Vulkan-Docs/chapters/dispatch.txt

380 lines
18 KiB
Plaintext

// Copyright (c) 2015-2018 Khronos Group. This work is licensed under a
// Creative Commons Attribution 4.0 International License; see
// http://creativecommons.org/licenses/by/4.0/
[[dispatch]]
= Dispatching Commands
_Dispatching commands_ (commands with ftext:Dispatch in the name) provoke
work in a compute pipeline.
Dispatching commands are recorded into a command buffer and when executed by
a queue, will produce work which executes according to the bound compute
pipeline.
A compute pipeline must: be bound to a command buffer before any dispatch
commands are recorded in that command buffer.
[open,refpage='vkCmdDispatch',desc='Dispatch compute work items',type='protos']
--
To record a dispatch, call:
include::../api/protos/vkCmdDispatch.txt[]
* pname:commandBuffer is the command buffer into which the command will be
recorded.
* pname:groupCountX is the number of local workgroups to dispatch in the X
dimension.
* pname:groupCountY is the number of local workgroups to dispatch in the Y
dimension.
* pname:groupCountZ is the number of local workgroups to dispatch in the Z
dimension.
When the command is executed, a global workgroup consisting of
[eq]#groupCountX {times} groupCountY {times} groupCountZ# local workgroups
is assembled.
.Valid Usage
****
* [[VUID-vkCmdDispatch-groupCountX-00386]]
pname:groupCountX must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[0]
* [[VUID-vkCmdDispatch-groupCountY-00387]]
pname:groupCountY must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[1]
* [[VUID-vkCmdDispatch-groupCountZ-00388]]
pname:groupCountZ must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[2]
* [[VUID-vkCmdDispatch-None-00389]]
For each set _n_ that is statically used by the sname:VkPipeline bound
to ename:VK_PIPELINE_BIND_POINT_COMPUTE, a descriptor set must: have
been bound to _n_ at ename:VK_PIPELINE_BIND_POINT_COMPUTE, with a
sname:VkPipelineLayout that is compatible for set _n_, with the
sname:VkPipelineLayout used to create the current sname:VkPipeline, as
described in <<descriptorsets-compatibility>>
* [[VUID-vkCmdDispatch-None-00390]]
Descriptors in each bound descriptor set, specified via
fname:vkCmdBindDescriptorSets, must: be valid if they are statically
used by the bound sname:VkPipeline object, specified via
fname:vkCmdBindPipeline
* [[VUID-vkCmdDispatch-None-00391]]
A valid compute pipeline must: be bound to the current command buffer
with ename:VK_PIPELINE_BIND_POINT_COMPUTE
* [[VUID-vkCmdDispatch-None-00392]]
For each push constant that is statically used by the sname:VkPipeline
bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE, a push constant value
must: have been set for ename:VK_PIPELINE_BIND_POINT_COMPUTE, with a
sname:VkPipelineLayout that is compatible for push constants with the
one used to create the current sname:VkPipeline, as described in
<<descriptorsets-compatibility>>
* [[VUID-vkCmdDispatch-None-00393]]
If any sname:VkSampler object that is accessed from a shader by the
sname:VkPipeline bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE uses
unnormalized coordinates, it must: not be used to sample from any
sname:VkImage with a sname:VkImageView of the type
ename:VK_IMAGE_VIEW_TYPE_3D, ename:VK_IMAGE_VIEW_TYPE_CUBE,
ename:VK_IMAGE_VIEW_TYPE_1D_ARRAY, ename:VK_IMAGE_VIEW_TYPE_2D_ARRAY or
ename:VK_IMAGE_VIEW_TYPE_CUBE_ARRAY, in any shader stage
* [[VUID-vkCmdDispatch-None-00394]]
If any sname:VkSampler object that is accessed from a shader by the
sname:VkPipeline bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE uses
unnormalized coordinates, it must: not be used with any of the SPIR-V
`OpImageSample*` or `OpImageSparseSample*` instructions with
code:ImplicitLod, code:Dref or code:Proj in their name, in any shader
stage
* [[VUID-vkCmdDispatch-None-00395]]
If any sname:VkSampler object that is accessed from a shader by the
sname:VkPipeline bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE uses
unnormalized coordinates, it must: not be used with any of the SPIR-V
`OpImageSample*` or `OpImageSparseSample*` instructions that includes a
LOD bias or any offset values, in any shader stage
* [[VUID-vkCmdDispatch-None-00396]]
If the <<features-features-robustBufferAccess,robust buffer access>>
feature is not enabled, and any shader stage in the sname:VkPipeline
object bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE accesses a uniform
buffer, it must: not access values outside of the range of that buffer
specified in the bound descriptor set
* [[VUID-vkCmdDispatch-None-00397]]
If the <<features-features-robustBufferAccess,robust buffer access>>
feature is not enabled, and any shader stage in the sname:VkPipeline
object bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE accesses a storage
buffer, it must: not access values outside of the range of that buffer
specified in the bound descriptor set
* [[VUID-vkCmdDispatch-None-02005]]
If a sname:VkImageView is sampled with with ename:VK_FILTER_LINEAR as a
result of this command, then the image view's
<<resources-image-view-format-features,format features>> must: contain
ename:VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT.
ifdef::VK_IMG_filter_cubic[]
* [[VUID-vkCmdDispatch-None-02006]]
If a sname:VkImageView is sampled with with ename:VK_FILTER_CUBIC_IMG as
a result of this command, then the image view's
<<resources-image-view-format-features,format features>> must: contain
ename:VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_CUBIC_BIT_IMG.
* [[VUID-vkCmdDispatch-None-00400]]
Any slink:VkImageView being sampled with ename:VK_FILTER_CUBIC_IMG as a
result of this command must: not have a elink:VkImageViewType of
ename:VK_IMAGE_VIEW_TYPE_3D, ename:VK_IMAGE_VIEW_TYPE_CUBE, or
ename:VK_IMAGE_VIEW_TYPE_CUBE_ARRAY
endif::VK_IMG_filter_cubic[]
ifdef::VK_VERSION_1_1[]
* [[VUID-vkCmdDispatch-commandBuffer-01844]]
If pname:commandBuffer is an unprotected command buffer, and any
pipeline stage in the sname:VkPipeline object bound to
ename:VK_PIPELINE_BIND_POINT_COMPUTE reads from or writes to any image
or buffer, that image or buffer must: not be a protected image or
protected buffer.
* [[VUID-vkCmdDispatch-commandBuffer-01845]]
If pname:commandBuffer is a protected command buffer, and any pipeline
stage in the sname:VkPipeline object bound to
ename:VK_PIPELINE_POINT_COMPUTE writes to any image or buffer, that
image or buffer must: not be an unprotected image or unprotected buffer.
* [[VUID-vkCmdDispatch-commandBuffer-01846]]
If pname:commandBuffer is a protected command buffer, and any pipeline
stage other than the compute pipeline stage in the sname:VkPipeline
object bound to ename:VK_PIPELINE_POINT_COMPUTE reads from any image or
buffer, the image or buffer must: not be a protected image or protected
buffer.
endif::VK_VERSION_1_1[]
****
include::../validity/protos/vkCmdDispatch.txt[]
--
[open,refpage='vkCmdDispatchIndirect',desc='Dispatch compute work items using indirect parameters',type='protos']
--
To record an indirect command dispatch, call:
include::../api/protos/vkCmdDispatchIndirect.txt[]
* pname:commandBuffer is the command buffer into which the command will be
recorded.
* pname:buffer is the buffer containing dispatch parameters.
* pname:offset is the byte offset into pname:buffer where parameters
begin.
fname:vkCmdDispatchIndirect behaves similarly to flink:vkCmdDispatch except
that the parameters are read by the device from a buffer during execution.
The parameters of the dispatch are encoded in a
slink:VkDispatchIndirectCommand structure taken from pname:buffer starting
at pname:offset.
.Valid Usage
****
* [[VUID-vkCmdDispatchIndirect-buffer-00401]]
If pname:buffer is non-sparse then it must: be bound completely and
contiguously to a single sname:VkDeviceMemory object
* [[VUID-vkCmdDispatchIndirect-None-00402]]
For each set _n_ that is statically used by the sname:VkPipeline bound
to ename:VK_PIPELINE_BIND_POINT_COMPUTE, a descriptor set must: have
been bound to _n_ at ename:VK_PIPELINE_BIND_POINT_COMPUTE, with a
sname:VkPipelineLayout that is compatible for set _n_, with the
sname:VkPipelineLayout used to create the current sname:VkPipeline, as
described in <<descriptorsets-compatibility>>
* [[VUID-vkCmdDispatchIndirect-None-00403]]
Descriptors in each bound descriptor set, specified via
fname:vkCmdBindDescriptorSets, must: be valid if they are statically
used by the bound sname:VkPipeline object, specified via
fname:vkCmdBindPipeline
* [[VUID-vkCmdDispatchIndirect-None-00404]]
A valid compute pipeline must: be bound to the current command buffer
with ename:VK_PIPELINE_BIND_POINT_COMPUTE
* [[VUID-vkCmdDispatchIndirect-buffer-00405]]
pname:buffer must: have been created with the
ename:VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT bit set
* [[VUID-vkCmdDispatchIndirect-offset-00406]]
pname:offset must: be a multiple of `4`
* [[VUID-vkCmdDispatchIndirect-offset-00407]]
The sum of pname:offset and the size of sname:VkDispatchIndirectCommand
must: be less than or equal to the size of pname:buffer
* [[VUID-vkCmdDispatchIndirect-None-00408]]
For each push constant that is statically used by the sname:VkPipeline
bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE, a push constant value
must: have been set for ename:VK_PIPELINE_BIND_POINT_COMPUTE, with a
sname:VkPipelineLayout that is compatible for push constants with the
one used to create the current sname:VkPipeline, as described in
<<descriptorsets-compatibility>>
* [[VUID-vkCmdDispatchIndirect-None-00409]]
If any sname:VkSampler object that is accessed from a shader by the
sname:VkPipeline bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE uses
unnormalized coordinates, it must: not be used to sample from any
sname:VkImage with a sname:VkImageView of the type
ename:VK_IMAGE_VIEW_TYPE_3D, ename:VK_IMAGE_VIEW_TYPE_CUBE,
ename:VK_IMAGE_VIEW_TYPE_1D_ARRAY, ename:VK_IMAGE_VIEW_TYPE_2D_ARRAY or
ename:VK_IMAGE_VIEW_TYPE_CUBE_ARRAY, in any shader stage
* [[VUID-vkCmdDispatchIndirect-None-00410]]
If any sname:VkSampler object that is accessed from a shader by the
sname:VkPipeline bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE uses
unnormalized coordinates, it must: not be used with any of the SPIR-V
`OpImageSample*` or `OpImageSparseSample*` instructions with
code:ImplicitLod, code:Dref or code:Proj in their name, in any shader
stage
* [[VUID-vkCmdDispatchIndirect-None-00411]]
If any sname:VkSampler object that is accessed from a shader by the
sname:VkPipeline bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE uses
unnormalized coordinates, it must: not be used with any of the SPIR-V
`OpImageSample*` or `OpImageSparseSample*` instructions that includes a
LOD bias or any offset values, in any shader stage
* [[VUID-vkCmdDispatchIndirect-None-00412]]
If the <<features-features-robustBufferAccess,robust buffer access>>
feature is not enabled, and any shader stage in the sname:VkPipeline
object bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE accesses a uniform
buffer, it must: not access values outside of the range of that buffer
specified in the bound descriptor set
* [[VUID-vkCmdDispatchIndirect-None-00413]]
If the <<features-features-robustBufferAccess,robust buffer access>>
feature is not enabled, and any shader stage in the sname:VkPipeline
object bound to ename:VK_PIPELINE_BIND_POINT_COMPUTE accesses a storage
buffer, it must: not access values outside of the range of that buffer
specified in the bound descriptor set
* [[VUID-vkCmdDispatchIndirect-None-02007]]
If a sname:VkImageView is sampled with with ename:VK_FILTER_LINEAR as a
result of this command, then the image view's
<<resources-image-view-format-features,format features>> must: contain
ename:VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT.
ifdef::VK_IMG_filter_cubic[]
* [[VUID-vkCmdDispatchIndirect-None-02008]]
If a sname:VkImageView is sampled with with ename:VK_FILTER_CUBIC_IMG as
a result of this command, then the image view's
<<resources-image-view-format-features,format features>> must: contain
ename:VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_CUBIC_BIT_IMG.
* [[VUID-vkCmdDispatchIndirect-None-00416]]
Any slink:VkImageView being sampled with ename:VK_FILTER_CUBIC_IMG as a
result of this command must: not have a elink:VkImageViewType of
ename:VK_IMAGE_VIEW_TYPE_3D, ename:VK_IMAGE_VIEW_TYPE_CUBE, or
ename:VK_IMAGE_VIEW_TYPE_CUBE_ARRAY
endif::VK_IMG_filter_cubic[]
ifdef::VK_VERSION_1_1[]
* [[VUID-vkCmdDispatchIndirect-commandBuffer-01847]]
If pname:commandBuffer is an unprotected command buffer, and any
pipeline stage in the sname:VkPipeline object bound to
ename:VK_PIPELINE_BIND_POINT_COMPUTE reads from or writes to any image
or buffer, that image or buffer must: not be a protected image or
protected buffer.
* [[VUID-vkCmdDispatchIndirect-commandBuffer-01848]]
If pname:commandBuffer is a protected command buffer, and any pipeline
stage in the sname:VkPipeline object bound to
ename:VK_PIPELINE_POINT_COMPUTE writes to any image or buffer, that
image or buffer must: not be an unprotected image or unprotected buffer.
* [[VUID-vkCmdDispatchIndirect-commandBuffer-01849]]
If pname:commandBuffer is a protected command buffer, and any pipeline
stage other than the compute pipeline stage in the sname:VkPipeline
object bound to ename:VK_PIPELINE_POINT_COMPUTE reads from any image or
buffer, the image or buffer must: not be a protected image or protected
buffer.
endif::VK_VERSION_1_1[]
****
include::../validity/protos/vkCmdDispatchIndirect.txt[]
--
[open,refpage='VkDispatchIndirectCommand',desc='Structure specifying a dispatch indirect command',type='structs',xrefs='vkCmdDispatchIndirect']
--
The sname:VkDispatchIndirectCommand structure is defined as:
include::../api/structs/VkDispatchIndirectCommand.txt[]
* pname:x is the number of local workgroups to dispatch in the X
dimension.
* pname:y is the number of local workgroups to dispatch in the Y
dimension.
* pname:z is the number of local workgroups to dispatch in the Z
dimension.
The members of sname:VkDispatchIndirectCommand have the same meaning as the
corresponding parameters of flink:vkCmdDispatch.
.Valid Usage
****
* [[VUID-VkDispatchIndirectCommand-x-00417]]
pname:x must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[0]
* [[VUID-VkDispatchIndirectCommand-y-00418]]
pname:y must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[1]
* [[VUID-VkDispatchIndirectCommand-z-00419]]
pname:z must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[2]
****
include::../validity/structs/VkDispatchIndirectCommand.txt[]
--
ifdef::VK_VERSION_1_1,VK_KHR_device_group[]
[open,refpage='vkCmdDispatchBase',desc='Dispatch compute work items',type='protos']
--
To record a dispatch using non-zero base values for the components of
code:WorkgroupId, call:
ifdef::VK_VERSION_1_1[]
include::../api/protos/vkCmdDispatchBase.txt[]
endif::VK_VERSION_1_1[]
ifdef::VK_VERSION_1_1+VK_KHR_device_group[or the equivalent command]
ifdef::VK_KHR_device_group[]
include::../api/protos/vkCmdDispatchBaseKHR.txt[]
endif::VK_KHR_device_group[]
* pname:commandBuffer is the command buffer into which the command will be
recorded.
* pname:baseGroupX is the start value for the X component of
code:WorkgroupId.
* pname:baseGroupY is the start value for the Y component of
code:WorkgroupId.
* pname:baseGroupZ is the start value for the Z component of
code:WorkgroupId.
* pname:groupCountX is the number of local workgroups to dispatch in the X
dimension.
* pname:groupCountY is the number of local workgroups to dispatch in the Y
dimension.
* pname:groupCountZ is the number of local workgroups to dispatch in the Z
dimension.
When the command is executed, a global workgroup consisting of
[eq]#groupCountX {times} groupCountY {times} groupCountZ# local workgroups
is assembled, with code:WorkgroupId values ranging from [eq]#[baseGroup,
baseGroup {plus} groupCount)# in each component.
flink:vkCmdDispatch is equivalent to
`vkCmdDispatchBase(0,0,0,groupCountX,groupCountY,groupCountZ)`.
.Valid Usage
****
* [[VUID-vkCmdDispatchBase-None-00420]]
All valid usage rules from flink:vkCmdDispatch apply
* [[VUID-vkCmdDispatchBase-baseGroupX-00421]]
pname:baseGroupX must: be less than
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[0]
* [[VUID-vkCmdDispatchBase-baseGroupX-00422]]
pname:baseGroupX must: be less than
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[1]
* [[VUID-vkCmdDispatchBase-baseGroupZ-00423]]
pname:baseGroupZ must: be less than
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[2]
* [[VUID-vkCmdDispatchBase-groupCountX-00424]]
pname:groupCountX must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[0] minus
pname:baseGroupX
* [[VUID-vkCmdDispatchBase-groupCountY-00425]]
pname:groupCountY must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[1] minus
pname:baseGroupY
* [[VUID-vkCmdDispatchBase-groupCountZ-00426]]
pname:groupCountZ must: be less than or equal to
sname:VkPhysicalDeviceLimits::pname:maxComputeWorkGroupCount[2] minus
pname:baseGroupZ
* [[VUID-vkCmdDispatchBase-baseGroupX-00427]]
If any of pname:baseGroupX, pname:baseGroupY, or pname:baseGroupZ are
not zero, then the bound compute pipeline must: have been created with
the ename:VK_PIPELINE_CREATE_DISPATCH_BASE flag.
****
include::../validity/protos/vkCmdDispatchBase.txt[]
--
endif::VK_VERSION_1_1,VK_KHR_device_group[]