================
@@ -2851,7 +2851,21 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
.constrainAllUses(TII, TRI, RBI);
break;
case Intrinsic::spv_thread_id:
- return selectSpvThreadId(ResVReg, ResType, I);
+ // The HLSL SV_DispatchThreadID semantic is lowered to llvm.spv.thread.id
+ // intrinsic in LLVM IR for SPIR-V backend.
+ //
+ // In SPIR-V backend, llvm.spv.thread.id is now correctly translated to a
+ // `GlobalInvocationId` builtin variable
----------------
tex3d wrote:
These comments seem redundant to me. We are in the case handling
`llvm.spv.thread.id` and the code translates it to use the `GlobalInvocationId`
builtin variable. What else needs to be said in this context?
"is now correctly translated to ..." as opposed to when? What does this
comment add/clarify?
As far as the HLSL context is concerned, that's defined elsewhere, and it's
presumably not in scope for the SPIR-V backend to comment on where an intrinsic
may come from in a particular source language. I can imagine similar
expository comments for many other intrinsics being lowered here, but I don't
think this is the best place to document such things.
Note: this is just an opinion, feel free to disagree, and I'm curious how
others more experienced with the backends feel as well.
https://github.com/llvm/llvm-project/pull/117781
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
