!62768 [MS] fix dtype cast bugs

Merge pull request !62768 from XianglongZeng/dyn_cpu

docs/api/api_python/nn/mindspore.nn.SiLU.rst

@@ -3,7 +3,7 @@ mindspore.nn.SiLU
 
 .. py:class:: mindspore.nn.SiLU
 
-    逐元素计算SiLU激活函数。
+    逐元素计算SiLU激活函数。有时也被称作Swish函数。
 
     SiLU函数定义为：
 

docs/api/api_python/ops/mindspore.ops.func_silu.rst

@@ -3,7 +3,7 @@ mindspore.ops.silu
 
 .. py:function:: mindspore.ops.silu(x)
 
-    按输入逐元素计算激活函数SiLU（Sigmoid Linear Unit）。该激活函数定义为：
+    按输入逐元素计算激活函数SiLU（Sigmoid Linear Unit）。有时也被称作Swish函数。该激活函数定义为：
 
     .. math::
         \text{SiLU}(x) = x * \sigma(x),

mindspore/ccsrc/transform/acl_ir/op_api_convert.h

@@ -400,6 +400,8 @@ inline aclScalar *ConvertType(const ScalarPtr &value) {
     converter.ConvertValue(value, AttrDeclType<bool>(), &acl_scalar);
   } else if (value->isa<Int64Imm>()) {
     converter.ConvertValue(value, AttrDeclType<int64_t>(), &acl_scalar);
+  } else if (value->isa<FP64Imm>()) {
+    converter.ConvertValue(value, AttrDeclType<double>(), &acl_scalar);
   } else if (value->isa<FP32Imm>()) {
     converter.ConvertValue(value, AttrDeclType<float>(), &acl_scalar);
   } else if (value->isa<Int32Imm>()) {

mindspore/ccsrc/utils/primfunc_utils.cc

@@ -140,14 +140,17 @@ bool ValidateArgsType(const AbstractBasePtr &abs_arg, OP_DTYPE type_arg) {
       return abs_arg->isa<abstract::AbstractScalar>() && (abs_type->isa<Float>() || abs_type->isa<BFloat>());
     }
     case OP_DTYPE::DT_NUMBER: {
-      return abs_arg->isa<abstract::AbstractScalar>() && (abs_type->isa<Number>());
+      return abs_arg->isa<abstract::AbstractScalar>() && abs_type->isa<Number>();
     }
     case OP_DTYPE::DT_STR: {
-      return abs_arg->isa<abstract::AbstractScalar>() && (abs_type->isa<String>());
+      return abs_arg->isa<abstract::AbstractScalar>() && abs_type->isa<String>();
     }
     case OP_DTYPE::DT_TENSOR: {
       return abs_arg->isa<abstract::AbstractTensor>();
     }
+    case OP_DTYPE::DT_TYPE: {
+      return abs_arg->isa<abstract::AbstractType>() && abs_type->isa<Type>();
+    }
     default: {
       return ValidateArgsSequenceType(abs_arg, type_arg);
     }

mindspore/core/ops/ops_def/silu_doc.yaml

@@ -1,6 +1,7 @@
 silu:
     description: |
-        Computes SiLU (Sigmoid Linear Unit activation function) of input tensors element-wise.
+        Computes SiLU (Sigmoid Linear Unit activation function) of input tensors element-wise. Also known as the Swish
+        function.
 
         Refer to :func:`mindspore.ops.silu` for more details.
 

mindspore/python/mindspore/ops_generate/arg_dtype_cast.py

@@ -29,25 +29,23 @@ def int_to_float(data):
     return float(data)
 
 
-def scalar_to_tuple(data, dst_type):
-    if dst_type == PY_DT_TUPLE_INT:
-        return (int(data),)
+def scalar_to_tuple(data):
     return (data,)
 
 
-def list_to_tuple(data, dst_type):
+def list_to_tuple(data):
     # tuple() currently does not support Any from JIT Fallback.
     res = ()
     for element in data:
-        if dst_type == PY_DT_TUPLE_INT:
-            res += (int(element),)
-        else:
-            res += (element,)
+        res += (element,)
     return res
 
 
-def tensor_to_tuple(data, dst_type):
-    if dst_type == PY_DT_TUPLE_INT:
+def tensor_to_tuple(data):
+    # Since tuple is not supported for precision conversion during KernelSelect, the original int32 tensor input cases
+    # would be failed. Thus, raise the tuple precision from int32 to int64 at frontend. But sequence data type cast
+    # must be adapted in future version.
+    if data.dtype == ms.int32:
         data = ops.cast(data, ms.int64)
     return tensor_to_tuple_(data)
 
@@ -66,8 +64,8 @@ def tuple_to_tensor(data):
     return ops.tuple_to_array(data)
 
 
-def list_to_tensor(data, dst_type):
-    return ops.tuple_to_array(list_to_tuple(data, dst_type))
+def list_to_tensor(data):
+    return ops.tuple_to_array(list_to_tuple(data))
 
 # type
 PY_DT_TYPE = OpDtype.PY_DT_TYPE.value
@@ -199,18 +197,18 @@ def do_type_cast(data, dst_type):
             return int_to_float(data)
     elif is_tuple(dst_type):
         if isinstance(data, (int, float, bool)):
-            return scalar_to_tuple(data, dst_type)
+            return scalar_to_tuple(data)
         if isinstance(data, list):
-            return list_to_tuple(data, dst_type)
+            return list_to_tuple(data)
         if isinstance(data, Tensor):
-            return tensor_to_tuple(data, dst_type)
+            return tensor_to_tuple(data)
     elif dst_type == PY_DT_TENSOR:
         if isinstance(data, (int, float, bool)):
             return scalar_to_tensor(data)
         if isinstance(data, tuple):
             return tuple_to_tensor(data)
         if isinstance(data, list):
-            return list_to_tensor(data, dst_type)
+            return list_to_tensor(data)
     elif is_number(dst_type):
         if isinstance(data, Tensor):
             if dst_type == PY_DT_INT: