================ @@ -163,7 +163,10 @@ BUILTIN(__builtin_amdgcn_raw_buffer_load_b64, "V2UiQbiiIi", "n") BUILTIN(__builtin_amdgcn_raw_buffer_load_b96, "V3UiQbiiIi", "n") BUILTIN(__builtin_amdgcn_raw_buffer_load_b128, "V4UiQbiiIi", "n") +TARGET_BUILTIN(__builtin_amdgcn_raw_buffer_load_lds, "vV4Uiv*3IUiiiIiIi", "t", "vmem-to-lds-load-insts") ---------------- JonChesterfield wrote:
I would say that's a difficult tradeoff. I'd expect the integer version to work more reliably and the pointer one to generate better code, as that's usually how int2ptr hackery works out. Would you prefer the integer intrinsic getting more use via the builtin or people continuing to use assembly? My understanding is that CK are using the compiler where they can and bypassing it where they can't, which rather suggests our compiler is not totally meeting their use cases. https://github.com/llvm/llvm-project/pull/137678 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
