junrushao commented on code in PR #14048:
URL: https://github.com/apache/tvm/pull/14048#discussion_r1111390041
##########
src/target/intrin_rule.cc:
##########
@@ -118,6 +119,22 @@ TVM_REGISTER_OP("tir.nearbyint")
TVM_REGISTER_OP("tir.pow").set_attr<FLowerIntrinsic>("default.FLowerIntrinsic",
DispatchPureExtern<FloatSuffix>);
+PrimExpr DispatchFastErf(const PrimExpr& e) {
+ LOG(WARNING) << "fast_erf will be used instead of erf";
+ const CallNode* call = e.as<CallNode>();
+ ICHECK(call != nullptr);
+ ICHECK_EQ(call->args.size(), 1);
+ PrimExpr arg = call->args[0];
+ int bits = arg.dtype().bits();
+ PrimExpr res;
+ if (arg.dtype().is_float() && (bits == 16 || bits == 32)) {
+ res = topi::fast_erf_float_expr(arg, bits);
Review Comment:
i know it's from some legacy impl of metal, but can we structure the folder
slightly better to avoid depending on topi? i.e. moving this method to tir
instead.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
