blob: 84956395be76fadc34206fb1da57a2cc000a9089 [file]
// Copyright (c) 2022-2026 The Khronos Group Inc.
// Copyright (c) 2022-2026 Valve Corporation
// Copyright (c) 2022-2026 LunarG, Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#version 460
#extension GL_GOOGLE_include_directive : enable
#extension GL_EXT_shader_explicit_arithmetic_types_int8 : require
#extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
#include "common.h"
#include "build_acceleration_structures.h"
layout(push_constant, scalar)
uniform PushConstants {
BLASValidationShaderPushData pc;
};
// CPU will try to dispatch `primitive_count` threads
layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
layout(buffer_reference, scalar) buffer ArrayU8 { uint8_t array[]; };
layout(buffer_reference, scalar) buffer ArrayU16 { uint16_t array[]; };
layout(buffer_reference, scalar) buffer ArrayU32 { uint array[]; };
layout(buffer_reference, scalar) buffer ArrayF32 { float array[]; };
// From VkIndexType
const uint VK_INDEX_TYPE_UINT16 = 0;
const uint VK_INDEX_TYPE_UINT32 = 1;
const uint VK_INDEX_TYPE_UINT8 = 1000265000;
const uint VK_INDEX_TYPE_NONE_KHR = 1000165000;
// From VkFormat
const uint VK_FORMAT_R32G32B32_SFLOAT = 106;
uint LoadIndex(uint64_t address, uint index_type, uint primitive_offset, uint i) {
if (pc.index_type == VK_INDEX_TYPE_NONE_KHR) {
return i;
} else if (index_type == VK_INDEX_TYPE_UINT16) {
ArrayU16 array_u16 = ArrayU16(address + primitive_offset);
return uint(array_u16.array[i]);
} else if (index_type == VK_INDEX_TYPE_UINT32) {
ArrayU32 array_u32 = ArrayU32(address + primitive_offset);
return array_u32.array[i];
} else if (index_type == VK_INDEX_TYPE_UINT8) {
ArrayU8 array_u8 = ArrayU8(address + primitive_offset);
return uint(array_u8.array[i]);
} else {
return 0;
}
}
void StoreIndex(uint64_t address, uint i, uint index_type, uint primitive_offset, uint value) {
if (index_type == VK_INDEX_TYPE_UINT16) {
ArrayU16 array_u16 = ArrayU16(address + primitive_offset);
array_u16.array[i] = uint16_t(value);
} else if (index_type == VK_INDEX_TYPE_UINT32) {
ArrayU32 array_u32 = ArrayU32(address + primitive_offset);
array_u32.array[i] = value;
} else if (index_type == VK_INDEX_TYPE_UINT8) {
ArrayU8 array_u8 = ArrayU8(address + primitive_offset);
array_u8.array[i] = uint8_t(value);
}
}
// Reference:
// https://docs.vulkan.org/refpages/latest/refpages/source/VkAccelerationStructureBuildRangeInfoKHR.html#_description
float LoadVertexX(uint64_t vertices_addr, uint64_t stride, uint index, uint format) {
float vertex_x = 0;
if (format == VK_FORMAT_R32G32B32_SFLOAT) {
// #ARNO_TODO check stride computations
ArrayF32 array_f32 = ArrayF32(vertices_addr + index * stride);
vertex_x = array_f32.array[0];
}
return vertex_x;
}
struct VkAabbPositionsKHR {
float minX;
float minY;
float minZ;
float maxX;
float maxY;
float maxZ;
};
layout(buffer_reference, scalar) buffer AabbsPtr { VkAabbPositionsKHR coords; };
layout(buffer_reference, scalar) buffer TransformPtr { mat3x4 mat; };
void main() {
const uint gid = gl_GlobalInvocationID.x;
if (pc.validation_mode == kBLASValidationMode_triangles_indices) {
if (gid >= (3 * pc.primitive_count)) {
return;
}
const uint64_t indices_addr = pc.address;
const uint fetched_index = LoadIndex(indices_addr, pc.index_type, pc.primitive_offset, gid);
if (pc.max_vertex < (pc.first_vertex + fetched_index)) {
// In practice an invalid index does not cause a device loss, so don't bother changing its value.
// Should someone add this back, a write barrier needs to be added on the CPU side.
// StoreIndex(gid, 0);
GpuavLogError4(kErrorGroup_GpuPreBuildAccelerationStructures,
kErrorSubCode_PreBuildAccelerationStructures_MaxFetchedIndex,
pc.error_info_i,
gid,
fetched_index,
0);
}
} else if (pc.validation_mode == kBLASValidationMode_active_triangles) {
if (gid >= (3 * pc.primitive_count)) {
return;
}
const uint64_t indices_addr = pc.address;
const uint fetched_index = LoadIndex(indices_addr, pc.index_type, pc.primitive_offset, gid);
uint64_t update_time_vertex_buffer_addr = pc.address_2 + pc.update_time_stride * pc.first_vertex;
if (pc.index_type == VK_INDEX_TYPE_NONE_KHR) {
update_time_vertex_buffer_addr += pc.primitive_offset;
}
const float update_time_vertex_x = LoadVertexX(update_time_vertex_buffer_addr, pc.update_time_stride, fetched_index, pc.vertex_format);
const bool is_update_time_vertex_nan = isnan(update_time_vertex_x);
const uint64_t build_time_vertex_buffer_addr = pc.address_3;
const float build_time_vertex_x = LoadVertexX(build_time_vertex_buffer_addr, pc.build_time_stride, fetched_index, pc.vertex_format);
const bool is_build_time_vertex_nan = isnan(build_time_vertex_x);
if (is_update_time_vertex_nan != is_build_time_vertex_nan) {
GpuavLogError4(kErrorGroup_GpuPreBuildAccelerationStructures,
kErrorSubCode_PreBuildAccelerationStructures_VertexBufferActiveStatusUpdated,
pc.error_info_i,
gid,
fetched_index,
uint(is_build_time_vertex_nan));
}
} else if (pc.validation_mode == kBLASValidationMode_aabbs) {
if (gid >= pc.primitive_count) {
return;
}
const uint64_t aabb_i = gid;
AabbsPtr aabb = AabbsPtr(pc.address + uint64_t(pc.primitive_offset) + aabb_i * pc.update_time_stride);
if (aabb.coords.minX > aabb.coords.maxX) {
GpuavLogError4(kErrorGroup_GpuPreBuildAccelerationStructures,
kErrorSubCode_PreBuildAccelerationStructures_MinMaxAabb_X,
pc.error_info_i,
gid,
floatBitsToUint(aabb.coords.minX),
floatBitsToUint(aabb.coords.maxX));
}
if (aabb.coords.minY > aabb.coords.maxY) {
GpuavLogError4(kErrorGroup_GpuPreBuildAccelerationStructures,
kErrorSubCode_PreBuildAccelerationStructures_MinMaxAabb_Y,
pc.error_info_i,
gid,
floatBitsToUint(aabb.coords.minY),
floatBitsToUint(aabb.coords.maxY));
}
if (aabb.coords.minZ > aabb.coords.maxZ) {
GpuavLogError4(kErrorGroup_GpuPreBuildAccelerationStructures,
kErrorSubCode_PreBuildAccelerationStructures_MinMaxAabb_Z,
pc.error_info_i,
gid,
floatBitsToUint(aabb.coords.minZ),
floatBitsToUint(aabb.coords.maxZ));
}
// If a copy of the AABBs supplied at build time is supplied,
// Validate that previously active/inactive AABBs did not change status
if (pc.address_2 != 0) {
AabbsPtr update_time_aabb = aabb;
AabbsPtr build_time_aabb = AabbsPtr(pc.address_2 + aabb_i * pc.build_time_stride);
const bool is_update_time_aabb_x_nan = isnan(update_time_aabb.coords.minX);
const bool is_build_time_aabb_x_nan = isnan(build_time_aabb.coords.minX);
if (is_update_time_aabb_x_nan != is_build_time_aabb_x_nan) {
GpuavLogError4(
kErrorGroup_GpuPreBuildAccelerationStructures,
kErrorSubCode_PreBuildAccelerationStructures_AabbBufferActiveStatusUpdated,
pc.error_info_i,
gid,
uint(is_build_time_aabb_x_nan),
0);
}
}
} else if (pc.validation_mode == kBLASValidationMode_transform_matrix) {
if (gid > 0) {
return;
}
TransformPtr transform = TransformPtr(pc.address + uint64_t(pc.primitive_offset));
const mat3 transform3x3 = mat3(
transform.mat[0].xyz,
transform.mat[1].xyz,
transform.mat[2].xyz
);
const float det = determinant(transform3x3);
if (abs(det) < 1e-6) {
GpuavLogError4(kErrorGroup_GpuPreBuildAccelerationStructures,
kErrorSubCode_PreBuildAccelerationStructures_Transform,
pc.error_info_i,
gid,
0,
0);
}
}
}