[bug #68049] imread() is slow

Please use the bug tracker to post updates to a bug report.  The mailing list is intended as a read-only notification stream.  Info posted to this mailing list address won't appear in the tracker database where it is most useful.

Follow-up Comment #1, bug #68049 (group octave):

With the attached CSET
(file bug68049_magic_read_fast_20260326_100.cset)
I see a factor ~3.5x speedup:

octave:2>  tic; a = imread("3.jpg"); toc
Elapsed time is 0.155641 seconds.
octave:3> 0.55/0.15
ans = 3.6667


CSET summary:

Optimize __magick_read__.cc (bug #68049).

The original code has performance bottleneck primarily caused by inefficient
memory access patterns in pixel loops iterated col-outer/row-inner. Swap to
row-outer/col-inner so reads from the PixelPacket array are sequential.
Also replace per-pixel floating-point division and octave_int(double)
construction with a lookup table for integral output types.

* magick_read.cc (image_region): Store m_row_inc and m_col_inc as members.
Add row_inc() and col_inc() accessors.

* magick_read.cc (read_indexed_images): Swap loop order to
row-outer/col-inner
with strided writes (dest_idx += nRows). Use col_inc for column stepping
instead of col_shift for backward stepping. Reset img_fvec pointer before
alpha computation since the frame loop now advances it.

* magick_read.cc (read_images): Build a lookup table (LUT) mapping each
Quantum value (0..MaxRGB) to the output pixel type when the output is
integral, MaxRGB <= 65535, and the image has more pixels than LUT entries.
Cast through the primitive integer type (P::val_type) during LUT fill to
bypass the octave_int(double) constructor. Swap all six image-type cases
(Grayscale, GrayscaleMatte, RGB, RGBA, CMYK, CMYKA) from col-outer/row-inner
to row-outer/col-inner with strided writes. Add if constexpr dispatch:
integral types use the LUT path or reciprocal multiplication with
static_cast<prim_type>(lrint(...)); float types use division to preserve
rounding semantics; bool uses division to avoid GCC warning.


Dmitri.
-- 


(file #58407)

    _______________________________________________________

Additional Item Attachment:

Name: bug68049_magic_read_fast_20260326_100.cset Size: 29KiB

<https://file.savannah.gnu.org/file/bug68049_magic_read_fast_20260326_100.cset?file_id=58407>


    AGPL NOTICE

These attachments are served by Savane. You can download the corresponding
source code of Savane at
https://savannah.gnu.org/source/savane-f290f6b25beb8cb99bbe243a6cd2c5fef79ffcde.tar.gz


    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?68049>

_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/

-----BEGIN PGP SIGNATURE-----

iHUEABYIAB0WIQQk97aszIMMAvLLwm6qLAuaBUf3TgUCacXbWgAKCRCqLAuaBUf3
TuDnAP9ScyTO+OSRtcgMMqy4BVUErn0xtPpfKNmGox9/UAvKigEAlpOlvOkPS4Ji
LVbMypiWAwzQlVB7Y+Yo/6X5qOHsggc=
=RD3T
-----END PGP SIGNATURE-----