Practice of Efficient CUDA Programming
V1 | V2 | V3 | V4 | V5 | V6 | V7 | |
---|---|---|---|---|---|---|---|
shared Memory | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | |
No Bank-conflict | ✓ | ✓ | ✓ | ✓ | ✓ | ||
Double Buffer | ✓ | ✓ | ✓ | ✓ | |||
Warp Tile | ✓ | ✓ | ✓ | ||||
Less Registers | ✓ | ✓ | |||||
L2 Swizzle | ✓ |
V1 | V2 | V3 | V4 | V5 | |
---|---|---|---|---|---|
Direct gmem-smem | ✓ | ✓ | ✓ | ✓ | |
Double Buffer | ✓ | ✓ | ✓ | ||
L2 Swizzle | ✓ | ✓ | |||
Unroll | ✓ |