!1153 update_headerfiles_0225

From: @shenwei41
Reviewed-by: @lilongfei15,@xsmq
Signed-off-by: @xsmq

inc/external/acl/acl_base.h

@@ -174,6 +174,12 @@ typedef enum {
     ACL_ERROR = 3,
 } aclLogLevel;
 
+typedef enum {
+    ACL_MEMTYPE_DEVICE = 0,
+    ACL_MEMTYPE_HOST = 1,
+} aclMemType;
+
+
 /**
  * @ingroup AscendCL
  * @brief Converts data of type aclFloat16 to data of type float
@@ -594,6 +600,18 @@ ACL_FUNC_VISIBILITY aclError aclSetTensorDynamicInput(aclTensorDesc *desc, const
  */
 ACL_FUNC_VISIBILITY aclError aclSetTensorConst(aclTensorDesc *desc, void *dataBuffer, size_t length);
 
+/**
+ * @ingroup AscendCL
+ * @brief Set tensor memory type specified by the tensor description
+ *
+ * @param  desc [OUT]      pointer to the instance of aclTensorDesc
+ * @param  memType [IN]       ACL_MEMTYPE_DEVICE means device, ACL_MEMTYPE_HOST means host
+ *
+ * @retval ACL_SUCCESS     The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorPlaceMent(aclTensorDesc *desc, aclMemType memType);
+
 /**
  * @ingroup AscendCL
  * @brief an interface for users to output  APP logs

inc/external/acl/acl_mdl.h

@@ -1203,6 +1203,18 @@ ACL_FUNC_VISIBILITY aclError aclmdlDestroyConfigHandle(aclmdlConfigHandle *handl
 ACL_FUNC_VISIBILITY aclError aclmdlSetConfigOpt(aclmdlConfigHandle *handle, aclmdlConfigAttr attr,
     const void *attrValue, size_t valueSize);
 
+/**
+ * @ingroup AscendCL
+ * @brief get real tensor name from modelDesc
+ *
+ * @param modelDesc [IN]  pointer to modelDesc
+ * @param name [IN]       tensor name
+ *
+ * @retval the pointer of real tensor name
+ * @retval Failure return NULL
+ */
+ACL_FUNC_VISIBILITY const char *aclmdlGetTensorRealName(const aclmdlDesc *modelDesc, const char *name);
+
 #ifdef __cplusplus
 }
 #endif

inc/external/acl/acl_rt.h

@@ -25,6 +25,8 @@
 extern "C" {
 #endif
 
+#define ACL_EVENT_TIME_LINE 0x00000008u
+
 typedef enum aclrtRunMode {
     ACL_DEVICE,
     ACL_HOST,
@@ -425,6 +427,18 @@ ACL_FUNC_VISIBILITY aclError aclrtGetDeviceCount(uint32_t *count);
  */
 ACL_FUNC_VISIBILITY aclError aclrtCreateEvent(aclrtEvent *event);
 
+/**
+ * @ingroup AscendCL
+ * @brief create event instance with flag
+ *
+ * @param event [OUT]   created event
+ * @param flag [IN]     event flag
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclrtCreateEventWithFlag(aclrtEvent *event, uint32_t flag);
+
 /**
  * @ingroup AscendCL
  * @brief destroy event instance

inc/external/hccl/hccl.h

@@ -66,15 +66,14 @@ extern HcclResult HcclCommInitRootInfo(uint32_t nRanks, const HcclRootInfo *root
  * @param sendBuf A pointer identifying the input data address of the operator.
  * @param recvBuf A pointer identifying the output data address of the operator.
  * @param count An integer(u64) identifying the number of the output data.
- * @param dataType The data type of the operator, must be one of the following types: int8, int16, int32, float16,
+ * @param dataType The data type of the operator, must be one of the following types: int8, int16, int32, float16, float32.
- * float32.
  * @param op The reduction type of the operator, must be one of the following types: sum, min, max, prod.
  * @param comm A pointer identifying the communication resource based on.
  * @param stream A pointer identifying the stream information.
  * @return HcclResult 
  */
-extern HcclResult HcclAllReduce(void *sendBuf, void *recvBuf, uint64_t count, HcclDataType dataType, HcclReduceOp op,
+extern HcclResult HcclAllReduce(void *sendBuf, void *recvBuf, uint64_t count, HcclDataType dataType, 
-                                HcclComm comm, aclrtStream stream);
+HcclReduceOp op, HcclComm comm, aclrtStream stream);
 
 /**
  * @brief Broadcast operator.
@@ -88,7 +87,7 @@ extern HcclResult HcclAllReduce(void *sendBuf, void *recvBuf, uint64_t count, Hc
  * @return HcclResult 
  */
 extern HcclResult HcclBroadcast(void *buf, uint64_t count, HcclDataType dataType, uint32_t root, HcclComm comm, 
-                                aclrtStream stream);
+aclrtStream stream);
 
 /**
  * @brief ReduceScatter operator.
@@ -103,7 +102,7 @@ extern HcclResult HcclBroadcast(void *buf, uint64_t count, HcclDataType dataType
  * @return HcclResult 
  */
 extern HcclResult HcclReduceScatter(void *sendBuf, void *recvBuf, uint64_t recvCount, HcclDataType dataType, 
-                                    HcclReduceOp op, HcclComm comm, aclrtStream stream);
+HcclReduceOp op, HcclComm comm, aclrtStream stream);
 
 /**
  * @brief AllGather operator.
@@ -116,8 +115,8 @@ extern HcclResult HcclReduceScatter(void *sendBuf, void *recvBuf, uint64_t recvC
  * @param stream A pointer identifying the stream information.
  * @return HcclResult 
  */
-extern HcclResult HcclAllGather(void *sendBuf, void *recvBuf, uint64_t sendCount, HcclDataType dataType, HcclComm comm,
+extern HcclResult HcclAllGather(void *sendBuf, void *recvBuf, uint64_t sendCount, HcclDataType dataType, 
-                                aclrtStream stream);
+HcclComm comm, aclrtStream stream);
 
 /**
  * @brief Destroy HCCL comm

third_party/fwkacllib/inc/ops/elewise_calculation_ops.h

@@ -3781,6 +3781,32 @@ REG_OP(ArgMaxGradD)
     .REQUIRED_ATTR(dimension, Int)
     .OP_END_FACTORY_REG(ArgMaxGradD)
 
+/**
+*@brief Returns cosine similarity between x1 and x2,computed along dim. \n
+
+*@par Inputs:
+*Two inputs, including:
+* @li input_x1: A tensor. Must be the following types:
+*     float32. \n
+
+*@par Inputs:
+*@li input_x2: A tensor. Must of the following types:
+*     float32. \n
+
+*@par Outputs:
+*@li output_y: A Tensor with the same type of input_x's. \n
+
+*@par Third-party framework compatibility
+*Compatible with the Pytorch operator CosineSimilarity. \n
+*/
+REG_OP(CosineSimilarity)
+    .INPUT(input_x1, TensorType({DT_FLOAT}))  /* "First operand." */
+    .INPUT(input_x2, TensorType({DT_FLOAT}))  /* "Second operand." */
+    .OUTPUT(output_y, TensorType({DT_FLOAT})) /* "Result, has same element type as two inputs" */
+    .ATTR(dim, Int, 1)
+    .ATTR(eps, Float, 1e-8)
+    .OP_END_FACTORY_REG(CosineSimilarity)
+
 }  // namespace ge
 
 #endif  // OPS_BUILT_IN_OP_PROTO_INC_ELEWISE_CALCULATION_OPS_H_

third_party/fwkacllib/inc/ops/math_ops.h

@@ -952,6 +952,36 @@ REG_OP(Angle)
     .ATTR(Tout, Type, DT_FLOAT)
     .OP_END_FACTORY_REG(Angle)
 
+/**
+*@brief Computes the gradient of SoftMarginLossGrad. \n
+
+*@par Inputs:
+*Three inputs, including:
+* @li predict: A tensor. Must be one of the following types:
+*     float16, float32. \n
+* @li label: A tensor with same shape of predict. Must be one of the following types:
+*     float16, float32. \n
+* @li dout: A tensor with same shpae of predcit. Must be one of the following types:
+*     float16, float32. \n
+
+*@par Attributes:
+* @li reduction: Specifies the reduction to apply to the output:
+*     'none' | 'mean' | 'sum'. Default: 'mean'. \n
+
+*@par Outputs:
+* gradient: A Tensor with the same type of predict. \n
+
+*@par Third-party framework compatibility
+*Compatible with the Pytorch operator SoftMarginLoss Backward. \n
+*/
+REG_OP(SoftMarginLossGrad)
+    .INPUT(predict, TensorType({DT_FLOAT16,DT_FLOAT}))
+    .INPUT(label, TensorType({DT_FLOAT16,DT_FLOAT}))
+    .INPUT(dout, TensorType({DT_FLOAT16,DT_FLOAT}))
+    .OUTPUT(gradient, TensorType({DT_FLOAT16,DT_FLOAT}))
+    .ATTR(reduction, String, "mean")
+    .OP_END_FACTORY_REG(SoftMarginLossGrad)
+
 }  // namespace ge
 
 #endif  // OPS_BUILT_IN_OP_PROTO_INC_MATH_OPS_H_

third_party/fwkacllib/inc/ops/nn_norm_ops.h

@@ -1205,6 +1205,34 @@ REG_OP(Centralization)
     .ATTR(axes, ListInt, {-1})
     .OP_END_FACTORY_REG(Centralization)
 
+/**
+ *@brief Calculate the loss. Creates a criterion that optimizes a two-class classification
+ logistic loss between input_x and input_y (containing 1 or -1). \n
+
+ *@par Inputs:
+ *One inputs, including:
+ * @li input_x: A tensor. Must be one of the following types:
+ *     float16, float32. \n
+ * @li input_y: A tensor. Must be one of the following types:
+ *     float16, float32. \n
+
+ *@par Attributes:
+ *@li lambd: An optional string.Defaults to "mean". \n
+
+ *@par Outputs:
+ *output_z: while reduction == "none", A Tensor with the same type and shape of input_x's. \n
+ *          while reduction == "sum" or "mean", A Tensor with the same type of input_x , shape of which is (1,)
+
+ *@par Third-party framework compatibility
+ *Compatible with the Pytorch operator SoftMarginLoss. \n
+ */
+REG_OP(SoftMarginLoss)
+    .INPUT(input_x, TensorType({DT_FLOAT, DT_FLOAT16}))
+    .INPUT(input_y, TensorType({DT_FLOAT, DT_FLOAT16}))
+    .ATTR(reduction, String, "mean")
+    .OUTPUT(output_z, TensorType({DT_FLOAT, DT_FLOAT16}))
+    .OP_END_FACTORY_REG(SoftMarginLoss)
+
 /**
 * @brief Computes gradients of sigmoid_cross_entropy_with_logits_v2.
 

third_party/fwkacllib/inc/ops/nonlinear_fuc_ops.h

@@ -792,6 +792,34 @@ REG_OP(HardShrink)
     .ATTR(lambd, Float, 0.5)
     .OP_END_FACTORY_REG(HardShrink)
 
+/**
+*@brief Calculate the hard shrink grad function. \n
+*
+* Computes the gradient for the HardShrink: if x > lambda or x < -lambda, x,otherwise 0
+*
+*@par Inputs:
+*Two inputs, including:
+* @li gradients: A tensor. Must be one of the following types:
+*     float16, float32. \n
+* @li features: A tensor. Must be one of the following types:
+*     float16, float32. \n
+*
+*@par Outputs:
+*backprops: A Tensor with the same type and shape of features's. \n
+*
+*@par Attributes:
+*@li lambda: An optional float.Defaults to 0.5. \n
+*
+*@par Third-party framework compatibility
+*Compatible with the Pytorch operator Hardshrink_backward. \n
+*/
+  REG_OP(HardShrinkGrad)
+  .INPUT(gradients, TensorType({DT_FLOAT16, DT_FLOAT}))
+  .INPUT(features, TensorType({DT_FLOAT16, DT_FLOAT}))
+  .OUTPUT(backprops, TensorType({DT_FLOAT16, DT_FLOAT}))
+  .ATTR(lambda, Float, 0.5)
+  .OP_END_FACTORY_REG(HardShrinkGrad)
+
 /**
 * @brief Calculate the hard sigmoid function. \n
 
@@ -884,6 +912,36 @@ REG_OP(LogSigmoid)
     .INPUT(x, TensorType({DT_FLOAT16, DT_FLOAT})) /* "input:x" */
     .OUTPUT(y, TensorType({DT_FLOAT16, DT_FLOAT}))  /* "output:y" */
     .OP_END_FACTORY_REG(LogSigmoid)
+
+/**
+*@brief Calculate the backward outputs of the function "hard_sigmoid" \n
+
+*@par Inputs:
+*One inputs, including:
+* @li grads: A tensor. Must be one of the following types:
+*       float16, float32. \n
+* @li input_x: A tensor. Must be one of the following types:
+*       float16, float32. \n
+
+*@par Outputs:
+*One outputs, including:
+* @li y: A tensor with the same type and shape of x's. \n
+
+* @par Attributes:
+* @li alpha: An optional float. Defaults to 0.16666666. \n
+* @li beta: An optional float. Defaults to 0.5. \n
+
+*@par Third-party framework compatibility
+*Compatible with the Pytorch operator LogSigmoidGrad. \n
+*/
+REG_OP(HardSigmoidGrad)
+    .INPUT(grads, TensorType({DT_FLOAT, DT_FLOAT16}))
+    .INPUT(input_x, TensorType({DT_FLOAT, DT_FLOAT16}))
+    .OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16}))
+    .ATTR(alpha, Float, 0.16666666)
+    .ATTR(beta, Float, 0.5)
+    .OP_END_FACTORY_REG(HardSigmoidGrad)
+
 } // namespace ge
 
 #endif  // OPS_BUILT_IN_OP_PROTO_INC_NONLINEAR_FUC_OPS_H_

third_party/fwkacllib/inc/ops/transformation_ops.h

@@ -737,14 +737,51 @@ where ho/wo is do = (output_d + 2*padding_d - dilation_d*(kernel_d - 1) - 1)//st
 * Compatible with Pytorch col2im/im2col_backward operator.
 */
 REG_OP(Col2im)
-    .INPUT(x, TensorType({DT_FLOAT}))
+    .INPUT(x, TensorType({DT_FLOAT, DT_FLOAT16}))
-    .INPUT(output_size, TensorType({DT_INT32}))
+    .INPUT(output_size, TensorType({DT_INT32, DT_INT32}))
-    .OUTPUT(y, TensorType({DT_FLOAT}))
+    .OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16}))
     .REQUIRED_ATTR(kernel_size, ListInt)
     .REQUIRED_ATTR(dilation, ListInt)
     .REQUIRED_ATTR(padding, ListInt)
     .REQUIRED_ATTR(stride, ListInt)
     .OP_END_FACTORY_REG(Col2im)
+
+/**
+*@brief Generates a 2D or 3D flow field (sampling grid), given a batch of affine
+matrices theta. \n
+
+*@par Inputs:
+*Input theta must be float16 or float, output_size must be int32 type.Inputs
+include:
+*@li theta: input batch of affine matrices with shape (N,2,3) for 2D or (N,3,4)
+for 3D
+*@li output_size: the target output image size. (N×C×H×W for 2D or N×C×D×H×W for
+3D) Example: torch.Size((32, 3, 24, 24)) . \n
+
+
+*@par Attributes:
+*align_corners: if True, consider -1 and 1 to refer to the centers of the corner
+pixels rather than the image corners.Refer to grid_sample() for a more complete
+description. A grid generated by affine_grid() should be passed to grid_sample()
+with the same setting for this option. Default: False \n
+
+*@par Outputs:
+*@li y: A 2-D integer tensor of shape [M] representing the
+selected indices from the boxes tensor, where M <= max_output_size. \n
+
+*@attention Constraints:
+*Input theta must be float16 or float, output_size must be int32 type . \n
+
+*@par Third-party framework compatibility
+*Compatible with Pytorch affine_grid operator.
+*/
+
+REG_OP(AffineGrid)
+    .INPUT(theta, TensorType({DT_FLOAT16, DT_FLOAT}))
+    .INPUT(output_size, TensorType({DT_INT32}))
+    .OUTPUT(y, TensorType({DT_FLOAT16, DT_FLOAT}))
+    .ATTR(align_corners, Bool, false)
+    .OP_END_FACTORY_REG(AffineGrid)
 }  // namespace ge
 
 #endif  // OPS_BUILT_IN_OP_PROTO_INC_TRANSFORMATION_OPS_H_

third_party/fwkacllib/inc/runtime/event.h

@@ -30,6 +30,11 @@ extern "C" {
 #define RT_EVENT_DEFAULT (0x00)
 #define RT_EVENT_WITH_FLAG (0x01)
 
+#define RT_EVENT_DDSYNC_NS    0x01U
+#define RT_EVENT_STREAM_MARK  0x02U
+#define RT_EVENT_DDSYNC       0x04U
+#define RT_EVENT_TIME_LINE    0x08U
+
 /**
  * @ingroup dvrt_event
  * @brief create event instance

third_party/fwkacllib/inc/toolchain/slog.h

@@ -141,7 +141,7 @@ enum {
   IDEDD,         /**< IDE daemon device */
   IDEDH,         /**< IDE daemon host */
   HCCL,          /**< HCCL */
-  FMK,           /**< Framework */
+  FMK,           /**< Adapter */
   HIAIENGINE,    /**< Matrix */
   DVPP,          /**< DVPP */
   RUNTIME,       /**< Runtime */
@@ -162,11 +162,11 @@ enum {
   MDCDEFAULT, /**< MDC undefine */
   MDCSC,      /**< MDC spatial cognition */
   MDCPNC,
-  MLL,
+  MLL,      /**< abandon */
   DEVMM,    /**< Dlog memory managent */
   KERNEL,   /**< Kernel */
   LIBMEDIA, /**< Libmedia */
-  CCECPU,   /**< ai cpu */
+  CCECPU,   /**< aicpu shedule */
   ASCENDDK, /**< AscendDK */
   ROS,      /**< ROS */
   HCCP,
@@ -179,7 +179,7 @@ enum {
   TSDUMP,    /**< TSDUMP module */
   AICPU,     /**< AICPU module */
   LP,        /**< LP module */
-  TDT,
+  TDT,       /**< tsdaemon or aicpu shedule */
   FE,
   MD,
   MB,