[RELAY] Refactor FoldConstant to skip TNonComputationalOps (apache#6720)

* add TNonComputational to qnn ops and change FoldConstant * remove comments * check if op in nonComputational map * forgot to mark device_copy op as TNonComputational * hacky fix to fuseops pass * fix typo * manually skip device_copy in fold_constant * Update src/relay/transforms/fold_constant.cc Co-authored-by: Junru Shao <junrushao1994@gmail.com> Co-authored-by: Junru Shao <junrushao1994@gmail.com>
neo-ai · Dec 4, 2020 · bed642e · bed642e
1 parent 9e27758
commit bed642e
Show file tree

Hide file tree

Showing 8 changed files with 13 additions and 3 deletions.
diff --git a/src/relay/qnn/op/concatenate.cc b/src/relay/qnn/op/concatenate.cc
@@ -207,6 +207,7 @@ RELAY_REGISTER_OP("qnn.concatenate")
                   "The quantization zero_point of the output tensor.")
     .set_support_level(11)
     .add_type_rel("QnnConcatenate", QnnConcatenateRel)
+    .set_attr<TNonComputational>("TNonComputational", true)
     .set_attr<FTVMLegalize>("FTVMQnnCanonicalize", ConcatenateQnnCanonicalize)
     .set_attr<FInferCorrectLayout>("FInferCorrectLayout", QnnConcatenateLayout);
 

diff --git a/src/relay/qnn/op/convolution.cc b/src/relay/qnn/op/convolution.cc
@@ -733,6 +733,7 @@ operator to understand how to scale back the int32 output to (u)int8.
                   "The quantization zero_point of the weight tensor.")
     .set_support_level(11)
     .add_type_rel("QnnConv2D", QnnConv2DRel)
+    .set_attr<TNonComputational>("TNonComputational", true)
     .set_attr<FTVMLegalize>("FTVMQnnCanonicalize", QnnConv2DCanonicalize)
     .set_attr<FInferCorrectLayout>("FInferCorrectLayout", QnnConvInferCorrectLayout);
 

diff --git a/src/relay/qnn/op/dense.cc b/src/relay/qnn/op/dense.cc
@@ -189,6 +189,7 @@ RELAY_REGISTER_OP("qnn.dense")
                   "The quantization zero_point of the weight tensor.")
     .set_support_level(11)
     .add_type_rel("QDense", QnnDenseRel)
+    .set_attr<TNonComputational>("TNonComputational", true)
     .set_attr<FTVMLegalize>("FTVMQnnCanonicalize", QnnDenseCanonicalize);
 
 TVM_REGISTER_GLOBAL("relay.qnn.op._make.dense").set_body_typed(MakeQuantizedDense);

diff --git a/src/relay/qnn/op/dequantize.cc b/src/relay/qnn/op/dequantize.cc
@@ -136,6 +136,7 @@ The input is always quantized (int8, uint8) and will be converted to float32 giv
     .add_argument("input_zero_point", "Tensor", "The quantization zero_point of the input tensor.")
     .set_support_level(11)
     .add_type_rel("Dequantize", DequantizeRel)
+    .set_attr<TNonComputational>("TNonComputational", true)
     .set_attr<FTVMLegalize>("FTVMQnnCanonicalize", DequantizeQnnCanonicalize);
 
 TVM_REGISTER_GLOBAL("relay.qnn.op._make.dequantize").set_body_typed(MakeDequantize);

diff --git a/src/relay/qnn/op/op_common.h b/src/relay/qnn/op/op_common.h
@@ -215,6 +215,7 @@ static inline bool QnnBroadcastRel(const Array<Type>& types, int num_inputs, con
       .add_argument("output_scale", "Tensor", "The scale of the output tensor.")                   \
       .add_argument("output_zero_point", "Tensor", "The zero_point of the output tensor.")         \
       .add_type_rel("QnnBroadcast", QnnBroadcastRel)                                               \
+      .set_attr<TNonComputational>("TNonComputational", true)                                      \
       .set_attr<FInferCorrectLayout>("FInferCorrectLayout", QnnBinaryBroadcastLayout)
 
 }  // namespace qnn

diff --git a/src/relay/qnn/op/quantize.cc b/src/relay/qnn/op/quantize.cc
@@ -150,6 +150,7 @@ scale and zero point.
                   "The quantization zero_point of the output tensor.")
     .set_support_level(11)
     .add_type_rel("Quantize", QuantizeRel)
+    .set_attr<TNonComputational>("TNonComputational", true)
     .set_attr<FTVMLegalize>("FTVMQnnCanonicalize", QuantizeQnnCanonicalize);
 
 TVM_REGISTER_GLOBAL("relay.qnn.op._make.quantize").set_body_typed(MakeQuantize);

diff --git a/src/relay/qnn/op/requantize.cc b/src/relay/qnn/op/requantize.cc
@@ -324,6 +324,7 @@ Q_output = zp_output +  (scale_input)/(scale_output) * (Q_input - zp_input)
                   "The quantization zero_point of the output tensor.")
     .set_support_level(11)
     .add_type_rel("Requantize", RequantizeRel)
+    .set_attr<TNonComputational>("TNonComputational", true)
     .set_attr<FTVMLegalize>("FTVMQnnCanonicalize", RequantizeQnnCanonicalize)
     .set_attr<FInferCorrectLayout>("FInferCorrectLayout", RequantizeInferCorrectLayout);
 

diff --git a/src/relay/transforms/fold_constant.cc b/src/relay/transforms/fold_constant.cc
@@ -151,9 +151,12 @@ class ConstantFolder : public MixedModeMutator {
     }
 
     // We should think about potentially constant evaluation over these ops too.
-    if (call->op == invoke_tvm_op_ || call->op == shape_func_op_ || call->op == alloc_tensor_op_ ||
-        call->op == alloc_storage_op_ || call->op == device_copy_op_) {
-      return GetRef<Call>(call);
+    static auto fnoncomputational = Op::GetAttrMap<TNonComputational>("TNonComputational");
+    if (const auto* call_node = call->op.as<OpNode>()) {
+      Op op = GetRef<Op>(call_node);
+      if ((fnoncomputational.count(op) && fnoncomputational[op]) || (call->op == device_copy_op_)) {
+        return GetRef<Call>(call);
+      }
     }
 
     bool all_const_args = true;