macurtis-amd wrote: @nikic Latest revision only promotes memsets where the region being set is within an existing vector in the underlying alloca (as determined by getTypePartition).
It is much more constrained as you can see by the much smaller number of tests that are now affected, while still improving the original gemm kernel. Is this okay? https://github.com/llvm/llvm-project/pull/133301 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
