You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It is possible to do SSE2 optimizations of the single-pass horizontal pass of IDWT 5x3. This has been partially implemented per 288f472 in the https://github.com/uclouvain/openjpeg/tree/opj_idwt53_h_cas0_SSE2 branch, during the investigations done in #957 but not committed yet .
The case here is restricted to cas == 0 (ie the origin of the array is on even coordinates) and where the length is a multiple of 8. This should be generalized to a length not a multiple of 8 (termination to modify), and to the cas == 1 (odd coordinates) as well. But the performance improvement in the situation implemented were rather small, so given the additional code complexity, not merged for now.
The text was updated successfully, but these errors were encountered:
rouault
changed the title
SSE2 optimization for IDWT 6x
SSE2 optimization for horizontal pass of IDWT 5x3
Jun 26, 2017
It is possible to do SSE2 optimizations of the single-pass horizontal pass of IDWT 5x3. This has been partially implemented per 288f472 in the https://github.com/uclouvain/openjpeg/tree/opj_idwt53_h_cas0_SSE2 branch, during the investigations done in #957 but not committed yet .
The case here is restricted to cas == 0 (ie the origin of the array is on even coordinates) and where the length is a multiple of 8. This should be generalized to a length not a multiple of 8 (termination to modify), and to the cas == 1 (odd coordinates) as well. But the performance improvement in the situation implemented were rather small, so given the additional code complexity, not merged for now.
The text was updated successfully, but these errors were encountered: