lhutton1 commented on a change in pull request #7206:
URL: https://github.com/apache/tvm/pull/7206#discussion_r551595603



##########
File path: python/tvm/relay/op/contrib/arm_compute_lib.py
##########
@@ -19,12 +19,15 @@
 import numpy as np
 import tvm
 
+import tvm._ffi

Review comment:
       Better to use more specific import
   ```suggestion
   from tvm._ffi import register_func
   ```

##########
File path: python/tvm/relay/op/contrib/arm_compute_lib.py
##########
@@ -71,6 +74,61 @@ def partition_for_arm_compute_lib(mod, params=None):
     return seq(mod)
 
 
+@tvm._ffi.register_func("relay.ext.arm_compute_lib.optimize")

Review comment:
       With above change
   ```suggestion
   @register_func("relay.ext.arm_compute_lib.optimize")
   ```

##########
File path: python/tvm/relay/op/contrib/arm_compute_lib.py
##########
@@ -71,6 +74,61 @@ def partition_for_arm_compute_lib(mod, params=None):
     return seq(mod)
 
 
+@tvm._ffi.register_func("relay.ext.arm_compute_lib.optimize")
+def preprocess_module(mod):
+    """
+        Pre-process a module containing functions ready for ACL codegen. For 
now we enforce OHWI
+        kernel layout and fold the transforms away.
+
+    `   Parameters

Review comment:
       Remove `

##########
File path: src/relay/backend/contrib/arm_compute_lib/codegen.cc
##########
@@ -126,7 +127,7 @@ class ACLJSONSerializer : public 
backend::contrib::JSONSerializer {
       nodes.activation = current_call;
       current_call = current_call->args[0].as<CallNode>();
     }
-    if (backend::IsOp(current_call, "nn.bias_add")) {
+    if (backend::IsOp(current_call, "add")) {

Review comment:
       I remember needing to change this but I didn't get to the bottom of why. 
Is there an explanation?




----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


Reply via email to