From 82af17beb16c10152e4ac96a4c33bd2408ef4dca Mon Sep 17 00:00:00 2001
From: kd-11 <karokidii@gmail.com>
Date: Thu, 16 Jan 2020 17:24:04 +0300
Subject: [PATCH] gl: Optimize image operations

- Avoid double transfers where a transfer to a temp image is done
without scaling and then a secondary transfer follows. Combines the two
steps into one whenever possible which can significantly alleviate
bandwidth problems at higher resolutions. Significant speedup, upto 90%
in some cases (PDF, PDF2)
---
 rpcs3/Emu/RSX/GL/GLHelpers.cpp    | 29 +++++++++++++++++++++++++++++
 rpcs3/Emu/RSX/GL/GLTextureCache.h | 11 +++++++++++
 2 files changed, 40 insertions(+)

diff --git a/rpcs3/Emu/RSX/GL/GLHelpers.cpp b/rpcs3/Emu/RSX/GL/GLHelpers.cpp
index 6dc8bbebb6..9c8ba3fd08 100644
--- a/rpcs3/Emu/RSX/GL/GLHelpers.cpp
+++ b/rpcs3/Emu/RSX/GL/GLHelpers.cpp
@@ -382,6 +382,35 @@ namespace gl
 		const gl::texture* real_src = src;
 		const gl::texture* real_dst = dst;
 
+		// Optimization pass; check for pass-through data transfer
+		if (!xfer_info.flip_horizontal && !xfer_info.flip_vertical && src_rect.height() == dst_rect.height())
+		{
+			auto src_w = src_rect.width();
+			auto dst_w = dst_rect.width();
+
+			if (xfer_info.src_is_typeless) src_w *= xfer_info.src_scaling_hint;
+			if (xfer_info.dst_is_typeless) dst_w *= xfer_info.dst_scaling_hint;
+
+			if (src_w == dst_w)
+			{
+				// Final dimensions are a match
+				if (xfer_info.src_is_typeless || xfer_info.dst_is_typeless)
+				{
+					const coord3i src_region = { { src_rect.x1, src_rect.y1, 0 }, { src_w, src_rect.height(), 1 } };
+					const coord3i dst_region = { { dst_rect.x1, dst_rect.y1, 0 }, { src_w, src_rect.height(), 1 } };
+					gl::copy_typeless(dst, src, dst_region, src_region);
+				}
+				else
+				{
+					glCopyImageSubData(src->id(), dst->id(), 0, src_rect.x1, src_rect.y1, 0,
+						dst->id(), GL_TEXTURE_2D, 0, dst_rect.x1, dst_rect.y1, 0,
+						src_w, src_rect.height(), 1);
+				}
+
+				return;
+			}
+		}
+
 		if (xfer_info.src_is_typeless)
 		{
 			const auto internal_fmt = xfer_info.src_native_format_override ?
diff --git a/rpcs3/Emu/RSX/GL/GLTextureCache.h b/rpcs3/Emu/RSX/GL/GLTextureCache.h
index ab28008155..3e7a033679 100644
--- a/rpcs3/Emu/RSX/GL/GLTextureCache.h
+++ b/rpcs3/Emu/RSX/GL/GLTextureCache.h
@@ -609,6 +609,17 @@ namespace gl
 					const u16 src_w2 = u16(src_w * src_bpp) / dst_bpp;
 					const u16 src_x2 = u16(src_x * src_bpp) / dst_bpp;
 
+					if (src_w2 == slice.dst_w && src_h == slice.dst_h && slice.level == 0)
+					{
+						// Optimization, avoid typeless copy to tmp followed by data copy to dst
+						// Combine the two transfers into one
+						const coord3u src_region = { { src_x, src_y, 0 }, { src_w, src_h, 1 } };
+						const coord3u dst_region = { { slice.dst_x, slice.dst_y, slice.dst_z }, { slice.dst_w, slice.dst_h, 1 } };
+						gl::copy_typeless(dst_image, slice.src, dst_region, src_region);
+
+						continue;
+					}
+
 					const coord3u src_region = { { src_x, src_y, 0 }, { src_w, src_h, 1 } };
 					const coord3u dst_region = { { src_x2, src_y, 0 }, { src_w2, src_h, 1 } };
 					gl::copy_typeless(src_image, slice.src, dst_region, src_region);