Adding support for vectorized masking for uchar/ushort.#15527
Merged
alalek merged 5 commits intoopencv:3.4from Oct 11, 2019
Merged
Adding support for vectorized masking for uchar/ushort.#15527alalek merged 5 commits intoopencv:3.4from
alalek merged 5 commits intoopencv:3.4from
Conversation
Contributor
Author
|
Just found a bug here, I was under the impression that the mask should return 0 but masking should just keep the value os dst as it is, also need to make some further improvements. |
the mask and tweaked for further performance improvements.
Contributor
Author
|
Fixed the problem, got event further performance improvements up to almost 4x |
terfendail
reviewed
Sep 27, 2019
modules/imgproc/src/accum.simd.hpp
Outdated
| int size = len * cn; | ||
| int y = -cVectorWidth; | ||
| int cc = 0; | ||
| for (; x <= size - cVectorWidth; x += cVectorWidth, cc = (cc+1) % cn) |
Contributor
There was a problem hiding this comment.
cc will always be equal to 0 since cn is 1 in this branch
terfendail
reviewed
Sep 27, 2019
modules/imgproc/src/accum.simd.hpp
Outdated
| v_store(dst + x + step * 3, v_dst11); | ||
| } | ||
| } else if ( cn == 1 ){ | ||
| //#include <stdio.h> |
terfendail
reviewed
Sep 27, 2019
modules/imgproc/src/accum.simd.hpp
Outdated
| v_float32 v_dst10 = vx_load(dst + x + step * 2); | ||
| v_float32 v_dst11 = vx_load(dst + x + step * 3); | ||
|
|
||
| v_dst00 = v_fma(v_fma(v_dst00, v_beta, v_cvt_f32(v_reinterpret_as_s32(v_src00)) * v_alpha), d0, (~d0)*v_mf00); |
Contributor
There was a problem hiding this comment.
Looks like for zero mask this line will produce result equal to (~d0)*v_mf00 that is definitely unrelated to expected initial dst value
Contributor
Author
There was a problem hiding this comment.
Yeah I'm fixing that in a bit, thanks for the heads up.
Contributor
There was a problem hiding this comment.
It's possible to use v_select(d0, v_fma(...), v_dst00) intrinsic to choose between updated and original destination values
Contributor
|
[355/1819]
Performance for SSE2 baseline
[249/1819]
Performance for SSE3 baseline
Performance for SSE4_2 baseline
Performance for AVX2 baseline
|
terfendail
approved these changes
Oct 8, 2019
Merged
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
3 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 whenever a mask is used at the accumulate function opencv reverts to the generic non-vectorized implementation. This PR adds masking support for uchar and ushort's implementation. Performance improvement for ppc64le are up to 89%.