void GSTextureCacheOGL::Read(Source* t, const GSVector4i& r)
{
const GIFRegTEX0& TEX0 = t->m_TEX0;
// FIXME Create a get function to avoid the useless copy
// Note: With openGL 4.5 you can use glGetTextureSubImage
if (GSTexture* offscreen = m_renderer->m_dev->CreateOffscreen(r.width(), r.height())) {
m_renderer->m_dev->CopyRect(t->m_texture, offscreen, r);
GSTexture::GSMap m;
GSVector4i r_offscreen(0, 0, r.width(), r.height());
if (offscreen->Map(m, &r_offscreen)) {
GSOffset* off = m_renderer->m_mem.GetOffset(TEX0.TBP0, TEX0.TBW, TEX0.PSM);
m_renderer->m_mem.WritePixel32(m.bits, m.pitch, off, r);
offscreen->Unmap();
}
// FIXME invalidate data
m_renderer->m_dev->Recycle(offscreen);
}
}
void GSDevice::Present(const GSVector4i& r, int shader)
{
GSVector4i cr = m_wnd->GetClientRect();
int w = std::max<int>(cr.width(), 1);
int h = std::max<int>(cr.height(), 1);
if(!m_backbuffer || m_backbuffer->GetWidth() != w || m_backbuffer->GetHeight() != h)
{
if(!Reset(w, h))
{
return;
}
}
GL_PUSH("Present");
ClearRenderTarget(m_backbuffer, 0);
if(m_current)
{
static int s_shader[5] = {ShaderConvert_COPY, ShaderConvert_SCANLINE,
ShaderConvert_DIAGONAL_FILTER, ShaderConvert_TRIANGULAR_FILTER,
ShaderConvert_COMPLEX_FILTER}; // FIXME
Present(m_current, m_backbuffer, GSVector4(r), s_shader[shader]);
}
Flip();
}
void GSDevice::Present(const GSVector4i& r, int shader)
{
GSVector4i cr = m_wnd->GetClientRect();
int w = std::max<int>(cr.width(), 1);
int h = std::max<int>(cr.height(), 1);
if(!m_backbuffer || m_backbuffer->GetWidth() != w || m_backbuffer->GetHeight() != h)
{
if(!Reset(w, h))
{
return;
}
}
GL_PUSH("Present");
ClearRenderTarget(m_backbuffer, 0);
if(m_current)
{
static int s_shader[5] = {0, 5, 6, 8, 9}; // FIXME
Present(m_current, m_backbuffer, GSVector4(r), s_shader[shader]);
}
Flip();
GL_POP();
}
bool GSRenderer::BeginCapture()
{
GSVector4i disp = m_wnd->GetClientRect().fit(m_aspectratio);
float aspect = (float)disp.width() / max(1, disp.height());
return m_capture.BeginCapture(GetTvRefreshRate(), GetInternalResolution(), aspect);
}
bool GSTexture9::Update(const GSVector4i& r, const void* data, int pitch, int layer)
{
if(m_surface)
{
D3DLOCKED_RECT lr;
if(SUCCEEDED(m_surface->LockRect(&lr, r, 0)))
{
uint8* src = (uint8*)data;
uint8* dst = (uint8*)lr.pBits;
int bytes = r.width() * sizeof(uint32);
switch(m_desc.Format)
{
case D3DFMT_A8: bytes >>= 2; break;
case D3DFMT_A1R5G5B5: bytes >>= 1; break;
default: ASSERT(m_desc.Format == D3DFMT_A8R8G8B8); break;
}
bytes = std::min(bytes, pitch);
bytes = std::min(bytes, lr.Pitch);
for(int i = 0, j = r.height(); i < j; i++, src += pitch, dst += lr.Pitch)
{
memcpy(dst, src, bytes);
}
m_surface->UnlockRect();
return true;
}
}
bool GSWndDX::Create(const string& title, int w, int h)
{
if(m_hWnd) return false;
m_managed = true;
WNDCLASS wc;
memset(&wc, 0, sizeof(wc));
wc.style = CS_HREDRAW | CS_VREDRAW | CS_DBLCLKS;
wc.lpfnWndProc = WndProc;
wc.hInstance = theApp.GetModuleHandle();
// TODO: wc.hIcon = ;
wc.hCursor = LoadCursor(NULL, IDC_ARROW);
wc.hbrBackground = (HBRUSH)GetStockObject(BLACK_BRUSH);
wc.lpszClassName = "GSWndDX";
if(!GetClassInfo(wc.hInstance, wc.lpszClassName, &wc))
{
if(!RegisterClass(&wc))
{
return false;
}
}
DWORD style = WS_CLIPCHILDREN | WS_CLIPSIBLINGS | WS_OVERLAPPEDWINDOW | WS_BORDER;
GSVector4i r;
GetWindowRect(GetDesktopWindow(), r);
bool remote = !!GetSystemMetrics(SM_REMOTESESSION);
if(w <= 0 || h <= 0 || remote)
{
w = r.width() / 3;
h = r.width() / 4;
if(!remote)
{
w *= 2;
h *= 2;
}
}
r.left = (r.left + r.right - w) / 2;
r.top = (r.top + r.bottom - h) / 2;
r.right = r.left + w;
r.bottom = r.top + h;
AdjustWindowRect(r, style, FALSE);
m_hWnd = CreateWindow(wc.lpszClassName, title.c_str(), style, r.left, r.top, r.width(), r.height(), NULL, NULL, wc.hInstance, (LPVOID)this);
return m_hWnd != NULL;
}
bool GSTextureOGL::Map(GSMap& m, const GSVector4i* _r)
{
GSVector4i r = _r ? *_r : GSVector4i(0, 0, m_size.x, m_size.y);
// LOTS OF CRAP CODE!!!! PLEASE FIX ME !!!
if (m_type == GSTexture::Offscreen) {
// The fastest way will be to use a PBO to read the data asynchronously. Unfortunately GSdx
// architecture is waiting the data right now.
#if 0
// Maybe it is as good as the code below. I don't know
// With openGL 4.5 you can use glGetTextureSubImage
glGetTextureSubImage(m_texture_id, GL_TEX_LEVEL_0, r.x, r.y, 0, r.width(), r.height(), 0, m_int_format, m_int_type, m_size.x * m_size.y * 4, m_local_buffer);
#else
// Bind the texture to the read framebuffer to avoid any disturbance
glBindFramebuffer(GL_READ_FRAMEBUFFER, m_fbo_read);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, m_texture_id, 0);
glReadPixels(r.x, r.y, r.width(), r.height(), m_int_format, m_int_type, m_local_buffer);
glBindFramebuffer(GL_READ_FRAMEBUFFER, 0);
#endif
m.bits = m_local_buffer;
m.pitch = m_size.x << m_int_shift;
return true;
} else if (m_type == GSTexture::Texture || m_type == GSTexture::RenderTarget) {
GL_PUSH_("Upload Texture %d", m_texture_id); // POP is in Unmap
m_clean = false;
uint32 row_byte = r.width() << m_int_shift;
uint32 map_size = r.height() * row_byte;
m.bits = (uint8*)PboPool::Map(map_size);
m.pitch = row_byte;
#ifdef ENABLE_OGL_DEBUG_MEM_BW
g_real_texture_upload_byte += map_size;
#endif
// Save the area for the unmap
m_r_x = r.x;
m_r_y = r.y;
m_r_w = r.width();
m_r_h = r.height();
return true;
}
return false;
}
GSTexture* GPURendererSW::GetOutput()
{
GSVector4i r = m_env.GetDisplayRect();
r.left <<= m_scale.x;
r.top <<= m_scale.y;
r.right <<= m_scale.x;
r.bottom <<= m_scale.y;
if(m_dev->ResizeTexture(&m_texture, r.width(), r.height()))
{
m_mem.ReadFrame32(r, m_output, !!m_env.STATUS.ISRGB24);
m_texture->Update(r.rsize(), m_output, m_mem.GetWidth() * sizeof(uint32));
}
return m_texture;
}
void GPULocalMemory::FillRect(const GSVector4i& r, uint16 c)
{
Invalidate(r);
uint16* RESTRICT dst = GetPixelAddressScaled(r.left, r.top);
int w = r.width() << m_scale.x;
int h = r.height() << m_scale.y;
int pitch = GetWidth();
for(int j = 0; j < h; j++, dst += pitch)
{
for(int i = 0; i < w; i++)
{
dst[i] = c;
}
}
}
void GPUDrawScanline::SetupPrim(const GSVertexSW* vertex, const uint32* index, const GSVertexSW& dscan)
{
GPUScanlineSelector sel = m_global.sel;
const GSVector4* shift = GPUSetupPrimCodeGenerator::m_shift;
if(sel.tme && !sel.twin)
{
if(sel.sprite)
{
GSVector4i t = (GSVector4i(vertex[index[1]].t) >> 8) - GSVector4i::x00000001();
t = t.ps32(t);
t = t.upl16(t);
m_local.twin[2].u = t.xxxx();
m_local.twin[2].v = t.yyyy();
}
else
{
bool GSTextureOGL::Update(const GSVector4i& r, const void* data, int pitch)
{
ASSERT(m_type != GSTexture::DepthStencil && m_type != GSTexture::Offscreen);
GL_PUSH("Upload Texture %d", m_texture_id);
m_dirty = true;
m_clean = false;
glPixelStorei(GL_UNPACK_ALIGNMENT, m_int_alignment);
char* src = (char*)data;
uint32 row_byte = r.width() << m_int_shift;
uint32 map_size = r.height() * row_byte;
char* map = PboPool::Map(map_size);
#ifdef ENABLE_OGL_DEBUG_MEM_BW
g_real_texture_upload_byte += map_size;
#endif
// PERF: slow path of the texture upload. Dunno if we could do better maybe check if TC can keep row_byte == pitch
// Note: row_byte != pitch
for (int h = 0; h < r.height(); h++) {
memcpy(map, src, row_byte);
map += row_byte;
src += pitch;
}
PboPool::Unmap();
glTextureSubImage2D(m_texture_id, GL_TEX_LEVEL_0, r.x, r.y, r.width(), r.height(), m_int_format, m_int_type, (const void*)PboPool::Offset());
// FIXME OGL4: investigate, only 1 unpack buffer always bound
PboPool::UnbindPbo();
PboPool::EndTransfer();
GL_POP();
return true;
// For reference, standard upload without pbo (Used to crash on FGLRX)
#if 0
// pitch is in byte wherease GL_UNPACK_ROW_LENGTH is in pixel
glPixelStorei(GL_UNPACK_ALIGNMENT, m_int_alignment);
glPixelStorei(GL_UNPACK_ROW_LENGTH, pitch >> m_int_shift);
glTextureSubImage2D(m_texture_id, GL_TEX_LEVEL_0, r.x, r.y, r.width(), r.height(), m_int_format, m_int_type, data);
// FIXME useful?
glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); // Restore default behavior
return true;
#endif
}
bool GSTextureSW::Update(const GSVector4i& r, const void* data, int pitch)
{
GSMap m;
if(m_data != NULL && Map(m, &r))
{
uint8* RESTRICT src = (uint8*)data;
uint8* RESTRICT dst = m.bits;
int rowbytes = r.width() << 2;
for(int h = r.height(); h > 0; h--, src += pitch, dst += m.pitch)
{
memcpy(dst, src, rowbytes);
}
Unmap();
return true;
}
return false;
}
void GSTextureCache11::Read(Source* t, const GSVector4i& r)
{
// FIXME: copy was copyied from openGL. It is unlikely to work.
const GIFRegTEX0& TEX0 = t->m_TEX0;
if (GSTexture* offscreen = m_renderer->m_dev->CreateOffscreen(r.width(), r.height())) {
m_renderer->m_dev->CopyRect(t->m_texture, offscreen, r);
GSTexture::GSMap m;
GSVector4i r_offscreen(0, 0, r.width(), r.height());
if (offscreen->Map(m, &r_offscreen)) {
GSOffset* off = m_renderer->m_mem.GetOffset(TEX0.TBP0, TEX0.TBW, TEX0.PSM);
m_renderer->m_mem.WritePixel32(m.bits, m.pitch, off, r);
offscreen->Unmap();
}
// FIXME invalidate data
m_renderer->m_dev->Recycle(offscreen);
}
}
void GSTextureCacheOGL::Read(Target* t, const GSVector4i& r)
{
if (!t->m_dirty.empty() || r.width() == 0 || r.height() == 0)
return;
const GIFRegTEX0& TEX0 = t->m_TEX0;
GLuint fmt;
int ps_shader;
switch (TEX0.PSM)
{
case PSM_PSMCT32:
case PSM_PSMCT24:
fmt = GL_RGBA8;
ps_shader = ShaderConvert_COPY;
break;
case PSM_PSMCT16:
case PSM_PSMCT16S:
fmt = GL_R16UI;
ps_shader = ShaderConvert_RGBA8_TO_16_BITS;
break;
case PSM_PSMZ32:
fmt = GL_R32UI;
ps_shader = ShaderConvert_FLOAT32_TO_32_BITS;
break;
case PSM_PSMZ24:
fmt = GL_R32UI;
ps_shader = ShaderConvert_FLOAT32_TO_32_BITS;
break;
case PSM_PSMZ16:
case PSM_PSMZ16S:
fmt = GL_R16UI;
ps_shader = ShaderConvert_FLOAT32_TO_32_BITS;
break;
default:
return;
}
// Yes lots of logging, but I'm not confident with this code
GL_PUSH("Texture Cache Read. Format(0x%x)", TEX0.PSM);
GL_PERF("TC: Read Back Target: %d (0x%x)[fmt: 0x%x]. Size %dx%d",
t->m_texture->GetID(), TEX0.TBP0, TEX0.PSM, r.width(), r.height());
GSVector4 src = GSVector4(r) * GSVector4(t->m_texture->GetScale()).xyxy() / GSVector4(t->m_texture->GetSize()).xyxy();
if(GSTexture* offscreen = m_renderer->m_dev->CopyOffscreen(t->m_texture, src, r.width(), r.height(), fmt, ps_shader))
{
GSTexture::GSMap m;
GSVector4i r_offscreen(0, 0, r.width(), r.height());
if(offscreen->Map(m, &r_offscreen))
{
// TODO: block level write
GSOffset* off = m_renderer->m_mem.GetOffset(TEX0.TBP0, TEX0.TBW, TEX0.PSM);
switch(TEX0.PSM)
{
case PSM_PSMCT32:
m_renderer->m_mem.WritePixel32(m.bits, m.pitch, off, r);
break;
case PSM_PSMCT24:
m_renderer->m_mem.WritePixel24(m.bits, m.pitch, off, r);
break;
case PSM_PSMCT16:
case PSM_PSMCT16S:
m_renderer->m_mem.WritePixel16(m.bits, m.pitch, off, r);
break;
case PSM_PSMZ32:
m_renderer->m_mem.WritePixel32(m.bits, m.pitch, off, r);
break;
case PSM_PSMZ24:
m_renderer->m_mem.WritePixel24(m.bits, m.pitch, off, r);
break;
case PSM_PSMZ16:
case PSM_PSMZ16S:
m_renderer->m_mem.WritePixel16(m.bits, m.pitch, off, r);
break;
default:
ASSERT(0);
}
offscreen->Unmap();
}
// FIXME invalidate data
m_renderer->m_dev->Recycle(offscreen);
}
}
void GSRenderer::VSync(int field)
{
GSPerfMonAutoTimer pmat(&m_perfmon);
m_perfmon.Put(GSPerfMon::Frame);
Flush();
if(!m_dev->IsLost(true))
{
if(!Merge(field ? 1 : 0))
{
return;
}
}
else
{
ResetDevice();
}
m_dev->AgePool();
// osd
if((m_perfmon.GetFrame() & 0x1f) == 0)
{
m_perfmon.Update();
double fps = 1000.0f / m_perfmon.Get(GSPerfMon::Frame);
GSVector4i r = GetDisplayRect();
string s;
#ifdef GSTITLEINFO_API_FORCE_VERBOSE
if (1)//force verbose reply
#else
if (m_wnd->IsManaged())
#endif
{
//GSdx owns the window's title, be verbose.
string s2 = m_regs->SMODE2.INT ? (string("Interlaced ") + (m_regs->SMODE2.FFMD ? "(frame)" : "(field)")) : "Progressive";
s = format(
"%lld | %d x %d | %.2f fps (%d%%) | %s - %s | %s | %d/%d/%d | %d%% CPU | %.2f | %.2f",
m_perfmon.GetFrame(), r.width(), r.height(), fps, (int)(100.0 * fps / GetFPS()),
s2.c_str(),
theApp.m_gs_interlace[m_interlace].name.c_str(),
theApp.m_gs_aspectratio[m_aspectratio].name.c_str(),
(int)m_perfmon.Get(GSPerfMon::SyncPoint),
(int)m_perfmon.Get(GSPerfMon::Prim),
(int)m_perfmon.Get(GSPerfMon::Draw),
m_perfmon.CPU(),
m_perfmon.Get(GSPerfMon::Swizzle) / 1024,
m_perfmon.Get(GSPerfMon::Unswizzle) / 1024
);
double fillrate = m_perfmon.Get(GSPerfMon::Fillrate);
if(fillrate > 0)
{
s += format(" | %.2f mpps", fps * fillrate / (1024 * 1024));
int sum = 0;
for(int i = 0; i < 16; i++)
{
sum += m_perfmon.CPU(GSPerfMon::WorkerDraw0 + i);
}
s += format(" | %d%% CPU", sum);
}
}
else
{
// Satisfy PCSX2's request for title info: minimal verbosity due to more external title text
s = format("%dx%d | %s", r.width(), r.height(), theApp.m_gs_interlace[m_interlace].name.c_str());
}
if(m_capture.IsCapturing())
{
s += " | Recording...";
}
if(m_wnd->IsManaged())
{
m_wnd->SetWindowText(s.c_str());
}
else
{
// note: do not use TryEnterCriticalSection. It is unnecessary code complication in
// an area that absolutely does not matter (even if it were 100 times slower, it wouldn't
// be noticeable). Besides, these locks are extremely short -- overhead of conditional
// is way more expensive than just waiting for the CriticalSection in 1 of 10,000,000 tries. --air
GSAutoLock lock(&m_pGSsetTitle_Crit);
strncpy(m_GStitleInfoBuffer, s.c_str(), countof(m_GStitleInfoBuffer) - 1);
m_GStitleInfoBuffer[sizeof(m_GStitleInfoBuffer) - 1] = 0; // make sure null terminated even if text overflows
}
}
else
{
// [TODO]
// We don't have window title rights, or the window has no title,
// so let's use actual OSD!
}
if(m_frameskip)
{
return;
}
// present
m_dev->Present(m_wnd->GetClientRect().fit(m_aspectratio), m_shader);
// snapshot
if(!m_snapshot.empty())
{
bool shift = false;
#ifdef _WINDOWS
shift = !!(::GetAsyncKeyState(VK_SHIFT) & 0x8000);
#else
shift = m_shift_key;
#endif
if(!m_dump && shift)
{
GSFreezeData fd;
fd.size = 0;
fd.data = NULL;
Freeze(&fd, true);
fd.data = new uint8[fd.size];
Freeze(&fd, false);
m_dump.Open(m_snapshot, m_crc, fd, m_regs);
delete [] fd.data;
}
if(GSTexture* t = m_dev->GetCurrent())
{
t->Save(m_snapshot + ".bmp");
}
m_snapshot.clear();
}
else
{
if(m_dump)
{
bool control = false;
#ifdef _WINDOWS
control = !!(::GetAsyncKeyState(VK_CONTROL) & 0x8000);
#else
control = m_control_key;
#endif
m_dump.VSync(field, !control, m_regs);
}
}
// capture
if(m_capture.IsCapturing())
{
if(GSTexture* current = m_dev->GetCurrent())
{
GSVector2i size = m_capture.GetSize();
if(GSTexture* offscreen = m_dev->CopyOffscreen(current, GSVector4(0, 0, 1, 1), size.x, size.y))
{
GSTexture::GSMap m;
if(offscreen->Map(m))
{
m_capture.DeliverFrame(m.bits, m.pitch, !m_dev->IsRBSwapped());
offscreen->Unmap();
}
m_dev->Recycle(offscreen);
}
}
}
}
//TODO: GSopen 1 => Drop?
// Used by GSReplay. At least for now.
// More or less copy pasted from GSWndDX::Create and GSWndWGL::Attach with a few
// modifications
bool GSWndWGL::Create(const string& title, int w, int h)
{
if(m_NativeWindow) return false;
m_managed = true;
WNDCLASS wc;
memset(&wc, 0, sizeof(wc));
wc.style = CS_HREDRAW | CS_VREDRAW | CS_DBLCLKS | CS_OWNDC;
wc.lpfnWndProc = WndProc;
wc.hInstance = theApp.GetModuleHandle();
wc.hCursor = LoadCursor(NULL, IDC_ARROW);
wc.hbrBackground = (HBRUSH)GetStockObject(BLACK_BRUSH);
wc.lpszClassName = "GSWndOGL";
if (!GetClassInfo(wc.hInstance, wc.lpszClassName, &wc))
{
if (!RegisterClass(&wc))
{
return false;
}
}
DWORD style = WS_CLIPCHILDREN | WS_CLIPSIBLINGS | WS_OVERLAPPEDWINDOW | WS_BORDER;
GSVector4i r;
GetWindowRect(GetDesktopWindow(), r);
// Old GSOpen ModeWidth and ModeHeight are not necessary with this.
bool remote = !!GetSystemMetrics(SM_REMOTESESSION);
if (w <= 0 || h <= 0 || remote)
{
w = r.width() / 3;
h = r.width() / 4;
if (!remote)
{
w *= 2;
h *= 2;
}
}
r.left = (r.left + r.right - w) / 2;
r.top = (r.top + r.bottom - h) / 2;
r.right = r.left + w;
r.bottom = r.top + h;
AdjustWindowRect(r, style, FALSE);
m_NativeWindow = CreateWindow(wc.lpszClassName, title.c_str(), style, r.left, r.top, r.width(), r.height(), NULL, NULL, wc.hInstance, (LPVOID)this);
if (m_NativeWindow == NULL) return false;
OpenWGLDisplay();
CreateContext(3, 3);
AttachContext();
m_swapinterval = (PFNWGLSWAPINTERVALEXTPROC)wglGetProcAddress("wglSwapIntervalEXT");
PopulateGlFunction();
return true;
}
bool GSRenderer::Merge(int field)
{
bool en[2];
GSVector4i fr[2];
GSVector4i dr[2];
int baseline = INT_MAX;
for(int i = 0; i < 2; i++)
{
en[i] = IsEnabled(i);
if(en[i])
{
fr[i] = GetFrameRect(i);
dr[i] = GetDisplayRect(i);
baseline = min(dr[i].top, baseline);
//printf("[%d]: %d %d %d %d, %d %d %d %d\n", i, fr[i].x,fr[i].y,fr[i].z,fr[i].w , dr[i].x,dr[i].y,dr[i].z,dr[i].w);
}
}
if(!en[0] && !en[1])
{
return false;
}
// try to avoid fullscreen blur, could be nice on tv but on a monitor it's like double vision, hurts my eyes (persona 4, guitar hero)
//
// NOTE: probably the technique explained in graphtip.pdf (Antialiasing by Supersampling / 4. Reading Odd/Even Scan Lines Separately with the PCRTC then Blending)
bool samesrc =
en[0] && en[1] &&
m_regs->DISP[0].DISPFB.FBP == m_regs->DISP[1].DISPFB.FBP &&
m_regs->DISP[0].DISPFB.FBW == m_regs->DISP[1].DISPFB.FBW &&
m_regs->DISP[0].DISPFB.PSM == m_regs->DISP[1].DISPFB.PSM;
// bool blurdetected = false;
if(samesrc /*&& m_regs->PMODE.SLBG == 0 && m_regs->PMODE.MMOD == 1 && m_regs->PMODE.ALP == 0x80*/)
{
if(fr[0].eq(fr[1] + GSVector4i(0, -1, 0, 0)) && dr[0].eq(dr[1] + GSVector4i(0, 0, 0, 1))
|| fr[1].eq(fr[0] + GSVector4i(0, -1, 0, 0)) && dr[1].eq(dr[0] + GSVector4i(0, 0, 0, 1)))
{
// persona 4:
//
// fr[0] = 0 0 640 448
// fr[1] = 0 1 640 448
// dr[0] = 159 50 779 498
// dr[1] = 159 50 779 497
//
// second image shifted up by 1 pixel and blended over itself
//
// god of war:
//
// fr[0] = 0 1 512 448
// fr[1] = 0 0 512 448
// dr[0] = 127 50 639 497
// dr[1] = 127 50 639 498
//
// same just the first image shifted
int top = min(fr[0].top, fr[1].top);
int bottom = max(dr[0].bottom, dr[1].bottom);
fr[0].top = top;
fr[1].top = top;
dr[0].bottom = bottom;
dr[1].bottom = bottom;
// blurdetected = true;
}
else if(dr[0].eq(dr[1]) && (fr[0].eq(fr[1] + GSVector4i(0, 1, 0, 1)) || fr[1].eq(fr[0] + GSVector4i(0, 1, 0, 1))))
{
// dq5:
//
// fr[0] = 0 1 512 445
// fr[1] = 0 0 512 444
// dr[0] = 127 50 639 494
// dr[1] = 127 50 639 494
int top = min(fr[0].top, fr[1].top);
int bottom = min(fr[0].bottom, fr[1].bottom);
fr[0].top = fr[1].top = top;
fr[0].bottom = fr[1].bottom = bottom;
// blurdetected = true;
}
//printf("samesrc = %d blurdetected = %d\n",samesrc,blurdetected);
}
GSVector2i fs(0, 0);
GSVector2i ds(0, 0);
GSTexture* tex[2] = {NULL, NULL};
if(samesrc && fr[0].bottom == fr[1].bottom)
{
tex[0] = GetOutput(0);
tex[1] = tex[0]; // saves one texture fetch
}
else
{
if(en[0]) tex[0] = GetOutput(0);
if(en[1]) tex[1] = GetOutput(1);
}
GSVector4 src[2];
GSVector4 dst[2];
for(int i = 0; i < 2; i++)
{
if(!en[i] || !tex[i]) continue;
GSVector4i r = fr[i];
// overscan hack
if(dr[i].height() > 512) // hmm
{
int y = GetDeviceSize(i).y;
if(m_regs->SMODE2.INT && m_regs->SMODE2.FFMD) y /= 2;
r.bottom = r.top + y;
}
GSVector4 scale = GSVector4(tex[i]->GetScale()).xyxy();
src[i] = GSVector4(r) * scale / GSVector4(tex[i]->GetSize()).xyxy();
GSVector2 o(0, 0);
if(dr[i].top - baseline >= 4) // 2?
{
o.y = tex[i]->GetScale().y * (dr[i].top - baseline);
if(m_regs->SMODE2.INT && m_regs->SMODE2.FFMD)
{
o.y /= 2;
}
}
dst[i] = GSVector4(o).xyxy() + scale * GSVector4(r.rsize());
fs.x = max(fs.x, (int)(dst[i].z + 0.5f));
fs.y = max(fs.y, (int)(dst[i].w + 0.5f));
}
ds = fs;
if(m_regs->SMODE2.INT && m_regs->SMODE2.FFMD)
{
ds.y *= 2;
}
bool slbg = m_regs->PMODE.SLBG;
bool mmod = m_regs->PMODE.MMOD;
if(tex[0] || tex[1])
{
if(tex[0] == tex[1] && !slbg && (src[0] == src[1] & dst[0] == dst[1]).alltrue())
{
// the two outputs are identical, skip drawing one of them (the one that is alpha blended)
tex[0] = NULL;
}
GSVector4 c = GSVector4((int)m_regs->BGCOLOR.R, (int)m_regs->BGCOLOR.G, (int)m_regs->BGCOLOR.B, (int)m_regs->PMODE.ALP) / 255;
m_dev->Merge(tex, src, dst, fs, slbg, mmod, c);
if(m_regs->SMODE2.INT && m_interlace > 0)
{
if (m_interlace == 7 && m_regs->SMODE2.FFMD == 1) // Auto interlace enabled / Odd frame interlace setting
{
int field2 = 0;
int mode = 2;
m_dev->Interlace(ds, field ^ field2, mode, tex[1] ? tex[1]->GetScale().y : tex[0]->GetScale().y);
}
else
{
int field2 = 1 - ((m_interlace - 1) & 1);
int mode = (m_interlace - 1) >> 1;
m_dev->Interlace(ds, field ^ field2, mode, tex[1] ? tex[1]->GetScale().y : tex[0]->GetScale().y);
}
}
if(m_shadeboost)
{
m_dev->ShadeBoost();
}
if (m_shaderfx)
{
m_dev->ExternalFX();
}
if(m_fxaa)
{
m_dev->FXAA();
}
}
return true;
}
void GSTextureCacheOGL::Read(Target* t, const GSVector4i& r)
{
if(t->m_type != RenderTarget)
{
ASSERT(0);
return;
}
const GIFRegTEX0& TEX0 = t->m_TEX0;
if(TEX0.PSM != PSM_PSMCT32
&& TEX0.PSM != PSM_PSMCT24
&& TEX0.PSM != PSM_PSMCT16
&& TEX0.PSM != PSM_PSMCT16S)
{
//ASSERT(0);
return;
}
if(!t->m_dirty.empty())
{
return;
}
GL_PUSH("Texture Cache Read");
// printf("GSRenderTarget::Read %d,%d - %d,%d (%08x)\n", r.left, r.top, r.right, r.bottom, TEX0.TBP0);
int w = r.width();
int h = r.height();
GSVector4 src = GSVector4(r) * GSVector4(t->m_texture->GetScale()).xyxy() / GSVector4(t->m_texture->GetSize()).xyxy();
GLuint format = TEX0.PSM == PSM_PSMCT16 || TEX0.PSM == PSM_PSMCT16S ? GL_R16UI : GL_RGBA8;
//if (format == GL_R16UI) fprintf(stderr, "Format 16 bits integer\n");
#if 0
DXGI_FORMAT format = TEX0.PSM == PSM_PSMCT16 || TEX0.PSM == PSM_PSMCT16S ? DXGI_FORMAT_R16_UINT : DXGI_FORMAT_R8G8B8A8_UNORM;
#endif
if(GSTexture* offscreen = m_renderer->m_dev->CopyOffscreen(t->m_texture, src, w, h, format))
{
GSTexture::GSMap m;
if(offscreen->Map(m))
{
// TODO: block level write
GSOffset* off = m_renderer->m_mem.GetOffset(TEX0.TBP0, TEX0.TBW, TEX0.PSM);
switch(TEX0.PSM)
{
case PSM_PSMCT32:
m_renderer->m_mem.WritePixel32(m.bits, m.pitch, off, r);
break;
case PSM_PSMCT24:
m_renderer->m_mem.WritePixel24(m.bits, m.pitch, off, r);
break;
case PSM_PSMCT16:
case PSM_PSMCT16S:
m_renderer->m_mem.WritePixel16(m.bits, m.pitch, off, r);
break;
default:
ASSERT(0);
}
offscreen->Unmap();
}
// FIXME invalidate data
m_renderer->m_dev->Recycle(offscreen);
}
GL_POP();
}
void GSTextureOGL::Clear(const void* data, const GSVector4i& area)
{
glClearTexSubImage(m_texture_id, GL_TEX_LEVEL_0, area.x, area.y, 0, area.width(), area.height(), 1, m_int_format, m_int_type, data);
}
bool GSTextureOGL::Update(const GSVector4i& r, const void* data, int pitch)
{
ASSERT(m_type != GSTexture::DepthStencil && m_type != GSTexture::Offscreen);
// Default upload path for the texture is the Map/Unmap
// This path is mostly used for palette. But also for texture that could
// overflow the pbo buffer
// Data upload is rather small typically 64B or 1024B. So don't bother with PBO
// and directly send the data to the GL synchronously
m_clean = false;
uint32 row_byte = r.width() << m_int_shift;
uint32 map_size = r.height() * row_byte;
#ifdef ENABLE_OGL_DEBUG_MEM_BW
g_real_texture_upload_byte += map_size;
#endif
#if 0
if (r.height() == 1) {
// Palette data. Transfer is small either 64B or 1024B.
// Sometimes it is faster, sometimes slower.
glTextureSubImage2D(m_texture_id, GL_TEX_LEVEL_0, r.x, r.y, r.width(), r.height(), m_int_format, m_int_type, data);
return true;
}
#endif
GL_PUSH("Upload Texture %d", m_texture_id);
// The easy solution without PBO
#if 0
// Likely a bad texture
glPixelStorei(GL_UNPACK_ROW_LENGTH, pitch >> m_int_shift);
glTextureSubImage2D(m_texture_id, GL_TEX_LEVEL_0, r.x, r.y, r.width(), r.height(), m_int_format, m_int_type, data);
glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); // Restore default behavior
#endif
// The complex solution with PBO
#if 1
char* src = (char*)data;
char* map = PboPool::Map(map_size);
// PERF: slow path of the texture upload. Dunno if we could do better maybe check if TC can keep row_byte == pitch
// Note: row_byte != pitch
for (int h = 0; h < r.height(); h++) {
memcpy(map, src, row_byte);
map += row_byte;
src += pitch;
}
PboPool::Unmap();
glTextureSubImage2D(m_texture_id, GL_TEX_LEVEL_0, r.x, r.y, r.width(), r.height(), m_int_format, m_int_type, (const void*)PboPool::Offset());
// FIXME OGL4: investigate, only 1 unpack buffer always bound
PboPool::UnbindPbo();
PboPool::EndTransfer();
#endif
return true;
}
void GSVertexTrace::FindMinMax(const void* vertex, const uint32* index, int count)
{
const GSDrawingContext* context = m_state->m_context;
int n = 1;
switch(primclass)
{
case GS_POINT_CLASS:
n = 1;
break;
case GS_LINE_CLASS:
case GS_SPRITE_CLASS:
n = 2;
break;
case GS_TRIANGLE_CLASS:
n = 3;
break;
}
GSVector4 tmin = s_minmax.xxxx();
GSVector4 tmax = s_minmax.yyyy();
GSVector4i cmin = GSVector4i::xffffffff();
GSVector4i cmax = GSVector4i::zero();
#if _M_SSE >= 0x401
GSVector4i pmin = GSVector4i::xffffffff();
GSVector4i pmax = GSVector4i::zero();
#else
GSVector4 pmin = s_minmax.xxxx();
GSVector4 pmax = s_minmax.yyyy();
#endif
const GSVertex* RESTRICT v = (GSVertex*)vertex;
for(int i = 0; i < count; i += n)
{
if(primclass == GS_POINT_CLASS)
{
GSVector4i c(v[index[i]].m[0]);
if(color)
{
cmin = cmin.min_u8(c);
cmax = cmax.max_u8(c);
}
if(tme)
{
if(!fst)
{
GSVector4 stq = GSVector4::cast(c);
GSVector4 q = stq.wwww();
stq = (stq.xyww() * q.rcpnr()).xyww(q);
tmin = tmin.min(stq);
tmax = tmax.max(stq);
}
else
{
GSVector4i uv(v[index[i]].m[1]);
GSVector4 st = GSVector4(uv.uph16()).xyxy();
tmin = tmin.min(st);
tmax = tmax.max(st);
}
}
GSVector4i xyzf(v[index[i]].m[1]);
GSVector4i xy = xyzf.upl16();
GSVector4i z = xyzf.yyyy();
#if _M_SSE >= 0x401
GSVector4i p = xy.blend16<0xf0>(z.uph32(xyzf));
pmin = pmin.min_u32(p);
pmax = pmax.max_u32(p);
#else
GSVector4 p = GSVector4(xy.upl64(z.srl32(1).upl32(xyzf.wwww())));
pmin = pmin.min(p);
pmax = pmax.max(p);
#endif
}
else if(primclass == GS_LINE_CLASS)
{
GSVector4i c0(v[index[i + 0]].m[0]);
GSVector4i c1(v[index[i + 1]].m[0]);
if(color)
{
if(iip)
{
cmin = cmin.min_u8(c0.min_u8(c1));
cmax = cmax.max_u8(c0.max_u8(c1));
}
else
{
cmin = cmin.min_u8(c1);
cmax = cmax.max_u8(c1);
}
}
if(tme)
{
if(!fst)
{
GSVector4 stq0 = GSVector4::cast(c0);
GSVector4 stq1 = GSVector4::cast(c1);
GSVector4 q = stq0.wwww(stq1).rcpnr();
stq0 = (stq0.xyww() * q.xxxx()).xyww(stq0);
stq1 = (stq1.xyww() * q.zzzz()).xyww(stq1);
tmin = tmin.min(stq0.min(stq1));
tmax = tmax.max(stq0.max(stq1));
}
else
{
GSVector4i uv0(v[index[i + 0]].m[1]);
GSVector4i uv1(v[index[i + 1]].m[1]);
GSVector4 st0 = GSVector4(uv0.uph16()).xyxy();
GSVector4 st1 = GSVector4(uv1.uph16()).xyxy();
tmin = tmin.min(st0.min(st1));
tmax = tmax.max(st0.max(st1));
}
}
GSVector4i xyzf0(v[index[i + 0]].m[1]);
GSVector4i xyzf1(v[index[i + 1]].m[1]);
GSVector4i xy0 = xyzf0.upl16();
GSVector4i z0 = xyzf0.yyyy();
GSVector4i xy1 = xyzf1.upl16();
GSVector4i z1 = xyzf1.yyyy();
#if _M_SSE >= 0x401
GSVector4i p0 = xy0.blend16<0xf0>(z0.uph32(xyzf0));
GSVector4i p1 = xy1.blend16<0xf0>(z1.uph32(xyzf1));
pmin = pmin.min_u32(p0.min_u32(p1));
pmax = pmax.max_u32(p0.max_u32(p1));
#else
GSVector4 p0 = GSVector4(xy0.upl64(z0.srl32(1).upl32(xyzf0.wwww())));
GSVector4 p1 = GSVector4(xy1.upl64(z1.srl32(1).upl32(xyzf1.wwww())));
pmin = pmin.min(p0.min(p1));
pmax = pmax.max(p0.max(p1));
#endif
}
else if(primclass == GS_TRIANGLE_CLASS)
{
GSVector4i c0(v[index[i + 0]].m[0]);
GSVector4i c1(v[index[i + 1]].m[0]);
GSVector4i c2(v[index[i + 2]].m[0]);
if(color)
{
if(iip)
{
cmin = cmin.min_u8(c2).min_u8(c0.min_u8(c1));
cmax = cmax.max_u8(c2).max_u8(c0.max_u8(c1));
}
else
{
cmin = cmin.min_u8(c2);
cmax = cmax.max_u8(c2);
}
}
if(tme)
{
if(!fst)
{
GSVector4 stq0 = GSVector4::cast(c0);
GSVector4 stq1 = GSVector4::cast(c1);
GSVector4 stq2 = GSVector4::cast(c2);
GSVector4 q = stq0.wwww(stq1).xzww(stq2).rcpnr();
stq0 = (stq0.xyww() * q.xxxx()).xyww(stq0);
stq1 = (stq1.xyww() * q.yyyy()).xyww(stq1);
stq2 = (stq2.xyww() * q.zzzz()).xyww(stq2);
tmin = tmin.min(stq2).min(stq0.min(stq1));
tmax = tmax.max(stq2).max(stq0.max(stq1));
}
else
{
GSVector4i uv0(v[index[i + 0]].m[1]);
GSVector4i uv1(v[index[i + 1]].m[1]);
GSVector4i uv2(v[index[i + 2]].m[1]);
GSVector4 st0 = GSVector4(uv0.uph16()).xyxy();
GSVector4 st1 = GSVector4(uv1.uph16()).xyxy();
GSVector4 st2 = GSVector4(uv2.uph16()).xyxy();
tmin = tmin.min(st2).min(st0.min(st1));
tmax = tmax.max(st2).max(st0.max(st1));
}
}
GSVector4i xyzf0(v[index[i + 0]].m[1]);
GSVector4i xyzf1(v[index[i + 1]].m[1]);
GSVector4i xyzf2(v[index[i + 2]].m[1]);
GSVector4i xy0 = xyzf0.upl16();
GSVector4i z0 = xyzf0.yyyy();
GSVector4i xy1 = xyzf1.upl16();
GSVector4i z1 = xyzf1.yyyy();
GSVector4i xy2 = xyzf2.upl16();
GSVector4i z2 = xyzf2.yyyy();
#if _M_SSE >= 0x401
GSVector4i p0 = xy0.blend16<0xf0>(z0.uph32(xyzf0));
GSVector4i p1 = xy1.blend16<0xf0>(z1.uph32(xyzf1));
GSVector4i p2 = xy2.blend16<0xf0>(z2.uph32(xyzf2));
pmin = pmin.min_u32(p2).min_u32(p0.min_u32(p1));
pmax = pmax.max_u32(p2).max_u32(p0.max_u32(p1));
#else
GSVector4 p0 = GSVector4(xy0.upl64(z0.srl32(1).upl32(xyzf0.wwww())));
GSVector4 p1 = GSVector4(xy1.upl64(z1.srl32(1).upl32(xyzf1.wwww())));
GSVector4 p2 = GSVector4(xy2.upl64(z2.srl32(1).upl32(xyzf2.wwww())));
pmin = pmin.min(p2).min(p0.min(p1));
pmax = pmax.max(p2).max(p0.max(p1));
#endif
}
else if(primclass == GS_SPRITE_CLASS)
{
GSVector4i c0(v[index[i + 0]].m[0]);
GSVector4i c1(v[index[i + 1]].m[0]);
if(color)
{
if(iip)
{
cmin = cmin.min_u8(c0.min_u8(c1));
cmax = cmax.max_u8(c0.max_u8(c1));
}
else
{
cmin = cmin.min_u8(c1);
cmax = cmax.max_u8(c1);
}
}
if(tme)
{
if(!fst)
{
GSVector4 stq0 = GSVector4::cast(c0);
GSVector4 stq1 = GSVector4::cast(c1);
GSVector4 q = stq1.wwww().rcpnr();
stq0 = (stq0.xyww() * q).xyww(stq1);
stq1 = (stq1.xyww() * q).xyww(stq1);
tmin = tmin.min(stq0.min(stq1));
tmax = tmax.max(stq0.max(stq1));
}
else
{
GSVector4i uv0(v[index[i + 0]].m[1]);
GSVector4i uv1(v[index[i + 1]].m[1]);
GSVector4 st0 = GSVector4(uv0.uph16()).xyxy();
GSVector4 st1 = GSVector4(uv1.uph16()).xyxy();
tmin = tmin.min(st0.min(st1));
tmax = tmax.max(st0.max(st1));
}
}
GSVector4i xyzf0(v[index[i + 0]].m[1]);
GSVector4i xyzf1(v[index[i + 1]].m[1]);
GSVector4i xy0 = xyzf0.upl16();
GSVector4i z0 = xyzf0.yyyy();
GSVector4i xy1 = xyzf1.upl16();
GSVector4i z1 = xyzf1.yyyy();
#if _M_SSE >= 0x401
GSVector4i p0 = xy0.blend16<0xf0>(z0.uph32(xyzf1));
GSVector4i p1 = xy1.blend16<0xf0>(z1.uph32(xyzf1));
pmin = pmin.min_u32(p0.min_u32(p1));
pmax = pmax.max_u32(p0.max_u32(p1));
#else
GSVector4 p0 = GSVector4(xy0.upl64(z0.srl32(1).upl32(xyzf1.wwww())));
GSVector4 p1 = GSVector4(xy1.upl64(z1.srl32(1).upl32(xyzf1.wwww())));
pmin = pmin.min(p0.min(p1));
pmax = pmax.max(p0.max(p1));
#endif
}
}
#if _M_SSE >= 0x401
pmin = pmin.blend16<0x30>(pmin.srl32(1));
pmax = pmax.blend16<0x30>(pmax.srl32(1));
#endif
GSVector4 o(context->XYOFFSET);
GSVector4 s(1.0f / 16, 1.0f / 16, 2.0f, 1.0f);
m_min.p = (GSVector4(pmin) - o) * s;
m_max.p = (GSVector4(pmax) - o) * s;
if(tme)
{
if(fst)
{
s = GSVector4(1.0f / 16, 1.0f).xxyy();
}
else
{
s = GSVector4(1 << context->TEX0.TW, 1 << context->TEX0.TH, 1, 1);
}
m_min.t = tmin * s;
m_max.t = tmax * s;
}
else
{
m_min.t = GSVector4::zero();
m_max.t = GSVector4::zero();
}
if(color)
{
m_min.c = cmin.zzzz().u8to32();
m_max.c = cmax.zzzz().u8to32();
}
else
{
m_min.c = GSVector4i::zero();
m_max.c = GSVector4i::zero();
}
}
bool GSRenderer::Merge(int field)
{
bool en[2];
GSVector4i fr[2];
GSVector4i dr[2];
GSVector2i display_baseline = { INT_MAX, INT_MAX };
GSVector2i frame_baseline = { INT_MAX, INT_MAX };
for(int i = 0; i < 2; i++)
{
en[i] = IsEnabled(i);
if(en[i])
{
fr[i] = GetFrameRect(i);
dr[i] = GetDisplayRect(i);
display_baseline.x = min(dr[i].left, display_baseline.x);
display_baseline.y = min(dr[i].top, display_baseline.y);
frame_baseline.x = min(fr[i].left, frame_baseline.x);
frame_baseline.y = min(fr[i].top, frame_baseline.y);
//printf("[%d]: %d %d %d %d, %d %d %d %d\n", i, fr[i].x,fr[i].y,fr[i].z,fr[i].w , dr[i].x,dr[i].y,dr[i].z,dr[i].w);
}
}
if(!en[0] && !en[1])
{
return false;
}
GL_PUSH("Renderer Merge %d (0: enabled %d 0x%x, 1: enabled %d 0x%x)", s_n, en[0], m_regs->DISP[0].DISPFB.Block(), en[1], m_regs->DISP[1].DISPFB.Block());
// try to avoid fullscreen blur, could be nice on tv but on a monitor it's like double vision, hurts my eyes (persona 4, guitar hero)
//
// NOTE: probably the technique explained in graphtip.pdf (Antialiasing by Supersampling / 4. Reading Odd/Even Scan Lines Separately with the PCRTC then Blending)
bool samesrc =
en[0] && en[1] &&
m_regs->DISP[0].DISPFB.FBP == m_regs->DISP[1].DISPFB.FBP &&
m_regs->DISP[0].DISPFB.FBW == m_regs->DISP[1].DISPFB.FBW &&
m_regs->DISP[0].DISPFB.PSM == m_regs->DISP[1].DISPFB.PSM;
if(samesrc /*&& m_regs->PMODE.SLBG == 0 && m_regs->PMODE.MMOD == 1 && m_regs->PMODE.ALP == 0x80*/)
{
// persona 4:
//
// fr[0] = 0 0 640 448
// fr[1] = 0 1 640 448
// dr[0] = 159 50 779 498
// dr[1] = 159 50 779 497
//
// second image shifted up by 1 pixel and blended over itself
//
// god of war:
//
// fr[0] = 0 1 512 448
// fr[1] = 0 0 512 448
// dr[0] = 127 50 639 497
// dr[1] = 127 50 639 498
//
// same just the first image shifted
//
// These kinds of cases are now fixed by the more generic frame_diff code below, as the code here was too specific and has become obsolete.
// NOTE: Persona 4 and God Of War are not rare exceptions, many games have the same(or very similar) offsets.
int topDiff = fr[0].top - fr[1].top;
if (dr[0].eq(dr[1]) && (fr[0].eq(fr[1] + GSVector4i(0, topDiff, 0, topDiff)) || fr[1].eq(fr[0] + GSVector4i(0, topDiff, 0, topDiff))))
{
// dq5:
//
// fr[0] = 0 1 512 445
// fr[1] = 0 0 512 444
// dr[0] = 127 50 639 494
// dr[1] = 127 50 639 494
int top = min(fr[0].top, fr[1].top);
int bottom = min(fr[0].bottom, fr[1].bottom);
fr[0].top = fr[1].top = top;
fr[0].bottom = fr[1].bottom = bottom;
}
}
GSVector2i fs(0, 0);
GSVector2i ds(0, 0);
GSTexture* tex[3] = {NULL, NULL, NULL};
int y_offset[3] = {0, 0, 0};
s_n++;
bool feedback_merge = m_regs->EXTWRITE.WRITE == 1;
if(samesrc && fr[0].bottom == fr[1].bottom && !feedback_merge)
{
tex[0] = GetOutput(0, y_offset[0]);
tex[1] = tex[0]; // saves one texture fetch
y_offset[1] = y_offset[0];
}
else
{
if(en[0]) tex[0] = GetOutput(0, y_offset[0]);
if(en[1]) tex[1] = GetOutput(1, y_offset[1]);
if(feedback_merge) tex[2] = GetFeedbackOutput();
}
GSVector4 src[2];
GSVector4 src_hw[2];
GSVector4 dst[2];
for(int i = 0; i < 2; i++)
{
if(!en[i] || !tex[i]) continue;
GSVector4i r = fr[i];
GSVector4 scale = GSVector4(tex[i]->GetScale()).xyxy();
src[i] = GSVector4(r) * scale / GSVector4(tex[i]->GetSize()).xyxy();
src_hw[i] = (GSVector4(r) + GSVector4 (0, y_offset[i], 0, y_offset[i])) * scale / GSVector4(tex[i]->GetSize()).xyxy();
GSVector2 off(0);
GSVector2i display_diff(dr[i].left - display_baseline.x, dr[i].top - display_baseline.y);
GSVector2i frame_diff(fr[i].left - frame_baseline.x, fr[i].top - frame_baseline.y);
// Time Crisis 2/3 uses two side by side images when in split screen mode.
// Though ignore cases where baseline and display rectangle offsets only differ by 1 pixel, causes blurring and wrong resolution output on FFXII
if(display_diff.x > 2)
{
off.x = tex[i]->GetScale().x * display_diff.x;
}
// If the DX offset is too small then consider the status of frame memory offsets, prevents blurring on Tenchu: Fatal Shadows, Worms 3D
else if(display_diff.x != frame_diff.x)
{
off.x = tex[i]->GetScale().x * frame_diff.x;
}
if(display_diff.y >= 4) // Shouldn't this be >= 2?
{
off.y = tex[i]->GetScale().y * display_diff.y;
if(m_regs->SMODE2.INT && m_regs->SMODE2.FFMD)
{
off.y /= 2;
}
}
else if(display_diff.y != frame_diff.y)
{
off.y = tex[i]->GetScale().y * frame_diff.y;
}
dst[i] = GSVector4(off).xyxy() + scale * GSVector4(r.rsize());
fs.x = max(fs.x, (int)(dst[i].z + 0.5f));
fs.y = max(fs.y, (int)(dst[i].w + 0.5f));
}
ds = fs;
if(m_regs->SMODE2.INT && m_regs->SMODE2.FFMD)
{
ds.y *= 2;
}
m_real_size = ds;
bool slbg = m_regs->PMODE.SLBG;
if(tex[0] || tex[1])
{
if(tex[0] == tex[1] && !slbg && (src[0] == src[1] & dst[0] == dst[1]).alltrue())
{
// the two outputs are identical, skip drawing one of them (the one that is alpha blended)
tex[0] = NULL;
}
GSVector4 c = GSVector4((int)m_regs->BGCOLOR.R, (int)m_regs->BGCOLOR.G, (int)m_regs->BGCOLOR.B, (int)m_regs->PMODE.ALP) / 255;
m_dev->Merge(tex, src_hw, dst, fs, m_regs->PMODE, m_regs->EXTBUF, c);
if(m_regs->SMODE2.INT && m_interlace > 0)
{
if(m_interlace == 7 && m_regs->SMODE2.FFMD) // Auto interlace enabled / Odd frame interlace setting
{
int field2 = 0;
int mode = 2;
m_dev->Interlace(ds, field ^ field2, mode, tex[1] ? tex[1]->GetScale().y : tex[0]->GetScale().y);
}
else
{
int field2 = 1 - ((m_interlace - 1) & 1);
int mode = (m_interlace - 1) >> 1;
m_dev->Interlace(ds, field ^ field2, mode, tex[1] ? tex[1]->GetScale().y : tex[0]->GetScale().y);
}
}
if(m_shadeboost)
{
m_dev->ShadeBoost();
}
if(m_shaderfx)
{
m_dev->ExternalFX();
}
if(m_fxaa)
{
m_dev->FXAA();
}
}
return true;
}
void GSTextureCache9::Read(Target* t, const GSVector4i& r)
{
if(t->m_type != RenderTarget)
{
// TODO
return;
}
const GIFRegTEX0& TEX0 = t->m_TEX0;
if(TEX0.PSM != PSM_PSMCT32
&& TEX0.PSM != PSM_PSMCT24
&& TEX0.PSM != PSM_PSMCT16
&& TEX0.PSM != PSM_PSMCT16S)
{
//ASSERT(0);
return;
}
if(!t->m_dirty.empty())
{
return;
}
// printf("GSRenderTarget::Read %d,%d - %d,%d (%08x)\n", r.left, r.top, r.right, r.bottom, TEX0.TBP0);
int w = r.width();
int h = r.height();
GSVector4 src = GSVector4(r) * GSVector4(t->m_texture->GetScale()).xyxy() / GSVector4(t->m_texture->GetSize()).xyxy();
if(GSTexture* offscreen = m_renderer->m_dev->CopyOffscreen(t->m_texture, src, w, h))
{
GSTexture::GSMap m;
if(offscreen->Map(m))
{
// TODO: block level write
GSOffset* o = m_renderer->m_mem.GetOffset(TEX0.TBP0, TEX0.TBW, TEX0.PSM);
switch(TEX0.PSM)
{
case PSM_PSMCT32:
m_renderer->m_mem.WritePixel32(m.bits, m.pitch, o, r);
break;
case PSM_PSMCT24:
m_renderer->m_mem.WritePixel24(m.bits, m.pitch, o, r);
break;
case PSM_PSMCT16:
case PSM_PSMCT16S:
m_renderer->m_mem.WriteFrame16(m.bits, m.pitch, o, r);
break;
default:
ASSERT(0);
}
offscreen->Unmap();
}
m_renderer->m_dev->Recycle(offscreen);
}
}
bool GSTextureCacheSW::Texture::Update(const GSVector4i& rect)
{
if(m_complete)
{
return true;
}
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[m_TEX0.PSM];
GSVector2i bs = psm.bs;
int shift = psm.pal == 0 ? 2 : 0;
int tw = std::max<int>(1 << m_TEX0.TW, bs.x);
int th = std::max<int>(1 << m_TEX0.TH, bs.y);
GSVector4i r = rect;
r = r.ralign<Align_Outside>(bs);
if(r.eq(GSVector4i(0, 0, tw, th)))
{
m_complete = true; // lame, but better than nothing
}
if(m_buff == NULL)
{
uint32 pitch = (1 << m_tw) << shift;
m_buff = _aligned_malloc(pitch * th * 4, 32);
if(m_buff == NULL)
{
return false;
}
}
GSLocalMemory& mem = m_state->m_mem;
const GSOffset* RESTRICT off = m_offset;
uint32 blocks = 0;
GSLocalMemory::readTextureBlock rtxbP = psm.rtxbP;
uint32 pitch = (1 << m_tw) << shift;
uint8* dst = (uint8*)m_buff + pitch * r.top;
int block_pitch = pitch * bs.y;
r = r.srl32(3);
bs.x >>= 3;
bs.y >>= 3;
shift += 3;
if(m_repeating)
{
for(int y = r.top; y < r.bottom; y += bs.y, dst += block_pitch)
{
uint32 base = off->block.row[y];
for(int x = r.left, i = (y << 7) + x; x < r.right; x += bs.x, i += bs.x)
{
uint32 block = (base + off->block.col[x]) % MAX_BLOCKS;
uint32 row = i >> 5;
uint32 col = 1 << (i & 31);
if((m_valid[row] & col) == 0)
{
m_valid[row] |= col;
(mem.*rtxbP)(block, &dst[x << shift], pitch, m_TEXA);
blocks++;
}
}
}
}
else
{
for(int y = r.top; y < r.bottom; y += bs.y, dst += block_pitch)
