NearestFunc SamplerJitCache::Compile(const SamplerID &id) {
BeginWrite();
const u8 *start = AlignCode16();
// Early exit on !srcPtr.
FixupBranch zeroSrc;
if (id.hasInvalidPtr) {
CMP(PTRBITS, R(srcReg), Imm8(0));
FixupBranch nonZeroSrc = J_CC(CC_NZ);
XOR(32, R(RAX), R(RAX));
zeroSrc = J(true);
SetJumpTarget(nonZeroSrc);
}
if (!Jit_ReadTextureFormat(id)) {
EndWrite();
SetCodePtr(const_cast<u8 *>(start));
return nullptr;
}
if (id.hasInvalidPtr) {
SetJumpTarget(zeroSrc);
}
RET();
EndWrite();
return (NearestFunc)start;
}
CLog& CLog::operator << (LOGLEVEL level)
{
if (level <= LOGLEVEL_NONE || level >= LOGLEVEL_MAX)
{
return *this;
}
//Lock(&m_StreamLock);
m_nLogLevel = level;
int nLen = strlen(m_szBuff);
if (nLen + 40 >= MAX_LOG)
{
EndWrite();
return *this;
}
time_t now = time(NULL);
tm* pNow = localtime(&now);
if (m_nMode & LOGMODE_YEAR)
sprintf(m_szBuff, "%s[%04d-%02d-%02d %02d:%02d:%02d]", m_LogLevelString[level],
1900 + pNow->tm_year, 1 + pNow->tm_mon, pNow->tm_mday, pNow->tm_hour, pNow->tm_min, pNow->tm_sec);
else
sprintf(m_szBuff, "%s[%02d-%02d %02d:%02d:%02d]", m_LogLevelString[level],
1 + pNow->tm_mon, pNow->tm_mday, pNow->tm_hour, pNow->tm_min, pNow->tm_sec);
return *this;
}
CLog& CLog::operator<< (std::ostream& (*op) (std::ostream&))
{
int nLen = strlen(m_szBuff);
if (nLen + 1 < MAX_LOG)
{
sprintf(m_szBuff + nLen, "\n");
}
EndWrite();
return *this;
}
CLog& CLog::operator << (unsigned short n)
{
int nLen = strlen(m_szBuff);
if (nLen + 6 >= MAX_LOG)
{
EndWrite();
return *this;
}
sprintf(m_szBuff + nLen, "%hu", n);
return *this;
}
CLog& CLog::operator << (unsigned char c)
{
int nLen = strlen(m_szBuff);
if (nLen + 2 >= MAX_LOG)
{
EndWrite();
return *this;
}
sprintf(m_szBuff + nLen, "%c", c);
return *this;
}
CLog& CLog::operator << (float f)
{
int nLen = strlen(m_szBuff);
if (nLen + 8 >= MAX_LOG)
{
EndWrite();
return *this;
}
sprintf(m_szBuff + nLen, "%f", f);
return *this;
}
/*!
\fn swLabel::_renderText( const std::String& str )
*/
bool swLabel::_renderText( const std::string& str )
{
swWriter* sw = StartWrite();
sw->Clear();
sw->Position(0,0);
/// @todo Have to render the text inside the label from the rules{ justification|wrapping|clipping } for now just write the text
(*sw) << swWriter::text << str;
EndWrite();
return true;
}
CLog& CLog::operator << (long n)
{
int nLen = strlen(m_szBuff);
if (nLen + 11 >= MAX_LOG)
{
EndWrite();
return *this;
}
sprintf(m_szBuff + nLen, "%ld", n);
return *this;
}
CLog& CLog::operator << (const int* p)
{
int nLen = strlen(m_szBuff);
if (nLen + 11 >= MAX_LOG)
{
EndWrite();
return *this;
}
sprintf(m_szBuff + nLen, "%p", p);
return *this;
}
CLog& CLog::operator << (double d)
{
int nLen = strlen(m_szBuff);
if (nLen + 20 >= MAX_LOG)
{
EndWrite();
return *this;
}
sprintf(m_szBuff + nLen, "%lf", d);
return *this;
}
CLog& CLog::operator << (const std::string& str)
{
int nLen = strlen(m_szBuff);
int nInputLen = str.size();
if (nLen + nInputLen >= MAX_LOG)
{
EndWrite();
return *this;
}
sprintf(m_szBuff + nLen, "%s", str.c_str());
return *this;
}
CLog& CLog::operator << (const char* str)
{
int nLen = strlen(m_szBuff);
int nInputLen = strlen(str);
if (nLen + nInputLen >= MAX_LOG)
{
EndWrite();
return *this;
}
sprintf(m_szBuff + nLen, "%s", str);
return *this;
}
void CPackFiles::Kill(void)
{
if (meMode == PFM_Read)
{
EndRead();
}
else if (meMode == PFM_Write)
{
EndWrite();
}
mcFile.Kill();
mcNames.Kill();
}
bool Pickle::WriteBytes(const void* data, int data_len)
{
DCHECK(capacity_ != kCapacityReadOnly) << "oops: pickle is readonly";
char* dest = BeginWrite(data_len);
if(!dest)
{
return false;
}
memcpy(dest, data, data_len);
EndWrite(dest, data_len);
return true;
}
CLog& CLog::operator << (bool b)
{
int nLen = strlen(m_szBuff);
if (nLen + 6 >= MAX_LOG)
{
EndWrite();
return *this;
}
if (b)
{
sprintf(m_szBuff + nLen, "true");
}
else
{
sprintf(m_szBuff + nLen, "false");
}
return *this;
}
char* Pickle::BeginWriteData(int length)
{
DCHECK_EQ(variable_buffer_offset_, 0U) <<
"There can only be one variable buffer in a Pickle";
if(length<0 || !WriteInt(length))
{
return NULL;
}
char* data_ptr = BeginWrite(length);
if(!data_ptr)
{
return NULL;
}
variable_buffer_offset_ = data_ptr -
reinterpret_cast<char*>(header_) - sizeof(int);
// 数据写入后不必要再调用EndWrite, 所以在这里调用进行数据对齐.
EndWrite(data_ptr, length);
return data_ptr;
}
HRESULT CGraphics::Render(void)
{
if(!m_Device)
return E_FAIL;
if(m_NeedReset)
return S_FALSE;
HRESULT hr;
if(FAILED(hr = m_Device->TestCooperativeLevel()))
{
TRACE(TEXT("Warning: Failed on test cooperative level.\n"));
// The device has been lost but cannot be reset at this time.
// Therefore, rendering is not possible and we'll have to return
// and try again at a later time.
if(hr == D3DERR_DEVICELOST)
{
TRACE(TEXT("Warning: Display device lost.\n"));
return S_FALSE;
}
// The device has been lost but it can be reset at this time.
if(hr == D3DERR_DEVICENOTRESET)
{
// Try to reset the device
m_NeedReset = TRUE;
return S_FALSE;
}
}
if(SUCCEEDED(m_Device->BeginScene()))
{
if(SUCCEEDED(BeginWrite()))
{
//m_Direct3DBeginEvent(NULL,TEXT("Rendering"));
if(!m_Blur)
{
m_Device->Clear(NULL,NULL,D3DCLEAR_TARGET,m_Background,1.0f,NULL);
if(GetVisualizationCount() && GetVisualization(m_Visualization))
GetVisualization(m_Visualization)->Draw(*this);
}
else
{
LPDIRECT3DSURFACE9 backbufferTarget = NULL;
m_Device->GetRenderTarget(0,&backbufferTarget);
LPDIRECT3DSURFACE9 blurTarget = NULL;
m_BlurTexture->GetSurfaceLevel(0,&blurTarget);
m_Device->SetRenderTarget(0,blurTarget);
m_Device->Clear(NULL,NULL,D3DCLEAR_TARGET,Color(0.0f,0.0f,0.0f,0.0f),1.0f,NULL);
// Maybe this is faster
//SpriteWrite(Rect(0.0f,0.0f,GetWidth()/2,GetHeight()/2),Color(1.0f,0.0f,0.0f,0.0f),NULL);
SpriteWrite(Rect(0.0f,0.0f,GetWidth()/2,GetHeight()/2),Color(1.0f,1.0f,1.0f,1.0f),m_VisualizationTexture);
SpriteFlush();
LineFlush();
LPDIRECT3DSURFACE9 visualizationTarget = NULL;
m_VisualizationTexture->GetSurfaceLevel(0,&visualizationTarget);
m_Device->SetRenderTarget(0,visualizationTarget);
SpriteWrite(Rect(0.0f,0.0f,GetWidth(),GetHeight()),Color(1.0f,1.0f,1.0f,1.0f),m_BlurTexture);
SpriteFlush();
m_Device->SetRenderState(D3DRS_SRCBLEND,D3DBLEND_SRCALPHA);
m_Device->SetRenderState(D3DRS_DESTBLEND,D3DBLEND_ONE);
m_Device->SetRenderState(D3DRS_BLENDOP,D3DBLENDOP_REVSUBTRACT);
SpriteWrite(Rect(0.0f,0.0f,GetWidth(),GetHeight()),Color(max(0.0f,min(1.0f,GetElapsedTime()/50.0f)),1.0f,1.0f,1.0f),NULL);
SpriteFlush();
m_Device->SetRenderState(D3DRS_SRCBLEND,D3DBLEND_SRCALPHA);
m_Device->SetRenderState(D3DRS_DESTBLEND,D3DBLEND_INVSRCALPHA);
m_Device->SetRenderState(D3DRS_BLENDOP,D3DBLENDOP_ADD);
if(GetVisualizationCount() && GetVisualization(m_Visualization))
GetVisualization(m_Visualization)->Draw(*this);
SpriteFlush();
LineFlush();
m_Device->SetRenderTarget(0,backbufferTarget);
visualizationTarget->Release();
blurTarget->Release();
backbufferTarget->Release();
SpriteWrite(Rect(0.0f,0.0f,GetWidth(),GetHeight()),Color(1.0f,1.0f,1.0f,1.0f),m_VisualizationTexture);
}
EndWrite();
//m_Direct3DEndEvent();
}
else
{
TRACE(TEXT("Error: Failed to begin sprite write.\n"));
}
m_Device->EndScene();
}
else
{
TRACE(TEXT("Error: Failed to begin scene.\n"));
}
if(FAILED(m_Device->Present(NULL,NULL,NULL,NULL)))
{
TRACE(TEXT("Error: Failed to present scene.\n"));
}
return S_OK;
}
LinearFunc SamplerJitCache::CompileLinear(const SamplerID &id) {
_assert_msg_(G3D, id.linear, "Linear should be set on sampler id");
BeginWrite();
// We'll first write the nearest sampler, which we will CALL.
// This may differ slightly based on the "linear" flag.
const u8 *nearest = AlignCode16();
if (!Jit_ReadTextureFormat(id)) {
EndWrite();
SetCodePtr(const_cast<u8 *>(nearest));
return nullptr;
}
RET();
// Now the actual linear func, which is exposed externally.
const u8 *start = AlignCode16();
// NOTE: This doesn't use the general register mapping.
// POSIX: arg1=uptr, arg2=vptr, arg3=frac_u, arg4=frac_v, arg5=src, arg6=bufw, stack+8=level
// Win64: arg1=uptr, arg2=vptr, arg3=frac_u, arg4=frac_v, stack+40=src, stack+48=bufw, stack+56=level
//
// We map these to nearest CALLs, with order: u, v, src, bufw, level
// Let's start by saving a bunch of registers.
PUSH(R15);
PUSH(R14);
PUSH(R13);
PUSH(R12);
// Won't need frac_u/frac_v for a while.
PUSH(arg4Reg);
PUSH(arg3Reg);
// Extra space to restore alignment and save resultReg for lerp.
// TODO: Maybe use XMMs instead?
SUB(64, R(RSP), Imm8(24));
MOV(64, R(R12), R(arg1Reg));
MOV(64, R(R13), R(arg2Reg));
#ifdef _WIN32
// First arg now starts at 24 (extra space) + 48 (pushed stack) + 8 (ret address) + 32 (shadow space)
const int argOffset = 24 + 48 + 8 + 32;
MOV(64, R(R14), MDisp(RSP, argOffset));
MOV(32, R(R15), MDisp(RSP, argOffset + 8));
// level is at argOffset + 16.
#else
MOV(64, R(R14), R(arg5Reg));
MOV(32, R(R15), R(arg6Reg));
// level is at 24 + 48 + 8.
#endif
// Early exit on !srcPtr.
FixupBranch zeroSrc;
if (id.hasInvalidPtr) {
CMP(PTRBITS, R(R14), Imm8(0));
FixupBranch nonZeroSrc = J_CC(CC_NZ);
XOR(32, R(RAX), R(RAX));
zeroSrc = J(true);
SetJumpTarget(nonZeroSrc);
}
// At this point:
// R12=uptr, R13=vptr, stack+24=frac_u, stack+32=frac_v, R14=src, R15=bufw, stack+X=level
auto doNearestCall = [&](int off) {
MOV(32, R(uReg), MDisp(R12, off));
MOV(32, R(vReg), MDisp(R13, off));
MOV(64, R(srcReg), R(R14));
MOV(32, R(bufwReg), R(R15));
// Leave level, we just always load from RAM. Separate CLUTs is uncommon.
CALL(nearest);
MOV(32, MDisp(RSP, off), R(resultReg));
};
doNearestCall(0);
doNearestCall(4);
doNearestCall(8);
doNearestCall(12);
// Convert TL, TR, BL, BR to floats for easier blending.
if (!cpu_info.bSSE4_1) {
PXOR(XMM0, R(XMM0));
}
MOVD_xmm(fpScratchReg1, MDisp(RSP, 0));
MOVD_xmm(fpScratchReg2, MDisp(RSP, 4));
MOVD_xmm(fpScratchReg3, MDisp(RSP, 8));
MOVD_xmm(fpScratchReg4, MDisp(RSP, 12));
if (cpu_info.bSSE4_1) {
PMOVZXBD(fpScratchReg1, R(fpScratchReg1));
PMOVZXBD(fpScratchReg2, R(fpScratchReg2));
PMOVZXBD(fpScratchReg3, R(fpScratchReg3));
PMOVZXBD(fpScratchReg4, R(fpScratchReg4));
} else {
PUNPCKLBW(fpScratchReg1, R(XMM0));
PUNPCKLBW(fpScratchReg2, R(XMM0));
PUNPCKLBW(fpScratchReg3, R(XMM0));
PUNPCKLBW(fpScratchReg4, R(XMM0));
PUNPCKLWD(fpScratchReg1, R(XMM0));
PUNPCKLWD(fpScratchReg2, R(XMM0));
PUNPCKLWD(fpScratchReg3, R(XMM0));
PUNPCKLWD(fpScratchReg4, R(XMM0));
}
CVTDQ2PS(fpScratchReg1, R(fpScratchReg1));
CVTDQ2PS(fpScratchReg2, R(fpScratchReg2));
CVTDQ2PS(fpScratchReg3, R(fpScratchReg3));
CVTDQ2PS(fpScratchReg4, R(fpScratchReg4));
// Okay, now multiply the R sides by frac_u, and L by (256 - frac_u)...
MOVD_xmm(fpScratchReg5, MDisp(RSP, 24));
CVTDQ2PS(fpScratchReg5, R(fpScratchReg5));
SHUFPS(fpScratchReg5, R(fpScratchReg5), _MM_SHUFFLE(0, 0, 0, 0));
if (RipAccessible(by256)) {
MULPS(fpScratchReg5, M(by256)); // rip accessible
} else {
Crash(); // TODO
}
MOVAPS(XMM0, M(ones));
SUBPS(XMM0, R(fpScratchReg5));
MULPS(fpScratchReg1, R(XMM0));
MULPS(fpScratchReg2, R(fpScratchReg5));
MULPS(fpScratchReg3, R(XMM0));
MULPS(fpScratchReg4, R(fpScratchReg5));
// Now set top=fpScratchReg1, bottom=fpScratchReg3.
ADDPS(fpScratchReg1, R(fpScratchReg2));
ADDPS(fpScratchReg3, R(fpScratchReg4));
// Next, time for frac_v.
MOVD_xmm(fpScratchReg5, MDisp(RSP, 32));
CVTDQ2PS(fpScratchReg5, R(fpScratchReg5));
SHUFPS(fpScratchReg5, R(fpScratchReg5), _MM_SHUFFLE(0, 0, 0, 0));
MULPS(fpScratchReg5, M(by256));
MOVAPS(XMM0, M(ones));
SUBPS(XMM0, R(fpScratchReg5));
MULPS(fpScratchReg1, R(XMM0));
MULPS(fpScratchReg3, R(fpScratchReg5));
// Still at the 255 scale, now we're interpolated.
ADDPS(fpScratchReg1, R(fpScratchReg3));
// Time to convert back to a single 32 bit value.
CVTPS2DQ(fpScratchReg1, R(fpScratchReg1));
PACKSSDW(fpScratchReg1, R(fpScratchReg1));
PACKUSWB(fpScratchReg1, R(fpScratchReg1));
MOVD_xmm(R(resultReg), fpScratchReg1);
if (id.hasInvalidPtr) {
SetJumpTarget(zeroSrc);
}
ADD(64, R(RSP), Imm8(24));
POP(arg3Reg);
POP(arg4Reg);
POP(R12);
POP(R13);
POP(R14);
POP(R15);
RET();
EndWrite();
return (LinearFunc)start;
}
