CUDA: make cuGetProcAddress to work with cuda 12

[shamisp]

What

CUDA: Adding support for CUDA-12

[shamisp]

@Akshay-Venkatesh can you please take a look ? Thanks

[Akshay-Venkatesh]

@shamisp changes look fine but we probably ought to consolidate code which gets split based on version. There are other instances similar to these where fxn signature changes but we should probably have a common function so that caller doesn't see all the ugly bits.

[shamisp]

@Akshay-Venkatesh In a longer term this makes sense but it is much more substantial overhaul of CUDA UCT. Majority of the ifdef flows were introduced in 11070 and I didn't want to change this.

[Akshay-Venkatesh]

@Akshay-Venkatesh In a longer term this makes sense but it is much more substantial overhaul of CUDA UCT. Majority of the ifdef flows were introduced in 11070 and I didn't want to change this.

Agreed that we should not make that overhaul in this PR.

Minor nitpick: the title of this PR is deceiving. Shouldn't it just be "make cuGetProcAddress work with cuda 12"?

[shamisp]

@Akshay-Venkatesh In a longer term this makes sense but it is much more substantial overhaul of CUDA UCT. Majority of the ifdef flows were introduced in 11070 and I didn't want to change this.

Agreed that we should not make that overhaul in this PR.

Minor nitpick: the title of this PR is deceiving. Shouldn't it just be "make cuGetProcAddress work with cuda 12"?

agreed

[shamisp]

@Akshay-Venkatesh can you please approve so I can merge it once it is ready ? thanks !

[yosefe]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 3 pipeline(s).

[yosefe]

@shamisp can you pls squash?

[shamisp]

@yosefe DONE. The failures are unrelated: fatal error: error writing to /tmp/ccn6FDdP.s: No space left on device