[SYCL][Doc] Update sub-group docs #1565
Conversation
Signed-off-by: John Pennycook <john.pennycook@intel.com>
Signed-off-by: John Pennycook <john.pennycook@intel.com>
Signed-off-by: John Pennycook <john.pennycook@intel.com>
Sub-group queries build on SYCL_device_specific_kernel_queries. Signed-off-by: John Pennycook <john.pennycook@intel.com>
|
I hadn't realized that SYCL_device_specific_kernel_queries had been merged in #1540. I've pushed a new commit to align the sub-group queries with the syntax there -- @Felipe-Intel, could you please take a look? |
LGTM! |
|
@AlexeySachkov Ping. |
| |+template <info::kernel_sub_group param>typename info::param_traits<info::kernel_sub_group, param>::return_type get_sub_group_info(const device &dev) const+ | ||
| |Query information from the sub-group from a kernel using the +info::kernel_sub_group+ descriptor for a specific device. | ||
| |+template <info::kernel_device_specific param>typename info::param_traits<info::kernel_device_specific, param>::return_type get_info(const device &dev, typename info::param_traits<info::kernel_device_specific, param>::input_type value) const+ | ||
| |Query information from a kernel using the +info::kernel_device_specific+ descriptor for a specific device and input parameter. The expected value of the input parameter depends on the information being queried. |
There was a problem hiding this comment.
Is it required to be mentioned that the second argument is optional when not needed?
The previous version had both versions of get_sub_group_info APIs mentioned.
https://github.com/intel/llvm/blob/fba2e0602550a86c74149d9875b788ad1117f8d3/sycl/doc/extensions/SubGroupNDRange/SubGroupNDRange.md
There was a problem hiding this comment.
I've added a reference to SYCL_device_specific_kernel_queries and highlighted that which version should be used depends on whether there is an input type specified for the query or not. Thanks!
Signed-off-by: John Pennycook <john.pennycook@intel.com>
|
Hi John, |
|
I had a call with John. He confirmed that my understanding was correct. |
|
@Pennycook Is this PR not merged for some reason. The PR corresponding to this change in SYCL : #1600 is ready to be merged. |
It cannot be merged without a review from one of the code owners for the extensions directory. @jbrodman, can you approve if there are no other concerns? @garimagu and I agree that what's specified here is correct, and we'll open a new issue if future updates are needed. |
Implements the change as per the spec update at https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/SubGroup/SYCL_INTEL_sub_group.asciidoc and update from : intel#1565 ABI update: Functions corresponding to the deleted APIs have been deleted. Edit info.cpp test to expect max sub-group size based on the spec update. Changed pi.h document and changed the plugin to always return a 32 bit value. ABI version update. Signed-off-by: Garima Gupta <garima.gupta@intel.com>
Implements the change as per the spec update at https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/SubGroup/SYCL_INTEL_sub_group.asciidoc and update from : #1565 ABI update: Functions corresponding to the deleted APIs have been deleted. Edit info.cpp test to expect max sub-group size based on the spec update. Changed pi.h document and changed the plugin to always return a 32 bit value. Signed-off-by: Garima Gupta <garima.gupta@intel.com>
Incompatible with memory model introduced by ExtendedAtomics extension. Signed-off-by: John Pennycook <john.pennycook@intel.com>
Reverts some changes made when switching from SubGroupNDRange to SubGroup extension:
max_sub_group_sizemust take a work-group size in order to support OpenCLbarriermember function was removed by mistakeshufflefunctions should be supported in addition to the higher-levelpermuteandshift_*functions from the SubGroupAlgorithms extensionThanks to @AlexeySachkov and @jasonsewall-intel for noticing some of these issues!