DOF blur, near and far fields..

2 files changed, 249 insertions, 27 deletions

diff --git a/drivers/gles3/shaders/effect_blur.glsl b/drivers/gles3/shaders/effect_blur.glsl
index 589af64d44..89afa12f60 100644
--- a/drivers/gles3/shaders/effect_blur.glsl
+++ b/drivers/gles3/shaders/effect_blur.glsl
@@ -41,6 +41,40 @@ uniform float glow_strength;
 
 #endif
 
+#if defined(DOF_FAR_BLUR) || defined (DOF_NEAR_BLUR)
+
+#ifdef DOF_QUALITY_LOW
+const int dof_kernel_size=5;
+const int dof_kernel_from=2;
+const float dof_kernel[5] = float[] (0.153388,0.221461,0.250301,0.221461,0.153388);
+#endif
+
+#ifdef DOF_QUALITY_MEDIUM
+const int dof_kernel_size=11;
+const int dof_kernel_from=5;
+const float dof_kernel[11] = float[] (0.055037,0.072806,0.090506,0.105726,0.116061,0.119726,0.116061,0.105726,0.090506,0.072806,0.055037);
+
+#endif
+
+#ifdef DOF_QUALITY_HIGH
+const int dof_kernel_size=21;
+const int dof_kernel_from=10;
+const float dof_kernel[21] = float[] (0.028174,0.032676,0.037311,0.041944,0.046421,0.050582,0.054261,0.057307,0.059587,0.060998,0.061476,0.060998,0.059587,0.057307,0.054261,0.050582,0.046421,0.041944,0.037311,0.032676,0.028174);
+#endif
+
+uniform sampler2D dof_source_depth; //texunit:1
+uniform float dof_begin;
+uniform float dof_end;
+uniform vec2 dof_dir;
+uniform float dof_radius;
+
+#ifdef DOF_NEAR_BLUR_MERGE
+
+uniform sampler2D source_dof_original; //texunit:2
+#endif
+
+#endif
+
 
 #ifdef GLOW_FIRST_PASS
 
@@ -60,16 +94,23 @@ uniform float glow_hdr_scale;
 
 #endif
 
+uniform float camera_z_far;
+uniform float camera_z_near;
+
 void main() {
 
 
 
 #ifdef GAUSSIAN_HORIZONTAL
-	vec4 color =textureLod( source_color,  uv_interp+vec2( 0.0, 0.0)*pixel_size,lod )*0.38774;
-	color+=textureLod( source_color,  uv_interp+vec2( 1.0, 0.0)*pixel_size,lod )*0.24477;
-	color+=textureLod( source_color,  uv_interp+vec2( 2.0, 0.0)*pixel_size,lod )*0.06136;
-	color+=textureLod( source_color,  uv_interp+vec2(-1.0, 0.0)*pixel_size,lod )*0.24477;
-	color+=textureLod( source_color,  uv_interp+vec2(-2.0, 0.0)*pixel_size,lod )*0.06136;
+	vec2 pix_size = pixel_size;
+	pix_size*=0.5; //reading from larger buffer, so use more samples
+	vec4 color =textureLod( source_color,  uv_interp+vec2( 0.0, 0.0)*pix_size,lod )*0.214607;
+	color+=textureLod( source_color,  uv_interp+vec2( 1.0, 0.0)*pix_size,lod )*0.189879;
+	color+=textureLod( source_color,  uv_interp+vec2( 2.0, 0.0)*pix_size,lod )*0.157305;
+	color+=textureLod( source_color,  uv_interp+vec2( 3.0, 0.0)*pix_size,lod )*0.071303;
+	color+=textureLod( source_color,  uv_interp+vec2(-1.0, 0.0)*pix_size,lod )*0.189879;
+	color+=textureLod( source_color,  uv_interp+vec2(-2.0, 0.0)*pix_size,lod )*0.157305;
+	color+=textureLod( source_color,  uv_interp+vec2(-3.0, 0.0)*pix_size,lod )*0.071303;
 	frag_color = color;
 #endif
 
@@ -82,32 +123,126 @@ void main() {
 	frag_color = color;
 #endif
 
-
+//glow uses larger sigma for a more rounded blur effect
 
 #ifdef GLOW_GAUSSIAN_HORIZONTAL
-	vec4 color =textureLod( source_color,  uv_interp+vec2( 0.0, 0.0)*pixel_size,lod )*0.174938;
-	color+=textureLod( source_color,  uv_interp+vec2( 1.0, 0.0)*pixel_size,lod )*0.165569;
-	color+=textureLod( source_color,  uv_interp+vec2( 2.0, 0.0)*pixel_size,lod )*0.140367;
-	color+=textureLod( source_color,  uv_interp+vec2( 3.0, 0.0)*pixel_size,lod )*0.106595;
-	color+=textureLod( source_color,  uv_interp+vec2(-1.0, 0.0)*pixel_size,lod )*0.165569;
-	color+=textureLod( source_color,  uv_interp+vec2(-2.0, 0.0)*pixel_size,lod )*0.140367;
-	color+=textureLod( source_color,  uv_interp+vec2(-3.0, 0.0)*pixel_size,lod )*0.165569;
+	vec2 pix_size = pixel_size;
+	pix_size*=0.5; //reading from larger buffer, so use more samples
+	vec4 color =textureLod( source_color,  uv_interp+vec2( 0.0, 0.0)*pix_size,lod )*0.174938;
+	color+=textureLod( source_color,  uv_interp+vec2( 1.0, 0.0)*pix_size,lod )*0.165569;
+	color+=textureLod( source_color,  uv_interp+vec2( 2.0, 0.0)*pix_size,lod )*0.140367;
+	color+=textureLod( source_color,  uv_interp+vec2( 3.0, 0.0)*pix_size,lod )*0.106595;
+	color+=textureLod( source_color,  uv_interp+vec2(-1.0, 0.0)*pix_size,lod )*0.165569;
+	color+=textureLod( source_color,  uv_interp+vec2(-2.0, 0.0)*pix_size,lod )*0.140367;
+	color+=textureLod( source_color,  uv_interp+vec2(-3.0, 0.0)*pix_size,lod )*0.106595;
 	color*=glow_strength;
 	frag_color = color;
 #endif
 
 #ifdef GLOW_GAUSSIAN_VERTICAL
-	vec4 color =textureLod( source_color,  uv_interp+vec2(0.0, 0.0)*pixel_size,lod )*0.174938;
-	color+=textureLod( source_color,  uv_interp+vec2(0.0, 1.0)*pixel_size,lod )*0.165569;
-	color+=textureLod( source_color,  uv_interp+vec2(0.0, 2.0)*pixel_size,lod )*0.140367;
-	color+=textureLod( source_color,  uv_interp+vec2(0.0, 3.0)*pixel_size,lod )*0.106595;
-	color+=textureLod( source_color,  uv_interp+vec2(0.0,-1.0)*pixel_size,lod )*0.165569;
-	color+=textureLod( source_color,  uv_interp+vec2(0.0,-2.0)*pixel_size,lod )*0.140367;
-	color+=textureLod( source_color,  uv_interp+vec2(0.0,-3.0)*pixel_size,lod )*0.165569;
+	vec4 color =textureLod( source_color,  uv_interp+vec2(0.0, 0.0)*pixel_size,lod )*0.288713;
+	color+=textureLod( source_color,  uv_interp+vec2(0.0, 1.0)*pixel_size,lod )*0.233062;
+	color+=textureLod( source_color,  uv_interp+vec2(0.0, 2.0)*pixel_size,lod )*0.122581;
+	color+=textureLod( source_color,  uv_interp+vec2(0.0,-1.0)*pixel_size,lod )*0.233062;
+	color+=textureLod( source_color,  uv_interp+vec2(0.0,-2.0)*pixel_size,lod )*0.122581;
 	color*=glow_strength;
 	frag_color = color;
 #endif
 
+#ifdef DOF_FAR_BLUR
+
+	vec4 color_accum = vec4(0.0);
+
+	float depth = textureLod( dof_source_depth, uv_interp, 0.0).r;
+	depth = depth * 2.0 - 1.0;
+	depth = 2.0 * camera_z_near * camera_z_far / (camera_z_far + camera_z_near - depth * (camera_z_far - camera_z_near));
+
+	float amount = smoothstep(dof_begin,dof_end,depth);
+	float k_accum=0.0;
+
+	for(int i=0;i<dof_kernel_size;i++) {
+
+		int int_ofs = i-dof_kernel_from;
+		vec2 tap_uv = uv_interp + dof_dir * float(int_ofs) * amount * dof_radius;
+
+		float tap_k = dof_kernel[i];
+
+		float tap_depth = texture( dof_source_depth, tap_uv, 0.0).r;
+		tap_depth = tap_depth * 2.0 - 1.0;
+		tap_depth = 2.0 * camera_z_near * camera_z_far / (camera_z_far + camera_z_near - tap_depth * (camera_z_far - camera_z_near));
+
+		float tap_amount = mix(smoothstep(dof_begin,dof_end,tap_depth),1.0,int_ofs==0);
+		tap_amount*=tap_amount*tap_amount; //prevent undesired glow effect
+
+		vec4 tap_color = textureLod( source_color, tap_uv, 0.0) * tap_k;
+
+		k_accum+=tap_k*tap_amount;
+		color_accum+=tap_color*tap_amount;
+
+
+	}
+
+	if (k_accum>0.0) {
+		color_accum/=k_accum;
+	}
+
+	frag_color = color_accum;///k_accum;
+
+#endif
+
+#ifdef DOF_NEAR_BLUR
+
+	vec4 color_accum = vec4(0.0);
+
+	float max_accum=0;
+
+	for(int i=0;i<dof_kernel_size;i++) {
+
+		int int_ofs = i-dof_kernel_from;
+		vec2 tap_uv = uv_interp + dof_dir * float(int_ofs) * dof_radius;
+		float ofs_influence = max(0.0,1.0-float(abs(int_ofs))/float(dof_kernel_from));
+
+		float tap_k = dof_kernel[i];
+
+		vec4 tap_color = textureLod( source_color, tap_uv, 0.0);
+
+		float tap_depth = texture( dof_source_depth, tap_uv, 0.0).r;
+		tap_depth = tap_depth * 2.0 - 1.0;
+		tap_depth = 2.0 * camera_z_near * camera_z_far / (camera_z_far + camera_z_near - tap_depth * (camera_z_far - camera_z_near));
+		float tap_amount = 1.0-smoothstep(dof_end,dof_begin,tap_depth);
+		tap_amount*=tap_amount*tap_amount; //prevent undesired glow effect
+
+#ifdef DOF_NEAR_FIRST_TAP
+
+		tap_color.a= 1.0-smoothstep(dof_end,dof_begin,tap_depth);
+
+#endif
+
+		max_accum=max(max_accum,tap_amount*ofs_influence);
+
+		color_accum+=tap_color*tap_k;
+
+	}
+
+	color_accum.a=max(color_accum.a,sqrt(max_accum));
+
+
+#ifdef DOF_NEAR_BLUR_MERGE
+
+	vec4 original = textureLod( source_dof_original, uv_interp, 0.0);
+	color_accum = mix(original,color_accum,color_accum.a);
+
+#endif
+
+#ifndef DOF_NEAR_FIRST_TAP
+	//color_accum=vec4(vec3(color_accum.a),1.0);
+#endif
+	frag_color = color_accum;
+
+#endif
+
+
+
 #ifdef GLOW_FIRST_PASS
 
 #ifdef GLOW_USE_AUTO_EXPOSURE
@@ -137,5 +272,7 @@ void main() {
 	frag_color = vec4( mix(color.rgb,color.rgb*mix(ssao_color.rgb,vec3(1.0),ssao),color.a), 1.0 );
 
 #endif
+
+
 }
 
diff --git a/drivers/gles3/shaders/tonemap.glsl b/drivers/gles3/shaders/tonemap.glsl
index 8ee51e9d0c..8f7e0c7be3 100644
--- a/drivers/gles3/shaders/tonemap.glsl
+++ b/drivers/gles3/shaders/tonemap.glsl
@@ -41,6 +41,90 @@ uniform highp float glow_intensity;
 
 layout(location = 0) out vec4 frag_color;
 
+#ifdef USE_GLOW_FILTER_BICUBIC
+
+// w0, w1, w2, and w3 are the four cubic B-spline basis functions
+float w0(float a)
+{
+    return (1.0/6.0)*(a*(a*(-a + 3.0) - 3.0) + 1.0);
+}
+
+float w1(float a)
+{
+    return (1.0/6.0)*(a*a*(3.0*a - 6.0) + 4.0);
+}
+
+float w2(float a)
+{
+    return (1.0/6.0)*(a*(a*(-3.0*a + 3.0) + 3.0) + 1.0);
+}
+
+float w3(float a)
+{
+    return (1.0/6.0)*(a*a*a);
+}
+
+// g0 and g1 are the two amplitude functions
+float g0(float a)
+{
+    return w0(a) + w1(a);
+}
+
+float g1(float a)
+{
+    return w2(a) + w3(a);
+}
+
+// h0 and h1 are the two offset functions
+float h0(float a)
+{
+    return -1.0 + w1(a) / (w0(a) + w1(a));
+}
+
+float h1(float a)
+{
+    return 1.0 + w3(a) / (w2(a) + w3(a));
+}
+
+uniform ivec2 glow_texture_size;
+
+vec4 texture2D_bicubic(sampler2D tex, vec2 uv,int p_lod)
+{
+	float lod=float(p_lod);
+	vec2 tex_size = vec2(glow_texture_size >> p_lod);
+	vec2 pixel_size =1.0/tex_size;
+	uv = uv*tex_size + 0.5;
+	vec2 iuv = floor( uv );
+	vec2 fuv = fract( uv );
+
+	float g0x = g0(fuv.x);
+	float g1x = g1(fuv.x);
+	float h0x = h0(fuv.x);
+	float h1x = h1(fuv.x);
+	float h0y = h0(fuv.y);
+	float h1y = h1(fuv.y);
+
+	vec2 p0 = (vec2(iuv.x + h0x, iuv.y + h0y) - 0.5) * pixel_size;
+	vec2 p1 = (vec2(iuv.x + h1x, iuv.y + h0y) - 0.5) * pixel_size;
+	vec2 p2 = (vec2(iuv.x + h0x, iuv.y + h1y) - 0.5) * pixel_size;
+	vec2 p3 = (vec2(iuv.x + h1x, iuv.y + h1y) - 0.5) * pixel_size;
+
+	return g0(fuv.y) * (g0x * textureLod(tex, p0,lod)  +
+			    g1x * textureLod(tex, p1,lod)) +
+			g1(fuv.y) * (g0x * textureLod(tex, p2,lod)  +
+				     g1x * textureLod(tex, p3,lod));
+}
+
+
+
+#define GLOW_TEXTURE_SAMPLE(m_tex,m_uv,m_lod) texture2D_bicubic(m_tex,m_uv,m_lod)
+
+#else
+
+#define GLOW_TEXTURE_SAMPLE(m_tex,m_uv,m_lod) textureLod(m_tex,m_uv,float(m_lod))
+
+#endif
+
 
 void main() {
 
@@ -60,31 +144,32 @@ void main() {
 	vec3 glow = vec3(0.0);
 
 #ifdef USE_GLOW_LEVEL1
-	glow+=textureLod(source_glow,uv_interp,1.0).rgb;
+
+	glow+=GLOW_TEXTURE_SAMPLE(source_glow,uv_interp,1).rgb;
 #endif
 
 #ifdef USE_GLOW_LEVEL2
-	glow+=textureLod(source_glow,uv_interp,2.0).rgb;
+	glow+=GLOW_TEXTURE_SAMPLE(source_glow,uv_interp,2).rgb;
 #endif
 
 #ifdef USE_GLOW_LEVEL3
-	glow+=textureLod(source_glow,uv_interp,3.0).rgb;
+	glow+=GLOW_TEXTURE_SAMPLE(source_glow,uv_interp,3).rgb;
 #endif
 
 #ifdef USE_GLOW_LEVEL4
-	glow+=textureLod(source_glow,uv_interp,4.0).rgb;
+	glow+=GLOW_TEXTURE_SAMPLE(source_glow,uv_interp,4).rgb;
 #endif
 
 #ifdef USE_GLOW_LEVEL5
-	glow+=textureLod(source_glow,uv_interp,5.0).rgb;
+	glow+=GLOW_TEXTURE_SAMPLE(source_glow,uv_interp,5).rgb;
 #endif
 
 #ifdef USE_GLOW_LEVEL6
-	glow+=textureLod(source_glow,uv_interp,6.0).rgb;
+	glow+=GLOW_TEXTURE_SAMPLE(source_glow,uv_interp,6).rgb;
 #endif
 
 #ifdef USE_GLOW_LEVEL7
-	glow+=textureLod(source_glow,uv_interp,7.0).rgb;
+	glow+=GLOW_TEXTURE_SAMPLE(source_glow,uv_interp,7).rgb;
 #endif