In ProgrammingGuide_3.0 G 3.2.2
G.3.2.2 Devices of Compute Capability 1.2 and 1.3
…
More precisely, the following protocol is used to determine the memory transactions
necessary to service all threads in a half-warp:
Find the memory segment that contains the address requested by the lowest
numbered active thread. The segment size depends on the size of the words
accessed by the threads:
[*] 32 bytes for 1-byte words,
[*] 64 bytes for 2-byte words,
[*] 128 bytes for 4-, 8- and 16-byte words.
I think it should be warp,not half-warp.so is it right?
In ProgrammingGuide_3.0 G 3.2.2
G.3.2.2 Devices of Compute Capability 1.2 and 1.3
…
More precisely, the following protocol is used to determine the memory transactions
necessary to service all threads in a half-warp:
Find the memory segment that contains the address requested by the lowest
numbered active thread. The segment size depends on the size of the words
accessed by the threads:
[*] 32 bytes for 1-byte words,
[*] 64 bytes for 2-byte words,
[*] 128 bytes for 4-, 8- and 16-byte words.
I think it should be warp,not half-warp.so is it right?
paulius
3
The text is correct - memory (smem and gmem) addresses are processed per half-warp prior to compute capability 2.0.
paulius
4
The text is correct - memory (smem and gmem) addresses are processed per half-warp prior to compute capability 2.0.