Rewrite YUV processing pipeline.

Use a 4-channel texture instead for packed YUV data so
every pixel in the texture has all the 4:2:2 color data
available.

data/yuv.frag

@@ -7,26 +7,24 @@ uniform mat3 color_matrix;
 uniform float inv_gamma;
 uniform float blacklevel;
 
-varying vec2 top_left_uv;
-varying vec2 top_right_uv;
-varying vec2 bottom_left_uv;
-varying vec2 bottom_right_uv;
+varying vec2 texture_coord;
 
 void
 main()
 {
-    // Note the coordinates for texture samples need to be a varying, as the
-    // Mali-400 has this as a fast path allowing 32-bit floats. Otherwise
-    // they end up as 16-bit floats and that's not accurate enough.
+    // Sample format: Y,U,Y,V
+    vec4 samples = texture2D(texture, texture_coord);
+    float y = samples.x;
+    float u = samples.y-0.5;
+    float v = samples.w-0.5;
 
-    vec4 samples = vec4(texture2D(texture, top_left_uv).r,
-    texture2D(texture, top_right_uv).r,
-    texture2D(texture, bottom_left_uv).r,
-    texture2D(texture, bottom_right_uv).r);
-    vec3 color = vec3(samples.z, samples.z, samples.z);
+    vec3 rgb;
+    rgb.r = y + (1.403 * v);
+    rgb.g = y - (0.344 * u) - (0.714 * v);
+    rgb.b = y + (1.770 * u);
 
     //color *= color_matrix;
-    vec3 gamma_color = pow(color, vec3(inv_gamma));
+    vec3 gamma_color = pow(rgb, vec3(inv_gamma));
 
     gl_FragColor = vec4(gamma_color, 1);
 }

data/yuv.vert

@@ -8,18 +8,11 @@ attribute vec2 tex_coord;
 uniform mat3 transform;
 uniform vec2 pixel_size;
 
-varying vec2 top_left_uv;
-varying vec2 top_right_uv;
-varying vec2 bottom_left_uv;
-varying vec2 bottom_right_uv;
+varying vec2 texture_coord;
 
 void
 main()
 {
-    top_left_uv = tex_coord - pixel_size / 2.0;
-    bottom_right_uv = tex_coord + pixel_size / 2.0;
-    top_right_uv = vec2(top_left_uv.x, bottom_right_uv.y);
-    bottom_left_uv = vec2(bottom_right_uv.x, top_left_uv.y);
-
+    texture_coord = vert.xy * vec2(0.5, 0.5) + vec2(0.5, 0.5);
     gl_Position = vec4(transform * vec3(vert, 1), 1);
 }

src/gles2_debayer.c

@@ -18,6 +18,7 @@ gles2_debayer_new(int format)
         uint32_t pixfmt = libmegapixels_format_to_v4l_pixfmt(format);
 
         int shader = 0;
+        int texture_mode = 0;
 
         switch (pixfmt) {
                 case V4L2_PIX_FMT_SBGGR8:
@@ -33,15 +34,18 @@ gles2_debayer_new(int format)
                 case V4L2_PIX_FMT_SGRBG12:
                 case V4L2_PIX_FMT_SRGGB12:
                         shader = SHADER_DEBAYER;
+                        texture_mode = TEXTURE_BAYER;
                         break;
                 case V4L2_PIX_FMT_SBGGR10P:
                 case V4L2_PIX_FMT_SGBRG10P:
                 case V4L2_PIX_FMT_SGRBG10P:
                 case V4L2_PIX_FMT_SRGGB10P:
                         shader = SHADER_DEBAYER_PACKED;
+                        texture_mode = TEXTURE_BAYER;
                         break;
                 case V4L2_PIX_FMT_YUYV:
                         shader = SHADER_YUV;
+                        texture_mode = TEXTURE_YUV;
                         break;
         }
 
@@ -86,6 +90,7 @@ gles2_debayer_new(int format)
         GLES2Debayer *self = malloc(sizeof(GLES2Debayer));
         self->format = format;
         self->shader = shader;
+        self->texture_mode = texture_mode;
 
         self->frame_buffer = frame_buffer;
         self->program = program;

src/gles2_debayer.h

@@ -8,10 +8,14 @@
 #define SHADER_DEBAYER_PACKED 2
 #define SHADER_YUV 3
 
+#define TEXTURE_BAYER 1
+#define TEXTURE_YUV 2
+
 typedef struct {
         int format;
         int shader;
         float forward_matrix[9];
+        int texture_mode;
 
         GLuint frame_buffer;
         GLuint program;

src/process_pipeline.c

@@ -406,7 +406,7 @@ static void
 auto_focus_step(const struct focus_stats *stats)
 {
 	static uint64_t best_sharp, best_focus;
-	static const int debug = 1;
+	static const int debug = 0;
 
 	if (focus_phase >= PH_DONE) {
 		if (debug>1) printf("Phase %d, sharp %d best %d ", focus_phase, (int)(stats->sharp / 10000), (int)(best_sharp/ 10000));
@@ -616,7 +616,7 @@ process_aaa()
                 update_control(&state_proc.gain);
                 update_control(&state_proc.dgain);
                 update_control(&state_proc.exposure);
-		summarize();
+		//summarize();
         }
 
         if (auto_balance) {
@@ -682,6 +682,8 @@ process_image_for_preview(const uint8_t *image)
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+        switch(gles2_debayer->texture_mode) {
+        case TEXTURE_BAYER:
                 glTexImage2D(
                         GL_TEXTURE_2D,
                         0,
@@ -695,6 +697,22 @@ process_image_for_preview(const uint8_t *image)
                         GL_LUMINANCE,
                         GL_UNSIGNED_BYTE,
                         image);
+                break;
+        case TEXTURE_YUV:
+                glTexImage2D(
+                        GL_TEXTURE_2D,
+                        0,
+                        GL_RGBA8,
+                        state_proc.mode->width/2 +
+                                libmegapixels_mode_width_to_padding(state_proc.mode->format,
+                                                                    state_proc.mode->width),
+                        state_proc.mode->height,
+                        0,
+                        GL_RGBA,
+                        GL_UNSIGNED_BYTE,
+                        image);
+                break;
+        }
         check_gl();
 
         gles2_debayer_process(