comaniac commented on a change in pull request #8234:
URL: https://github.com/apache/tvm/pull/8234#discussion_r660801654
##########
File path: python/tvm/relay/op/strategy/x86.py
##########
@@ -370,6 +370,55 @@ def conv1d_strategy_cpu(attrs, inputs, out_type, target):
return strategy
+@matmul_strategy.register("cpu")
+def matmul_strategy_cpu(attrs, inputs, out_type, target):
+ """matmul x86 strategy"""
+ strategy = _op.OpStrategy()
+ if is_auto_scheduler_enabled():
+ strategy.add_implementation(
+ wrap_compute_matmul(topi.nn.matmul,
need_auto_scheduler_layout=True),
+ naive_schedule,
+ name="matmul.generic",
+ plevel=11,
+ )
+ else:
+ logger.warning("Matmul other than NT format is not optimized for x86.")
+ strategy.add_implementation(
+ wrap_compute_matmul(topi.nn.matmul),
+ naive_schedule,
+ name="matmul.generic",
+ )
+
+ same_type = inputs[0].dtype == inputs[1].dtype == out_type.dtype
+ dtype = inputs[0].dtype
+ u8s8s32 = dtype == "uint8" and inputs[1].dtype == "int8" and
out_type.dtype == "int32"
+ if "cblas" in target.libs:
Review comment:
We can simply check if `strategy.specializations` is empty at the end of
this function.
##########
File path: include/tvm/relay/attrs/nn.h
##########
@@ -961,6 +961,32 @@ struct AvgPool3DAttrs : public
tvm::AttrsNode<AvgPool3DAttrs> {
}
};
+/*! \brief Attributes for matmul operator */
+struct MatmulAttrs : public tvm::AttrsNode<MatmulAttrs> {
+ IndexExpr units;
+ DataType out_dtype;
+ bool data_transposed;
+ bool weight_transposed;
+ tvm::String auto_scheduler_rewritten_layout; // The layout after
auto-scheduler's layout rewrite
Review comment:
This is used by auto-scheduler since it may want to rewrite the layout
after applying a schedule. It's not the best solution but cleanest workaround
for now, and it has been used by many ops already if you search
"auto_scheduler_rewritten_layout" in the code base.
##########
File path: python/tvm/relay/frontend/tensorflow.py
##########
@@ -1204,7 +1208,7 @@ def from_tensorflow(self, graph, layout="NHWC",
shape=None, outputs=None):
return func, self._params
-def from_tensorflow(graph, layout="NHWC", shape=None, outputs=None):
+def from_tensorflow(graph, layout="NHWC", shape=None, outputs=None,
use_dense_op=True):
Review comment:
This PR already uses dense schedule for matmul_nt in the case of
lowering to TOPI. On the other hand, as @jcf94 mentioned in the PR comment,
doing so will affect much more places in the codebase and we better gradually
convert them instead of in a single PR. It sounds reasonable to me.
--
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]
