[XPU][TritonIntelGPUToLLVM] Do not generate duplicated string constants

[victor-eds]

Do not generate duplicated string constants as result of tt.assert or tt.print operations. Use a string constant cache to avoid duplication.

Closes #3054.

[victor-eds]

Fixing this issue would result in overhead in the base case (no tt.assert or tt.print operations used). As we may not have these operations in most cases, I'd rather close the issue as won't fix.

Alternatively, we can merge this PR.

[anmyachev]

Fixing this issue would result in overhead in the base case (no tt.assert or tt.print operations used).

Hi @victor-eds, is the overhead significant? Maybe it is acceptable, considering that we seem to be saving memory?

[victor-eds]

Hi @victor-eds, is the overhead significant? Maybe it is acceptable, considering that we seem to be saving memory?

I don't think the overhead is significant, but I wonder if we want to go with any additional overhead to save memory on "debug" scenarios (asserts don't have effect unless TRITON_DEBUG is set and I wouldn't expect prints on release kernels either). I'm fine with this option to save memory when debugging kernels, but:

• We would improve debug mode performance in exchange of performance mode performance (potentially small overhead, but, still)
• How many times outside artificial test will we find kernels with several asserts or prints to a point avoiding string duplication is worth it?