diff --git a/pcsx2-gsrunner/test_check_dumps.py b/pcsx2-gsrunner/test_check_dumps.py
index 5c48b7a557b2f..9759d52d4f6ca 100644
--- a/pcsx2-gsrunner/test_check_dumps.py
+++ b/pcsx2-gsrunner/test_check_dumps.py
@@ -114,7 +114,7 @@ def check_regression_test(baselinedir, testdir, name):
if not os.path.isfile(path2):
print("--- Frame %u for %s is missing in test set" % (framenum, name))
write("
{}
".format(name))
- write("--- Frame %u for %s is missing in test set" % (framenum, name))
+ write("--- Frame %u for %s is missing in test set
" % (framenum, name))
return False
if not compare_frames(path1, path2):
diff --git a/pcsx2/GS/Renderers/HW/GSRendererHW.cpp b/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
index 70a5e8af66cab..f7400420334da 100644
--- a/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
+++ b/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
@@ -952,11 +952,11 @@ GSVector2i GSRendererHW::GetValidSize(const GSTextureCache::Source* tex)
return GSVector2i(width, height);
}
-GSVector2i GSRendererHW::GetTargetSize(const GSTextureCache::Source* tex)
+GSVector2i GSRendererHW::GetTargetSize(const GSTextureCache::Source* tex, const bool can_expand)
{
const GSVector2i valid_size = GetValidSize(tex);
- return g_texture_cache->GetTargetSize(m_cached_ctx.FRAME.Block(), m_cached_ctx.FRAME.FBW, m_cached_ctx.FRAME.PSM, valid_size.x, valid_size.y);
+ return g_texture_cache->GetTargetSize(m_cached_ctx.FRAME.Block(), m_cached_ctx.FRAME.FBW, m_cached_ctx.FRAME.PSM, valid_size.x, valid_size.y, can_expand);
}
bool GSRendererHW::IsPossibleChannelShuffle() const
@@ -2553,7 +2553,7 @@ void GSRendererHW::Draw()
const u32 color_mask = (m_vt.m_max.c > GSVector4i::zero()).mask();
const bool texture_function_color = m_cached_ctx.TEX0.TFX == TFX_DECAL || (color_mask & 0xFFF) || (m_cached_ctx.TEX0.TFX > TFX_DECAL && (color_mask & 0xF000));
const bool texture_function_alpha = m_cached_ctx.TEX0.TFX != TFX_MODULATE || (color_mask & 0xF000);
- const bool req_color = texture_function_color && (!PRIM->ABE || (PRIM->ABE && IsUsingCsInBlend())) && (possible_shuffle || (m_cached_ctx.FRAME.FBMSK & (fm_mask & 0x00FFFFFF)) != (fm_mask & 0x00FFFFFF)) || need_aem_color;
+ const bool req_color = (texture_function_color && (!PRIM->ABE || (PRIM->ABE && IsUsingCsInBlend())) && (possible_shuffle || (m_cached_ctx.FRAME.FBMSK & (fm_mask & 0x00FFFFFF)) != (fm_mask & 0x00FFFFFF))) || need_aem_color;
const bool alpha_used = (GSUtil::GetChannelMask(m_context->TEX0.PSM) == 0x8 || (m_context->TEX0.TCC && texture_function_alpha)) && ((PRIM->ABE && IsUsingAsInBlend()) || (m_cached_ctx.TEST.ATE && m_cached_ctx.TEST.ATST > ATST_ALWAYS) || (possible_shuffle || (m_cached_ctx.FRAME.FBMSK & (fm_mask & 0xFF000000)) != (fm_mask & 0xFF000000)));
const bool req_alpha = (GSUtil::GetChannelMask(m_context->TEX0.PSM) & 0x8) && alpha_used;
@@ -2612,8 +2612,14 @@ void GSRendererHW::Draw()
}
}
+ // Urban Reign trolls by scissoring a draw to a target at 0x0-0x117F to 378x449 which ends up the size being rounded up to 640x480
+ // causing the buffer to expand to around 0x1400, which makes a later framebuffer at 0x1180 to fail to be created correctly.
+ // We can cheese this by checking if the Z is masked and the resultant colour is going to be black anyway.
+ const bool output_black = PRIM->ABE && ((m_context->ALPHA.A == 1 && m_context->ALPHA.B == 0 && GetAlphaMinMax().min >= 128) || m_context->ALPHA.IsBlack()) && m_draw_env->COLCLAMP.CLAMP == 1;
+ const bool can_expand = !(m_cached_ctx.ZBUF.ZMSK && output_black);
+
// Estimate size based on the scissor rectangle and height cache.
- const GSVector2i t_size = GetTargetSize(src);
+ const GSVector2i t_size = GetTargetSize(src, can_expand);
const GSVector4i t_size_rect = GSVector4i::loadh(t_size);
// Ensure draw rect is clamped to framebuffer size. Necessary for updating valid area.
@@ -3404,20 +3410,18 @@ void GSRendererHW::Draw()
std::string s;
- if (GSConfig.SaveRT && s_n >= GSConfig.SaveN)
+ if (rt && GSConfig.SaveRT && s_n >= GSConfig.SaveN)
{
- s = GetDrawDumpPath("%05d_f%lld_rt1_%05x_%s.bmp", s_n, frame, m_cached_ctx.FRAME.Block(), psm_str(m_cached_ctx.FRAME.PSM));
+ s = GetDrawDumpPath("%05d_f%lld_rt1_%05x_(%05x)_%s.bmp", s_n, frame, m_cached_ctx.FRAME.Block(), rt->m_TEX0.TBP0, psm_str(m_cached_ctx.FRAME.PSM));
- if (rt)
- rt->m_texture->Save(s);
+ rt->m_texture->Save(s);
}
- if (GSConfig.SaveDepth && s_n >= GSConfig.SaveN)
+ if (ds && GSConfig.SaveDepth && s_n >= GSConfig.SaveN)
{
- s = GetDrawDumpPath("%05d_f%lld_rz1_%05x_%s.bmp", s_n, frame, m_cached_ctx.ZBUF.Block(), psm_str(m_cached_ctx.ZBUF.PSM));
+ s = GetDrawDumpPath("%05d_f%lld_rz1_%05x_(%05x)_%s.bmp", s_n, frame, m_cached_ctx.ZBUF.Block(), ds->m_TEX0.TBP0, psm_str(m_cached_ctx.ZBUF.PSM));
- if (ds)
- ds->m_texture->Save(s);
+ ds->m_texture->Save(s);
}
if (GSConfig.SaveL > 0 && (s_n - GSConfig.SaveN) > GSConfig.SaveL)
@@ -3429,6 +3433,8 @@ void GSRendererHW::Draw()
if (rt)
rt->m_last_draw = s_n;
+ if (ds)
+ ds->m_last_draw = s_n;
#ifdef DISABLE_HW_TEXTURE_CACHE
if (rt)
g_texture_cache->Read(rt, real_rect);
@@ -7407,9 +7413,10 @@ ClearType GSRendererHW::IsConstantDirectWriteMemClear()
&& !(m_draw_env->SCANMSK.MSK & 2) && !m_cached_ctx.TEST.ATE // no alpha test
&& !m_cached_ctx.TEST.DATE // no destination alpha test
&& (!m_cached_ctx.TEST.ZTE || m_cached_ctx.TEST.ZTST == ZTST_ALWAYS) // no depth test
- && (m_vt.m_eq.rgba == 0xFFFF || m_vertex.next == 2)) // constant color write
+ && (m_vt.m_eq.rgba == 0xFFFF || m_vertex.next == 2) // constant color write
+ && (!PRIM->FGE || m_vt.m_min.p.w == 255.0f)) // No fog effect
{
- if (PRIM->ABE && (!m_context->ALPHA.IsOpaque() || m_cached_ctx.FRAME.FBMSK))
+ if ((PRIM->ABE && !m_context->ALPHA.IsOpaque()) || (m_cached_ctx.FRAME.FBMSK & GSLocalMemory::m_psm[m_cached_ctx.FRAME.PSM].fmsk))
return ClearWithDraw;
return NormalClear;
diff --git a/pcsx2/GS/Renderers/HW/GSRendererHW.h b/pcsx2/GS/Renderers/HW/GSRendererHW.h
index 1ed0d96d92cb0..fced6cb8bdec6 100644
--- a/pcsx2/GS/Renderers/HW/GSRendererHW.h
+++ b/pcsx2/GS/Renderers/HW/GSRendererHW.h
@@ -214,7 +214,7 @@ class GSRendererHW : public GSRenderer
void MergeSprite(GSTextureCache::Source* tex);
float GetTextureScaleFactor() override;
GSVector2i GetValidSize(const GSTextureCache::Source* tex = nullptr);
- GSVector2i GetTargetSize(const GSTextureCache::Source* tex = nullptr);
+ GSVector2i GetTargetSize(const GSTextureCache::Source* tex = nullptr, const bool can_expand = true);
void Reset(bool hardware_reset) override;
void UpdateSettings(const Pcsx2Config::GSOptions& old_config) override;
diff --git a/pcsx2/GS/Renderers/HW/GSTextureCache.cpp b/pcsx2/GS/Renderers/HW/GSTextureCache.cpp
index 25a7df769a28d..cd1b87024e181 100644
--- a/pcsx2/GS/Renderers/HW/GSTextureCache.cpp
+++ b/pcsx2/GS/Renderers/HW/GSTextureCache.cpp
@@ -2569,7 +2569,7 @@ bool GSTextureCache::PreloadTarget(GIFRegTEX0 TEX0, const GSVector2i& size, cons
{
const GSVector4i save_rect = preserve_target ? newrect : eerect;
- if(!hw_clear)
+ if (!hw_clear)
dst->UpdateValidity(save_rect);
GL_INS("Preloading the RT DATA from updated GS Memory");
AddDirtyRectTarget(dst, save_rect, TEX0.PSM, TEX0.TBW, rgba, GSLocalMemory::m_psm[TEX0.PSM].trbpp >= 16);
@@ -2605,12 +2605,14 @@ bool GSTextureCache::PreloadTarget(GIFRegTEX0 TEX0, const GSVector2i& size, cons
auto j = i;
Target* t = *j;
- if (dst != t && t->m_TEX0.TBW == dst->m_TEX0.TBW && t->m_TEX0.PSM == dst->m_TEX0.PSM && t->m_TEX0.TBW > 4)
+ if (dst != t && t->m_TEX0.PSM == dst->m_TEX0.PSM/* && t->m_TEX0.TBW == dst->m_TEX0.TBW*/)
if (t->Overlaps(dst->m_TEX0.TBP0, dst->m_TEX0.TBW, dst->m_TEX0.PSM, dst->m_valid))
{
+ const u32 buffer_width = std::max(1U, dst->m_TEX0.TBW);
+
// If the two targets are misaligned, it's likely a relocation, so we can just kill the old target.
// Kill targets that are overlapping new targets, but ignore the copy if the old target is dirty because we favour GS memory.
- if (((((t->m_TEX0.TBP0 - dst->m_TEX0.TBP0) >> 5) % dst->m_TEX0.TBW) != 0) && !t->m_dirty.empty())
+ if (((((t->m_TEX0.TBP0 - dst->m_TEX0.TBP0) >> 5) % buffer_width) != 0) && !t->m_dirty.empty())
{
InvalidateSourcesFromTarget(t);
i = list.erase(j);
@@ -2629,64 +2631,84 @@ bool GSTextureCache::PreloadTarget(GIFRegTEX0 TEX0, const GSVector2i& size, cons
return hw_clear.value_or(false);
}
// The new texture is behind it but engulfs the whole thing, shrink the new target so it grows in the HW Draw resize.
- else if (dst->m_TEX0.TBP0 < t->m_TEX0.TBP0 && (dst->UnwrappedEndBlock() + 1) > t->m_TEX0.TBP0 && dst->m_TEX0.TBP0 < (t->UnwrappedEndBlock() + 1))
+ else if (dst->m_TEX0.TBP0 < t->m_TEX0.TBP0 && (dst->UnwrappedEndBlock() + 1) > t->m_TEX0.TBP0)
{
const int rt_pages = ((t->UnwrappedEndBlock() + 1) - t->m_TEX0.TBP0) >> 5;
const int overlapping_pages = std::min(rt_pages, static_cast((dst->UnwrappedEndBlock() + 1) - t->m_TEX0.TBP0) >> 5);
- const int overlapping_pages_height = (overlapping_pages / dst->m_TEX0.TBW) * GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y;
+ const int overlapping_pages_height = ((overlapping_pages + (buffer_width - 1)) / buffer_width) * GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y;
- if (overlapping_pages_height == 0 || (overlapping_pages % dst->m_TEX0.TBW))
+ if (overlapping_pages_height == 0 || (overlapping_pages % buffer_width))
{
// No overlap top copy or the widths don't match.
i++;
continue;
}
- const int dst_offset_width = (((t->m_TEX0.TBP0 - dst->m_TEX0.TBP0) >> 5) % dst->m_TEX0.TBW) * GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.x;
- const int dst_offset_height = ((((t->m_TEX0.TBP0 - dst->m_TEX0.TBP0) >> 5) / dst->m_TEX0.TBW) * GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y);
+ const int dst_offset_height = ((((t->m_TEX0.TBP0 - dst->m_TEX0.TBP0) >> 5) / buffer_width) * GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y);
+ const int texture_height = (dst->m_TEX0.TBW == t->m_TEX0.TBW) ? (dst_offset_height + t->m_valid.w) : (dst_offset_height + overlapping_pages_height);
+
+ if (texture_height > dst->m_unscaled_size.y && !dst->ResizeTexture(dst->m_unscaled_size.x, texture_height, true))
+ {
+ // Resize failed, probably ran out of VRAM, better luck next time. Fall back to CPU.
+ DevCon.Warning("Failed to resize target on preload? Draw %d", GSState::s_n);
+ i++;
+ continue;
+ }
+
+ const int dst_offset_width = (((t->m_TEX0.TBP0 - dst->m_TEX0.TBP0) >> 5) % buffer_width) * GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.x;
const int dst_offset_scaled_width = dst_offset_width * dst->m_scale;
const int dst_offset_scaled_height = dst_offset_height * dst->m_scale;
- const GSVector4i dst_rect_scale = GSVector4i(t->m_valid.x, dst_offset_height, t->m_valid.z, dst_offset_height + overlapping_pages_height);
+ const GSVector4i dst_rect_scale = GSVector4i(t->m_valid.x, dst_offset_height, t->m_valid.z, texture_height);
+
if (((!hw_clear && (preserve_target || preload)) || dst_rect_scale.rintersect(draw_rect).rempty()) && dst->GetScale() == t->GetScale())
{
- const int copy_width = ((t->m_texture->GetWidth()) > (dst->m_texture->GetWidth()) ? (dst->m_texture->GetWidth()) : t->m_texture->GetWidth()) - dst_offset_scaled_width;
- const int copy_height = overlapping_pages_height * t->m_scale;
+ int copy_width = ((t->m_texture->GetWidth()) > (dst->m_texture->GetWidth()) ? (dst->m_texture->GetWidth()) : t->m_texture->GetWidth()) - dst_offset_scaled_width;
+ int copy_height = (texture_height - dst_offset_height) * t->m_scale;
GL_INS("RT double buffer copy from FBP 0x%x, %dx%d => %d,%d", t->m_TEX0.TBP0, copy_width, copy_height, 0, dst_offset_scaled_height);
// Clear the dirty first
t->Update();
dst->Update();
+
+ // Clamp it if it gets too small, shouldn't happen but stranger things have happened.
+ if (copy_width < 0)
+ {
+ copy_width = 0;
+ }
+
// Invalidate has been moved to after DrawPrims(), because we might kill the current sources' backing.
if (!t->m_valid_rgb || !(t->m_valid_alpha_high || t->m_valid_alpha_low) || t->m_scale != dst->m_scale)
{
const GSVector4 src_rect = GSVector4(0, 0, copy_width, copy_height) / (GSVector4(t->m_texture->GetSize()).xyxy());
- const GSVector4 dst_rect = GSVector4(dst_offset_scaled_width, dst_offset_scaled_height, dst_offset_scaled_width + copy_width, dst_offset_scaled_width + copy_height);
+ const GSVector4 dst_rect = GSVector4(dst_offset_scaled_width, dst_offset_scaled_height, dst_offset_scaled_width + copy_width, dst_offset_scaled_height + copy_height);
g_gs_device->StretchRect(t->m_texture, src_rect, dst->m_texture, dst_rect, t->m_valid_rgb, t->m_valid_rgb, t->m_valid_rgb, t->m_valid_alpha_high || t->m_valid_alpha_low);
}
else
{
- // Invalidate has been moved to after DrawPrims(), because we might kill the current sources' backing.
+ if ((copy_width + dst_offset_scaled_width) > (dst->m_unscaled_size.x * dst->m_scale) || (copy_height + dst_offset_scaled_height) > (dst->m_unscaled_size.y * dst->m_scale))
+ {
+ copy_width = std::min(copy_width, static_cast((dst->m_unscaled_size.x * dst->m_scale) - dst_offset_scaled_width));
+ copy_height = std::min(copy_height, static_cast((dst->m_unscaled_size.y * dst->m_scale) - dst_offset_scaled_height));
+ }
+
g_gs_device->CopyRect(t->m_texture, dst->m_texture, GSVector4i(0, 0, copy_width, copy_height), dst_offset_scaled_width, dst_offset_scaled_height);
}
}
- if ((overlapping_pages < rt_pages) || (src && src->m_target && src->m_from_target == t))
- {
- // This should never happen as we're making a new target so the src should never be something it overlaps, but just incase..
- GSVector4i new_valid = t->m_valid;
- new_valid.y = std::max(new_valid.y - overlapping_pages_height, 0);
- new_valid.w = std::max(new_valid.w - overlapping_pages_height, 0);
- t->m_TEX0.TBP0 += (overlapping_pages_height / GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y) << 5;
- t->ResizeValidity(new_valid);
- }
- else
+ // src is using this target, so point it at the new copy.
+ if (src && src->m_target && src->m_from_target == t)
{
- InvalidateSourcesFromTarget(t);
- i = list.erase(j);
- delete t;
+ src->m_from_target = dst;
+ src->m_texture = dst->m_texture;
+ src->m_region.SetY(src->m_region.GetMinY() + dst_offset_height, src->m_region.GetMaxY() + dst_offset_height);
+ src->m_region.SetX(src->m_region.GetMinX() + dst_offset_width, src->m_region.GetMaxX() + dst_offset_width);
}
- return hw_clear.value_or(false);
+
+ InvalidateSourcesFromTarget(t);
+ i = list.erase(j);
+ delete t;
+ continue;
}
}
i++;
@@ -3736,6 +3758,19 @@ bool GSTextureCache::Move(u32 SBP, u32 SBW, u32 SPSM, int sx, int sy, u32 DBP, u
if (alpha_only && (!dst || GSLocalMemory::m_psm[dst->m_TEX0.PSM].bpp != 32))
return false;
+ // Beware of the case where a game might create a larger texture by moving a bunch of chunks around.
+ if (dst && DBP == SBP && dy > dst->m_unscaled_size.y)
+ {
+ const u32 new_DBP = DBP + (((dy / GSLocalMemory::m_psm[dst->m_TEX0.PSM].pgs.y) * DBW) << 5);
+
+ dst = nullptr;
+
+ DBP = new_DBP;
+ dy = 0;
+
+ dst = GetExactTarget(DBP, DBW, dpsm_s.depth ? DepthStencil : RenderTarget, DBP);
+ }
+
// Beware of the case where a game might create a larger texture by moving a bunch of chunks around.
// We use dx/dy == 0 and the TBW check as a safeguard to make sure these go through to local memory.
// We can also recreate the target if it's previously been created in the height cache with a valid size.
@@ -4182,7 +4217,7 @@ GSTextureCache::Target* GSTextureCache::FindOverlappingTarget(u32 BP, u32 BW, u3
return FindOverlappingTarget(BP, end_bp);
}
-GSVector2i GSTextureCache::GetTargetSize(u32 bp, u32 fbw, u32 psm, s32 min_width, s32 min_height)
+GSVector2i GSTextureCache::GetTargetSize(u32 bp, u32 fbw, u32 psm, s32 min_width, s32 min_height, bool can_expand)
{
TargetHeightElem search = {};
search.bp = bp;
@@ -4196,14 +4231,17 @@ GSVector2i GSTextureCache::GetTargetSize(u32 bp, u32 fbw, u32 psm, s32 min_width
TargetHeightElem& elem = const_cast(*it);
if (elem.bits == search.bits)
{
- if (elem.width < min_width || elem.height < min_height)
+ if (can_expand)
{
- DbgCon.WriteLn("Expand size at %x %u %u from %ux%u to %ux%u", bp, fbw, psm, elem.width, elem.height,
- min_width, min_height);
- }
+ if (elem.width < min_width || elem.height < min_height)
+ {
+ DbgCon.WriteLn("Expand size at %x %u %u from %ux%u to %ux%u", bp, fbw, psm, elem.width, elem.height,
+ min_width, min_height);
+ }
- elem.width = std::max(elem.width, min_width);
- elem.height = std::max(elem.height, min_height);
+ elem.width = std::max(elem.width, min_width);
+ elem.height = std::max(elem.height, min_height);
+ }
m_target_heights.MoveFront(it.Index());
elem.age = 0;
@@ -4211,7 +4249,7 @@ GSVector2i GSTextureCache::GetTargetSize(u32 bp, u32 fbw, u32 psm, s32 min_width
}
}
- DbgCon.WriteLn("New size at %x %u %u: %ux%u", bp, fbw, psm, min_width, min_height);
+ DbgCon.WriteLn("New size at %x %u %u: %ux%u draw %d", bp, fbw, psm, min_width, min_height, GSState::s_n);
m_target_heights.push_front(search);
return GSVector2i(min_width, min_height);
}
diff --git a/pcsx2/GS/Renderers/HW/GSTextureCache.h b/pcsx2/GS/Renderers/HW/GSTextureCache.h
index c2551f4b399d2..756ecd881e53f 100644
--- a/pcsx2/GS/Renderers/HW/GSTextureCache.h
+++ b/pcsx2/GS/Renderers/HW/GSTextureCache.h
@@ -504,7 +504,7 @@ class GSTextureCache
Target* FindOverlappingTarget(u32 BP, u32 end_bp) const;
Target* FindOverlappingTarget(u32 BP, u32 BW, u32 PSM, GSVector4i rc) const;
- GSVector2i GetTargetSize(u32 bp, u32 fbw, u32 psm, s32 min_width, s32 min_height);
+ GSVector2i GetTargetSize(u32 bp, u32 fbw, u32 psm, s32 min_width, s32 min_height, bool can_expand = true);
bool HasTargetInHeightCache(u32 bp, u32 fbw, u32 psm, u32 max_age = std::numeric_limits::max(), bool move_front = true);
bool Has32BitTarget(u32 bp);