fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-25 13:00:09 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

anv/rt: multithread writing of invalid leaves

Browse source 
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36937>
This commit is contained in:
Felix DeGrood 2025-10-29 21:10:29 +00:00 • committed by Marge Bot
parent 09c218e8aa
commit 768bb1c7a3
1 changed files with 25 additions and 45 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

70
src/intel/vulkan/bvh/encode.comp
View file

@ -391,18 +391,18 @@ encode_internal_node(uint32_t child, uint32_t child_block_offset_from_internal_n
type == vk_ir_node_instance ? uint8_t(2) : uint8_t(1);
child_aabb = conservative_aabb(child_aabb);
vk_aabb quantize_aabb = quantize_bounds(child_aabb, conservative_child_aabb.min, exp_i8);
bool valid_leaf = cluster.idx < child_count;
DEREF(dst_node).lower_x[cluster.idx] = uint8_t(quantize_aabb.min.x);
DEREF(dst_node).lower_y[cluster.idx] = uint8_t(quantize_aabb.min.y);
DEREF(dst_node).lower_z[cluster.idx] = uint8_t(quantize_aabb.min.z);
DEREF(dst_node).upper_x[cluster.idx] = uint8_t(quantize_aabb.max.x);
DEREF(dst_node).upper_y[cluster.idx] = uint8_t(quantize_aabb.max.y);
DEREF(dst_node).upper_z[cluster.idx] = uint8_t(quantize_aabb.max.z);
DEREF(dst_node).lower_x[cluster.idx] = valid_leaf ? uint8_t(quantize_aabb.min.x) : uint8_t(0x80);
DEREF(dst_node).lower_y[cluster.idx] = valid_leaf ? uint8_t(quantize_aabb.min.y) : uint8_t(0);
DEREF(dst_node).lower_z[cluster.idx] = valid_leaf ? uint8_t(quantize_aabb.min.z) : uint8_t(0);
DEREF(dst_node).upper_x[cluster.idx] = valid_leaf ? uint8_t(quantize_aabb.max.x) : uint8_t(0);
DEREF(dst_node).upper_y[cluster.idx] = valid_leaf ? uint8_t(quantize_aabb.max.y) : uint8_t(0);
DEREF(dst_node).upper_z[cluster.idx] = valid_leaf ? uint8_t(quantize_aabb.max.z) : uint8_t(0);
/* for a mixed node, encode type of each children in startPrim in childdata */
if (DEREF(dst_node).node_type == uint8_t(ANV_NODE_TYPE_MIXED)){
if (node_type == uint8_t(ANV_NODE_TYPE_MIXED)){
uint32_t type = ir_id_to_type(child);
switch (type){
case vk_ir_node_triangle:
@ -419,27 +419,6 @@ encode_internal_node(uint32_t child, uint32_t child_block_offset_from_internal_n
break;
}
}
if (cluster.idx == 0) {
for (uint32_t i = child_count; i < 6; i++) {
/* Invalid Child Nodes: For invalid child nodes, the MSBs of lower and upper
* x planes are flipped. In other words:
* bool valid(int i) const {
* return !(lower_x[i] & 0x80) || (upper_x[i] & 0x80);
* }
*/
DEREF(dst_node).lower_x[i] = uint8_t(0x80);
DEREF(dst_node).lower_y[i] = uint8_t(0);
DEREF(dst_node).lower_z[i] = uint8_t(0);
DEREF(dst_node).upper_x[i] = uint8_t(0);
DEREF(dst_node).upper_y[i] = uint8_t(0);
DEREF(dst_node).upper_z[i] = uint8_t(0);
/* in case HW also references blockIncr to do something, we zero out the data. */
DEREF(dst_node).data[i].block_incr_and_start_prim = uint8_t(0);
DEREF(dst_node).data[i].block_incr_and_start_prim |= (uint8_t(ANV_NODE_TYPE_INVALID) << 2);
}
}
}
/* Collapse nodes until reaching 6 children, which typically can be
@ -702,32 +681,33 @@ main()
uint32_t idx = (READY_TO_WRITE(bvh_block_offset)) ? cluster.idx : -1;
idx = subgroupClusteredMin(idx, 8);
/* Propagate src child and dest blocks of next simd lane to other lanes */
IR_NODE child = VK_BVH_INVALID_NODE;
BLOCK child_block = anv_shuffle(cluster, idx, first_child_block);
BLOCK internal_node_block = anv_shuffle(cluster, idx, bvh_block_offset);
vk_aabb child_aabb = {vec3(INFINITY), vec3(-INFINITY)};
bvh_block_offset = (cluster.idx == idx) ? VK_NULL_BVH_OFFSET
: bvh_block_offset;
if (cluster.idx >= anv_shuffle(cluster, idx, found_child_count))
if (cluster.idx >= 6)
continue;
for (uint32_t i = 0; ; i++) {
child = anv_shuffle(cluster, idx, children[i]);
if (i == cluster.idx)
break;
if (cluster.idx < anv_shuffle(cluster, idx, found_child_count)) {
for (uint32_t i = 0; ; i++) {
child = anv_shuffle(cluster, idx, children[i]);
if (i == cluster.idx)
break;
uint32_t type = ir_id_to_type(child);
child_block += (type == vk_ir_node_instance) ? 2 : 1;
}
if (child != VK_BVH_INVALID_NODE)
child_aabb = DEREF(REF(vk_ir_node)NODE_OFFSET(child)).aabb;
uint32_t type = ir_id_to_type(child);
child_block += (type == vk_ir_node_instance) ? 2 : 1;
if (child != VK_BVH_INVALID_NODE && type != vk_ir_node_internal)
encode_leaf_node(type, NODE_OFFSET(child),
BLOCK_OFFSET(child_block), header);
}
vk_aabb child_aabb = {vec3(INFINITY), vec3(-INFINITY)};
if (child != VK_BVH_INVALID_NODE)
child_aabb = DEREF(REF(vk_ir_node)NODE_OFFSET(child)).aabb;
uint32_t type = ir_id_to_type(child);
if (child != VK_BVH_INVALID_NODE && type != vk_ir_node_internal)
encode_leaf_node(type, NODE_OFFSET(child),
BLOCK_OFFSET(child_block), header);
BLOCK child_block_offset =
anv_shuffle(cluster, 0, child_block) - internal_node_block;
encode_internal_node(child, child_block_offset,