After the initial generation, when a frame is requested,
the test pattern generator rotates the image left by 1 column.
The current approach has two shortcomings:
(1) it allocates a temporary buffer to hold one column;
(2) it swaps two columns at a time.
The test patterns are simple ARGB images, in row-major order,
so doing (2) works against memory prefetching. This can be
addressed by doing the rotation one row at a time as that way
the image is addressed in a purely linear fashion. Doing so also
eliminates the need for a dynamically allocated temporary buffer,
as the required buffer now only needs to hold one sample,
which is 4 bytes in this case.
In an optimized build, this results in about a 2x increase in the
number of frames per second as reported by `cam`. In an unoptimized,
ASAN and UBSAN intrumented build, the difference is even bigger,
which is useful for running lc-compliance in CI in a reasonable time.
Signed-off-by: Barnabás Pőcze <pobrn@protonmail.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Barnabás Pőcze
4f9d8a6301