aaupov wrote: > Do you have benchmarks on how well this performs relative to standard IRPGO > (or on top of it)?
Couple of things to untangle here: 1. In general, CSSPGO is meant as a more practical alternative for IRPGO thanks to i) the use of sampled profile, ii) context-sensitivity. However, IRPGO is still expected to provide better performance in all cases where it's applicable thanks to accurate profile information. 2. Due to that, I don't think it makes sense to apply CSSPGO on top of IRPGO, and CMake automation would not permit mixing the two (both implemented as LLVM_BUILD_INSTRUMENTED exclusive options). 3. [CSSPGO RFC](https://lists.llvm.org/pipermail/llvm-dev/2020-August/144101.html) has direct perf comparison of early implementation of CSSPGO vs IRPGO on Spec06: <img width="1491" alt="Screenshot 2024-01-29 at 9 43 17 PM" src="https://github.com/llvm/llvm-project/assets/876514/a921e324-6b5d-4660-9928-0fb0c994649a";> 4. With CMake stuff implemented in this diff, it's expected that the amount of profile information collected would be miniscule (in-tree perf-training only has a single hello world source file) and inadequate for getting much perf boost with CSSPGO (it even prints the warning that it needs 6985000.0x more profile). 5. But I'll kick out a perf run anyway to see what's the starting point. If it's zero, I'll try building different targets (in-tree or from LLVM Test Suite) to identify a viable profiling workload, but it's outside of scope of this diff. https://github.com/llvm/llvm-project/pull/79942 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
