SWT speed up... #58
SWT speed up... #58
Conversation
…n by using a pre-calculated lookup table.
…location by using a pre-calculated lookup table.
|
I have found some bugs in the implementation. I am looking to see if it's something that can be corrected. |
|
I made some bad assumptions about the algorithm and so most everything in this pull request is invalid. The speed ups I saw were due to short circuiting the algorithm. One of the commits should still be valid where I reordered the memory accesses but is rather small on whole. |
|
After my embarrassing last try I looked at it again and managed to improve performance while faithfully maintaining the algorithm's integrity. Hopefully I got it right this time. I have reopened the pull request and hopefully the new commits will show up. Basically there are two tight loops in ccv_swt() that do memory location calculations. It so happens that these loops can be modified in such a way that part of the memory location can be pre-calculated and then use a lookup table with some simple addition to gain some speed. |
…d a memory access violation. Accidentally used the wrong variable as the base of the offset. Also renamed an offset variable to properly reflect its origin.
|
+1 |
I removed a subtract and a multiply from calculating the memory location by using a pre-calculated lookup table. When testing I found this decreased the processing time of up to 35%.