fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-07 15:48:36 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

radv/bvh: Update leaf nodes before refitting

Browse source 
This should reduce latency between refitting nodes and their parent
nodes.

Reviewed-by: Natalie Vock <natalie.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35445>
This commit is contained in:
Konstantin Seurer 2025-07-19 17:33:17 +02:00 • committed by Marge Bot
parent 33a694fe9b
commit b20ab07e4a
1 changed files with 34 additions and 32 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

66
src/amd/vulkan/bvh/update_gfx12.comp
View file

@ -81,6 +81,27 @@ main()
child_offset += SIZEOF(radv_gfx12_box_node) * child_index;
}
vk_aabb bounds;
bounds.min = vec3(INFINITY);
bounds.max = vec3(-INFINITY);
if (is_leaf) {
VOID_REF src_leaf_node = OFFSET(src_bvh, child_offset);
uint32_t geometry_index =
DEREF(REF(uint32_t)(src_leaf_node + RADV_GFX12_UPDATABLE_PRIMITIVE_NODE_INDICES_OFFSET)) >> 4;
uint32_t primitive_index =
DEREF(REF(uint32_t)(src_leaf_node + RADV_GFX12_UPDATABLE_PRIMITIVE_NODE_INDICES_OFFSET + 4)) & 0xfffffff;
vk_bvh_geometry_data geom_data = DEREF(INDEX(vk_bvh_geometry_data, args.geom_data, geometry_index));
VOID_REF dst_leaf_node = OFFSET(dst_bvh, child_offset);
if (geom_data.geometry_type == VK_GEOMETRY_TYPE_TRIANGLES_KHR) {
radv_build_triangle(bounds, dst_leaf_node, geom_data, primitive_index, true);
} else {
VOID_REF src_ptr = OFFSET(geom_data.data, primitive_index * geom_data.stride);
radv_build_aabb(bounds, src_ptr, dst_leaf_node, geometry_index, primitive_index, true);
}
}
uint32_t child_index = (child_offset - internal_nodes_offset) / SIZEOF(radv_gfx12_box_node);
bool is_ready = is_leaf_or_invalid;
@ -91,27 +112,7 @@ main()
if (radv_ballot(cluster, is_ready) != 0xff)
continue;
vk_aabb bounds;
bounds.min = vec3(INFINITY);
bounds.max = vec3(-INFINITY);
if (is_leaf) {
VOID_REF src_leaf_node = OFFSET(src_bvh, child_offset);
uint32_t geometry_index =
DEREF(REF(uint32_t)(src_leaf_node + RADV_GFX12_UPDATABLE_PRIMITIVE_NODE_INDICES_OFFSET)) >> 4;
uint32_t primitive_index =
DEREF(REF(uint32_t)(src_leaf_node + RADV_GFX12_UPDATABLE_PRIMITIVE_NODE_INDICES_OFFSET + 4)) & 0xfffffff;
vk_bvh_geometry_data geom_data = DEREF(INDEX(vk_bvh_geometry_data, args.geom_data, geometry_index));
VOID_REF dst_leaf_node = OFFSET(dst_bvh, child_offset);
if (geom_data.geometry_type == VK_GEOMETRY_TYPE_TRIANGLES_KHR) {
radv_build_triangle(bounds, dst_leaf_node, geom_data, primitive_index, true);
} else {
VOID_REF src_ptr = OFFSET(geom_data.data, primitive_index * geom_data.stride);
radv_build_aabb(bounds, src_ptr, dst_leaf_node, geometry_index, primitive_index, true);
}
} else if (is_valid) {
if (!is_leaf_or_invalid) {
bounds = DEREF(INDEX(vk_aabb, args.bounds, child_index));
}
@ -123,13 +124,20 @@ main()
total_bounds.max.y = subgroupClusteredMax(bounds.max.y, 8);
total_bounds.max.z = subgroupClusteredMax(bounds.max.z, 8);
if (!is_root_node) {
DEREF(INDEX(vk_aabb, args.bounds, node_index - 1)) = total_bounds;
if (cluster.invocation_index == 0) {
if (is_root_node) {
DEREF(args.dst).aabb = total_bounds;
} else {
DEREF(INDEX(vk_aabb, args.bounds, node_index - 1)) = total_bounds;
memoryBarrier(gl_ScopeDevice, gl_StorageSemanticsBuffer,
gl_SemanticsAcquireRelease | gl_SemanticsMakeAvailable | gl_SemanticsMakeVisible);
memoryBarrier(gl_ScopeDevice, gl_StorageSemanticsBuffer,
gl_SemanticsAcquireRelease | gl_SemanticsMakeAvailable | gl_SemanticsMakeVisible);
DEREF(INDEX(uint32_t, args.internal_ready_count, node_index - 1)) = 1;
DEREF(INDEX(uint32_t, args.internal_ready_count, node_index - 1)) = 1;
memoryBarrier(gl_ScopeDevice, gl_StorageSemanticsBuffer,
gl_SemanticsAcquireRelease | gl_SemanticsMakeAvailable | gl_SemanticsMakeVisible);
}
}
vec3 origin = total_bounds.min;
@ -174,12 +182,6 @@ main()
DEREF(dst_node).children[cluster.invocation_index] = null_child;
}
if (is_root_node)
DEREF(args.dst).aabb = total_bounds;
/* Make changes to internal_ready_count available to the other invocations. */
memoryBarrier(gl_ScopeDevice, gl_StorageSemanticsBuffer,
gl_SemanticsAcquireRelease | gl_SemanticsMakeAvailable | gl_SemanticsMakeVisible);
break;
}
}