================ ---------------- Keenuts wrote:
I think what Steven suggested was to have the scope + operation being an immediate operand to the builtin. Meaning thew CGHLSLBuiltin code would have to emit something like `llvm.spv.wave.product(i32 3, i32 0, i32 %iexpr)` so the Scope & Operation operands are something we can do in the FE, and it reduces the amount of intrinsics needed (as here, if we implement all SPIR-V intrinsics combinaisons, we'd have to add 8 (operations) x 7 (scope) intrinsincs to cover all combinaisons. https://github.com/llvm/llvm-project/pull/165109 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
