Metal: Resolve artifact in EEVEE Next Film Cryptomatte

Cryptomatte passes would generate a feathered outline
in Metal due to missing texture fence in chained
read->modify->write->read->... patterns.

Added imageFence function to explicitly state that
imageStore's should be visible to future imageLoad's.

Authored by Apple: Michael Parkin-White

Pull Request: https://projects.blender.org/blender/blender/pulls/119163

source/blender/draw/engines/eevee_next/shaders/eevee_cryptomatte_lib.glsl

@@ -34,6 +34,8 @@ void cryptomatte_clear_samples(FilmSample dst)
   int layer_len = imageSize(cryptomatte_img).z;
   for (int i = 0; i < layer_len; i++) {
     imageStore(cryptomatte_img, ivec3(dst.texel, i), vec4(0.0));
+    /* Ensure stores are visible to later reads. */
+    imageFence(cryptomatte_img);
   }
 }
 
@@ -71,4 +73,6 @@ void cryptomatte_store_film_sample(FilmSample dst,
     imageStore(cryptomatte_img, img_co, sample_pair);
     break;
   }
+  /* Ensure stores are visible to later reads. */
+  imageFence(cryptomatte_img);
 }

source/blender/draw/engines/eevee_next/shaders/eevee_film_cryptomatte_post_comp.glsl

@@ -61,6 +61,8 @@ void cryptomatte_store_samples(ivec2 texel, int layer, vec2 samples[CRYPTOMATTE_
     pass_sample.zw = samples[p * 2 + 1];
     imageStore(cryptomatte_img, ivec3(texel, p + layer_id), pass_sample);
   }
+  /* Ensure stores are visible to later reads. */
+  imageFence(cryptomatte_img);
 }
 
 void main()

source/blender/draw/engines/eevee_next/shaders/eevee_film_lib.glsl

@@ -142,17 +142,10 @@ void film_sample_accum_combined(FilmSample samp, inout vec4 accum, inout float w
   weight_accum += weight;
 }
 
-#ifdef GPU_METAL
-void film_sample_cryptomatte_accum(FilmSample samp,
-                                   int layer,
-                                   sampler2D tex,
-                                   thread vec2 *crypto_samples)
-#else
 void film_sample_cryptomatte_accum(FilmSample samp,
                                    int layer,
                                    sampler2D tex,
                                    inout vec2 crypto_samples[4])
-#endif
 {
   float hash = texelFetch(tex, samp.texel, 0)[layer];
   /* Find existing entry. */
@@ -247,11 +240,7 @@ vec2 film_pixel_history_motion_vector(ivec2 texel_sample)
 /* \a t is inter-pixel position. 0 means perfectly on a pixel center.
  * Returns weights in both dimensions.
  * Multiply each dimension weights to get final pixel weights. */
-#ifdef GPU_METAL
-void film_get_catmull_rom_weights(vec2 t, thread vec2 *weights)
-#else
 void film_get_catmull_rom_weights(vec2 t, out vec2 weights[4])
-#endif
 {
   vec2 t2 = t * t;
   vec2 t3 = t2 * t;

source/blender/gpu/shaders/metal/mtl_shader_defines.msl

@@ -366,6 +366,9 @@ struct SStruct {
 #define imageStore(_tex, _coord, _value) _texture_write_internal(_tex, _coord, _value)
 #define imageStoreFast(_tex, _coord, _value) _texture_write_internal_fast(_tex, _coord, _value)
 
+/* Texture synchronization functions. */
+#define imageFence(image) image.texture->fence()
+
 /* Singular return values from texture functions of type DEPTH are often indexed with either .r or
  * .x. This is a lightweight wrapper type for handling this syntax. */
 union _msl_return_float {

source/blender/gpu/shaders/opengl/glsl_shader_defines.glsl

@@ -83,6 +83,9 @@
 #define isampler2DAtomic isampler2D
 #define isampler3DAtomic isampler3D
 
+/* Pass through functions. */
+#define imageFence(image)
+
 /* Backend Functions. */
 #define select(A, B, mask) mix(A, B, mask)