blob: 44caada27aacb413505ace18bdae1e869af78049 [file] [log] [blame]
// Copyright (c) 2016 The vulkano developers
// Licensed under the Apache License, Version 2.0
// <LICENSE-APACHE or
// https://www.apache.org/licenses/LICENSE-2.0> or the MIT
// license <LICENSE-MIT or https://opensource.org/licenses/MIT>,
// at your option. All files in the project carrying such
// notice may not be copied, modified, or distributed except
// according to those terms.
//! Gather information about rendering, held in query pools.
//!
//! In Vulkan, queries are not created individually. Instead you manipulate **query pools**, which
//! represent a collection of queries. Whenever you use a query, you have to specify both the query
//! pool and the slot id within that query pool.
use crate::check_errors;
use crate::device::Device;
use crate::device::DeviceOwned;
use crate::DeviceSize;
use crate::Error;
use crate::OomError;
use crate::Success;
use crate::VulkanObject;
use std::error;
use std::ffi::c_void;
use std::fmt;
use std::mem::MaybeUninit;
use std::ops::Range;
use std::ptr;
use std::sync::Arc;
/// A collection of one or more queries of a particular type.
#[derive(Debug)]
pub struct QueryPool {
pool: ash::vk::QueryPool,
device: Arc<Device>,
num_slots: u32,
ty: QueryType,
}
impl QueryPool {
/// Builds a new query pool.
pub fn new(
device: Arc<Device>,
ty: QueryType,
num_slots: u32,
) -> Result<QueryPool, QueryPoolCreationError> {
let statistics = match ty {
QueryType::PipelineStatistics(flags) => {
if !device.enabled_features().pipeline_statistics_query {
return Err(QueryPoolCreationError::PipelineStatisticsQueryFeatureNotEnabled);
}
flags.into()
}
QueryType::Occlusion | QueryType::Timestamp => {
ash::vk::QueryPipelineStatisticFlags::empty()
}
};
let pool = unsafe {
let infos = ash::vk::QueryPoolCreateInfo {
flags: ash::vk::QueryPoolCreateFlags::empty(),
query_type: ty.into(),
query_count: num_slots,
pipeline_statistics: statistics,
..Default::default()
};
let mut output = MaybeUninit::uninit();
let fns = device.fns();
check_errors(fns.v1_0.create_query_pool(
device.internal_object(),
&infos,
ptr::null(),
output.as_mut_ptr(),
))?;
output.assume_init()
};
Ok(QueryPool {
pool,
device,
num_slots,
ty,
})
}
/// Returns the [`QueryType`] that this query pool was created with.
#[inline]
pub fn ty(&self) -> QueryType {
self.ty
}
/// Returns the number of query slots of this query pool.
#[inline]
pub fn num_slots(&self) -> u32 {
self.num_slots
}
/// Returns a reference to a single query slot, or `None` if the index is out of range.
#[inline]
pub fn query(&self, index: u32) -> Option<Query> {
if index < self.num_slots() {
Some(Query { pool: self, index })
} else {
None
}
}
/// Returns a reference to a range of queries, or `None` if out of range.
///
/// # Panic
///
/// Panics if the range is empty.
#[inline]
pub fn queries_range(&self, range: Range<u32>) -> Option<QueriesRange> {
assert!(!range.is_empty());
if range.end <= self.num_slots() {
Some(QueriesRange { pool: self, range })
} else {
None
}
}
}
unsafe impl VulkanObject for QueryPool {
type Object = ash::vk::QueryPool;
#[inline]
fn internal_object(&self) -> ash::vk::QueryPool {
self.pool
}
}
unsafe impl DeviceOwned for QueryPool {
#[inline]
fn device(&self) -> &Arc<Device> {
&self.device
}
}
impl Drop for QueryPool {
#[inline]
fn drop(&mut self) {
unsafe {
let fns = self.device.fns();
fns.v1_0
.destroy_query_pool(self.device.internal_object(), self.pool, ptr::null());
}
}
}
/// Error that can happen when creating a query pool.
#[derive(Clone, Debug, PartialEq, Eq)]
pub enum QueryPoolCreationError {
/// Not enough memory.
OomError(OomError),
/// A pipeline statistics pool was requested but the corresponding feature wasn't enabled.
PipelineStatisticsQueryFeatureNotEnabled,
}
impl error::Error for QueryPoolCreationError {
#[inline]
fn source(&self) -> Option<&(dyn error::Error + 'static)> {
match *self {
QueryPoolCreationError::OomError(ref err) => Some(err),
_ => None,
}
}
}
impl fmt::Display for QueryPoolCreationError {
#[inline]
fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
write!(
fmt,
"{}",
match *self {
QueryPoolCreationError::OomError(_) => "not enough memory available",
QueryPoolCreationError::PipelineStatisticsQueryFeatureNotEnabled => {
"a pipeline statistics pool was requested but the corresponding feature \
wasn't enabled"
}
}
)
}
}
impl From<OomError> for QueryPoolCreationError {
#[inline]
fn from(err: OomError) -> QueryPoolCreationError {
QueryPoolCreationError::OomError(err)
}
}
impl From<Error> for QueryPoolCreationError {
#[inline]
fn from(err: Error) -> QueryPoolCreationError {
match err {
err @ Error::OutOfHostMemory => QueryPoolCreationError::OomError(OomError::from(err)),
err @ Error::OutOfDeviceMemory => QueryPoolCreationError::OomError(OomError::from(err)),
_ => panic!("unexpected error: {:?}", err),
}
}
}
/// A reference to a single query slot.
///
/// This is created through [`QueryPool::query`].
#[derive(Clone, Debug)]
pub struct Query<'a> {
pool: &'a QueryPool,
index: u32,
}
impl<'a> Query<'a> {
/// Returns a reference to the query pool.
#[inline]
pub fn pool(&self) -> &'a QueryPool {
&self.pool
}
/// Returns the index of the query represented.
#[inline]
pub fn index(&self) -> u32 {
self.index
}
}
/// A reference to a range of queries.
///
/// This is created through [`QueryPool::queries_range`].
#[derive(Clone, Debug)]
pub struct QueriesRange<'a> {
pool: &'a QueryPool,
range: Range<u32>,
}
impl<'a> QueriesRange<'a> {
/// Returns a reference to the query pool.
#[inline]
pub fn pool(&self) -> &'a QueryPool {
&self.pool
}
/// Returns the range of queries represented.
#[inline]
pub fn range(&self) -> Range<u32> {
self.range.clone()
}
/// Copies the results of this range of queries to a buffer on the CPU.
///
/// [`self.pool().ty().result_size()`](QueryType::result_size) elements
/// will be written for each query in the range, plus 1 extra element per query if
/// [`QueryResultFlags::with_availability`] is enabled.
/// The provided buffer must be large enough to hold the data.
///
/// `true` is returned if every result was available and written to the buffer. `false`
/// is returned if some results were not yet available; these will not be written to the buffer.
///
/// See also [`copy_query_pool_results`](crate::command_buffer::AutoCommandBufferBuilder::copy_query_pool_results).
pub fn get_results<T>(
&self,
destination: &mut [T],
flags: QueryResultFlags,
) -> Result<bool, GetResultsError>
where
T: QueryResultElement,
{
let stride = self.check_query_pool_results::<T>(
destination.as_ptr() as DeviceSize,
destination.len() as DeviceSize,
flags,
)?;
let result = unsafe {
let fns = self.pool.device.fns();
check_errors(fns.v1_0.get_query_pool_results(
self.pool.device.internal_object(),
self.pool.internal_object(),
self.range.start,
self.range.end - self.range.start,
std::mem::size_of_val(destination),
destination.as_mut_ptr() as *mut c_void,
stride,
ash::vk::QueryResultFlags::from(flags) | T::FLAG,
))?
};
Ok(match result {
Success::Success => true,
Success::NotReady => false,
s => panic!("unexpected success value: {:?}", s),
})
}
pub(crate) fn check_query_pool_results<T>(
&self,
buffer_start: DeviceSize,
buffer_len: DeviceSize,
flags: QueryResultFlags,
) -> Result<DeviceSize, GetResultsError>
where
T: QueryResultElement,
{
assert!(buffer_len > 0);
debug_assert!(buffer_start % std::mem::size_of::<T>() as DeviceSize == 0);
let count = self.range.end - self.range.start;
let per_query_len = self.pool.ty.result_size() + flags.with_availability as DeviceSize;
let required_len = per_query_len * count as DeviceSize;
if buffer_len < required_len {
return Err(GetResultsError::BufferTooSmall {
required_len: required_len as DeviceSize,
actual_len: buffer_len as DeviceSize,
});
}
match self.pool.ty {
QueryType::Occlusion => (),
QueryType::PipelineStatistics(_) => (),
QueryType::Timestamp => {
if flags.partial {
return Err(GetResultsError::InvalidFlags);
}
}
}
Ok(per_query_len * std::mem::size_of::<T>() as DeviceSize)
}
}
/// Error that can happen when calling [`QueriesRange::get_results`].
#[derive(Clone, Debug, PartialEq, Eq)]
pub enum GetResultsError {
/// The buffer is too small for the operation.
BufferTooSmall {
/// Required number of elements in the buffer.
required_len: DeviceSize,
/// Actual number of elements in the buffer.
actual_len: DeviceSize,
},
/// The connection to the device has been lost.
DeviceLost,
/// The provided flags are not allowed for this type of query.
InvalidFlags,
/// Not enough memory.
OomError(OomError),
}
impl From<Error> for GetResultsError {
#[inline]
fn from(err: Error) -> Self {
match err {
Error::OutOfHostMemory | Error::OutOfDeviceMemory => {
Self::OomError(OomError::from(err))
}
Error::DeviceLost => Self::DeviceLost,
_ => panic!("unexpected error: {:?}", err),
}
}
}
impl From<OomError> for GetResultsError {
#[inline]
fn from(err: OomError) -> Self {
Self::OomError(err)
}
}
impl fmt::Display for GetResultsError {
#[inline]
fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
write!(
fmt,
"{}",
match *self {
Self::BufferTooSmall { .. } => {
"the buffer is too small for the operation"
}
Self::DeviceLost => "the connection to the device has been lost",
Self::InvalidFlags => {
"the provided flags are not allowed for this type of query"
}
Self::OomError(_) => "not enough memory available",
}
)
}
}
impl error::Error for GetResultsError {
#[inline]
fn source(&self) -> Option<&(dyn error::Error + 'static)> {
match *self {
Self::OomError(ref err) => Some(err),
_ => None,
}
}
}
/// A trait for elements of buffers that can be used as a destination for query results.
///
/// # Safety
/// This is implemented for `u32` and `u64`. Unless you really know what you're doing, you should
/// not implement this trait for any other type.
pub unsafe trait QueryResultElement {
const FLAG: ash::vk::QueryResultFlags;
}
unsafe impl QueryResultElement for u32 {
const FLAG: ash::vk::QueryResultFlags = ash::vk::QueryResultFlags::empty();
}
unsafe impl QueryResultElement for u64 {
const FLAG: ash::vk::QueryResultFlags = ash::vk::QueryResultFlags::TYPE_64;
}
/// The type of query that a query pool should perform.
#[derive(Debug, Copy, Clone)]
pub enum QueryType {
/// Tracks the number of samples that pass per-fragment tests (e.g. the depth test).
Occlusion,
/// Tracks statistics on pipeline invocations and their input data.
PipelineStatistics(QueryPipelineStatisticFlags),
/// Writes timestamps at chosen points in a command buffer.
Timestamp,
}
impl QueryType {
/// Returns the number of [`QueryResultElement`]s that are needed to hold the result of a
/// single query of this type.
///
/// - For `Occlusion` and `Timestamp` queries, this returns 1.
/// - For `PipelineStatistics` queries, this returns the number of statistics flags enabled.
///
/// If the results are retrieved with [`QueryResultFlags::with_availability`] enabled, then
/// an additional element is required per query.
#[inline]
pub const fn result_size(&self) -> DeviceSize {
match self {
Self::Occlusion | Self::Timestamp => 1,
Self::PipelineStatistics(flags) => flags.count(),
}
}
}
impl From<QueryType> for ash::vk::QueryType {
#[inline]
fn from(value: QueryType) -> Self {
match value {
QueryType::Occlusion => ash::vk::QueryType::OCCLUSION,
QueryType::PipelineStatistics(_) => ash::vk::QueryType::PIPELINE_STATISTICS,
QueryType::Timestamp => ash::vk::QueryType::TIMESTAMP,
}
}
}
/// Flags that control how a query is to be executed.
#[derive(Clone, Copy, Debug, Default)]
pub struct QueryControlFlags {
/// For occlusion queries, specifies that the result must reflect the exact number of
/// tests passed. If not enabled, the query may return a result of 1 even if more fragments
/// passed the test.
pub precise: bool,
}
impl From<QueryControlFlags> for ash::vk::QueryControlFlags {
#[inline]
fn from(value: QueryControlFlags) -> Self {
let mut result = ash::vk::QueryControlFlags::empty();
if value.precise {
result |= ash::vk::QueryControlFlags::PRECISE;
}
result
}
}
/// For pipeline statistics queries, the statistics that should be gathered.
#[derive(Clone, Copy, Debug, Default)]
pub struct QueryPipelineStatisticFlags {
/// Count the number of vertices processed by the input assembly.
pub input_assembly_vertices: bool,
/// Count the number of primitives processed by the input assembly.
pub input_assembly_primitives: bool,
/// Count the number of times a vertex shader is invoked.
pub vertex_shader_invocations: bool,
/// Count the number of times a geometry shader is invoked.
pub geometry_shader_invocations: bool,
/// Count the number of primitives generated by geometry shaders.
pub geometry_shader_primitives: bool,
/// Count the number of times the clipping stage is invoked on a primitive.
pub clipping_invocations: bool,
/// Count the number of primitives that are output by the clipping stage.
pub clipping_primitives: bool,
/// Count the number of times a fragment shader is invoked.
pub fragment_shader_invocations: bool,
/// Count the number of patches processed by a tessellation control shader.
pub tessellation_control_shader_patches: bool,
/// Count the number of times a tessellation evaluation shader is invoked.
pub tessellation_evaluation_shader_invocations: bool,
/// Count the number of times a compute shader is invoked.
pub compute_shader_invocations: bool,
}
impl QueryPipelineStatisticFlags {
#[inline]
pub fn none() -> QueryPipelineStatisticFlags {
QueryPipelineStatisticFlags {
input_assembly_vertices: false,
input_assembly_primitives: false,
vertex_shader_invocations: false,
geometry_shader_invocations: false,
geometry_shader_primitives: false,
clipping_invocations: false,
clipping_primitives: false,
fragment_shader_invocations: false,
tessellation_control_shader_patches: false,
tessellation_evaluation_shader_invocations: false,
compute_shader_invocations: false,
}
}
/// Returns the number of flags that are set to `true`.
#[inline]
pub const fn count(&self) -> DeviceSize {
let &Self {
input_assembly_vertices,
input_assembly_primitives,
vertex_shader_invocations,
geometry_shader_invocations,
geometry_shader_primitives,
clipping_invocations,
clipping_primitives,
fragment_shader_invocations,
tessellation_control_shader_patches,
tessellation_evaluation_shader_invocations,
compute_shader_invocations,
} = self;
input_assembly_vertices as DeviceSize
+ input_assembly_primitives as DeviceSize
+ vertex_shader_invocations as DeviceSize
+ geometry_shader_invocations as DeviceSize
+ geometry_shader_primitives as DeviceSize
+ clipping_invocations as DeviceSize
+ clipping_primitives as DeviceSize
+ fragment_shader_invocations as DeviceSize
+ tessellation_control_shader_patches as DeviceSize
+ tessellation_evaluation_shader_invocations as DeviceSize
+ compute_shader_invocations as DeviceSize
}
/// Returns `true` if any flags referring to compute operations are set to `true`.
#[inline]
pub const fn is_compute(&self) -> bool {
let &Self {
compute_shader_invocations,
..
} = self;
compute_shader_invocations
}
/// Returns `true` if any flags referring to graphics operations are set to `true`.
#[inline]
pub const fn is_graphics(&self) -> bool {
let &Self {
input_assembly_vertices,
input_assembly_primitives,
vertex_shader_invocations,
geometry_shader_invocations,
geometry_shader_primitives,
clipping_invocations,
clipping_primitives,
fragment_shader_invocations,
tessellation_control_shader_patches,
tessellation_evaluation_shader_invocations,
..
} = self;
input_assembly_vertices
|| input_assembly_primitives
|| vertex_shader_invocations
|| geometry_shader_invocations
|| geometry_shader_primitives
|| clipping_invocations
|| clipping_primitives
|| fragment_shader_invocations
|| tessellation_control_shader_patches
|| tessellation_evaluation_shader_invocations
}
}
impl From<QueryPipelineStatisticFlags> for ash::vk::QueryPipelineStatisticFlags {
fn from(value: QueryPipelineStatisticFlags) -> ash::vk::QueryPipelineStatisticFlags {
let mut result = ash::vk::QueryPipelineStatisticFlags::empty();
if value.input_assembly_vertices {
result |= ash::vk::QueryPipelineStatisticFlags::INPUT_ASSEMBLY_VERTICES;
}
if value.input_assembly_primitives {
result |= ash::vk::QueryPipelineStatisticFlags::INPUT_ASSEMBLY_PRIMITIVES;
}
if value.vertex_shader_invocations {
result |= ash::vk::QueryPipelineStatisticFlags::VERTEX_SHADER_INVOCATIONS;
}
if value.geometry_shader_invocations {
result |= ash::vk::QueryPipelineStatisticFlags::GEOMETRY_SHADER_INVOCATIONS;
}
if value.geometry_shader_primitives {
result |= ash::vk::QueryPipelineStatisticFlags::GEOMETRY_SHADER_PRIMITIVES;
}
if value.clipping_invocations {
result |= ash::vk::QueryPipelineStatisticFlags::CLIPPING_INVOCATIONS;
}
if value.clipping_primitives {
result |= ash::vk::QueryPipelineStatisticFlags::CLIPPING_PRIMITIVES;
}
if value.fragment_shader_invocations {
result |= ash::vk::QueryPipelineStatisticFlags::FRAGMENT_SHADER_INVOCATIONS;
}
if value.tessellation_control_shader_patches {
result |= ash::vk::QueryPipelineStatisticFlags::TESSELLATION_CONTROL_SHADER_PATCHES;
}
if value.tessellation_evaluation_shader_invocations {
result |=
ash::vk::QueryPipelineStatisticFlags::TESSELLATION_EVALUATION_SHADER_INVOCATIONS;
}
if value.compute_shader_invocations {
result |= ash::vk::QueryPipelineStatisticFlags::COMPUTE_SHADER_INVOCATIONS;
}
result
}
}
/// Flags to control how the results of a query should be retrieved.
///
/// `VK_QUERY_RESULT_64_BIT` is not included, as it is determined automatically via the
/// [`QueryResultElement`] trait.
#[derive(Clone, Copy, Debug, Default)]
pub struct QueryResultFlags {
/// Wait for the results to become available before writing the results.
pub wait: bool,
/// Write an additional element to the end of each query's results, indicating the availability
/// of the results:
/// - Nonzero: The results are available, and have been written to the element(s) preceding.
/// - Zero: The results are not yet available, and have not been written.
pub with_availability: bool,
/// Allow writing partial results to the buffer, instead of waiting until they are fully
/// available.
pub partial: bool,
}
impl From<QueryResultFlags> for ash::vk::QueryResultFlags {
#[inline]
fn from(value: QueryResultFlags) -> Self {
let mut result = ash::vk::QueryResultFlags::empty();
if value.wait {
result |= ash::vk::QueryResultFlags::WAIT;
}
if value.with_availability {
result |= ash::vk::QueryResultFlags::WITH_AVAILABILITY;
}
if value.partial {
result |= ash::vk::QueryResultFlags::PARTIAL;
}
result
}
}
#[cfg(test)]
mod tests {
use crate::query::QueryPipelineStatisticFlags;
use crate::query::QueryPool;
use crate::query::QueryPoolCreationError;
use crate::query::QueryType;
#[test]
fn pipeline_statistics_feature() {
let (device, _) = gfx_dev_and_queue!();
let ty = QueryType::PipelineStatistics(QueryPipelineStatisticFlags::none());
match QueryPool::new(device, ty, 256) {
Err(QueryPoolCreationError::PipelineStatisticsQueryFeatureNotEnabled) => (),
_ => panic!(),
};
}
}