From 904e1e4be361cccabdc70fc71473baaf6a84cbe8 Mon Sep 17 00:00:00 2001
From: Vicki Pfau <vi@endrift.com>
Date: Sun, 11 Mar 2018 22:13:13 -0700
Subject: [PATCH 1/2] Improve speed of framebuffer tiling

---
 nx/source/gfx/gfx.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/nx/source/gfx/gfx.c b/nx/source/gfx/gfx.c
index 8bfa2208..5d80c07a 100644
--- a/nx/source/gfx/gfx.c
+++ b/nx/source/gfx/gfx.c
@@ -559,10 +559,23 @@ void gfxFlushBuffers(void) {
         size_t width = g_gfx_framebuf_display_width;
         size_t height = g_gfx_framebuf_display_height;
         u32 *in_framebuf = (u32*)g_gfxFramebufLinear;
+        u128 *out_framebuf = (u128*)actual_framebuf;
 
         for (y=0; y<height; y++) {
-            for (x=0; x<width; x+=4) {
-                *((u128*)&actual_framebuf[gfxGetFramebufferDisplayOffset(x, y)]) = *((u128*)&in_framebuf[y * width + x]);
+            size_t row = y;
+            if (g_gfx_drawflip) row = g_gfx_framebuf_display_height-1-row;
+
+            size_t y_offset = (row & 0x70) * 4 + (row & ~0x7F) * (g_gfx_framebuf_aligned_width / 16) * 4;
+            y_offset += (row & 8) * 4 + (row & 6) * 2 + (row & 1);
+
+            for (x=0; x<width; x+=16) {
+                size_t x_offset = (x & ~0xF) * 0x20;
+                u128 *in_pos = (u128*)&in_framebuf[y * width + x];
+                u128 *out_pos = &out_framebuf[y_offset + x_offset];
+                out_pos[0] = in_pos[0];
+                out_pos[2] = in_pos[1];
+                out_pos[16] = in_pos[2];
+                out_pos[18] = in_pos[3];
             }
         }
     }

From 84ae00facf698c454515fc1889044dc56f696470 Mon Sep 17 00:00:00 2001
From: Vicki Pfau <vi@endrift.com>
Date: Tue, 13 Mar 2018 18:50:31 -0700
Subject: [PATCH 2/2] Hoisting and slight simplification

---
 nx/source/gfx/gfx.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/nx/source/gfx/gfx.c b/nx/source/gfx/gfx.c
index 5d80c07a..3d5720cf 100644
--- a/nx/source/gfx/gfx.c
+++ b/nx/source/gfx/gfx.c
@@ -563,19 +563,22 @@ void gfxFlushBuffers(void) {
 
         for (y=0; y<height; y++) {
             size_t row = y;
+            size_t y_base = y * width;
             if (g_gfx_drawflip) row = g_gfx_framebuf_display_height-1-row;
 
-            size_t y_offset = (row & 0x70) * 4 + (row & ~0x7F) * (g_gfx_framebuf_aligned_width / 16) * 4;
-            y_offset += (row & 8) * 4 + (row & 6) * 2 + (row & 1);
+            size_t y_offset = (row & 0x78) + (row & ~0x7F) * (g_gfx_framebuf_aligned_width / 16);
+            y_offset = y_offset * 4 + (row & 6) * 2 + (row & 1);
+
+            size_t x_offset = 0;
 
             for (x=0; x<width; x+=16) {
-                size_t x_offset = (x & ~0xF) * 0x20;
-                u128 *in_pos = (u128*)&in_framebuf[y * width + x];
+                u128 *in_pos = (u128*)&in_framebuf[y_base + x];
                 u128 *out_pos = &out_framebuf[y_offset + x_offset];
                 out_pos[0] = in_pos[0];
                 out_pos[2] = in_pos[1];
                 out_pos[16] = in_pos[2];
                 out_pos[18] = in_pos[3];
+                x_offset += 0x200;
             }
         }
     }