Add ARM NEON intrinsics to unpack_yuy2 #13270
Conversation
|
Can one of the admins verify this patch? |
fateshelled
changed the title
|
I explained too little, my apologies. |
|
Hi @fateshelled , |
|
Hi @Nir-Az , |
|
It will take us some time to review and validate this PR on dedicated HW. |
fateshelled
force-pushed
the
neon-yuy2-support
branch
from
0e54f81
to
a13850d
Compare
|
Thanks for the reply. |
| { | ||
| // Load 16 pixels | ||
| const uint8x8x4_t yuyv = vld4_u8(reinterpret_cast<const uint8_t *>(&src[i * 2])); | ||
| // yuyv.val[0] = y0, yuyv.val[1] = u, yuyv.val[2] = y1, yuyv.val[3] = v |
There was a problem hiding this comment.
It wasn't necessary, so I fixed it.
|
|
||
| void unpack_yuy2_neon_y8(uint8_t * const d[], const uint8_t * s, int n) | ||
| { | ||
| unpack_yuy2_neon<RS2_FORMAT_Y8>(d, s, n); |
There was a problem hiding this comment.
Why not calling the templated function at the first place instead of adding functions that call it?
There was a problem hiding this comment.
I used the code image-avx.cpp as a reference, but is this wrong?
librealsense/src/image-avx.cpp
Line 253 in e1688cc
There was a problem hiding this comment.
Not wrong.. just redundant.|
But since our code is a reference let's keep as is
|
@fateshelled appreciate your contribution, we are always happy to integrate community pull requests :) Thank you :) |
|
Thank you for merging the PR. |
Changes:
unpack_yuy2_neonfunctionyuy2toy8,y16,rgb8,rgba8,bgr8andbgra8format.