess) {
_dbg_assert_msg_(CPU,0,"vsbz not implemented");
}
PC += 4;
EatPrefixes();
}
2012-11-01 15:19:01 +00:00
}