fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-26 02:00:12 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

radv: use vk_query

Browse source 
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24416>
This commit is contained in:
Samuel Pitoiset 2023-08-01 08:47:07 +02:00 • committed by Marge Bot
parent 97c926cf17
commit a7d6edfb36
3 changed files with 32 additions and 36 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
src/amd/vulkan/radv_private.h
View file

@ -65,6 +65,7 @@
#include "vk_instance.h"
#include "vk_log.h"
#include "vk_physical_device.h"
#include "vk_query_pool.h"
#include "vk_queue.h"
#include "vk_sampler.h"
#include "vk_shader_module.h"
@ -2939,14 +2940,12 @@ VkResult radv_device_init_meta(struct radv_device *device);
void radv_device_finish_meta(struct radv_device *device);
struct radv_query_pool {
struct vk_object_base base;
struct vk_query_pool vk;
struct radeon_winsys_bo *bo;
uint32_t stride;
uint32_t availability_offset;
uint64_t size;
char *ptr;
VkQueryType type;
uint32_t pipeline_stats_mask;
bool uses_gds; /* For NGG GS on GFX10+ */
};
@ -3661,7 +3660,7 @@ VK_DEFINE_NONDISP_HANDLE_CASTS(radv_indirect_command_layout, base, VkIndirectCom
VK_OBJECT_TYPE_INDIRECT_COMMANDS_LAYOUT_NV)
VK_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline, base, VkPipeline, VK_OBJECT_TYPE_PIPELINE)
VK_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout, base, VkPipelineLayout, VK_OBJECT_TYPE_PIPELINE_LAYOUT)
VK_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool, base, VkQueryPool, VK_OBJECT_TYPE_QUERY_POOL)
VK_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool, vk.base, VkQueryPool, VK_OBJECT_TYPE_QUERY_POOL)
VK_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler, vk.base, VkSampler, VK_OBJECT_TYPE_SAMPLER)
VK_DEFINE_NONDISP_HANDLE_CASTS(radv_video_session, vk.base, VkVideoSessionKHR, VK_OBJECT_TYPE_VIDEO_SESSION_KHR)

55
src/amd/vulkan/radv_query.c
View file

@ -1027,7 +1027,7 @@ static void
radv_destroy_query_pool(struct radv_device *device, const VkAllocationCallbacks *pAllocator,
struct radv_query_pool *pool)
{
if (pool->type == VK_QUERY_TYPE_PERFORMANCE_QUERY_KHR)
if (pool->vk.query_type == VK_QUERY_TYPE_PERFORMANCE_QUERY_KHR)
radv_pc_deinit_query_pool((struct radv_pc_query_pool *)pool);
if (pool->bo) {
@ -1036,7 +1036,7 @@ radv_destroy_query_pool(struct radv_device *device, const VkAllocationCallbacks
}
radv_rmv_log_resource_destroy(device, (uint64_t)radv_query_pool_to_handle(pool));
vk_object_base_finish(&pool->base);
vk_query_pool_finish(&pool->vk);
vk_free2(&device->vk.alloc, pAllocator, pool);
}
@ -1055,10 +1055,7 @@ radv_create_query_pool(struct radv_device *device, const VkQueryPoolCreateInfo *
if (!pool)
return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY);
vk_object_base_init(&device->vk, &pool->base, VK_OBJECT_TYPE_QUERY_POOL);
pool->type = pCreateInfo->queryType;
pool->pipeline_stats_mask = pCreateInfo->pipelineStatistics;
vk_query_pool_init(&device->vk, &pool->vk, pCreateInfo);
/* The number of primitives generated by geometry shader invocations is only counted by the
* hardware if GS uses the legacy path. When NGG GS is used, the hardware can't know the number
@ -1070,8 +1067,8 @@ radv_create_query_pool(struct radv_device *device, const VkQueryPoolCreateInfo *
*/
pool->uses_gds =
(device->physical_device->emulate_ngg_gs_query_pipeline_stat &&
(pool->pipeline_stats_mask & (VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_PRIMITIVES_BIT |
VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_INVOCATIONS_BIT))) ||
(pool->vk.pipeline_statistics & (VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_PRIMITIVES_BIT |
VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_INVOCATIONS_BIT))) ||
(device->physical_device->use_ngg && pCreateInfo->queryType == VK_QUERY_TYPE_PRIMITIVES_GENERATED_EXT);
switch (pCreateInfo->queryType) {
@ -1175,7 +1172,7 @@ radv_GetQueryPoolResults(VkDevice _device, VkQueryPool queryPool, uint32_t first
char *src = pool->ptr + query * pool->stride;
uint32_t available;
switch (pool->type) {
switch (pool->vk.query_type) {
case VK_QUERY_TYPE_TIMESTAMP:
case VK_QUERY_TYPE_ACCELERATION_STRUCTURE_COMPACTED_SIZE_KHR:
case VK_QUERY_TYPE_ACCELERATION_STRUCTURE_SERIALIZATION_SIZE_KHR:
@ -1258,9 +1255,9 @@ radv_GetQueryPoolResults(VkDevice _device, VkQueryPool queryPool, uint32_t first
const uint64_t *stop = (uint64_t *)(src + pipelinestat_block_size);
if (flags & VK_QUERY_RESULT_64_BIT) {
uint64_t *dst = (uint64_t *)dest;
dest += util_bitcount(pool->pipeline_stats_mask) * 8;
dest += util_bitcount(pool->vk.pipeline_statistics) * 8;
for (int i = 0; i < ARRAY_SIZE(pipeline_statistics_indices); ++i) {
if (pool->pipeline_stats_mask & (1u << i)) {
if (pool->vk.pipeline_statistics & (1u << i)) {
if (available || (flags & VK_QUERY_RESULT_PARTIAL_BIT)) {
*dst = stop[pipeline_statistics_indices[i]] - start[pipeline_statistics_indices[i]];
}
@ -1270,9 +1267,9 @@ radv_GetQueryPoolResults(VkDevice _device, VkQueryPool queryPool, uint32_t first
} else {
uint32_t *dst = (uint32_t *)dest;
dest += util_bitcount(pool->pipeline_stats_mask) * 4;
dest += util_bitcount(pool->vk.pipeline_statistics) * 4;
for (int i = 0; i < ARRAY_SIZE(pipeline_statistics_indices); ++i) {
if (pool->pipeline_stats_mask & (1u << i)) {
if (pool->vk.pipeline_statistics & (1u << i)) {
if (available || (flags & VK_QUERY_RESULT_PARTIAL_BIT)) {
*dst = stop[pipeline_statistics_indices[i]] - start[pipeline_statistics_indices[i]];
}
@ -1417,7 +1414,7 @@ static size_t
radv_query_result_size(const struct radv_query_pool *pool, VkQueryResultFlags flags)
{
unsigned values = (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) ? 1 : 0;
switch (pool->type) {
switch (pool->vk.query_type) {
case VK_QUERY_TYPE_TIMESTAMP:
case VK_QUERY_TYPE_ACCELERATION_STRUCTURE_COMPACTED_SIZE_KHR:
case VK_QUERY_TYPE_ACCELERATION_STRUCTURE_SERIALIZATION_SIZE_KHR:
@ -1427,7 +1424,7 @@ radv_query_result_size(const struct radv_query_pool *pool, VkQueryResultFlags fl
values += 1;
break;
case VK_QUERY_TYPE_PIPELINE_STATISTICS:
values += util_bitcount(pool->pipeline_stats_mask);
values += util_bitcount(pool->vk.pipeline_statistics);
break;
case VK_QUERY_TYPE_TRANSFORM_FEEDBACK_STREAM_EXT:
values += 2;
@ -1477,7 +1474,7 @@ radv_CmdCopyQueryPoolResults(VkCommandBuffer commandBuffer, VkQueryPool queryPoo
*/
emit_query_flush(cmd_buffer, pool);
switch (pool->type) {
switch (pool->vk.query_type) {
case VK_QUERY_TYPE_OCCLUSION:
radv_query_shader(cmd_buffer, &cmd_buffer->device->meta_state.query.occlusion_query_pipeline, pool->bo,
dst_buffer->bo, firstQuery * pool->stride, dst_buffer->offset + dstOffset, pool->stride, stride,
@ -1498,7 +1495,7 @@ radv_CmdCopyQueryPoolResults(VkCommandBuffer commandBuffer, VkQueryPool queryPoo
}
radv_query_shader(cmd_buffer, &cmd_buffer->device->meta_state.query.pipeline_statistics_query_pipeline, pool->bo,
dst_buffer->bo, firstQuery * pool->stride, dst_buffer->offset + dstOffset, pool->stride, stride,
dst_size, queryCount, flags, pool->pipeline_stats_mask,
dst_size, queryCount, flags, pool->vk.pipeline_statistics,
pool->availability_offset + 4 * firstQuery, false);
break;
case VK_QUERY_TYPE_TIMESTAMP:
@ -1587,7 +1584,7 @@ radv_CmdResetQueryPool(VkCommandBuffer commandBuffer, VkQueryPool queryPool, uin
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
RADV_FROM_HANDLE(radv_query_pool, pool, queryPool);
uint32_t value = query_clear_value(pool->type);
uint32_t value = query_clear_value(pool->vk.query_type);
uint32_t flush_bits = 0;
/* Make sure to sync all previous work if the given command buffer has
@ -1599,7 +1596,7 @@ radv_CmdResetQueryPool(VkCommandBuffer commandBuffer, VkQueryPool queryPool, uin
flush_bits |= radv_fill_buffer(cmd_buffer, NULL, pool->bo, radv_buffer_get_va(pool->bo) + firstQuery * pool->stride,
queryCount * pool->stride, value);
if (pool->type == VK_QUERY_TYPE_PIPELINE_STATISTICS) {
if (pool->vk.query_type == VK_QUERY_TYPE_PIPELINE_STATISTICS) {
flush_bits |=
radv_fill_buffer(cmd_buffer, NULL, pool->bo,
radv_buffer_get_va(pool->bo) + pool->availability_offset + firstQuery * 4, queryCount * 4, 0);
@ -1617,14 +1614,14 @@ radv_ResetQueryPool(VkDevice _device, VkQueryPool queryPool, uint32_t firstQuery
{
RADV_FROM_HANDLE(radv_query_pool, pool, queryPool);
uint32_t value = query_clear_value(pool->type);
uint32_t value = query_clear_value(pool->vk.query_type);
uint32_t *data = (uint32_t *)(pool->ptr + firstQuery * pool->stride);
uint32_t *data_end = (uint32_t *)(pool->ptr + (firstQuery + queryCount) * pool->stride);
for (uint32_t *p = data; p != data_end; ++p)
*p = value;
if (pool->type == VK_QUERY_TYPE_PIPELINE_STATISTICS) {
if (pool->vk.query_type == VK_QUERY_TYPE_PIPELINE_STATISTICS) {
memset(pool->ptr + pool->availability_offset + firstQuery * 4, 0, queryCount * 4);
}
}
@ -1737,13 +1734,13 @@ emit_begin_query(struct radv_cmd_buffer *cmd_buffer, struct radv_query_pool *poo
radeon_emit(cs, va >> 32);
if (pool->uses_gds) {
if (pool->pipeline_stats_mask & VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_PRIMITIVES_BIT) {
if (pool->vk.pipeline_statistics & VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_PRIMITIVES_BIT) {
uint32_t gs_prim_offset =
radv_get_pipelinestat_query_offset(VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_PRIMITIVES_BIT);
gfx10_copy_gds_query(cmd_buffer, RADV_SHADER_QUERY_GS_PRIM_EMIT_OFFSET, va + gs_prim_offset);
} else {
assert(pool->pipeline_stats_mask & VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_INVOCATIONS_BIT);
assert(pool->vk.pipeline_statistics & VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_INVOCATIONS_BIT);
uint32_t gs_invoc_offset =
radv_get_pipelinestat_query_offset(VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_INVOCATIONS_BIT);
@ -1886,13 +1883,13 @@ emit_end_query(struct radv_cmd_buffer *cmd_buffer, struct radv_query_pool *pool,
EOP_DATA_SEL_VALUE_32BIT, avail_va, 1, cmd_buffer->gfx9_eop_bug_va);
if (pool->uses_gds) {
if (pool->pipeline_stats_mask & VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_PRIMITIVES_BIT) {
if (pool->vk.pipeline_statistics & VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_PRIMITIVES_BIT) {
uint32_t gs_prim_offset =
radv_get_pipelinestat_query_offset(VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_PRIMITIVES_BIT);
gfx10_copy_gds_query(cmd_buffer, RADV_SHADER_QUERY_GS_PRIM_EMIT_OFFSET, va + gs_prim_offset);
} else {
assert(pool->pipeline_stats_mask & VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_INVOCATIONS_BIT);
assert(pool->vk.pipeline_statistics & VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_INVOCATIONS_BIT);
uint32_t gs_invoc_offset =
radv_get_pipelinestat_query_offset(VK_QUERY_PIPELINE_STATISTIC_GEOMETRY_SHADER_INVOCATIONS_BIT);
@ -1991,7 +1988,7 @@ radv_CmdBeginQueryIndexedEXT(VkCommandBuffer commandBuffer, VkQueryPool queryPoo
va += pool->stride * query;
emit_begin_query(cmd_buffer, pool, va, pool->type, flags, index);
emit_begin_query(cmd_buffer, pool, va, pool->vk.query_type, flags, index);
}
VKAPI_ATTR void VKAPI_CALL
@ -2006,7 +2003,7 @@ radv_CmdEndQueryIndexedEXT(VkCommandBuffer commandBuffer, VkQueryPool queryPool,
/* Do not need to add the pool BO to the list because the query must
* currently be active, which means the BO is already in the list.
*/
emit_end_query(cmd_buffer, pool, va, avail_va, pool->type, index);
emit_end_query(cmd_buffer, pool, va, avail_va, pool->vk.query_type, index);
/*
* For multiview we have to emit a query for each bit in the mask,
@ -2020,8 +2017,8 @@ radv_CmdEndQueryIndexedEXT(VkCommandBuffer commandBuffer, VkQueryPool queryPool,
for (unsigned i = 1; i < util_bitcount(cmd_buffer->state.render.view_mask); i++) {
va += pool->stride;
avail_va += 4;
emit_begin_query(cmd_buffer, pool, va, pool->type, 0, 0);
emit_end_query(cmd_buffer, pool, va, avail_va, pool->type, 0);
emit_begin_query(cmd_buffer, pool, va, pool->vk.query_type, 0, 0);
emit_end_query(cmd_buffer, pool, va, avail_va, pool->vk.query_type, 0);
}
}
}

6
src/amd/vulkan/radv_rmv.c
View file

@ -607,8 +607,8 @@ radv_rmv_log_query_pool_create(struct radv_device *device, VkQueryPool _pool, bo
RADV_FROM_HANDLE(radv_query_pool, pool, _pool);
if (pool->type != VK_QUERY_TYPE_OCCLUSION && pool->type != VK_QUERY_TYPE_PIPELINE_STATISTICS &&
pool->type != VK_QUERY_TYPE_TRANSFORM_FEEDBACK_STREAM_EXT)
if (pool->vk.query_type != VK_QUERY_TYPE_OCCLUSION && pool->vk.query_type != VK_QUERY_TYPE_PIPELINE_STATISTICS &&
pool->vk.query_type != VK_QUERY_TYPE_TRANSFORM_FEEDBACK_STREAM_EXT)
return;
radv_rmv_log_bo_allocate(device, pool->bo, pool->size, is_internal);
@ -618,7 +618,7 @@ radv_rmv_log_query_pool_create(struct radv_device *device, VkQueryPool _pool, bo
create_token.is_driver_internal = is_internal;
create_token.resource_id = vk_rmv_get_resource_id_locked(&device->vk, (uint64_t)_pool);
create_token.type = VK_RMV_RESOURCE_TYPE_QUERY_HEAP;
create_token.query_pool.type = pool->type;
create_token.query_pool.type = pool->vk.query_type;
create_token.query_pool.has_cpu_access = true;
vk_rmv_emit_token(&device->vk.memory_trace_data, VK_RMV_TOKEN_TYPE_RESOURCE_CREATE, &create_token);