[SYCL] Remove redundant build options processing #3342
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
After 86b0e8d patch extra operations with device images happen before check if it is present in in-memory cache. For applications with small kernels which are executed multiple times noticeable performance degradation is seen for host code. That was done to get build options stored in the kernel image and use them as in-memory cache key. At the same time kernel image (where these options are taken from) is used in cache key so it is reasonable to use only build options which are specified in SYCL API and/or environment variables as separate cache key. Getting build options from kernel image is moved back to build operation which happens only if built program is missed in in-memory cache.
|
/summary:run |
|
|
As discussed offline TODO: mark was added to ensure that the mechanism will work once added posibillity to access several versions of the same kernel built with different options. |
|
@vladimirlaz, please address this one too:
|
The problem was caught by benchmarks and they can be used as regression test. There is no benchmark-style test in LIT infra. |
|
@vladimirlaz, can a non-benchmark test be created, which enables PI tracing and checks that recompilation and/or reusing from cache happens expected number of times? This should belong to llvm-test-suite though, as it is end-to-end. So not a blocker for this PR. |
kbobrovs
approved these changes
Mar 12, 2021
alexbatashev
pushed a commit
to alexbatashev/llvm
that referenced
this pull request
Mar 15, 2021
* upstream/sycl: Add ITT stubs and wrappers for SPIR-V devices (intel#3279) [SYCL] Add zero argument version of buffer::reinterpret() for SYCL 2020 (intel#3333) [SYCL] Restore old behavior of get() method (intel#3356) [Driver][SYCL][FPGA] Improve FPGA AOT when using Triple (intel#3330) [SYCL] Revert support for pinned_host_memory extension in Level-Zero backend. Make it a NOP (intel#3349) [SYCL] Remove redundant build options processing (intel#3342) [SYCL][NFC] Factor out GenXIntrinsics git repo tag. (intel#3351) [SYCL] Add coarse-grained debug aid for finding imbalance in Level-Zero alloc/free (intel#3334) [SYCL] Sub-group load/store for raw pointers (intel#3255) [SYCL] Fix more unused variables warnings (intel#3352) [BuildBot] Uplift CPU/FPGAEMU RT version for CI Process (intel#3338)
alexbatashev
pushed a commit
to alexbatashev/llvm
that referenced
this pull request
Mar 19, 2021
* upstream/sycl: (1804 commits) [SYCL] SYCL 2020 backend interoperability part 1 (intel#3354) [Driver][SYCL] Address issue when unbundling for non-FPGA archive (intel#3366) [SYCL][Doc] Update compiler options descriptions (intel#3340) [SYCL] Update ABI dump tool to disable checks with libcxx by default (intel#3370) [SYCL] Update the way we handle -sycl-std= based on community review feedback (intel#3371) [SYCL] Move tests to llvm-test-suite (intel#3359) [SYCL][PI][L0] Revert copy batching from intel#3232 (intel#3363) [SYCL] Remove unsupported option. [SYCL] Fix pragma setting in sycl-post-link (intel#3358) [SYCL] Add ITT annotation instructions (intel#3299) [SYCL] Propagate attributes of original kernel to wrapper kernel generated for range-rounding (intel#3306) [BuildBot] Uplift GPU RT version for Linux to 21.09.19150 (intel#3316) [SYCL] Retain PI events until they have signaled (intel#3350) [SYCL] Add caching when using interop constructor (intel#3327) Add ITT stubs and wrappers for SPIR-V devices (intel#3279) [SYCL] Add zero argument version of buffer::reinterpret() for SYCL 2020 (intel#3333) [SYCL] Restore old behavior of get() method (intel#3356) [Driver][SYCL][FPGA] Improve FPGA AOT when using Triple (intel#3330) [SYCL] Revert support for pinned_host_memory extension in Level-Zero backend. Make it a NOP (intel#3349) [SYCL] Remove redundant build options processing (intel#3342) ...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
After 86b0e8d patch extra operations with device images happen before
check if it is present in in-memory cache. For applications with small
kernels which are executed multiple times noticeable performance
degradation is seen for host code. That was done to get build options
stored in the kernel image and use them as in-memory cache key. At the
same time kernel image (where these options are taken from) is used in
cache key so it is reasonable to use only build options which are
specified in SYCL API and/or environment variables as separate cache
key.
Getting build options from kernel image is moved back to build
operation which happens only if built program is missed in in-memory
cache.