Pre Merge pull request !2115 from yanghaoran/r1.7

inc/external/acl/error_codes/rt_error_codes.h

@@ -45,6 +45,7 @@ static const int32_t ACL_ERROR_RT_INVALID_MEMORY_TYPE = 107016;       // invalid
 static const int32_t ACL_ERROR_RT_INVALID_HANDLE = 107017;            // invalid handle
 static const int32_t ACL_ERROR_RT_INVALID_MALLOC_TYPE = 107018;       // invalid malloc type
 static const int32_t ACL_ERROR_RT_WAIT_TIMEOUT = 107019;              // wait timeout
+static const int32_t ACL_ERROR_RT_TASK_TIMEOUT = 107020;              // task timeout
 
 static const int32_t ACL_ERROR_RT_FEATURE_NOT_SUPPORT = 207000;  // feature not support
 static const int32_t ACL_ERROR_RT_MEMORY_ALLOCATION = 207001;    // memory allocation error

inc/external/hccl/hccl_types.h

@@ -61,8 +61,7 @@ typedef enum {
  * @brief handle to HCCL communicator
  */
 typedef void *HcclComm;
-typedef void *HcclMessage;
-typedef void *HcclRequest;
+
 /**
  * @brief HCCL Reduction opperation
  */
@@ -88,14 +87,6 @@ typedef enum {
   HCCL_DATA_TYPE_RESERVED    /**< reserved */
 } HcclDataType;
 
-typedef struct {
-  int srcRank;  // 接收/探测到的msg/信封的发送端rank_id，MPI标准定义，调用者可以访问
-  int tag;      // 接收/探测到的msg/信封的tag，MPI标准定义，调用者可以访问
-  int error;  // 接收/探测的错误码0：no error，others：传输过程出错，MPI标准定义，调用者可以访问
-  int cancelled;  // 指定实现，不建议调用者访问
-  int count;      // 接收/探测到的payload大小，指定实现，不建议调用者访问
-} HcclStatus;
-
 const uint32_t HCCL_ROOT_INFO_BYTES = 4108;  // 4108: root info length
 /**
  * @brief HCCL root info
@@ -104,7 +95,6 @@ typedef struct HcclRootInfoDef {
   char internal[HCCL_ROOT_INFO_BYTES];
 } HcclRootInfo;
 
-#define HCCL_REQUEST_NULL NULL
 #ifdef __cplusplus
 }
 #endif  // __cplusplus

inc/external/runtime/rt_error_codes.h

@@ -45,6 +45,7 @@ static const int32_t ACL_ERROR_RT_INVALID_MEMORY_TYPE = 107016;       // invalid
 static const int32_t ACL_ERROR_RT_INVALID_HANDLE = 107017;            // invalid handle
 static const int32_t ACL_ERROR_RT_INVALID_MALLOC_TYPE = 107018;       // invalid malloc type
 static const int32_t ACL_ERROR_RT_WAIT_TIMEOUT = 107019;              // wait timeout
+static const int32_t ACL_ERROR_RT_TASK_TIMEOUT = 107020;              // task timeout
 
 static const int32_t ACL_ERROR_RT_FEATURE_NOT_SUPPORT = 207000;  // feature not support
 static const int32_t ACL_ERROR_RT_MEMORY_ALLOCATION = 207001;    // memory allocation error

inc/framework/common/profiling_definitions.h

@@ -72,6 +72,14 @@ enum {
   kSelectBranch,
   kExecuteSubGraph,
   kInitSubGraphExecutor,
+  // fuzz compile
+  kSelectBin,
+  kFindCompileCache,
+  kAddCompileCache,
+  kFuzzCompileOp,
+  kCalcRuningParam,
+  kGenTask,
+  kRegisterBin,
 
   // Add new definitions here
   kProfilingIndexEnd

metadef

@@ -1 +1 @@
-Subproject commit e0efffc740a79d49ba0553478b51d9d3481771cb
+Subproject commit 22309b14838a763d41dccd636fec567dae3720fd

third_party/fwkacllib/inc/hccl/base.h

@@ -197,6 +197,20 @@ typedef struct tagCommAttr {
     WorkMode mode;  // 通信域内的probe工作模式
     uint32_t deviceId = 0;
 } CommAttr;
+
+typedef void* HcclMessage;
+typedef void* HcclRequest;
+
+typedef struct {
+    int srcRank;    // 接收/探测到的msg/信封的发送端rank_id，MPI标准定义，调用者可以访问
+    int tag;        // 接收/探测到的msg/信封的tag，MPI标准定义，调用者可以访问
+    int error;      // 接收/探测的错误码0：no error，others：传输过程出错，MPI标准定义，调用者可以访问
+    int cancelled;  // 指定实现，不建议调用者访问
+    int count;      // 接收/探测到的payload大小，指定实现，不建议调用者访问
+} HcclStatus;
+
+#define HCCL_REQUEST_NULL   NULL
+
 #ifdef __cplusplus
 }
 #endif // __cplusplus

third_party/fwkacllib/inc/ops/elewise_calculation_ops.h

@@ -3488,7 +3488,7 @@ REG_OP(Addcmul)
 REG_OP(AxpyV2)
     .INPUT(x1, TensorType({DT_FLOAT16, DT_FLOAT, DT_INT32}))
     .INPUT(x2, TensorType({DT_FLOAT16, DT_FLOAT, DT_INT32}))
-    .INPUT(alpha, TensorType({DT_FLOAT16, DT_FLOAT}))
+    .INPUT(alpha, TensorType({DT_FLOAT16, DT_FLOAT, DT_INT32}))
     .OUTPUT(y, TensorType({DT_FLOAT16, DT_FLOAT, DT_INT32}))
     .OP_END_FACTORY_REG(AxpyV2)
 

third_party/fwkacllib/inc/ops/image_ops.h

@@ -1737,17 +1737,17 @@ round_prefer_ceil, floor, ceil. Only used by nearest interpolation.
 */
 
 REG_OP(Resize)
-    .INPUT(x, TensorType({DT_INT8, DT_UINT8, DT_INT16, DT_UINT16, DT_INT32,
-                                DT_INT64, DT_FLOAT16, DT_FLOAT, DT_DOUBLE}))
-    .INPUT(roi, TensorType({DT_FLOAT16, DT_FLOAT, DT_DOUBLE}))
-    .INPUT(scales, TensorType({DT_FLOAT}))
-    .OPTIONAL_INPUT(sizes, TensorType({DT_INT64}))
-    .OUTPUT(y, TensorType({DT_INT8, DT_UINT8, DT_INT16, DT_UINT16, DT_INT32,
-                                DT_INT64, DT_FLOAT16, DT_FLOAT, DT_DOUBLE}))
+    .INPUT(x, TensorType({DT_INT8,DT_UINT8,DT_INT16,DT_UINT16,DT_INT32,
+                          DT_INT64,DT_FLOAT16,DT_FLOAT,DT_DOUBLE}))
+    .OPTIONAL_INPUT(roi, TensorType({DT_FLOAT16,DT_FLOAT,DT_DOUBLE}))
+    .OPTIONAL_INPUT(scales, TensorType({DT_FLOAT}))
+    .OPTIONAL_INPUT(sizes, TensorType({DT_INT64,DT_INT32}))
+    .OUTPUT(y, TensorType({DT_INT8,DT_UINT8,DT_INT16,DT_UINT16,DT_INT32,
+                           DT_INT64,DT_FLOAT16,DT_FLOAT,DT_DOUBLE}))
     .ATTR(coordinate_transformation_mode, String, "half_pixel")
     .ATTR(cubic_coeff_a, Float, -0.75)
     .ATTR(exclude_outside, Int, 0)
-    .ATTR(extrapolation_value, Float, 0)
+    .ATTR(extrapolation_value, Float, 0.0)
     .ATTR(mode, String, "nearest")
     .ATTR(nearest_mode, String, "round_prefer_floor")
     .OP_END_FACTORY_REG(Resize)

third_party/fwkacllib/inc/ops/nn_norm_ops.h

@@ -1801,7 +1801,7 @@ REG_OP(SoftmaxCrossEntropyLoss)
 * @par Restrictions:
 * Warning: THIS FUNCTION IS EXPERIMENTAL. Please do not use.
 */
-REG_OP(AxpyWithSoftmaxAndDropoutdomask)
+REG_OP(AxpyWithSoftmaxAndDropOutDoMask)
     .INPUT(x1, TensorType({DT_FLOAT16}))
     .INPUT(x2, TensorType({DT_FLOAT16}))
     .INPUT(mask, TensorType({DT_UINT8}))
@@ -1810,6 +1810,6 @@ REG_OP(AxpyWithSoftmaxAndDropoutdomask)
     .REQUIRED_ATTR(alpha, Float)
     .REQUIRED_ATTR(input_keep_prob, Float)
     .ATTR(axis, ListInt, {-1})
-    .OP_END_FACTORY_REG(AxpyWithSoftmaxAndDropoutdomask)
+    .OP_END_FACTORY_REG(AxpyWithSoftmaxAndDropOutDoMask)
 }  // namespace ge
 #endif  // OPS_BUILT_IN_OP_PROTO_INC_NN_NORM_OPS_H_

third_party/fwkacllib/inc/ops/nn_ops.h

@@ -116,6 +116,7 @@ REG_OP(FusedBatchNormV2)
  * @par Outputs:
  * One output, including:
  * output_proposal: A Tensor. Datatype and format is same as input_data. Proposal sorted for each channel.
+ * @par Restrictions:
  * Warning: THIS FUNCTION IS EXPERIMENTAL. Please do not use.
  */
 REG_OP(SegmentSort)
@@ -137,6 +138,7 @@ REG_OP(SegmentSort)
  * Two output, including:
  * output_proposal: A Tensor. Datatype and format is same as input_data. Proposal sorted for each channel.
  * output_index: A Tensor.If include_index is true, output index.
+ * @par Restrictions:
  * Warning: THIS FUNCTION IS EXPERIMENTAL. Please do not use.
  */
 REG_OP(MultiMerge)
@@ -148,6 +150,27 @@ REG_OP(MultiMerge)
     .OP_END_FACTORY_REG(MultiMerge)
 
 /**
+ * @brief Large amount of data sort.Third operator of TopK.
+ * @par Inputs:
+ * One input, including:
+ * input_proposal: A Tensor. Proposal sorted for each channel. Support float16
+ * @par Attributes:
+ * k_num: Int.Number to be sorted.
+ * @par Outputs:
+ * Two output, including:
+ * @li output_data: A Tensor. Datatype and format is same as input_data. Data sorted.
+ * @li output_index: A Tensor. int32. Data index.
+ * @par Restrictions:
+ * Warning: THIS FUNCTION IS EXPERIMENTAL. Please do not use.
+ */
+REG_OP(SingleMerge)
+    .INPUT(input_proposal, TensorType({DT_FLOAT16}))
+    .OUTPUT(output_data, TensorType({DT_FLOAT16}))
+    .OUTPUT(output_index, TensorType({DT_INT32}))
+    .REQUIRED_ATTR(k_num, Int)
+    .OP_END_FACTORY_REG(SingleMerge)
+
+/**
  * @brief MultiHeadAttention.
  * @par Inputs:
  * thirteen input, including:

third_party/fwkacllib/inc/ops/reduce_ops.h

@@ -663,6 +663,9 @@ REG_OP(ReduceProdD)
 *keep_dims: A bool or NoneType.
 * - If true, retains reduced dimensions with length 1.
 * - If false, the rank of the tensor is reduced by 1 for each entry in axis.
+*noop_with_empty_axes: A bool.
+* - If true, when axes = [], not reduce.
+* - If false, when axes = [], reduce all.
 *@par Outputs:
 *y: A Tensor. Has the same type as "x" . \n
 
@@ -674,6 +677,7 @@ REG_OP(ReduceMean)
     .INPUT(axes, TensorType::IndexNumberType())
     .OUTPUT(y, TensorType::NumberType())
     .ATTR(keep_dims, Bool, false)
+    .ATTR(noop_with_empty_axes, Bool, true)
     .OP_END_FACTORY_REG(ReduceMean)
 
 /**