Hi Thomas,
Thomas Schwinge wrote:
On 2023-09-08T11:04:24+0200, Tobias Burnus <tob...@codesourcery.com> wrote:
On 02.08.23 19:00, Andrew Stubbs wrote:
The use of the PTX dynamic_smem_size feature means that low-latency allocator
will not work with the PTX 3.1 multilib.
Right: PTX '%dynamic_smem_size' was "Introduced in PTX ISA version 4.1",
and "Requires 'sm_20' or higher".
I don't understand the "Right" part, but it is good news that
GCC-supported sm_XX support it :-)
Thanks also for correcting the documentation, i.e. removing the
@code{sm_53} bit.
If so, I think we should mention that
nvptx GCC has to be configured with with-arch=sm_... >= sm_53 (=
supported version >=4.1) and, during compilation, no -march= < that
configure-time value may be specified. (Cf. also
https://gcc.gnu.org/install/specific.html#nvptx-x-none )
Given that GCC/nvptx generally supports 'sm_20', only the PTX ISA version
matters here, and that's all fine if just using GCC's defaults. OK to push
"Clarify libgomp nvptx 'omp_low_lat_mem_space' documentation", see
attached?
Regarding the patch, I don't really see the point of '@c -mptx=4.1' but
I also do not mind keeping it.
In any case, the user-visible text LGTM.
Thanks,
Tobias
PS: This will be the 200th GCC 15 commit for BayLibre.
