Downscaling

1 files changed, 54 insertions, 41 deletions

diff --git a/runtime.c b/runtime.c
index 144fbc2..5444139 100644
--- a/runtime.c
+++ b/runtime.c
@@ -152,6 +152,7 @@
 //    - Self-contained impl
 //    - Anti-aliasing
 //  - System
+//    - Dynamic libraries
 //    - Window - X11, Web
 //    - Screenshot - X11, Wayland
 //    - Clipboard
@@ -1846,11 +1847,7 @@ static void convert_pixels_for_window_(void) {
     return;
   }
 
-  //  TODO, PERF: Improve performance.
-
   //  Downscale pixels from vec4_f32 pixel buffer to internal u32 pixel buffer.
-  //
-  //  FIXME: Check if the scale is 1:1.
   {
     u32 dst_width  = g_platform.frame_width  / aa_scale;
     u32 dst_height = g_platform.frame_height / aa_scale;
@@ -1873,49 +1870,62 @@ static void convert_pixels_for_window_(void) {
 
     PROFILER_begin(PROFILE_DOWNSCALE);
 
-    f32       k = 1.f / (aa_scale * aa_scale);
-    vec4_f32 *s = g_platform.pixels;
-    u32      *d = _internal_pixels;
+    if (aa_scale == 1) {
+      vec4_f32 *s     = g_platform.pixels;
+      u32      *d     = _internal_pixels;
+      u32      *d_end = d + dst_width * dst_height;
 
-    for (u32 j = 0; j < dst_height; ++j) {
-      mem_set_(_internal_row, 0, dst_width * sizeof *_internal_row);
+      if (g_platform.enable_gamma)
+        for (; d < d_end; ++d, ++s)
+          *d = rgb_u32_from_f32_((vec3_f32) { gamma_(s->x), gamma_(s->y), gamma_(s->z) }) | 0xff000000;
+      else
+        for (; d < d_end; ++d, ++s)
+          *d = rgb_u32_from_f32_((vec3_f32) { s->x, s->y, s->z }) | 0xff000000;
+    } else {
+      f32       k = 1.f / (aa_scale * aa_scale);
+      vec4_f32 *s = g_platform.pixels;
+      u32      *d = _internal_pixels;
 
-      for (u32 jj = 0; jj < aa_scale; ++jj) {
-        vec3_f32 *im     = _internal_row;
-        vec3_f32 *im_end = im + dst_width;
+      for (u32 j = 0; j < dst_height; ++j) {
+        mem_set_(_internal_row, 0, dst_width * sizeof *_internal_row);
+
+        for (u32 jj = 0; jj < aa_scale; ++jj) {
+          vec3_f32 *im     = _internal_row;
+          vec3_f32 *im_end = im + dst_width;
 
-        for (; im < im_end; ++im) {
-          vec4_f32 *s_end = s + aa_scale;
+          for (; im < im_end; ++im) {
+            vec4_f32 *s_end = s + aa_scale;
 
-          for (; s < s_end; ++s) {
-            im->x += s->x;
-            im->y += s->y;
-            im->z += s->z;
+            for (; s < s_end; ++s) {
+              im->x += s->x;
+              im->y += s->y;
+              im->z += s->z;
+            }
           }
         }
-      }
 
-      u32      *d_end  = d + dst_width;
-      vec3_f32 *im     = _internal_row;
-      vec3_f32 *im_end = _internal_row + dst_width;
+        u32      *d_end  = d + dst_width;
+        vec3_f32 *im     = _internal_row;
+        vec3_f32 *im_end = _internal_row + dst_width;
 
-      if (g_platform.enable_gamma)
-        for (; im < im_end; ++im) {
-          im->x = gamma_(im->x * k);
-          im->y = gamma_(im->y * k);
-          im->z = gamma_(im->z * k);
-        }
-      else
-        for (; im < im_end; ++im) {
-          im->x *= k;
-          im->y *= k;
-          im->z *= k;
-        }
+        if (g_platform.enable_gamma)
+          for (; im < im_end; ++im) {
+            im->x = gamma_(im->x * k);
+            im->y = gamma_(im->y * k);
+            im->z = gamma_(im->z * k);
+          }
+        else
+          for (; im < im_end; ++im) {
+            im->x *= k;
+            im->y *= k;
+            im->z *= k;
+          }
 
-      im = _internal_row;
+        im = _internal_row;
 
-      for (; d < d_end; ++d, ++im)
-        *d = rgb_u32_from_f32_(*im) | 0xff000000;
+        for (; d < d_end; ++d, ++im)
+          *d = rgb_u32_from_f32_(*im) | 0xff000000;
+      }
     }
 
     PROFILER_end(PROFILE_DOWNSCALE);
@@ -1945,12 +1955,15 @@ static void convert_pixels_for_window_(void) {
     u32 x_ratio = (src_width  << 16) / dst_width;
     u32 y_ratio = (src_height << 16) / dst_height;
 
+    u32 x_half = x_ratio / 2;
+    u32 y_half = y_ratio / 2;
+
     if (src_len < dst_len)
       for (u32 j = dst_height - 1;; --j) {
-        u32 jj_w = ((j * y_ratio) >> 16) * src_width;
+        u32 jj_w = ((j * y_ratio + y_half) >> 16) * src_width;
         u32 j_w  = j * dst_width;
         for (u32 i = dst_width - 1;; --i) {
-          u32 ii = (i * x_ratio) >> 16;
+          u32 ii = (i * x_ratio + x_half) >> 16;
           _internal_pixels[j_w + i] = _internal_pixels[jj_w + ii];
           if (i == 0) break;
         }
@@ -1958,10 +1971,10 @@ static void convert_pixels_for_window_(void) {
       }
     else
       for (u32 j = 0; j < dst_height; ++j) {
-        u32 jj_w = ((j * y_ratio) >> 16) * src_width;
+        u32 jj_w = ((j * y_ratio + y_half) >> 16) * src_width;
         u32 j_w  = j * dst_width;
         for (u32 i = 0; i < dst_width; ++i) {
-          u32 ii = (i * x_ratio) >> 16;
+          u32 ii = (i * x_ratio + x_half) >> 16;
           _internal_pixels[j_w + i] = _internal_pixels[jj_w + ii];
         }
       }