16.2. Query Operation

The operation of queries is controlled by the commands vkCmdBeginQuery, vkCmdEndQuery, vkCmdResetQueryPool, vkCmdCopyQueryPoolResults, and vkCmdWriteTimestamp.

In order for a VkCommandBuffer to record query management commands, the queue family for which its VkCommandPool was created must support the appropriate type of operations (graphics, compute) suitable for the query type of a given query pool.

Each query in a query pool has a status that is either unavailable or available, and also has state to store the numerical results of a query operation of the type requested when the query pool was created. Resetting a query via vkCmdResetQueryPool sets the status to unavailable and makes the numerical results undefined. Performing a query operation with vkCmdBeginQuery and vkCmdEndQuery changes the status to available when the query finishes, and updates the numerical results. Both the availability status and numerical results are retrieved by calling either vkGetQueryPoolResults or vkCmdCopyQueryPoolResults.

All query commands execute in order and are guaranteed to see the effects of each other’s memory accesses, with one significant exception: vkCmdCopyQueryPoolResults may execute before the results of vkCmdEndQuery are available. However, if VK_QUERY_RESULT_WAIT_BIT is used, then vkCmdCopyQueryPoolResults must reflect the result of any previously executed queries. Other sequences of commands, such as vkCmdResetQueryPool followed by vkCmdBeginQuery, must make the effects of the first command visible to the second command.

After query pool creation, each query is in an undefined state and must be reset prior to use. Queries must also be reset between uses. Using a query that has not been reset will result in undefined behavior.

To reset a range of queries in a query pool, call:

 

void vkCmdResetQueryPool(
    VkCommandBuffer                             commandBuffer,
    VkQueryPool                                 queryPool,
    uint32_t                                    firstQuery,
    uint32_t                                    queryCount);

When executed on a queue, this command sets the status of query indices $firstQuery,firstQuery+queryCount-1$ to unavailable.

Once queries are reset and ready for use, query commands can be issued to a command buffer. Occlusion queries and pipeline statistics queries count events - drawn samples and pipeline stage invocations, respectively - resulting from commands that are recorded between a vkCmdBeginQuery command and a vkCmdEndQuery command within a specified command buffer, effectively scoping a set of drawing and/or compute commands. Timestamp queries write timestamps to a query pool.

A query must begin and end in the same command buffer, although if it is a primary command buffer, and the inherited queries feature is enabled, it can execute secondary command buffers during the query operation. For a secondary command buffer to be executed while a query is active, it must set the occlusionQueryEnable, queryFlags, and/or pipelineStatistics members of VkCommandBufferInheritanceInfo to conservative values, as described in the Command Buffer Recording section. A query must either begin and end inside the same subpass of a render pass instance, or must both begin and end outside of a render pass instance (i.e. contain entire render pass instances).

Begin a query by calling:

 

void vkCmdBeginQuery(
    VkCommandBuffer                             commandBuffer,
    VkQueryPool                                 queryPool,
    uint32_t                                    query,
    VkQueryControlFlags                         flags);

 

typedef enum VkQueryControlFlagBits {
    VK_QUERY_CONTROL_PRECISE_BIT = 0x00000001,
} VkQueryControlFlagBits;

If the queryType of the pool is VK_QUERY_TYPE_OCCLUSION and flags contains VK_QUERY_CONTROL_PRECISE_BIT, an implementation must return a result that matches the actual number of samples passed. This is described in more detail in Occlusion Queries.

After beginning a query, that query is considered active within the command buffer it was called in until that same query is ended. Queries active in a primary command buffer when secondary command buffers are executed are considered active for those secondary command buffers.

After the set of desired draw or dispatch commands, end a query by calling:

 

void vkCmdEndQuery(
    VkCommandBuffer                             commandBuffer,
    VkQueryPool                                 queryPool,
    uint32_t                                    query);

As queries operate asynchronously, ending a query does not immediately set the query’s status to available. A query is considered finished when the final results of the query are ready to be retrieved by vkGetQueryPoolResults and vkCmdCopyQueryPoolResults, and this is when the query’s status is set to available.

Once a query is ended the query must finish in finite time, unless the state of the query is changed using other commands, e.g. by issuing a reset of the query.

An application can retrieve results either by requesting they be written into application-provided memory, or by requesting they be copied into a VkBuffer. In either case, the layout in memory is defined as follows:

To retrieve status and results for a set of queries, call:

 

VkResult vkGetQueryPoolResults(
    VkDevice                                    device,
    VkQueryPool                                 queryPool,
    uint32_t                                    firstQuery,
    uint32_t                                    queryCount,
    size_t                                      dataSize,
    void*                                       pData,
    VkDeviceSize                                stride,
    VkQueryResultFlags                          flags);

Valid bits in flags include:

 

typedef enum VkQueryResultFlagBits {
    VK_QUERY_RESULT_64_BIT = 0x00000001,
    VK_QUERY_RESULT_WAIT_BIT = 0x00000002,
    VK_QUERY_RESULT_WITH_AVAILABILITY_BIT = 0x00000004,
    VK_QUERY_RESULT_PARTIAL_BIT = 0x00000008,
} VkQueryResultFlagBits;

These bits have the following meanings:

If no bits are set in flags, and all requested queries are in the available state, results are written as an array of 32-bit unsigned integer values. The behavior when not all queries are available, is described below.

If VK_QUERY_RESULT_64_BIT is not set and the result overflows a 32-bit value, the value may either wrap or saturate. Similarly, if VK_QUERY_RESULT_64_BIT is set and the result overflows a 64-bit value, the value may either wrap or saturate.

If VK_QUERY_RESULT_WAIT_BIT is set, Vulkan will wait for each query to be in the available state before retrieving the numerical results for that query. In this case, vkGetQueryPoolResults is guaranteed to succeed and return VK_SUCCESS if the queries become available in a finite time (i.e. if they have been issued and not reset). If queries will never finish (e.g. due to being reset but not issued), then vkGetQueryPoolResults may not return in finite time.

If VK_QUERY_RESULT_WAIT_BIT and VK_QUERY_RESULT_PARTIAL_BIT are both not set then no result values are written to pData for queries that are in the unavailable state at the time of the call, and vkGetQueryPoolResults returns VK_NOT_READY. However, availability state is still written to pData for those queries if VK_QUERY_RESULT_WITH_AVAILABILITY_BIT is set.

[Note]Note

Applications must take care to ensure that use of the VK_QUERY_RESULT_WAIT_BIT bit has the desired effect.

For example, if a query has been used previously and a command buffer records the commands vkCmdResetQueryPool, vkCmdBeginQuery, and vkCmdEndQuery for that query, then the query will remain in the available state until the vkCmdResetQueryPool command executes on a queue. Applications can use fences or events to ensure that an query has already been reset before checking for its results or availability status. Otherwise, a stale value could be returned from a previous use of the query.

The above also applies when VK_QUERY_RESULT_WAIT_BIT is used in combination with VK_QUERY_RESULT_WITH_AVAILABILITY_BIT. In this case, the returned availability status may reflect the result of a previous use of the query unless the vkCmdResetQueryPool command has been executed since the last use of the query.

[Note]Note

Applications can double-buffer query pool usage, with a pool per frame, and reset queries at the end of the frame in which they are read.

If VK_QUERY_RESULT_PARTIAL_BIT is set, VK_QUERY_RESULT_WAIT_BIT is not set, and the query’s status is unavailable, an intermediate result value between zero and the final result value is written to pData for that query.

VK_QUERY_RESULT_PARTIAL_BIT must not be used if the pool’s queryType is VK_QUERY_TYPE_TIMESTAMP.

If VK_QUERY_RESULT_WITH_AVAILABILITY_BIT is set, the final integer value written for each query is non-zero if the query’s status was available or zero if the status was unavailable. When VK_QUERY_RESULT_WITH_AVAILABILITY_BIT is used, implementations must guarantee that if they return a non-zero availability value then the numerical results must be valid, assuming the results are not reset by a subsequent command.

[Note]Note

Satisfying this guarantee may require careful ordering by the application, e.g. to read the availability status before reading the results.

To copy query statuses and numerical results directly to buffer memory, call:

 

void vkCmdCopyQueryPoolResults(
    VkCommandBuffer                             commandBuffer,
    VkQueryPool                                 queryPool,
    uint32_t                                    firstQuery,
    uint32_t                                    queryCount,
    VkBuffer                                    dstBuffer,
    VkDeviceSize                                dstOffset,
    VkDeviceSize                                stride,
    VkQueryResultFlags                          flags);

vkCmdCopyQueryPoolResults is guaranteed to see the effect of previous uses of vkCmdResetQueryPool in the same queue, without any additional synchronization. Thus, the results will always reflect the most recent use of the query.

flags has the same possible values described above for the flags parameter of vkGetQueryPoolResults, but the different style of execution causes some subtle behavioral differences. Because vkCmdCopyQueryPoolResults executes in order with respect to other query commands, there is less ambiguity about which use of a query is being requested.

If no bits are set in flags, results for all requested queries in the available state are written as 32-bit unsigned integer values, and nothing is written for queries in the unavailable state.

If VK_QUERY_RESULT_64_BIT is set, the results are written as an array of 64-bit unsigned integer values as described for vkGetQueryPoolResults.

If VK_QUERY_RESULT_WAIT_BIT is set, the implementation will wait for each query’s status to be in the available state before retrieving the numerical results for that query. This is guaranteed to reflect the most recent use of the query on the same queue, assuming that the query is not being simultaneously used by other queues. If the query does not become available in a finite amount of time (e.g. due to not issuing a query since the last reset), a VK_ERROR_DEVICE_LOST error may occur.

Similarly, if VK_QUERY_RESULT_WITH_AVAILABILITY_BIT is set and VK_QUERY_RESULT_WAIT_BIT is not set, the availability is guaranteed to reflect the most recent use of the query on the same queue, assuming that the query is not being simultaneously used by other queues. As with vkGetQueryPoolResults, implementations must guarantee that if they return a non-zero availability value, then the numerical results are valid.

If VK_QUERY_RESULT_PARTIAL_BIT is set, VK_QUERY_RESULT_WAIT_BIT is not set, and the query’s status is unavailable, an intermediate result value between zero and the final result value is written for that query.

VK_QUERY_RESULT_PARTIAL_BIT must not be used if the pool’s queryType is VK_QUERY_TYPE_TIMESTAMP.

vkCmdCopyQueryPoolResults is considered to be a transfer operation, and its writes to buffer memory must be synchronized using VK_PIPELINE_STAGE_TRANSFER_BIT and VK_ACCESS_TRANSFER_WRITE_BIT before using the results.

Rendering operations such as clears, MSAA resolves, attachment load/store operations, and blits may count towards the results of queries. This behavior is implementation-dependent and may vary depending on the path used within an implementation. For example, some implementations have several types of clears, some of which may include vertices and some not.