Skype call

1. resovled the segmetation faults in the p-cblk()
2. the col_grp() function, use the Intel vector instructions
3. use ASM directive to insert the assebly code in the c source code
4. operating 64 bits (Quard word) instead of 32 bits at a time
5. compilation: #gcc -O2 -o testcopy testcopy.c vectorcopy.s
6. loop unrolling, instead of MOV, ADD(4), we do 4 MOV, and ADD(16)
7. gcc could do loop unroolling with certain argument when compile, look up that
8. Tasks
9. - split_col_grp() mainly data copying, try to enhance
10. - try gcc unrolling on jasper, and profile/record the execution time
11. - try vector instructions (ASM directive)
12. - publis a paper by the end of year

Research Progress

1. Fixed the "segmentation faults" when executing the jasper with more than 2 thread. (error was caused by the incorrect workload distributing index incrementation in 2 nested loops within the enc_cblks() routine.
2. The fused loop also give some indication of performance enhancement
3. try to fuse more loops if possible
   

Skype Call today

1. clarification on the report
2. run the pc+pd 4 for multiple time, see if the result is always bad
   
3. run the pc+pd 2 for multiple time, see if the result is always good
4. quote or paraphrase the literature from other source
5. segmentation fault related to the image quality?
6. diving into the code to correct the seg. fault for multithreading..
7. cache issue reported in the previous paper, dose the new version of japser fix the issue, updated, or changed?
8. p-dwt, poor performance may caused by the cache read & write between processor => overhead
9. whether it's possible to improve the dwt performance on the single core, in the c code level, with special instruction for example.
10. looking the dwt 2 loops for the horizontal and vertical filtering, and come up some ideas for improvement, and discuss it in the next week meeting.
11. next Tuesday will have the discussion for the course material

Research Progress

1. talked about the p-dwt performance, hoping to get a better one
2. the elec871 course reading, chapter 3 cache coherance top, and Cullar book related to chapter 5
3. checked the threading in the p-dwt functions. ok
4. checked the p-cblk functions.
5. thread safety, run jasper with 4 threads, segmentation faults happened a lot, the run is ok with 2 threads.
6. updated jasper code in cvs..