diff options
author | Clément Foucault <foucault.clem@gmail.com> | 2018-04-17 23:25:53 +0300 |
---|---|---|
committer | Clément Foucault <foucault.clem@gmail.com> | 2018-04-17 23:25:59 +0300 |
commit | b40f5be01f8d0aa50af46ac7a79cfe88efdcbe18 (patch) | |
tree | 3a64d3c4fc0fe4d0e0383a26111904b66db725d2 /source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl | |
parent | 38399c0b10e73cec5772a71809d4074332bb6231 (diff) |
Eevee: Use textureGather for minmaxZbuffer downsampling.
I haven't noticed any performance improvement but it could be more important
for other hardware. At least it's not slower!
Diffstat (limited to 'source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl')
-rw-r--r-- | source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl | 61 |
1 files changed, 41 insertions, 20 deletions
diff --git a/source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl b/source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl index 05fef73b159..c11dbdd59ad 100644 --- a/source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl +++ b/source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl @@ -4,6 +4,8 @@ * Adapted from http://rastergrid.com/blog/2010/10/hierarchical-z-map-based-occlusion-culling/ **/ +#extension GL_ARB_texture_gather : enable + #ifdef LAYERED uniform sampler2DArray depthBuffer; uniform int depthLayer; @@ -12,15 +14,21 @@ uniform sampler2D depthBuffer; #endif #ifdef LAYERED -#define sampleLowerMip(t) texelFetch(depthBuffer, ivec3(t, depthLayer), 0).r +# define sampleLowerMip(t) texelFetch(depthBuffer, ivec3(t, depthLayer), 0).r +# define gatherLowerMip(t) textureGather(depthBuffer, vec3(t, depthLayer)) #else -#define sampleLowerMip(t) texelFetch(depthBuffer, t, 0).r +# define sampleLowerMip(t) texelFetch(depthBuffer, t, 0).r +# define gatherLowerMip(t) textureGather(depthBuffer, t) #endif #ifdef MIN_PASS -#define minmax(a, b) min(a, b) +#define minmax2(a, b) min(a, b) +#define minmax3(a, b, c) min(min(a, b), c) +#define minmax4(a, b, c, d) min(min(min(a, b), c), d) #else /* MAX_PASS */ -#define minmax(a, b) max(a, b) +#define minmax2(a, b) max(a, b) +#define minmax3(a, b, c) max(max(a, b), c) +#define minmax4(a, b, c, d) max(max(max(a, b), c), d) #endif /* On some AMD card / driver conbination, it is needed otherwise, @@ -38,32 +46,45 @@ void main() texelPos *= 2; #endif +#ifdef COPY_DEPTH float val = sampleLowerMip(texelPos); -#ifndef COPY_DEPTH - float val2 = sampleLowerMip(texelPos + ivec2(1, 0)); - float val3 = sampleLowerMip(texelPos + ivec2(1, 1)); - float val4 = sampleLowerMip(texelPos + ivec2(0, 1)); - val = minmax(val, val2); - val = minmax(val, val3); - val = minmax(val, val4); +#else + vec4 samp; +# ifdef GL_ARB_texture_gather + samp = gatherLowerMip(vec2(texelPos) / vec2(mipsize)); +# else + samp.x = sampleLowerMip(texelPos); + samp.y = sampleLowerMip(texelPos + ivec2(1, 0)); + samp.z = sampleLowerMip(texelPos + ivec2(1, 1)); + samp.w = sampleLowerMip(texelPos + ivec2(0, 1)); +# endif + + float val = minmax4(samp.x, samp.y, samp.z, samp.w); /* if we are reducing an odd-width texture then fetch the edge texels */ if (((mipsize.x & 1) != 0) && (texelPos.x == mipsize.x - 3)) { /* if both edges are odd, fetch the top-left corner texel */ if (((mipsize.y & 1) != 0) && (texelPos.y == mipsize.y - 3)) { - val = minmax(val, sampleLowerMip(texelPos + ivec2(2, 2))); + samp.x = sampleLowerMip(texelPos + ivec2(2, 2)); + val = minmax2(val, samp.x); } - float val2 = sampleLowerMip(texelPos + ivec2(2, 0)); - float val3 = sampleLowerMip(texelPos + ivec2(2, 1)); - val = minmax(val, val2); - val = minmax(val, val3); +# ifdef GL_ARB_texture_gather + samp = gatherLowerMip((vec2(texelPos) + vec2(1.0, 0.0)) / vec2(mipsize)); +# else + samp.y = sampleLowerMip(texelPos + ivec2(2, 0)); + samp.z = sampleLowerMip(texelPos + ivec2(2, 1)); +# endif + val = minmax3(val, samp.y, samp.z); } /* if we are reducing an odd-height texture then fetch the edge texels */ if (((mipsize.y & 1) != 0) && (texelPos.y == mipsize.y - 3)) { - float val2 = sampleLowerMip(texelPos + ivec2(0, 2)); - float val3 = sampleLowerMip(texelPos + ivec2(1, 2)); - val = minmax(val, val2); - val = minmax(val, val3); +# ifdef GL_ARB_texture_gather + samp = gatherLowerMip((vec2(texelPos) + vec2(0.0, 1.0)) / vec2(mipsize)); +# else + samp.x = sampleLowerMip(texelPos + ivec2(0, 2)); + samp.y = sampleLowerMip(texelPos + ivec2(1, 2)); +# endif + val = minmax3(val, samp.x, samp.y); } #endif |