panfrost: Allow R/RG/RGB varyings

This can be a bandwidth savings.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

diff --git a/src/gallium/drivers/panfrost/pan_assemble.c b/src/gallium/drivers/panfrost/pan_assemble.c
index d569b26732e89b11238662b981d89cd778eca05d..a274a5bc0bb259df933df2bd4854c5cb0ecae8bc 100644 (file)
--- a/src/gallium/drivers/panfrost/pan_assemble.c
+++ b/src/gallium/drivers/panfrost/pan_assemble.c
@@ -77,17 +77,27 @@ pan_format_from_nir_size(nir_alu_type base, unsigned size)
 }
 
 static enum mali_format
-pan_format_from_glsl(const struct glsl_type *type)
+pan_format_from_glsl(const struct glsl_type *type, unsigned frac)
 {
-        enum glsl_base_type glsl_base = glsl_get_base_type(glsl_without_array(type));
+        const struct glsl_type *column = glsl_without_array_or_matrix(type);
+        enum glsl_base_type glsl_base = glsl_get_base_type(column);
         nir_alu_type t = nir_get_nir_type_for_glsl_base_type(glsl_base);
+        unsigned chan = glsl_get_components(column);
+
+        /* If we have a fractional location added, we need to increase the size
+         * so it will fit, i.e. a vec3 in YZW requires us to allocate a vec4.
+         * We could do better but this is an edge case as it is, normally
+         * packed varyings will be aligned. */
+        chan += frac;
+
+        assert(chan >= 1 && chan <= 4);
 
         unsigned base = nir_alu_type_get_base_type(t);
         unsigned size = nir_alu_type_get_type_size(t);
 
         return pan_format_from_nir_base(base) |
                 pan_format_from_nir_size(base, size) |
-                MALI_NR_CHANNELS(4);
+                MALI_NR_CHANNELS(chan);
 }
 
 static enum bifrost_shader_type
@@ -267,7 +277,7 @@ panfrost_shader_compile(struct panfrost_context *ctx,
 
                 for (int c = 0; c < sz; ++c) {
                         p_varyings[loc + c] = var->data.location + c;
-                        p_varying_type[loc + c] = pan_format_from_glsl(var->type);
+                        p_varying_type[loc + c] = pan_format_from_glsl(var->type, var->data.location_frac);
                 }
         }
 

diff --git a/src/panfrost/include/panfrost-job.h b/src/panfrost/include/panfrost-job.h
index d6f1fc861da0bafaef3d301adf35fe9d60a63935..0fef5e420b5835654621c99db9ccc7cc1cdd4a76 100644 (file)
--- a/src/panfrost/include/panfrost-job.h
+++ b/src/panfrost/include/panfrost-job.h
@@ -257,6 +257,7 @@ struct mali_channel_swizzle {
  * there are.
  */
 #define MALI_NR_CHANNELS(n) ((n - 1) << 3)
+#define MALI_EXTRACT_CHANNELS(fmt) ((((fmt) >> 3) & 3) + 1)
 
 /* If the high 3 bits are 3 to 6, then the low 3 bits say how big each
  * component is, except the special MALI_CHANNEL_FLOAT which overrides what the