87f7f0f039
Updating header using blorp code path involves setting up the render surface state. Header (CPU) update code path involves compute_w_to_host_r barrier which involves heavy flushing. Switching to completely shader based header update avoid all that overhead. Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com> Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39971>
49 lines
1.4 KiB
Text
49 lines
1.4 KiB
Text
/* Copyright © 2024 Valve Corporation
|
|
* Copyright © 2024 Intel Corporation
|
|
* SPDX-License-Identifier: MIT
|
|
*/
|
|
|
|
#version 460
|
|
|
|
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
|
|
|
#include "anv_build_interface.h"
|
|
|
|
layout(push_constant) uniform CONSTS
|
|
{
|
|
header_args args;
|
|
};
|
|
|
|
void
|
|
main(void)
|
|
{
|
|
uint64_t compacted_size = args.bvh_size;
|
|
if (args.is_compacted == uint8_t(1)) {
|
|
compacted_size =
|
|
args.bvh_offset + DEREF(args.src).dst_node_offset * ANV_RT_BLOCK_SIZE;
|
|
}
|
|
|
|
uint64_t serialization_size = compacted_size +
|
|
SIZEOF(vk_accel_struct_serialization_header) + SIZEOF(uint64_t) *
|
|
args.instance_count;
|
|
|
|
DEREF(args.dst).compacted_size = compacted_size;
|
|
DEREF(args.dst).size = compacted_size;
|
|
DEREF(args.dst).serialization_size = serialization_size;
|
|
DEREF(args.dst).self_ptr = uint64_t(args.dst);
|
|
DEREF(args.dst).instance_count = args.instance_count;
|
|
DEREF(args.dst).instance_leaves_offset = args.instance_leaves_offset;
|
|
|
|
/* 128 is local_size_x in copy.comp shader, 8 is the amount of data
|
|
* copied by each iteration of that shader's loop
|
|
*/
|
|
DEREF(args.dst).copy_dispatch_size[0] =
|
|
uint32_t(DIV_ROUND_UP(compacted_size, 8 * 128));
|
|
DEREF(args.dst).copy_dispatch_size[1] = 1;
|
|
DEREF(args.dst).copy_dispatch_size[2] = 1;
|
|
#if GFX_VERx10 >= 300
|
|
DEREF(args.dst).enable_64b_rt = 1;
|
|
#else
|
|
DEREF(args.dst).enable_64b_rt = 0;
|
|
#endif
|
|
}
|