From b1a50ad80538d57d917c3f399053e1a22d1aa749 Mon Sep 17 00:00:00 2001 From: Kinwailo Date: Thu, 4 Nov 2021 11:59:33 +0800 Subject: [PATCH] fix post procressing with transparent bg and keep alpha channel of the tonemap shader output --- drivers/gles2/rasterizer_scene_gles2.cpp | 5 +- drivers/gles2/shaders/effect_blur.glsl | 19 ++++-- drivers/gles2/shaders/tonemap.glsl | 65 +++++++++++-------- drivers/gles3/rasterizer_scene_gles3.cpp | 2 +- drivers/gles3/shaders/effect_blur.glsl | 19 ++++-- drivers/gles3/shaders/tonemap.glsl | 81 ++++++++++++++---------- 6 files changed, 117 insertions(+), 74 deletions(-) diff --git a/drivers/gles2/rasterizer_scene_gles2.cpp b/drivers/gles2/rasterizer_scene_gles2.cpp index a85264835b32..ec463c95ad2b 100644 --- a/drivers/gles2/rasterizer_scene_gles2.cpp +++ b/drivers/gles2/rasterizer_scene_gles2.cpp @@ -2794,9 +2794,8 @@ void RasterizerSceneGLES2::_post_process(Environment *env, const CameraMatrix &p glDepthFunc(GL_LEQUAL); glColorMask(1, 1, 1, 1); - //no post process on small, transparent or render targets without an env - bool use_post_process = env && !storage->frame.current_rt->flags[RasterizerStorage::RENDER_TARGET_TRANSPARENT]; - use_post_process = use_post_process && storage->frame.current_rt->width >= 4 && storage->frame.current_rt->height >= 4; + //no post process on small or render targets without an env + bool use_post_process = env && storage->frame.current_rt->width >= 4 && storage->frame.current_rt->height >= 4; use_post_process = use_post_process && storage->frame.current_rt->mip_maps_allocated; if (env) { diff --git a/drivers/gles2/shaders/effect_blur.glsl b/drivers/gles2/shaders/effect_blur.glsl index 16cbd839ca05..51fe0774e600 100644 --- a/drivers/gles2/shaders/effect_blur.glsl +++ b/drivers/gles2/shaders/effect_blur.glsl @@ -250,7 +250,7 @@ void main() { #endif float amount = smoothstep(dof_begin, dof_end, depth); - float k_accum = 0.0; + vec4 k_accum = vec4(0.0); for (int i = 0; i < dof_kernel_size; i++) { int int_ofs = i - dof_kernel_from; @@ -267,14 +267,16 @@ void main() { #endif float tap_amount = int_ofs == 0 ? 1.0 : smoothstep(dof_begin, dof_end, tap_depth); tap_amount *= tap_amount * tap_amount; //prevent undesired glow effect + tap_amount *= tap_k; - vec4 tap_color = texture2DLod(source_color, tap_uv, 0.0) * tap_k; + vec4 tap_color = texture2DLod(source_color, tap_uv, 0.0); - k_accum += tap_k * tap_amount; - color_accum += tap_color * tap_amount; + vec4 w = vec4(tap_amount) * vec4(vec3(tap_color.a), 1.0); + k_accum += w; + color_accum += tap_color * w; } - if (k_accum > 0.0) { + if (k_accum.r > 0.0) { color_accum /= k_accum; } @@ -287,6 +289,7 @@ void main() { vec4 color_accum = vec4(0.0); float max_accum = 0.0; + float k_accum = 0.0; for (int i = 0; i < dof_kernel_size; i++) { int int_ofs = i - dof_kernel_from; @@ -296,6 +299,7 @@ void main() { float tap_k = dof_kernel[i]; vec4 tap_color = texture2DLod(source_color, tap_uv, 0.0); + float w = tap_color.a; float tap_depth = texture2D(dof_source_depth, tap_uv, 0.0).r; tap_depth = tap_depth * 2.0 - 1.0; @@ -315,9 +319,14 @@ void main() { max_accum = max(max_accum, tap_amount * ofs_influence); + k_accum += w; + tap_color.rgb *= w; color_accum += tap_color * tap_k; } + if (k_accum > 0.0) { + color_accum.rgb /= k_accum / dof_kernel_size; + } color_accum.a = max(color_accum.a, sqrt(max_accum)); gl_FragColor = color_accum; diff --git a/drivers/gles2/shaders/tonemap.glsl b/drivers/gles2/shaders/tonemap.glsl index 6d3120980274..f550b46f4f1e 100644 --- a/drivers/gles2/shaders/tonemap.glsl +++ b/drivers/gles2/shaders/tonemap.glsl @@ -180,13 +180,13 @@ vec4 texture2D_bicubic(sampler2D tex, vec2 uv, int p_lod) { #define GLOW_TEXTURE_SAMPLE(m_tex, m_uv, m_lod) texture2DLod(m_tex, m_uv, float(m_lod)) #endif //GL_EXT_gpu_shader4 -vec3 apply_glow(vec3 color, vec3 glow) { // apply glow using the selected blending mode +vec4 apply_glow(vec4 color, vec3 glow) { // apply glow using the selected blending mode #ifdef USE_GLOW_REPLACE - color = glow; + color.rgb = glow; #endif #ifdef USE_GLOW_SCREEN - color = max((color + glow) - (color * glow), vec3(0.0)); + color.rgb = max((color.rgb + glow) - (color.rgb * glow), vec3(0.0)); #endif #ifdef USE_GLOW_SOFTLIGHT @@ -198,7 +198,18 @@ vec3 apply_glow(vec3 color, vec3 glow) { // apply glow using the selected blendi #endif #if !defined(USE_GLOW_SCREEN) && !defined(USE_GLOW_SOFTLIGHT) && !defined(USE_GLOW_REPLACE) // no other selected -> additive - color += glow; + color.rgb += glow; +#endif + +#ifndef USE_GLOW_SOFTLIGHT // softlight has no effect on black color + // compute the alpha from glow + float a = max(max(glow.r, glow.g), glow.b); + color.a = a + color.a * (1 - a); + if (color.a == 0.0) { + color.rgb = vec3(0.0); + } else if (color.a < 1.0) { + color.rgb /= color.a; + } #endif return color; @@ -220,22 +231,22 @@ vec3 apply_color_correction(vec3 color, sampler2D correction_tex) { return color; } -vec3 apply_fxaa(vec3 color, vec2 uv_interp, vec2 pixel_size) { +vec4 apply_fxaa(vec4 color, vec2 uv_interp, vec2 pixel_size) { const float FXAA_REDUCE_MIN = (1.0 / 128.0); const float FXAA_REDUCE_MUL = (1.0 / 8.0); const float FXAA_SPAN_MAX = 8.0; - vec3 rgbNW = texture2DLod(source, uv_interp + vec2(-1.0, -1.0) * pixel_size, 0.0).xyz; - vec3 rgbNE = texture2DLod(source, uv_interp + vec2(1.0, -1.0) * pixel_size, 0.0).xyz; - vec3 rgbSW = texture2DLod(source, uv_interp + vec2(-1.0, 1.0) * pixel_size, 0.0).xyz; - vec3 rgbSE = texture2DLod(source, uv_interp + vec2(1.0, 1.0) * pixel_size, 0.0).xyz; - vec3 rgbM = color; + vec4 rgbNW = texture2DLod(source, uv_interp + vec2(-1.0, -1.0) * pixel_size, 0.0); + vec4 rgbNE = texture2DLod(source, uv_interp + vec2(1.0, -1.0) * pixel_size, 0.0); + vec4 rgbSW = texture2DLod(source, uv_interp + vec2(-1.0, 1.0) * pixel_size, 0.0); + vec4 rgbSE = texture2DLod(source, uv_interp + vec2(1.0, 1.0) * pixel_size, 0.0); + vec3 rgbM = color.rgb; vec3 luma = vec3(0.299, 0.587, 0.114); - float lumaNW = dot(rgbNW, luma); - float lumaNE = dot(rgbNE, luma); - float lumaSW = dot(rgbSW, luma); - float lumaSE = dot(rgbSE, luma); - float lumaM = dot(rgbM, luma); + float lumaNW = dot(rgbNW.rgb, luma) - ((1 - rgbNW.a) / 8.0); + float lumaNE = dot(rgbNE.rgb, luma) - ((1 - rgbNE.a) / 8.0); + float lumaSW = dot(rgbSW.rgb, luma) - ((1 - rgbSW.a) / 8.0); + float lumaSE = dot(rgbSE.rgb, luma) - ((1 - rgbSE.a) / 8.0); + float lumaM = dot(rgbM, luma) - (color.a / 8.0); float lumaMin = min(lumaM, min(min(lumaNW, lumaNE), min(lumaSW, lumaSE))); float lumaMax = max(lumaM, max(max(lumaNW, lumaNE), max(lumaSW, lumaSE))); @@ -253,19 +264,21 @@ vec3 apply_fxaa(vec3 color, vec2 uv_interp, vec2 pixel_size) { dir * rcpDirMin)) * pixel_size; - vec3 rgbA = 0.5 * (texture2DLod(source, uv_interp + dir * (1.0 / 3.0 - 0.5), 0.0).xyz + texture2DLod(source, uv_interp + dir * (2.0 / 3.0 - 0.5), 0.0).xyz); - vec3 rgbB = rgbA * 0.5 + 0.25 * (texture2DLod(source, uv_interp + dir * -0.5, 0.0).xyz + texture2DLod(source, uv_interp + dir * 0.5, 0.0).xyz); + vec4 rgbA = 0.5 * (texture2DLod(source, uv_interp + dir * (1.0 / 3.0 - 0.5), 0.0) + texture2DLod(source, uv_interp + dir * (2.0 / 3.0 - 0.5), 0.0)); + vec4 rgbB = rgbA * 0.5 + 0.25 * (texture2DLod(source, uv_interp + dir * -0.5, 0.0) + texture2DLod(source, uv_interp + dir * 0.5, 0.0)); - float lumaB = dot(rgbB, luma); - if ((lumaB < lumaMin) || (lumaB > lumaMax)) { - return rgbA; - } else { - return rgbB; + float lumaB = dot(rgbB.rgb, luma) - ((1 - rgbB.a) / 8.0); + vec4 color_output = ((lumaB < lumaMin) || (lumaB > lumaMax)) ? rgbA : rgbB; + if (color_output.a == 0.0) { + color_output.rgb = vec3(0.0); + } else if (color_output.a < 1.0) { + color_output.rgb /= color_output.a; } + return color_output; } void main() { - vec3 color = texture2DLod(source, uv_interp, 0.0).rgb; + vec4 color = texture2DLod(source, uv_interp, 0.0); #ifdef USE_FXAA color = apply_fxaa(color, uv_interp, pixel_size); @@ -336,12 +349,12 @@ void main() { // Additional effects #ifdef USE_BCS - color = apply_bcs(color, bcs); + color.rgb = apply_bcs(color.rgb, bcs); #endif #ifdef USE_COLOR_CORRECTION - color = apply_color_correction(color, color_correction); + color.rgb = apply_color_correction(color.rgb, color_correction); #endif - gl_FragColor = vec4(color, 1.0); + gl_FragColor = color; } diff --git a/drivers/gles3/rasterizer_scene_gles3.cpp b/drivers/gles3/rasterizer_scene_gles3.cpp index cbb3a3f132b2..acbf943c3940 100644 --- a/drivers/gles3/rasterizer_scene_gles3.cpp +++ b/drivers/gles3/rasterizer_scene_gles3.cpp @@ -3593,7 +3593,7 @@ void RasterizerSceneGLES3::_post_process(Environment *env, const CameraMatrix &p glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0); } - if ((!env || storage->frame.current_rt->flags[RasterizerStorage::RENDER_TARGET_TRANSPARENT] || storage->frame.current_rt->width < 4 || storage->frame.current_rt->height < 4) && !storage->frame.current_rt->use_fxaa && !storage->frame.current_rt->use_debanding && storage->frame.current_rt->sharpen_intensity < 0.001) { //no post process on small render targets + if ((!env || storage->frame.current_rt->width < 4 || storage->frame.current_rt->height < 4) && !storage->frame.current_rt->use_fxaa && !storage->frame.current_rt->use_debanding && storage->frame.current_rt->sharpen_intensity < 0.001) { //no post process on small render targets //no environment or transparent render, simply return and convert to SRGB if (storage->frame.current_rt->external.fbo != 0) { glBindFramebuffer(GL_FRAMEBUFFER, storage->frame.current_rt->external.fbo); diff --git a/drivers/gles3/shaders/effect_blur.glsl b/drivers/gles3/shaders/effect_blur.glsl index ef8005d47ce4..1167da49054c 100644 --- a/drivers/gles3/shaders/effect_blur.glsl +++ b/drivers/gles3/shaders/effect_blur.glsl @@ -201,7 +201,7 @@ void main() { #endif float amount = smoothstep(dof_begin, dof_end, depth); - float k_accum = 0.0; + vec4 k_accum = vec4(0.0); for (int i = 0; i < dof_kernel_size; i++) { int int_ofs = i - dof_kernel_from; @@ -218,14 +218,16 @@ void main() { #endif 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 + tap_amount *= tap_k; - vec4 tap_color = textureLod(source_color, tap_uv, 0.0) * tap_k; + vec4 tap_color = textureLod(source_color, tap_uv, 0.0); - k_accum += tap_k * tap_amount; - color_accum += tap_color * tap_amount; + vec4 w = vec4(tap_amount) * vec4(vec3(tap_color.a), 1.0); + k_accum += w; + color_accum += tap_color * w; } - if (k_accum > 0.0) { + if (k_accum.r > 0.0) { color_accum /= k_accum; } @@ -238,6 +240,7 @@ void main() { vec4 color_accum = vec4(0.0); float max_accum = 0.0; + float k_accum = 0.0; for (int i = 0; i < dof_kernel_size; i++) { int int_ofs = i - dof_kernel_from; @@ -247,6 +250,7 @@ void main() { float tap_k = dof_kernel[i]; vec4 tap_color = textureLod(source_color, tap_uv, 0.0); + float w = tap_color.a; float tap_depth = texture(dof_source_depth, tap_uv, 0.0).r; tap_depth = tap_depth * 2.0 - 1.0; @@ -266,9 +270,14 @@ void main() { max_accum = max(max_accum, tap_amount * ofs_influence); + k_accum += w; + tap_color.rgb *= w; color_accum += tap_color * tap_k; } + if (k_accum > 0.0) { + color_accum.rgb /= k_accum / dof_kernel_size; + } color_accum.a = max(color_accum.a, sqrt(max_accum)); #ifdef DOF_NEAR_BLUR_MERGE diff --git a/drivers/gles3/shaders/tonemap.glsl b/drivers/gles3/shaders/tonemap.glsl index 05260093b937..81ed00fa97c8 100644 --- a/drivers/gles3/shaders/tonemap.glsl +++ b/drivers/gles3/shaders/tonemap.glsl @@ -264,20 +264,20 @@ vec3 gather_glow(sampler2D tex, vec2 uv) { // sample all selected glow levels return glow; } -vec3 apply_glow(vec3 color, vec3 glow) { // apply glow using the selected blending mode +vec4 apply_glow(vec4 color, vec3 glow) { // apply glow using the selected blending mode #ifdef USE_GLOW_REPLACE - color = glow; + color.rgb = glow; #endif #ifdef USE_GLOW_SCREEN //need color clamping - color = clamp(color, vec3(0.0f), vec3(1.0f)); - color = max((color + glow) - (color * glow), vec3(0.0)); + color.rgb = clamp(color.rgb, vec3(0.0f), vec3(1.0f)); + color.rgb = max((color.rgb + glow) - (color.rgb * glow), vec3(0.0)); #endif #ifdef USE_GLOW_SOFTLIGHT //need color clamping - color = clamp(color, vec3(0.0f), vec3(1.0)); + color.rgb = clamp(color.rgb, vec3(0.0f), vec3(1.0)); glow = glow * vec3(0.5f) + vec3(0.5f); color.r = (glow.r <= 0.5f) ? (color.r - (1.0f - 2.0f * glow.r) * color.r * (1.0f - color.r)) : (((glow.r > 0.5f) && (color.r <= 0.25f)) ? (color.r + (2.0f * glow.r - 1.0f) * (4.0f * color.r * (4.0f * color.r + 1.0f) * (color.r - 1.0f) + 7.0f * color.r)) : (color.r + (2.0f * glow.r - 1.0f) * (sqrt(color.r) - color.r))); @@ -286,7 +286,18 @@ vec3 apply_glow(vec3 color, vec3 glow) { // apply glow using the selected blendi #endif #if !defined(USE_GLOW_SCREEN) && !defined(USE_GLOW_SOFTLIGHT) && !defined(USE_GLOW_REPLACE) // no other selected -> additive - color += glow; + color.rgb += glow; +#endif + +#ifndef USE_GLOW_SOFTLIGHT // softlight has no effect on black color + // compute the alpha from glow + float a = max(max(glow.r, glow.g), glow.b); + color.a = a + color.a * (1 - a); + if (color.a == 0.0) { + color.rgb = vec3(0.0); + } else if (color.a < 1.0) { + color.rgb /= color.a; + } #endif return color; @@ -308,22 +319,22 @@ vec3 apply_color_correction(vec3 color, sampler2D correction_tex) { return color; } -vec3 apply_fxaa(vec3 color, float exposure, vec2 uv_interp, vec2 pixel_size) { +vec4 apply_fxaa(vec4 color, float exposure, vec2 uv_interp, vec2 pixel_size) { const float FXAA_REDUCE_MIN = (1.0 / 128.0); const float FXAA_REDUCE_MUL = (1.0 / 8.0); const float FXAA_SPAN_MAX = 8.0; - vec3 rgbNW = textureLod(source, uv_interp + vec2(-1.0, -1.0) * pixel_size, 0.0).xyz * exposure; - vec3 rgbNE = textureLod(source, uv_interp + vec2(1.0, -1.0) * pixel_size, 0.0).xyz * exposure; - vec3 rgbSW = textureLod(source, uv_interp + vec2(-1.0, 1.0) * pixel_size, 0.0).xyz * exposure; - vec3 rgbSE = textureLod(source, uv_interp + vec2(1.0, 1.0) * pixel_size, 0.0).xyz * exposure; - vec3 rgbM = color; + vec4 rgbNW = textureLod(source, uv_interp + vec2(-1.0, -1.0) * pixel_size, 0.0); + vec4 rgbNE = textureLod(source, uv_interp + vec2(1.0, -1.0) * pixel_size, 0.0); + vec4 rgbSW = textureLod(source, uv_interp + vec2(-1.0, 1.0) * pixel_size, 0.0); + vec4 rgbSE = textureLod(source, uv_interp + vec2(1.0, 1.0) * pixel_size, 0.0); + vec3 rgbM = color.rgb; vec3 luma = vec3(0.299, 0.587, 0.114); - float lumaNW = dot(rgbNW, luma); - float lumaNE = dot(rgbNE, luma); - float lumaSW = dot(rgbSW, luma); - float lumaSE = dot(rgbSE, luma); - float lumaM = dot(rgbM, luma); + float lumaNW = dot(rgbNW.rgb * exposure, luma) - ((1 - rgbNW.a) / 8.0); + float lumaNE = dot(rgbNE.rgb * exposure, luma) - ((1 - rgbNE.a) / 8.0); + float lumaSW = dot(rgbSW.rgb * exposure, luma) - ((1 - rgbSW.a) / 8.0); + float lumaSE = dot(rgbSE.rgb * exposure, luma) - ((1 - rgbSE.a) / 8.0); + float lumaM = dot(rgbM * exposure, luma) - (color.a / 8.0); float lumaMin = min(lumaM, min(min(lumaNW, lumaNE), min(lumaSW, lumaSE))); float lumaMax = max(lumaM, max(max(lumaNW, lumaNE), max(lumaSW, lumaSE))); @@ -341,15 +352,17 @@ vec3 apply_fxaa(vec3 color, float exposure, vec2 uv_interp, vec2 pixel_size) { dir * rcpDirMin)) * pixel_size; - vec3 rgbA = 0.5 * exposure * (textureLod(source, uv_interp + dir * (1.0 / 3.0 - 0.5), 0.0).xyz + textureLod(source, uv_interp + dir * (2.0 / 3.0 - 0.5), 0.0).xyz); - vec3 rgbB = rgbA * 0.5 + 0.25 * exposure * (textureLod(source, uv_interp + dir * -0.5, 0.0).xyz + textureLod(source, uv_interp + dir * 0.5, 0.0).xyz); + vec4 rgbA = 0.5 * exposure * (textureLod(source, uv_interp + dir * (1.0 / 3.0 - 0.5), 0.0) + textureLod(source, uv_interp + dir * (2.0 / 3.0 - 0.5), 0.0)); + vec4 rgbB = rgbA * 0.5 + 0.25 * exposure * (textureLod(source, uv_interp + dir * -0.5, 0.0) + textureLod(source, uv_interp + dir * 0.5, 0.0)); - float lumaB = dot(rgbB, luma); - if ((lumaB < lumaMin) || (lumaB > lumaMax)) { - return rgbA; - } else { - return rgbB; + float lumaB = dot(rgbB.rgb, luma) - ((1 - rgbB.a) / 8.0); + vec4 color_output = ((lumaB < lumaMin) || (lumaB > lumaMax)) ? rgbA : rgbB; + if (color_output.a == 0.0) { + color_output.rgb = vec3(0.0); + } else if (color_output.a < 1.0) { + color_output.rgb /= color_output.a; } + return color_output; } // From http://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_GDC2015.pdf @@ -413,7 +426,7 @@ vec3 apply_cas(vec3 color, float exposure, vec2 uv_interp, float sharpen_intensi } void main() { - vec3 color = textureLod(source, uv_interp, 0.0f).rgb; + vec4 color = textureLod(source, uv_interp, 0.0f); // Exposure float full_exposure = exposure; @@ -422,7 +435,7 @@ void main() { full_exposure /= texelFetch(source_auto_exposure, ivec2(0, 0), 0).r / auto_exposure_grey; #endif - color *= full_exposure; + color.rgb *= full_exposure; #ifdef USE_FXAA // FXAA must be applied before tonemapping. @@ -432,24 +445,24 @@ void main() { #ifdef USE_SHARPENING // CAS gives best results when applied after tonemapping, but `source` isn't tonemapped. // As a workaround, apply CAS before tonemapping so that the image still has a correct appearance when tonemapped. - color = apply_cas(color, full_exposure, uv_interp, sharpen_intensity); + color.rgb = apply_cas(color.rgb, full_exposure, uv_interp, sharpen_intensity); #endif #ifdef USE_DEBANDING // For best results, debanding should be done before tonemapping. // Otherwise, we're adding noise to an already-quantized image. - color += screen_space_dither(gl_FragCoord.xy); + color.rgb += screen_space_dither(gl_FragCoord.xy); #endif // Early Tonemap & SRGB Conversion; note that Linear tonemapping does not clamp to [0, 1]; some operations below expect a [0, 1] range and will clamp - color = apply_tonemapping(color, white); + color.rgb = apply_tonemapping(color.rgb, white); #ifdef KEEP_3D_LINEAR // leave color as is (-> don't convert to SRGB) #else //need color clamping - color = clamp(color, vec3(0.0f), vec3(1.0f)); - color = linear_to_srgb(color); // regular linear -> SRGB conversion (needs clamped values) + color.rgb = clamp(color.rgb, vec3(0.0f), vec3(1.0f)); + color.rgb = linear_to_srgb(color.rgb); // regular linear -> SRGB conversion (needs clamped values) #endif // Glow @@ -468,12 +481,12 @@ void main() { // Additional effects #ifdef USE_BCS - color = apply_bcs(color, bcs); + color.rgb = apply_bcs(color.rgb, bcs); #endif #ifdef USE_COLOR_CORRECTION - color = apply_color_correction(color, color_correction); + color.rgb = apply_color_correction(color.rgb, color_correction); #endif - frag_color = vec4(color, 1.0f); + frag_color = color; }