fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-03-11 17:50:32 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

radv/rt: Fix cases in which the bound BVH build pipeline gets clobbered

Browse source 
The most egregious case was AS updates, in which case radv_copy_memory
would decide to use compute, which overwrites the bound pipeline with
a copy shader. Subsequent dispatches assumed the update pipeline to be
bound, but dispatched another copy shader instead.

There is also a chance of this happening for geometry info copying for
RRA. Instead of adding another pass (which would make us have to bump
MAX_ENCODE_PASSES again), just rebind the pipeline - RRA captures aren't
the most performance-sensitive thing in the world.
This commit is contained in:
Natalie Vock 2026-02-23 11:51:13 +01:00
parent 99617e4e15
commit dee7455281
1 changed files with 55 additions and 33 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

88
src/amd/vulkan/radv_acceleration_structure.c
View file

@ -5,6 +5,7 @@
*/
#include "meta/radv_meta.h"
#include "radv_buffer.h"
#include "radv_cs.h"
#include "radv_entrypoints.h"
@ -328,7 +329,7 @@ radv_get_build_config(VkDevice _device, struct vk_acceleration_structure_build_s
if (device->meta_state.accel_struct_build.build_args.propagate_cull_flags)
update_key |= VK_BUILD_FLAG_PROPAGATE_CULL_FLAGS;
state->config.update_key[0] = update_key;
state->config.update_key[1] = update_key;
}
static void
@ -702,6 +703,7 @@ radv_init_header(VkCommandBuffer commandBuffer, const struct vk_acceleration_str
sizeof(header) - base);
if (layout.geometry_info_offset != RADV_OFFSET_UNUSED) {
VK_FROM_HANDLE(radv_buffer, dst_buffer, vk_buffer_to_handle(dst->buffer));
uint64_t geometry_infos_size = state->build_info->geometryCount * sizeof(struct radv_accel_struct_geometry_info);
struct radv_accel_struct_geometry_info *geometry_infos = malloc(geometry_infos_size);
@ -716,8 +718,14 @@ radv_init_header(VkCommandBuffer commandBuffer, const struct vk_acceleration_str
geometry_infos[i].primitive_count = state->build_range_infos[i].primitiveCount;
}
radv_CmdUpdateBuffer(commandBuffer, vk_buffer_to_handle(dst->buffer), dst->offset + layout.geometry_info_offset,
geometry_infos_size, geometry_infos);
radv_update_memory(cmd_buffer, vk_acceleration_structure_get_va(dst) + layout.geometry_info_offset,
geometry_infos_size, geometry_infos, radv_get_copy_flags_from_bo(dst_buffer->bo));
/* CmdUpdateBuffer might use compute, which clobbers the pipeline bind point. If that happens, rebind the header
* pipeline to restore the state expected at entry of radv_init_header.
*/
radv_bvh_build_bind_pipeline(commandBuffer, RADV_META_OBJECT_KEY_BVH_HEADER, header_spv, sizeof(header_spv),
sizeof(struct header_args), 0);
free(geometry_infos);
}
@ -740,7 +748,7 @@ radv_init_update_scratch(VkCommandBuffer commandBuffer, const struct vk_accelera
layout.size - layout.internal_ready_count_offset, 0x0, RADV_COPY_FLAGS_DEVICE_LOCAL);
/* geometryCount == 1 passes the data as push constant. */
if (radv_use_bvh8(pdev) && !(state->config.update_key[0] & RADV_BUILD_FLAG_UPDATE_SINGLE_GEOMETRY)) {
if (radv_use_bvh8(pdev) && !(state->config.update_key[1] & RADV_BUILD_FLAG_UPDATE_SINGLE_GEOMETRY)) {
uint32_t data_size = sizeof(struct vk_bvh_geometry_data) * state->build_info->geometryCount;
struct vk_bvh_geometry_data *data = malloc(data_size);
if (!data) {
@ -767,11 +775,40 @@ radv_init_update_scratch(VkCommandBuffer commandBuffer, const struct vk_accelera
}
}
static void
radv_update_copy_prepare(VkCommandBuffer commandBuffer, const struct vk_acceleration_structure_build_state *state,
bool flushed_cp_after_init_update_scratch, bool flushed_compute_after_init_update_scratch)
{
}
static void
radv_update_copy(VkCommandBuffer commandBuffer, const struct vk_acceleration_structure_build_state *state)
{
VK_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
VK_FROM_HANDLE(vk_acceleration_structure, src, state->build_info->srcAccelerationStructure);
VK_FROM_HANDLE(vk_acceleration_structure, dst, state->build_info->dstAccelerationStructure);
struct radv_device *device = radv_cmd_buffer_device(cmd_buffer);
if (src != dst) {
struct acceleration_structure_layout layout;
radv_get_acceleration_structure_layout(device, state, &layout);
/* Copy header/metadata */
const uint64_t src_va = vk_acceleration_structure_get_va(src);
const uint64_t dst_va = vk_acceleration_structure_get_va(dst);
radv_copy_memory(cmd_buffer, src_va, dst_va, layout.bvh_offset, RADV_COPY_FLAGS_DEVICE_LOCAL,
RADV_COPY_FLAGS_DEVICE_LOCAL);
}
}
static void
radv_update_prepare(VkCommandBuffer commandBuffer, const struct vk_acceleration_structure_build_state *state,
bool flushed_cp_after_init_update_scratch, bool flushed_compute_after_init_update_scratch)
{
VK_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
VK_FROM_HANDLE(vk_acceleration_structure, src, state->build_info->srcAccelerationStructure);
VK_FROM_HANDLE(vk_acceleration_structure, dst, state->build_info->dstAccelerationStructure);
struct radv_device *device = radv_cmd_buffer_device(cmd_buffer);
const struct radv_physical_device *pdev = radv_device_physical(device);
@ -784,7 +821,14 @@ radv_update_prepare(VkCommandBuffer commandBuffer, const struct vk_acceleration_
cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_INV_L2;
}
uint32_t flags = state->config.update_key[0];
uint32_t flags = state->config.update_key[1];
/* If we copied anything, we need to wait for that copy. */
if (!(flags & RADV_BUILD_FLAG_UPDATE_IN_PLACE)) {
cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_CS_PARTIAL_FLUSH | RADV_CMD_FLAG_INV_VCACHE |
radv_src_access_flush(cmd_buffer, VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT,
VK_ACCESS_2_SHADER_WRITE_BIT, 0, NULL, NULL);
}
if (radv_use_bvh8(pdev)) {
radv_bvh_build_bind_pipeline(commandBuffer, RADV_META_OBJECT_KEY_BVH_UPDATE, update_gfx12_spv,
@ -813,18 +857,6 @@ radv_update_as(VkCommandBuffer commandBuffer, const struct vk_acceleration_struc
VK_FROM_HANDLE(vk_acceleration_structure, dst, state->build_info->dstAccelerationStructure);
struct radv_device *device = radv_cmd_buffer_device(cmd_buffer);
if (src != dst) {
struct acceleration_structure_layout layout;
radv_get_acceleration_structure_layout(device, state, &layout);
/* Copy header/metadata */
const uint64_t src_va = vk_acceleration_structure_get_va(src);
const uint64_t dst_va = vk_acceleration_structure_get_va(dst);
radv_copy_memory(cmd_buffer, src_va, dst_va, layout.bvh_offset, RADV_COPY_FLAGS_DEVICE_LOCAL,
RADV_COPY_FLAGS_DEVICE_LOCAL);
}
struct update_scratch_layout layout;
radv_get_update_scratch_layout(device, state, &layout);
@ -860,18 +892,6 @@ radv_update_as_gfx12(VkCommandBuffer commandBuffer, const struct vk_acceleration
VK_FROM_HANDLE(vk_acceleration_structure, dst, state->build_info->dstAccelerationStructure);
struct radv_device *device = radv_cmd_buffer_device(cmd_buffer);
if (src != dst) {
struct acceleration_structure_layout layout;
radv_get_acceleration_structure_layout(device, state, &layout);
/* Copy header/metadata */
const uint64_t src_va = vk_acceleration_structure_get_va(src);
const uint64_t dst_va = vk_acceleration_structure_get_va(dst);
radv_copy_memory(cmd_buffer, src_va, dst_va, layout.bvh_offset, RADV_COPY_FLAGS_DEVICE_LOCAL,
RADV_COPY_FLAGS_DEVICE_LOCAL);
}
struct update_scratch_layout layout;
radv_get_update_scratch_layout(device, state, &layout);
@ -884,7 +904,7 @@ radv_update_as_gfx12(VkCommandBuffer commandBuffer, const struct vk_acceleration
.leaf_node_count = state->leaf_node_count,
};
if (state->config.update_key[0] & RADV_BUILD_FLAG_UPDATE_SINGLE_GEOMETRY) {
if (state->config.update_key[1] & RADV_BUILD_FLAG_UPDATE_SINGLE_GEOMETRY) {
const VkAccelerationStructureGeometryKHR *geom =
state->build_info->pGeometries ? &state->build_info->pGeometries[0] : state->build_info->ppGeometries[0];
update_consts.geom_data0 = vk_fill_geometry_data(state->build_info->type, 0, 0, geom, state->build_range_infos);
@ -968,11 +988,13 @@ radv_device_init_accel_struct_build_state(struct radv_device *device)
.get_as_size = radv_get_as_size,
.get_update_scratch_size = radv_get_update_scratch_size,
.init_update_scratch = radv_init_update_scratch,
.update_prepare[0] = radv_update_prepare,
.update_prepare[0] = radv_update_copy_prepare,
.update_as[0] = radv_update_copy,
.update_prepare[1] = radv_update_prepare,
};
if (radv_use_bvh8(pdev)) {
device->meta_state.accel_struct_build.build_ops.update_as[0] = radv_update_as_gfx12;
device->meta_state.accel_struct_build.build_ops.update_as[1] = radv_update_as_gfx12;
device->meta_state.accel_struct_build.build_ops.get_encode_scratch_size = radv_get_encode_scratch_size;
device->meta_state.accel_struct_build.build_ops.encode_prepare[0] = radv_encode_prepare_gfx12;
device->meta_state.accel_struct_build.build_ops.encode_as[0] = radv_encode_as_gfx12;
@ -983,7 +1005,7 @@ radv_device_init_accel_struct_build_state(struct radv_device *device)
device->meta_state.accel_struct_build.build_ops.encode_prepare[3] = radv_init_header_prepare;
device->meta_state.accel_struct_build.build_ops.encode_as[3] = radv_init_header;
} else {
device->meta_state.accel_struct_build.build_ops.update_as[0] = radv_update_as;
device->meta_state.accel_struct_build.build_ops.update_as[1] = radv_update_as;
device->meta_state.accel_struct_build.build_ops.encode_prepare[0] = radv_encode_prepare;
device->meta_state.accel_struct_build.build_ops.encode_as[0] = radv_encode_as;
device->meta_state.accel_struct_build.build_ops.encode_prepare[1] = radv_init_header_prepare;