GS: Only copy texture coverage area for current FRAME/Z sources

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -162,6 +162,13 @@ void GSRendererHW::PurgeTextureCache()
 	m_tc->RemoveAll();
 }
 
+bool GSRendererHW::IsPossibleTextureShuffle(GSTextureCache::Source* src) const
+{
+	return (PRIM->TME && m_vt.m_primclass == GS_SPRITE_CLASS &&
+		src->m_32_bits_fmt && GSLocalMemory::m_psm[src->m_TEX0.PSM].bpp == 16 &&
+		GSLocalMemory::m_psm[m_context->FRAME.PSM].bpp == 16);
+}
+
 void GSRendererHW::SetGameCRC(u32 crc, int options)
 {
 	GSRenderer::SetGameCRC(crc, options);
@@ -1711,6 +1718,11 @@ void GSRendererHW::Draw()
 
 	//
 
+	// Temporary source *must* be invalidated before normal, because otherwise it'll be double freed.
+	m_tc->InvalidateTemporarySource();
+
+	//
+
 	if ((fm & fm_mask) != fm_mask && rt)
 	{
 		//rt->m_valid = rt->m_valid.runion(r);

pcsx2/GS/Renderers/HW/GSRendererHW.h

@@ -186,4 +186,7 @@ public:
 
 	// Called by the texture cache to know if current texture is useful
 	virtual bool IsDummyTexture() const { return false; }
+
+	// Called by the texture cache when optimizing the copy range for sources
+	bool IsPossibleTextureShuffle(GSTextureCache::Source* src) const;
 };

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -391,7 +391,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const GIFRegTEX0& TEX0, con
 			GL_CACHE("TC: src miss (0x%x, 0x%x, %s)", TEX0.TBP0, psm_s.pal > 0 ? TEX0.CBP : 0, psm_str(TEX0.PSM));
 		}
 #endif
-		src = CreateSource(TEX0, TEXA, dst, half_right, x_offset, y_offset, lod);
+		src = CreateSource(TEX0, TEXA, dst, half_right, x_offset, y_offset, lod, &r);
 		new_source = true;
 	}
 	else
@@ -1165,7 +1165,7 @@ void GSTextureCache::IncAge()
 }
 
 //Fixme: Several issues in here. Not handling depth stencil, pitch conversion doesnt work.
-GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Target* dst, bool half_right, int x_offset, int y_offset, const GSVector2i* lod)
+GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Target* dst, bool half_right, int x_offset, int y_offset, const GSVector2i* lod, const GSVector4i* src_range)
 {
 	const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[TEX0.PSM];
 	Source* src = new Source(m_renderer, TEX0, TEXA, false);
@@ -1340,8 +1340,10 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
 
 		GSVector2 scale = is_8bits ? GSVector2(1, 1) : dst->m_texture->GetScale();
 
-		GSVector4i sRect(0, 0, w, h);
 		const bool use_texture = shader == ShaderConvert::COPY;
+		GSVector4i sRect(0, 0, w, h);
+		int destX = 0;
+		int destY = 0;
 
 		if (half_right)
 		{
@@ -1361,6 +1363,27 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
 				DevCon.Error("Invalid half-right copy with width %d from %dx%d texture", half_width * 2, w, h);
 			}
 		}
+		else if (src_range && dst->m_TEX0.TBW == TEX0.TBW && !is_8bits)
+		{
+			// optimization for TBP == FRAME
+			const GSDrawingContext* const context = m_renderer->m_context;
+			if (context->FRAME.Block() == TEX0.TBP0 || context->ZBUF.Block() == TEX0.TBP0)
+			{
+				// if it looks like a texture shuffle, we might read up to +/- 8 pixels on either side.
+				GSVector4 adjusted_src_range(*src_range);
+				if (static_cast<GSRendererHW*>(m_renderer)->IsPossibleTextureShuffle(src))
+					adjusted_src_range += GSVector4(-8.0f, 0.0f, 8.0f, 0.0f);
+
+				// don't forget to scale the copy range
+				adjusted_src_range = adjusted_src_range * GSVector4(scale).xyxy();
+				sRect = sRect.rintersect(GSVector4i(adjusted_src_range));
+				destX = sRect.x;
+				destY = sRect.y;
+
+				// clean up immediately afterwards
+				m_temporary_source = src;
+			}
+		}
 
 		// Don't be fooled by the name. 'dst' is the old target (hence the input)
 		// 'src' is the new texture cache entry (hence the output)
@@ -1380,7 +1403,7 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
 
 		if (use_texture)
 		{
-			g_gs_device->CopyRect(sTex, dTex, sRect, 0, 0);
+			g_gs_device->CopyRect(sTex, dTex, sRect, destX, destY);
 		}
 		else
 		{
@@ -1388,7 +1411,7 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
 			sRectF.z /= sTex->GetWidth();
 			sRectF.w /= sTex->GetHeight();
 
-			g_gs_device->StretchRect(sTex, sRectF, dTex, GSVector4(0, 0, w, h), shader, false);
+			g_gs_device->StretchRect(sTex, sRectF, dTex, GSVector4(destX, destY, w, h), shader, false);
 		}
 
 		if (src->m_texture)
@@ -2439,6 +2462,15 @@ GSTextureCache::SurfaceOffset GSTextureCache::ComputeSurfaceOffset(const Surface
 	return so;
 }
 
+void GSTextureCache::InvalidateTemporarySource()
+{
+	if (!m_temporary_source)
+		return;
+
+	m_src.RemoveAt(m_temporary_source);
+	m_temporary_source = nullptr;
+}
+
 void GSTextureCache::InjectHashCacheTexture(const HashCacheKey& key, GSTexture* tex)
 {
 	auto it = m_hash_cache.find(key);

pcsx2/GS/Renderers/HW/GSTextureCache.h

@@ -285,8 +285,9 @@ protected:
 	static u8* m_temp;
 	constexpr static size_t S_SURFACE_OFFSET_CACHE_MAX_SIZE = std::numeric_limits<u16>::max();
 	std::unordered_map<SurfaceOffsetKey, SurfaceOffset, SurfaceOffsetKeyHash, SurfaceOffsetKeyEqual> m_surface_offset_cache;
+	Source* m_temporary_source = nullptr; // invalidated after the draw
 
-	Source* CreateSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Target* t = NULL, bool half_right = false, int x_offset = 0, int y_offset = 0, const GSVector2i* lod = nullptr);
+	Source* CreateSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Target* t = NULL, bool half_right = false, int x_offset = 0, int y_offset = 0, const GSVector2i* lod = nullptr, const GSVector4i* src_range = nullptr);
 	Target* CreateTarget(const GIFRegTEX0& TEX0, int w, int h, int type, const bool clear);
 
 	HashCacheEntry* LookupHashCache(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, bool& paltex, const u32* clut, const GSVector2i* lod);
@@ -333,6 +334,9 @@ public:
 	SurfaceOffset ComputeSurfaceOffset(const uint32_t bp, const uint32_t bw, const uint32_t psm, const GSVector4i& r, const Target* t);
 	SurfaceOffset ComputeSurfaceOffset(const SurfaceOffsetKey& sok);
 
+	/// Invalidates a temporary source, a partial copy only created from the current RT/DS for the current draw.
+	void InvalidateTemporarySource();
+
 	/// Injects a texture into the hash cache, by using GSTexture::Swap(), transitively applying to all sources. Ownership of tex is transferred.
 	void InjectHashCacheTexture(const HashCacheKey& key, GSTexture* tex);
 };