EEVEE: Use Fullscreen maxZBuffer instead of halfres

This removes the need for per mipmap scalling factor and trilinear interpolation
issues. We pad the texture so that all mipmaps have pixels in the next mip.

This simplifies the downsampling shader too.

This also change the SSR radiance buffer as well in the same fashion.

9 files changed, 50 insertions, 100 deletions

diff --git a/source/blender/draw/engines/eevee/shaders/ambient_occlusion_lib.glsl b/source/blender/draw/engines/eevee/shaders/ambient_occlusion_lib.glsl
index 473990e1683..0b01b186b1b 100644
--- a/source/blender/draw/engines/eevee/shaders/ambient_occlusion_lib.glsl
+++ b/source/blender/draw/engines/eevee/shaders/ambient_occlusion_lib.glsl
@@ -94,8 +94,7 @@ float search_horizon(vec3 vI,
     vec2 uv = uv_start + uv_dir * t;
     float lod = min(MAX_LOD, max(i - noise, 0.0) * aoQuality);
 
-    int mip = int(lod) + hizMipOffset;
-    float depth = textureLod(depth_tx, uv * mipRatio[mip].xy, floor(lod)).r;
+    float depth = textureLod(depth_tx, uv * hizUvScale.xy, floor(lod)).r;
 
     /* Bias depth a bit to avoid self shadowing issues. */
     const float bias = 2.0 * 2.4e-7;
diff --git a/source/blender/draw/engines/eevee/shaders/common_uniforms_lib.glsl b/source/blender/draw/engines/eevee/shaders/common_uniforms_lib.glsl
index a6c9eebaff2..24de4520207 100644
--- a/source/blender/draw/engines/eevee/shaders/common_uniforms_lib.glsl
+++ b/source/blender/draw/engines/eevee/shaders/common_uniforms_lib.glsl
@@ -2,7 +2,7 @@
 layout(std140) uniform common_block
 {
   mat4 pastViewProjectionMatrix;
-  vec2 mipRatio[10]; /* To correct mip level texel misalignment */
+  vec4 hizUvScale; /* To correct mip level texel misalignment */
   /* Ambient Occlusion */
   vec4 aoParameters[2];
   /* Volumetric */
@@ -37,15 +37,15 @@ layout(std140) uniform common_block
   int prbIrradianceVisSize;
   float prbIrradianceSmooth;
   float prbLodCubeMax;
-  float prbLodPlanarMax;
   /* Misc*/
-  int hizMipOffset;
   int rayType;
   float rayDepth;
   float alphaHashOffset;
   float alphaHashScale;
+  float pad6;
   float pad7;
   float pad8;
+  float pad9;
 };
 
 /* rayType (keep in sync with ray_type) */
@@ -69,9 +69,3 @@ layout(std140) uniform common_block
 #define ssrQuality ssrParameters.x
 #define ssrThickness ssrParameters.y
 #define ssrPixelSize ssrParameters.zw
-
-vec2 mip_ratio_interp(float mip)
-{
-  float low_mip = floor(mip);
-  return mix(mipRatio[int(low_mip)], mipRatio[int(low_mip + 1.0)], mip - low_mip);
-}
diff --git a/source/blender/draw/engines/eevee/shaders/effect_dof_lib.glsl b/source/blender/draw/engines/eevee/shaders/effect_dof_lib.glsl
index 88d83cd913a..3ad3c90d27a 100644
--- a/source/blender/draw/engines/eevee/shaders/effect_dof_lib.glsl
+++ b/source/blender/draw/engines/eevee/shaders/effect_dof_lib.glsl
@@ -62,12 +62,6 @@ const vec2 quad_offsets[4] = vec2[4](
 
 #define dof_coc_from_zdepth(d) calculate_coc(linear_depth(d))
 
-vec4 safe_color(vec4 c)
-{
-  /* Clamp to avoid black square artifacts if a pixel goes NaN. */
-  return clamp(c, vec4(0.0), vec4(1e20)); /* 1e20 arbitrary. */
-}
-
 float dof_hdr_color_weight(vec4 color)
 {
   /* From UE4. Very fast "luma" weighting. */
diff --git a/source/blender/draw/engines/eevee/shaders/effect_downsample_frag.glsl b/source/blender/draw/engines/eevee/shaders/effect_downsample_frag.glsl
index a4637b9df91..5bf850fe229 100644
--- a/source/blender/draw/engines/eevee/shaders/effect_downsample_frag.glsl
+++ b/source/blender/draw/engines/eevee/shaders/effect_downsample_frag.glsl
@@ -1,5 +1,9 @@
+
+#pragma BLENDER_REQUIRE(common_math_lib.glsl)
+
 /**
- * Simple down-sample shader. Takes the average of the 4 texels of lower mip.
+ * Simple down-sample shader.
+ * Do a gaussian filter using 4 bilinear texture samples.
  */
 
 uniform sampler2D source;
@@ -14,24 +18,25 @@ float brightness(vec3 c)
 
 void main()
 {
-#if 0
-  /* Reconstructing Target uvs like this avoid missing pixels if NPO2 */
-  vec2 uvs = gl_FragCoord.xy * 2.0 / vec2(textureSize(source, 0));
+  vec2 texel_size = 1.0 / vec2(textureSize(source, 0));
+  vec2 uvs = gl_FragCoord.xy * texel_size;
 
+#ifdef COPY_SRC
   FragColor = textureLod(source, uvs, 0.0);
+  FragColor = safe_color(FragColor);
+
+  /* Clamped brightness. */
+  float luma = max(1e-8, brightness(FragColor.rgb));
+  FragColor *= 1.0 - max(0.0, luma - fireflyFactor) / luma;
+
 #else
-  vec2 texel_size = 1.0 / vec2(textureSize(source, 0));
-  vec2 uvs = gl_FragCoord.xy * 2.0 * texel_size;
   vec4 ofs = texel_size.xyxy * vec4(0.75, 0.75, -0.75, -0.75);
+  uvs *= 2.0;
 
   FragColor = textureLod(source, uvs + ofs.xy, 0.0);
   FragColor += textureLod(source, uvs + ofs.xw, 0.0);
   FragColor += textureLod(source, uvs + ofs.zy, 0.0);
   FragColor += textureLod(source, uvs + ofs.zw, 0.0);
   FragColor *= 0.25;
-
-  /* Clamped brightness. */
-  float luma = max(1e-8, brightness(FragColor.rgb));
-  FragColor *= 1.0 - max(0.0, luma - fireflyFactor) / luma;
 #endif
 }
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 b99037b1e80..ccb65d2e5a6 100644
--- a/source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl
+++ b/source/blender/draw/engines/eevee/shaders/effect_minmaxz_frag.glsl
@@ -2,6 +2,9 @@
  * Shader that down-sample depth buffer,
  * saving min and max value of each texel in the above mipmaps.
  * Adapted from http://rastergrid.com/blog/2010/10/hierarchical-z-map-based-occlusion-culling/
+ *
+ * Major simplification has been made since we pad the buffer to always be bigger than input to
+ * avoid mipmapping misalignement.
  */
 
 #ifdef LAYERED
@@ -12,10 +15,10 @@ uniform sampler2D depthBuffer;
 #endif
 
 #ifdef LAYERED
-#  define sampleLowerMip(t) texelFetch(depthBuffer, ivec3(t, depthLayer), 0).r
+#  define sampleLowerMip(t) texture(depthBuffer, vec3(t, depthLayer)).r
 #  define gatherLowerMip(t) textureGather(depthBuffer, vec3(t, depthLayer))
 #else
-#  define sampleLowerMip(t) texelFetch(depthBuffer, t, 0).r
+#  define sampleLowerMip(t) texture(depthBuffer, t).r
 #  define gatherLowerMip(t) textureGather(depthBuffer, t)
 #endif
 
@@ -37,54 +40,27 @@ out vec4 fragColor;
 
 void main()
 {
-  ivec2 texelPos = ivec2(gl_FragCoord.xy);
-  ivec2 mipsize = textureSize(depthBuffer, 0).xy;
-
-#ifndef COPY_DEPTH
-  texelPos *= 2;
-#endif
+  vec2 texel = gl_FragCoord.xy;
+  vec2 texel_size = 1.0 / vec2(textureSize(depthBuffer, 0).xy);
 
 #ifdef COPY_DEPTH
-  float val = sampleLowerMip(texelPos);
+  vec2 uv = texel * texel_size;
+
+  float val = sampleLowerMip(uv);
 #else
+  vec2 uv = texel * 2.0 * texel_size;
+
   vec4 samp;
 #  ifdef GPU_ARB_texture_gather
-  /* + 1.0 to gather at the center of target 4 texels. */
-  samp = gatherLowerMip((vec2(texelPos) + 1.0) / vec2(mipsize));
+  samp = gatherLowerMip(uv);
 #  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));
+  samp.x = sampleLowerMip(uv + vec2(-0.5, -0.5) * texel_size);
+  samp.y = sampleLowerMip(uv + vec2(-0.5, 0.5) * texel_size);
+  samp.z = sampleLowerMip(uv + vec2(0.5, -0.5) * texel_size);
+  samp.w = sampleLowerMip(uv + vec2(0.5, 0.5) * texel_size);
 #  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)) {
-      samp.x = sampleLowerMip(texelPos + ivec2(2, 2));
-      val = minmax2(val, samp.x);
-    }
-#  ifdef GPU_ARB_texture_gather
-    samp = gatherLowerMip((vec2(texelPos) + vec2(2.0, 1.0)) / 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)) {
-#  ifdef GPU_ARB_texture_gather
-    samp = gatherLowerMip((vec2(texelPos) + vec2(1.0, 2.0)) / 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
 
 #if defined(GPU_INTEL) || defined(GPU_ATI)
diff --git a/source/blender/draw/engines/eevee/shaders/effect_ssr_frag.glsl b/source/blender/draw/engines/eevee/shaders/effect_ssr_frag.glsl
index ecff28dcd38..f2117d8ee67 100644
--- a/source/blender/draw/engines/eevee/shaders/effect_ssr_frag.glsl
+++ b/source/blender/draw/engines/eevee/shaders/effect_ssr_frag.glsl
@@ -104,6 +104,11 @@ void do_ssr(vec3 V, vec3 N, vec3 T, vec3 B, vec3 vP, float a2, vec4 rand)
 
   pdfData = min(1024e32, pdf_ggx_reflect(NH, a2)); /* Theoretical limit of 16bit float */
 
+  /* TODO(fclem) This bias should use depth precision and the dot product between
+   * ray direction and geom normal. */
+  float rays_bias = 0.005;
+  vP = vP + vNg * rays_bias;
+
   vec3 hit_pos = raycast(-1, vP, R * 1e16, ssrThickness, rand.y, ssrQuality, a2, true);
 
   hitData = encode_hit_data(hit_pos.xy, (hit_pos.z > 0.0), false);
@@ -325,10 +330,10 @@ vec3 get_hit_vector(vec3 hit_pos,
 vec3 get_scene_color(vec2 ref_uvs, float mip, float planar_index, bool is_planar)
 {
   if (is_planar) {
-    return textureLod(probePlanars, vec3(ref_uvs, planar_index), min(mip, prbLodPlanarMax)).rgb;
+    return textureLod(probePlanars, vec3(ref_uvs, planar_index), mip).rgb;
   }
   else {
-    return textureLod(prevColorBuffer, ref_uvs, mip).rgb;
+    return textureLod(prevColorBuffer, ref_uvs * hizUvScaleBias.xy + hizUvScaleBias.zw, mip).rgb;
   }
 }
 
@@ -403,12 +408,6 @@ vec4 get_ssr_samples(vec4 hit_pdf,
   vec4 mip = log2(cone_footprint * max_v2(vec2(textureSize(depthBuffer, 0))));
   mip = clamp(mip, 0.0, MAX_MIP);
 
-  /* Correct UVs for mipmaping mis-alignment */
-  hit_co[0].xy *= mip_ratio_interp(mip.x);
-  hit_co[0].zw *= mip_ratio_interp(mip.y);
-  hit_co[1].xy *= mip_ratio_interp(mip.z);
-  hit_co[1].zw *= mip_ratio_interp(mip.w);
-
   /* Slide 54 */
   vec4 bsdf;
   bsdf.x = bsdf_ggx(N, hit_pos[0], V, roughnessSquared);
diff --git a/source/blender/draw/engines/eevee/shaders/effect_temporal_aa.glsl b/source/blender/draw/engines/eevee/shaders/effect_temporal_aa.glsl
index 28947e971d2..165aed2a46f 100644
--- a/source/blender/draw/engines/eevee/shaders/effect_temporal_aa.glsl
+++ b/source/blender/draw/engines/eevee/shaders/effect_temporal_aa.glsl
@@ -10,12 +10,6 @@ uniform mat4 prevViewProjectionMatrix;
 
 out vec4 FragColor;
 
-vec4 safe_color(vec4 c)
-{
-  /* Clamp to avoid black square artifacts if a pixel goes NaN. */
-  return clamp(c, vec4(0.0), vec4(1e20)); /* 1e20 arbitrary. */
-}
-
 #ifdef USE_REPROJECTION
 
 /**
diff --git a/source/blender/draw/engines/eevee/shaders/raytrace_lib.glsl b/source/blender/draw/engines/eevee/shaders/raytrace_lib.glsl
index aebd1c3aef3..fbc13d832bd 100644
--- a/source/blender/draw/engines/eevee/shaders/raytrace_lib.glsl
+++ b/source/blender/draw/engines/eevee/shaders/raytrace_lib.glsl
@@ -16,10 +16,7 @@ float sample_depth(vec2 uv, int index, float lod)
   }
   else {
 #endif
-    lod = clamp(floor(lod), 0.0, 8.0);
-    /* Correct UVs for mipmaping mis-alignment */
-    uv *= mipRatio[int(lod) + hizMipOffset];
-    return textureLod(maxzBuffer, uv, lod).r;
+    return textureLod(maxzBuffer, uv * hizUvScale.xy, floor(lod)).r;
 #ifdef PLANAR_PROBE_RAYTRACE
   }
 #endif
@@ -37,10 +34,10 @@ vec4 sample_depth_grouped(vec4 uv1, vec4 uv2, int index, float lod)
   }
   else {
 #endif
-    depths.x = textureLod(maxzBuffer, uv1.xy, lod).r;
-    depths.y = textureLod(maxzBuffer, uv1.zw, lod).r;
-    depths.z = textureLod(maxzBuffer, uv2.xy, lod).r;
-    depths.w = textureLod(maxzBuffer, uv2.zw, lod).r;
+    depths.x = textureLod(maxzBuffer, uv1.xy * hizUvScale.xy, lod).r;
+    depths.y = textureLod(maxzBuffer, uv1.zw * hizUvScale.xy, lod).r;
+    depths.z = textureLod(maxzBuffer, uv2.xy * hizUvScale.xy, lod).r;
+    depths.w = textureLod(maxzBuffer, uv2.zw * hizUvScale.xy, lod).r;
 #ifdef PLANAR_PROBE_RAYTRACE
   }
 #endif
@@ -131,9 +128,6 @@ void prepare_raycast(vec3 ray_origin,
 #endif
   ss_ray = ss_start * m.xyyy + 0.5;
   ss_step *= m.xyyy;
-
-  /* take the center of the texel. */
-  // ss_ray.xy += sign(ss_ray.xy) * m * ssrPixelSize * (1.0 + hizMipOffset);
 }
 
 /* See times_and_deltas. */
@@ -175,9 +169,7 @@ vec3 raycast(int index,
   bool hit = false;
   float iter;
   for (iter = 1.0; !hit && (ray_time < max_time) && (iter < MAX_STEP); iter++) {
-    /* Minimum stride of 2 because we are using half res minmax zbuffer. */
-    /* WORKAROUND: Factor is a bit higher than 2 to avoid some banding. To investigate. */
-    float stride = max(1.0, iter * trace_quality) * (2.0 + 0.05);
+    float stride = max(1.01, iter * trace_quality);
     float lod = log2(stride * 0.5 * trace_quality) * lod_fac;
     ray_time += stride;
 
diff --git a/source/blender/draw/engines/eevee/shaders/ssr_lib.glsl b/source/blender/draw/engines/eevee/shaders/ssr_lib.glsl
index 15c28efe622..907a9cf17f8 100644
--- a/source/blender/draw/engines/eevee/shaders/ssr_lib.glsl
+++ b/source/blender/draw/engines/eevee/shaders/ssr_lib.glsl
@@ -68,10 +68,7 @@ vec4 screen_space_refraction(vec3 vP, vec3 N, vec3 V, float ior, float roughness
     vec2 texture_size = vec2(textureSize(colorBuffer, 0).xy);
     float mip = clamp(log2(cone_footprint * max(texture_size.x, texture_size.y)), 0.0, 9.0);
 
-    /* Correct UVs for mipmaping mis-alignment */
-    hit_uvs *= mip_ratio_interp(mip);
-
-    vec3 spec = textureLod(colorBuffer, hit_uvs, mip).xyz;
+    vec3 spec = textureLod(colorBuffer, hit_uvs * hizUvScale.xy, mip).xyz;
     float mask = screen_border_mask(hit_uvs);
 
     return vec4(spec, mask);