added detection & dispatching of some modern NEON instructions (NEON_FP16, NEON_BF16)#24420
Merged
vpisarev merged 6 commits intoopencv:4.xfrom Oct 18, 2023
Merged
added detection & dispatching of some modern NEON instructions (NEON_FP16, NEON_BF16)#24420vpisarev merged 6 commits intoopencv:4.xfrom
vpisarev merged 6 commits intoopencv:4.xfrom
Conversation
… method to detect various NEON extensions, such as FP16 SIMD arithmetics, BF16 SIMD arithmetics, SIMD dotprod etc. It could be propagated to other instruction sets if necessary.
Contributor
|
Looks like the mentioned features are supported in |
Contributor
Link: https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/signal?view=msvc-170 |
Contributor
There is a header of actual auxv flags / values: https://elixir.bootlin.com/linux/latest/source/arch/arm64/include/uapi/asm/hwcap.h#L106 Also some libraries parse |
| ocv_update(CPU_FP16_IMPLIES "NEON") | ||
| else() | ||
| ocv_update(CPU_KNOWN_OPTIMIZATIONS "NEON;FP16;NEON_DOTPROD") | ||
| ocv_update(CPU_KNOWN_OPTIMIZATIONS "NEON;FP16;NEON_DOTPROD;FP16_SIMD;BF16_SIMD") |
Contributor
There was a problem hiding this comment.
FP16_SIMD;BF16_SIMD
How does scope of these instructions correlate with other platforms?
... and available universal intrinsics?
Until it is unclear, it is better to add ARM_ prefix.
…ions without signal() * renamed FP16_SIMD & BF16_SIMD to NEON_FP16 and NEON_BF16, respectively
Contributor
Author
|
@asmorkalov, @opencv-alalek, thank you for the comments and for the links! all your concerns have been addressed :) |
Merged
tomoaki0705
reviewed
Dec 6, 2023
| ocv_update(CPU_NEON_DOTPROD_IMPLIES "NEON") | ||
| ocv_update(CPU_NEON_FP16_FLAGS_ON "-march=armv8.2-a+fp16") | ||
| ocv_update(CPU_NEON_FP16_IMPLIES "NEON") | ||
| ocv_update(CPU_NEON_BF16_FLAGS_ON "-march=armv8.2-a+fp16+bf16") |
Contributor
IskXCr
pushed a commit
to Haosonn/opencv
that referenced
this pull request
Dec 20, 2023
…FP16, NEON_BF16) (opencv#24420) * added more or less cross-platform (based on POSIX signal() semantics) method to detect various NEON extensions, such as FP16 SIMD arithmetics, BF16 SIMD arithmetics, SIMD dotprod etc. It could be propagated to other instruction sets if necessary. * hopefully fixed compile errors * continue to fix CI * another attempt to fix build on Linux aarch64 * * reverted to the original method to detect special arm neon instructions without signal() * renamed FP16_SIMD & BF16_SIMD to NEON_FP16 and NEON_BF16, respectively * removed extra whitespaces
thewoz
pushed a commit
to thewoz/opencv
that referenced
this pull request
Jan 4, 2024
…FP16, NEON_BF16) (opencv#24420) * added more or less cross-platform (based on POSIX signal() semantics) method to detect various NEON extensions, such as FP16 SIMD arithmetics, BF16 SIMD arithmetics, SIMD dotprod etc. It could be propagated to other instruction sets if necessary. * hopefully fixed compile errors * continue to fix CI * another attempt to fix build on Linux aarch64 * * reverted to the original method to detect special arm neon instructions without signal() * renamed FP16_SIMD & BF16_SIMD to NEON_FP16 and NEON_BF16, respectively * removed extra whitespaces
thewoz
pushed a commit
to thewoz/opencv
that referenced
this pull request
May 29, 2024
…FP16, NEON_BF16) (opencv#24420) * added more or less cross-platform (based on POSIX signal() semantics) method to detect various NEON extensions, such as FP16 SIMD arithmetics, BF16 SIMD arithmetics, SIMD dotprod etc. It could be propagated to other instruction sets if necessary. * hopefully fixed compile errors * continue to fix CI * another attempt to fix build on Linux aarch64 * * reverted to the original method to detect special arm neon instructions without signal() * renamed FP16_SIMD & BF16_SIMD to NEON_FP16 and NEON_BF16, respectively * removed extra whitespaces
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Currently, platform-specific (NEON) code is required to make use of those instructions. Later on, maybe universal intrinsics for FP16 and BF16 arithmetics will be added. Note that even modern ARM platforms don't have full set of BF16 operations. This is mostly instructions to implement BF16xBF16 to FP32 matrix multiplication.
Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
Patch to opencv_extra has the same branch name.