libavfilter/x86/vf_gblur: correct the order of loop step
authorWu Jianhua <jianhua.wu@intel.com>
Thu, 16 Sep 2021 07:34:08 +0000 (15:34 +0800)
committerPaul B Mahol <onemda@gmail.com>
Sat, 18 Sep 2021 10:38:01 +0000 (12:38 +0200)
commit7bbad32d5ab69cb52bc92a5ec30c7b9838daa08a
tree9260a90a0962514b7a456ff929aaeeb6be709733
parentfcf10c925dc4a4ab883b4ad14ea61bbdaa6f8521
libavfilter/x86/vf_gblur: correct the order of loop step

The problem was caused by if the width of the processed block
minus 1 is a multiple of the aligned number the instruction
jle .bscale_scalar would skip the Optimized Loop Step, which
will lead to an incorrect sampling when specifying steps more
than 1. Move the Optimized Loop Step after .bscale_scalar to
ensure the loop step is enabled.

Signed-off-by: Wu Jianhua <jianhua.wu@intel.com>
libavfilter/x86/vf_gblur.asm