!15244 Add logsoftmax C++ infer.

From: @liangzhibo
Reviewed-by: @zh_qh,@ginfung,@ginfung,@zh_qh
Signed-off-by: @zh_qh,@zh_qh

mindspore/core/abstract/infer_functions.h

@@ -304,8 +304,6 @@ AbstractBasePtr InferImplArgMaxWithValue(const AnalysisEnginePtr &, const Primit
                                          const AbstractBasePtrList &args_spec_list);
 AbstractBasePtr InferImplSparseSoftmaxCrossEntropyWithLogits(const AnalysisEnginePtr &, const PrimitivePtr &primitive,
                                                              const AbstractBasePtrList &args_spec_list);
-AbstractBasePtr InferImplDType(const AnalysisEnginePtr &, const PrimitivePtr &primitive,
-                               const AbstractBasePtrList &args_spec_list);
 AbstractBasePtr InferImplLoad(const AnalysisEnginePtr &, const PrimitivePtr &primitive,
                               const AbstractBasePtrList &args_spec_list);
 AbstractBasePtr InferImplAssign(const AnalysisEnginePtr &, const PrimitivePtr &primitive,

mindspore/core/abstract/prim_others.cc

@@ -545,22 +545,6 @@ AbstractBasePtr InferImplGpuConvertToDynamicShape(const AnalysisEnginePtr &, con
   return std::make_shared<AbstractTensor>(input->element(), shape);
 }
 
-AbstractBasePtr InferImplDType(const AnalysisEnginePtr &, const PrimitivePtr &primitive,
-                               const AbstractBasePtrList &args_spec_list) {
-  MS_EXCEPTION_IF_NULL(primitive);
-  auto op_name = primitive->name();
-  CheckArgsSize(op_name, args_spec_list, 1);
-  MS_EXCEPTION_IF_NULL(args_spec_list[0]);
-  auto type = args_spec_list[0]->BuildType();
-  MS_EXCEPTION_IF_NULL(type);
-  auto tensor_type = type->cast<TensorTypePtr>();
-  MS_EXCEPTION_IF_NULL(tensor_type);
-  auto value = tensor_type->element();
-  auto abstract = std::make_shared<abstract::AbstractType>(value);
-  abstract->set_value(value);
-  return abstract;
-}
-
 AbstractBasePtr InferImplLoad(const AnalysisEnginePtr &, const PrimitivePtr &primitive,
                               const AbstractBasePtrList &args_spec_list) {
   // Inputs: Ref/Tensor, universal

mindspore/core/ops/log_softmax.cc

@@ -34,22 +34,33 @@ void LogSoftmax::Init(const int64_t axis) { this->set_axis(axis); }
 
 abstract::ShapePtr LogSoftmaxInferShape(const PrimitivePtr &primitive, const std::vector<AbstractBasePtr> &input_args) {
   MS_EXCEPTION_IF_NULL(primitive);
-  auto LogSoftmax_prim = primitive->cast<PrimLogSoftmaxPtr>();
+  auto op_name = primitive->name();
-  MS_EXCEPTION_IF_NULL(LogSoftmax_prim);
+  auto axis = GetValue<int64_t>(primitive->GetAttr(kAxis));
-  auto op_name = LogSoftmax_prim->name();
+  CheckAndConvertUtils::CheckInteger("log_softmax infer", input_args.size(), kEqual, 1, op_name);
-  auto axis = LogSoftmax_prim->get_axis();
+  MS_EXCEPTION_IF_NULL(input_args[0]);
-  auto in_shape = CheckAndConvertUtils::ConvertShapePtrToShape("input_shape", input_args[0]->GetShapeTrack(), op_name);
+  auto shape_map = CheckAndConvertUtils::ConvertShapePtrToShapeMap(input_args[0]->BuildShape());
+  if (shape_map.empty()) {
+    // Scalar input, has no shape
+    return std::make_shared<abstract::Shape>(std::vector<int64_t>());
+  }
+  auto in_shape = shape_map[kShape];
+  auto min_shape = shape_map[kMinShape];
+  auto max_shape = shape_map[kMaxShape];
   auto rank = SizeToLong(in_shape.size());
   CheckAndConvertUtils::CheckInRange<int64_t>("axis", axis, kIncludeLeft, {-rank, rank}, op_name);
+  if (min_shape.size() != 0 && max_shape.size() != 0) {
+    return std::make_shared<abstract::Shape>(in_shape, min_shape, max_shape);
+  }
   return std::make_shared<abstract::Shape>(in_shape);
 }
 
 TypePtr LogSoftmaxInferType(const PrimitivePtr &prim, const std::vector<AbstractBasePtr> &input_args) {
-  if (std::any_of(input_args.begin(), input_args.end(), [](const AbstractBasePtr &a) { return a == nullptr; })) {
+  MS_EXCEPTION_IF_NULL(prim);
-    MS_LOG(EXCEPTION) << "nullptr";
+  auto op_name = prim->name();
-  }
+  CheckAndConvertUtils::CheckInteger("log_softmax infer", input_args.size(), kEqual, 1, op_name);
+  MS_EXCEPTION_IF_NULL(input_args[0]);
   const std::set<TypePtr> valid_types = {kFloat16, kFloat32};
-  return CheckAndConvertUtils::CheckTensorTypeValid("x", input_args[0]->BuildType(), valid_types, prim->name());
+  return CheckAndConvertUtils::CheckTensorTypeValid("x", input_args[0]->BuildType(), valid_types, op_name);
 }
 
 AbstractBasePtr LogSoftmaxInfer(const abstract::AnalysisEnginePtr &, const PrimitivePtr &primitive,
@@ -57,6 +68,6 @@ AbstractBasePtr LogSoftmaxInfer(const abstract::AnalysisEnginePtr &, const Primi
   return std::make_shared<abstract::AbstractTensor>(LogSoftmaxInferType(primitive, input_args),
                                                     LogSoftmaxInferShape(primitive, input_args)->shape());
 }
-REGISTER_PRIMITIVE_C(kNameLogSoftmax, LogSoftmax);
+REGISTER_PRIMITIVE_EVAL_IMPL(LogSoftmax, prim::kPrimLogSoftmax, LogSoftmaxInfer, nullptr, true);
 }  // namespace ops
 }  // namespace mindspore

mindspore/core/ops/log_softmax.h

@@ -37,9 +37,6 @@ class LogSoftmax : public PrimitiveC {
   void set_axis(const int64_t axis);
   int64_t get_axis() const;
 };
-
-AbstractBasePtr LogSoftmaxInfer(const abstract::AnalysisEnginePtr &, const PrimitivePtr &primitive,
-                                const std::vector<AbstractBasePtr> &input_args);
 using PrimLogSoftmaxPtr = std::shared_ptr<LogSoftmax>;
 }  // namespace ops
 }  // namespace mindspore

mindspore/ops/operations/nn_ops.py

@@ -178,7 +178,7 @@ class Softmax(Primitive):
             validator.check_value_type("item of axis", item, [int], self.name)
 
 
-class LogSoftmax(PrimitiveWithInfer):
+class LogSoftmax(Primitive):
     r"""
     Log Softmax activation function.
 
@@ -220,15 +220,6 @@ class LogSoftmax(PrimitiveWithInfer):
     def __init__(self, axis=-1):
         validator.check_value_type("axis", axis, [int], self.name)
 
-    def infer_shape(self, logits):
-        rank = len(logits)
-        validator.check_int_range(self.axis, -rank, rank, Rel.INC_LEFT, 'axis', self.name)
-        return logits
-
-    def infer_dtype(self, logits):
-        validator.check_tensor_dtype_valid("logits", logits, (mstype.float16, mstype.float32), self.name)
-        return logits
-
 
 class Softplus(Primitive):
     r"""