ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

GS/HW: Combine dirty rectangles for target updates 

Fewer texture uploads makes Intel GPUs happy.
This commit is contained in:
Stenzek 2023-02-11 20:34:08 +10:00 committed by refractionpcsx2
parent 245b03e208
commit f0798f6510
3 changed files with 55 additions and 57 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
pcsx2/GS/Renderers/Common/GSDirtyRect.cpp
View File

@ -31,7 +31,7 @@ GSDirtyRect::GSDirtyRect(GSVector4i& r, u32 psm, u32 bw) :
{
}
GSVector4i GSDirtyRect::GetDirtyRect(GIFRegTEX0& TEX0)
GSVector4i GSDirtyRect::GetDirtyRect(GIFRegTEX0 TEX0) const
{
GSVector4i _r;
@ -54,7 +54,7 @@ GSVector4i GSDirtyRect::GetDirtyRect(GIFRegTEX0& TEX0)
return _r;
}
GSVector4i GSDirtyRectList::GetTotalRect(GIFRegTEX0& TEX0, const GSVector2i& size)
GSVector4i GSDirtyRectList::GetTotalRect(GIFRegTEX0 TEX0, const GSVector2i& size) const
{
if (!empty())
{
@ -73,31 +73,12 @@ GSVector4i GSDirtyRectList::GetTotalRect(GIFRegTEX0& TEX0, const GSVector2i& siz
return GSVector4i::zero();
}
GSVector4i GSDirtyRectList::GetDirtyRect(GIFRegTEX0& TEX0, const GSVector2i& size, bool clear)
GSVector4i GSDirtyRectList::GetDirtyRect(size_t index, GIFRegTEX0 TEX0, const GSVector4i& clamp) const
{
if (!empty())
{
const std::vector<GSDirtyRect>::iterator &it = begin();
const GSVector4i r = it[0].GetDirtyRect(TEX0);
const GSVector4i r = (*this)[index].GetDirtyRect(TEX0);
if (clear)
erase(it);
GSVector2i bs = GSLocalMemory::m_psm[TEX0.PSM].bs;
GSVector2i bs = GSLocalMemory::m_psm[TEX0.PSM].bs;
return r.ralign<Align_Outside>(bs).rintersect(GSVector4i(0, 0, size.x, size.y));
}
return GSVector4i::zero();
return r.ralign<Align_Outside>(bs).rintersect(clamp);
}
GSVector4i GSDirtyRectList::GetDirtyRectAndClear(GIFRegTEX0& TEX0, const GSVector2i& size)
{
const GSVector4i r = GetDirtyRect(TEX0, size, true);
return r;
}
void GSDirtyRectList::ClearDirty()
{
clear();
}

8
pcsx2/GS/Renderers/Common/GSDirtyRect.h
View File

@ -26,15 +26,13 @@ public:
GSDirtyRect();
GSDirtyRect(GSVector4i& r, u32 psm, u32 bw);
GSVector4i GetDirtyRect(GIFRegTEX0& TEX0);
GSVector4i GetDirtyRect(GIFRegTEX0 TEX0) const;
};
class GSDirtyRectList : public std::vector<GSDirtyRect>
{
public:
GSDirtyRectList() {}
GSVector4i GetTotalRect(GIFRegTEX0& TEX0, const GSVector2i& size);
GSVector4i GetDirtyRect(GIFRegTEX0& TEX0, const GSVector2i& size, bool clear = false);
GSVector4i GetDirtyRectAndClear(GIFRegTEX0& TEX0, const GSVector2i& size);
void ClearDirty();
GSVector4i GetTotalRect(GIFRegTEX0 TEX0, const GSVector2i& size) const;
GSVector4i GetDirtyRect(size_t index, GIFRegTEX0 TEX0, const GSVector4i& clamp) const;
};

73
pcsx2/GS/Renderers/HW/GSTextureCache.cpp
View File

@ -3073,48 +3073,67 @@ void GSTextureCache::Target::Update(bool reset_age)
{
// do the most likely thing a direct write would do, clear it
GL_INS("ERROR: Update DepthStencil dummy");
m_dirty.ClearDirty();
m_dirty.clear();
return;
}
GIFRegTEXA TEXA = {};
const GSVector2i unscaled_size(static_cast<int>(m_texture->GetWidth() / m_texture->GetScale().x),
static_cast<int>(m_texture->GetHeight() / m_texture->GetScale().y));
const GSVector4i total_rect(m_dirty.GetTotalRect(m_TEX0, unscaled_size));
if (total_rect.rempty())
{
GL_INS("ERROR: Nothing to update?");
m_dirty.clear();
return;
}
const GSVector4i t_offset(total_rect.xyxy());
const GSVector4i t_size(total_rect - t_offset);
const GSVector4 t_sizef(t_size.zwzw());
// This'll leave undefined data in pixels that we're not reading from... shouldn't hurt anything.
GSTexture* const t = g_gs_device->CreateTexture(t_size.z, t_size.w, 1, GSTexture::Format::Color);
GSTexture::GSMap m;
const bool mapped = t->Map(m);
GIFRegTEXA TEXA = {};
TEXA.AEM = 1;
TEXA.TA0 = 0;
TEXA.TA1 = 0x80;
const GSOffset off = g_gs_renderer->m_mem.GetOffset(m_TEX0.TBP0, m_TEX0.TBW, m_TEX0.PSM);
GSTexture::GSMap m;
const GSVector4i unscaled_size = GSVector4i(GSVector4(m_texture->GetSize()) / GSVector4(m_texture->GetScale()));
const GSVector4i rect_size = m_dirty.GetTotalRect(m_TEX0, GSVector2i(unscaled_size.x, unscaled_size.y));
GSTexture* t = g_gs_device->CreateTexture(rect_size.z, rect_size.w, 1, GSTexture::Format::Color);
while (m_dirty.size() > 0)
const GSOffset off(g_gs_renderer->m_mem.GetOffset(m_TEX0.TBP0, m_TEX0.TBW, m_TEX0.PSM));
for (size_t i = 0; i < m_dirty.size(); i++)
{
const GSVector4i r = m_dirty.GetDirtyRectAndClear(m_TEX0, GSVector2i(unscaled_size.x, unscaled_size.y));
const GSVector4i r(m_dirty.GetDirtyRect(i, m_TEX0, total_rect));
if (r.rempty())
continue;
if (t->Map(m))
const GSVector4i t_r(r - t_offset);
if (mapped)
{
g_gs_renderer->m_mem.ReadTexture(off, r, m.bits, m.pitch, TEXA);
t->Unmap();
g_gs_renderer->m_mem.ReadTexture(
off, r, m.bits + t_r.y * static_cast<u32>(m.pitch) + (t_r.x * sizeof(u32)), m.pitch, TEXA);
}
else
{
const int pitch = Common::AlignUpPow2(r.width() * sizeof(u32), 32);
g_gs_renderer->m_mem.ReadTexture(off, r, s_unswizzle_buffer, pitch, TEXA);
t->Update(r, s_unswizzle_buffer, pitch);
t->Update(t_r, s_unswizzle_buffer, pitch);
}
}
const GSVector4 sRect = GSVector4(0.0f, 0.0f, static_cast<float>(r.width()) / rect_size.z, static_cast<float>(r.height()) / rect_size.w);
const GSVector4 dest_rect = (GSVector4(r) * GSVector4(m_texture->GetScale()).xyxy());
if (mapped)
t->Unmap();
for (size_t i = 0; i < m_dirty.size(); i++)
{
const GSVector4i r(m_dirty.GetDirtyRect(i, m_TEX0, total_rect));
if (r.rempty())
continue;
const GSVector4 sRect(GSVector4(r - t_offset) / t_sizef);
const GSVector4 dRect(GSVector4(r) * GSVector4(m_texture->GetScale()).xyxy());
// Copy the new GS memory content into the destination texture.
if (m_type == RenderTarget)
@ -3122,21 +3141,21 @@ void GSTextureCache::Target::Update(bool reset_age)
GL_INS("ERROR: Update RenderTarget 0x%x bw:%d (%d,%d => %d,%d)", m_TEX0.TBP0, m_TEX0.TBW, r.x, r.y, r.z, r.w);
// Bilinear filtering this is probably not a good thing, at least in native, but upscaling Nearest can be gross and messy.
g_gs_device->StretchRect(t, sRect, m_texture, dest_rect,ShaderConvert::COPY, g_gs_renderer->CanUpscale());
g_gs_device->StretchRect(t, sRect, m_texture, dRect, ShaderConvert::COPY, g_gs_renderer->CanUpscale());
}
else if (m_type == DepthStencil)
{
GL_INS("ERROR: Update DepthStencil 0x%x", m_TEX0.TBP0);
// FIXME linear or not?
const GSVector4 sRect = GSVector4(0.0f, 0.0f, static_cast<float>(r.width()) / rect_size.z, static_cast<float>(r.height()) / rect_size.w);
g_gs_device->StretchRect(t, sRect, m_texture, dest_rect, ShaderConvert::RGBA8_TO_FLOAT32_BILN);
g_gs_device->StretchRect(t, sRect, m_texture, dRect, ShaderConvert::RGBA8_TO_FLOAT32_BILN);
}
}
// m_renderer->m_perfmon.Put(GSPerfMon::Unswizzle, w * h * 4);
g_gs_device->Recycle(t);
UpdateValidity(rect_size);
UpdateValidity(total_rect);
g_gs_device->Recycle(t);
m_dirty.clear();
}
void GSTextureCache::Target::UpdateIfDirtyIntersects(const GSVector4i& rc)