[llvm] [NVPTX] Add intrinsics for prefetch.* (PR #125887)

[Durgadoss R via llvm-commits]

durga4github wrote:

> Hi @abhilash1910, the new test fails for me as
> 
> ```
> ptxas warning : incompatible redefinition for option 'gpu-name', the last value of this option was used
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 19; error   : Unknown modifier '.evictnormal'
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 20; error   : Unknown modifier '.evictnormal'
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 33; error   : Unknown modifier '.evictnormal'
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 34; error   : Unknown modifier '.evictnormal'
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 35; error   : Unknown modifier '.evictlast'
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 36; error   : Unknown modifier '.evictlast'
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 49; error   : Unknown modifier '.evictnormal'
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 50; error   : Unknown modifier '.evictnormal'
> ptxas /tmp/tmpxft_000a5e18_00000000-0_stdin, line 63; error   : Unknown modifier '.evictnormal'
> ptxas fatal   : Ptx assembly aborted due to errors
> ```
> 
> ptxas --version is
> 
> ```
> ptxas: NVIDIA (R) Ptx optimizing assembler
> Copyright (c) 2005-2024 NVIDIA Corporation
> Built on Thu_Mar_28_02:14:54_PDT_2024
> Cuda compilation tools, release 12.4, V12.4.131
> Build cuda_12.4.r12.4/compiler.34097967_0
> ```
> 
> I have attached [prefetch.txt](https://github.com/user-attachments/files/18763610/prefetch.txt) generated by llc
> 
> Running as `cat prefetch.txt | ptxas -arch=sm_60 -c - -arch=sm_90`
> 
> What might be the issue?

Yes, I could repro the issue locally. Thank you for reporting!

At first sight, all the evict qualifiers need an underscore '_' to match the Spec.
will investigate further and update..


https://github.com/llvm/llvm-project/pull/125887