rsx/gl/scaling: Fixes

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -692,6 +692,15 @@ bool GLGSRender::load_program()
 		else
 			surface = m_rtts.get_texture_from_depth_stencil_if_applicable(texaddr);
 
+		if (!surface)
+		{
+			auto rsc = m_rtts.get_surface_subresource_if_applicable(texaddr, 0, 0, tex.pitch());
+			if (!rsc.surface || rsc.is_depth_surface != is_depth)
+				return std::make_tuple(false, 0);
+
+			surface = rsc.surface;
+		}
+
 		return std::make_tuple(true, surface->get_native_pitch());
 	};
 

rpcs3/Emu/RSX/GL/GLRenderTargets.h

@@ -298,11 +298,12 @@ struct surface_subresource
 
 	bool is_bound = false;
 	bool is_depth_surface = false;
+	bool is_clipped = false;
 
 	surface_subresource() {}
 
-	surface_subresource(gl::render_target *src, u16 X, u16 Y, u16 W, u16 H, bool _Bound, bool _Depth)
+	surface_subresource(gl::render_target *src, u16 X, u16 Y, u16 W, u16 H, bool _Bound, bool _Depth, bool _Clipped = false)
-		: surface(src), x(X), y(Y), w(W), h(H), is_bound(_Bound), is_depth_surface(_Depth)
+		: surface(src), x(X), y(Y), w(W), h(H), is_bound(_Bound), is_depth_surface(_Depth), is_clipped(_Clipped)
 	{}
 };
 
@@ -361,7 +362,7 @@ private:
 	}
 
 public:
-	surface_subresource get_surface_subresource_if_applicable(u32 texaddr, u16 requested_width, u16 requested_height, u16 requested_pitch, bool scale_to_fit =false)
+	surface_subresource get_surface_subresource_if_applicable(u32 texaddr, u16 requested_width, u16 requested_height, u16 requested_pitch, bool scale_to_fit =false, bool crop=false)
 	{
 		gl::render_target *surface = nullptr;
 		bool is_subslice = false;
@@ -390,18 +391,18 @@ public:
 					return{ surface, x_offset, y_offset, requested_width, requested_height, is_bound(this_address, false), false };
 				else
 				{
-					if (scale_to_fit) //Forcefully fit the requested region by clipping and scaling
+					if (crop) //Forcefully fit the requested region by clipping and scaling
 					{
 						u16 remaining_width = dims.first - x_offset;
 						u16 remaining_height = dims.second - y_offset;
 
-						return{ surface, x_offset, y_offset, remaining_width, remaining_height, is_bound(this_address, false), false };
+						return{ surface, x_offset, y_offset, remaining_width, remaining_height, is_bound(this_address, false), false, true };
 					}
 
 					if (dims.first >= requested_width && dims.second >= requested_height)
 					{
 						LOG_WARNING(RSX, "Overlapping surface exceeds bounds; returning full surface region");
-						return{ surface, 0, 0, requested_width, requested_height, is_bound(this_address, false), false };
+						return{ surface, 0, 0, requested_width, requested_height, is_bound(this_address, false), false, true };
 					}
 				}
 			}
@@ -430,18 +431,18 @@ public:
 					return{ surface, x_offset, y_offset, requested_width, requested_height, is_bound(this_address, true), true };
 				else
 				{
-					if (scale_to_fit) //Forcefully fit the requested region by clipping and scaling
+					if (crop) //Forcefully fit the requested region by clipping and scaling
 					{
 						u16 remaining_width = dims.first - x_offset;
 						u16 remaining_height = dims.second - y_offset;
 
-						return{ surface, x_offset, y_offset, remaining_width, remaining_height, is_bound(this_address, false), false };
+						return{ surface, x_offset, y_offset, remaining_width, remaining_height, is_bound(this_address, true), true, true };
 					}
 
 					if (dims.first >= requested_width && dims.second >= requested_height)
 					{
 						LOG_WARNING(RSX, "Overlapping depth surface exceeds bounds; returning full surface region");
-						return{ surface, 0, 0, requested_width, requested_height, is_bound(this_address, true), true };
+						return{ surface, 0, 0, requested_width, requested_height, is_bound(this_address, true), true, true };
 					}
 				}
 			}

rpcs3/Emu/RSX/GL/GLTextureCache.h

@@ -754,7 +754,11 @@ namespace gl
 			 * a bound render target. We can bypass the expensive download in this case
 			 */
 
-			surface_subresource rsc = m_rtts.get_surface_subresource_if_applicable(texaddr, tex.width(), tex.height(), tex.pitch(), true);
+			const u32 format = tex.format() & ~(CELL_GCM_TEXTURE_LN | CELL_GCM_TEXTURE_UN);
+			const f32 internal_scale = (f32)tex.pitch() / (tex.width() * get_format_block_size_in_bytes(format));
+			const u32 internal_width = tex.width() * internal_scale;
+
+			const surface_subresource rsc = m_rtts.get_surface_subresource_if_applicable(texaddr, internal_width, tex.height(), tex.pitch(), true);
 			if (rsc.surface)
 			{
 				//Check that this region is not cpu-dirty before doing a copy
@@ -782,8 +786,6 @@ namespace gl
 					}
 					else
 					{
-						const u32 format = tex.format() & ~(CELL_GCM_TEXTURE_LN | CELL_GCM_TEXTURE_UN);
-
 						GLenum src_format = (GLenum)rsc.surface->get_internal_format();
 						GLenum dst_format = std::get<0>(get_format_type(format));
 
@@ -1075,7 +1077,7 @@ namespace gl
 			const u32 dst_address = (u32)((u64)dst.pixels - (u64)vm::base(0));
 
 			//Check if src/dst are parts of render targets
-			surface_subresource src_subres = m_rtts.get_surface_subresource_if_applicable(src_address, src.width, src.slice_h, src.pitch, true);
+			surface_subresource src_subres = m_rtts.get_surface_subresource_if_applicable(src_address, src.width, src.slice_h, src.pitch, true, true);
 			src_is_render_target = src_subres.surface != nullptr;
 
 			float scale_x = (f32)dst.width / src.width;
@@ -1147,7 +1149,7 @@ namespace gl
 				source_texture = src_subres.surface->id();
 			}
 
-			surface_subresource dst_subres = m_rtts.get_surface_subresource_if_applicable(dst_address, dst.width, dst.clip_height, dst.pitch, true);
+			surface_subresource dst_subres = m_rtts.get_surface_subresource_if_applicable(dst_address, dst.width, dst.clip_height, dst.pitch, true, true);
 			dst_is_render_target = dst_subres.surface != nullptr;
 
 			if (!dst_is_render_target)