Diferență între revizuiri ale paginii „PC Lab 4”
De la WikiLabs
Jump to navigationJump to searchCbira (discuție | contribuții) (Pagină nouă: '''Session 4''' Speed optimization over i5/i7 x64 arch: Compute distance between two vectors of points in 128-D space (128 coordinates). The purpose is to find the maximum dista...) |
Cbira (discuție | contribuții) |
||
| (Nu s-au afișat 7 versiuni intermediare efectuate de același utilizator) | |||
| Linia 8: | Linia 8: | ||
As a distance, implement L1 (SAD) and L2 (SSD) norm over the 8-bit unsigned data type | As a distance, implement L1 (SAD) and L2 (SSD) norm over the 8-bit unsigned data type | ||
| − | [[Fișier: | + | [[Fișier:lnorm.png]] |
Points (out of 10) vs. expected performance (SSD & SAD): | Points (out of 10) vs. expected performance (SSD & SAD): | ||
| Linia 18: | Linia 18: | ||
7 for better than 20 & 40 MD/s | 7 for better than 20 & 40 MD/s | ||
| − | 8 for better than | + | 8 for better than 35 & 70 MD/s |
| − | 9 for better than | + | 9 for better than 50 & 100 MD/s |
| − | 10 for better than | + | 10 for better than 75 & 150 MD/s |
Versiunea curentă din 19 aprilie 2018 16:03
Session 4
Speed optimization over i5/i7 x64 arch:
Compute distance between two vectors of points in 128-D space (128 coordinates). The purpose is to find the maximum distance from any one point.
As a distance, implement L1 (SAD) and L2 (SSD) norm over the 8-bit unsigned data type
Points (out of 10) vs. expected performance (SSD & SAD):
5 for better than 5 & 10 MD/s
6 for better than 10 & 20 MD/s
7 for better than 20 & 40 MD/s
8 for better than 35 & 70 MD/s
9 for better than 50 & 100 MD/s
10 for better than 75 & 150 MD/s
