[XPU] Add docs for gm_default_size and fix type (#1962)

Add description of gm_default_size and change kunlunxin_gm_default_size‘s type to int64_t
PaddlePaddle · May 18, 2023 · c8ff8b6 · c8ff8b6
1 parent 04ef882
commit c8ff8b6
Show file tree

Hide file tree

Showing 6 changed files with 7 additions and 3 deletions.
diff --git a/c_api/fastdeploy_capi/runtime/runtime_option.cc b/c_api/fastdeploy_capi/runtime/runtime_option.cc
@@ -108,7 +108,7 @@ void FD_C_RuntimeOptionWrapperUseKunlunXin(
                                std::string(precision), 
                                bool(adaptive_seqlen),
                                bool(enable_multi_stream),
-                               gm_default_size);
+                               int64_t(gm_default_size));
 }
 
 void FD_C_RuntimeOptionWrapperUseSophgo(

diff --git a/c_api/fastdeploy_capi/runtime/runtime_option.h b/c_api/fastdeploy_capi/runtime/runtime_option.h
@@ -126,6 +126,7 @@ FASTDEPLOY_CAPI_EXPORT extern void FD_C_RuntimeOptionWrapperUseAscend(
 /// \param[in] adaptive_seqlen Is the input of multi_encoder variable length
 /// \param[in] enable_multi_stream Whether to enable the multi stream of
 ///        KunlunXin XPU.
+/// \param[in] gm_default_size The default size of context global memory of KunlunXin XPU.
 ///
 FASTDEPLOY_CAPI_EXPORT extern void FD_C_RuntimeOptionWrapperUseKunlunXin(
     __fd_keep FD_C_RuntimeOptionWrapper* fd_c_runtime_option_wrapper,

diff --git a/csharp/fastdeploy/runtime_option.cs b/csharp/fastdeploy/runtime_option.cs
@@ -103,6 +103,7 @@ public void UseAscend() {
   /// \param adaptive_seqlen Is the input of multi_encoder variable length
   /// \param enable_multi_stream Whether to enable the multi stream of
   ///        KunlunXin XPU.
+  /// \param gm_default_size The default size of context global memory of KunlunXin XPU.
   ///
   public void
   UseKunlunXin(int kunlunxin_id = 0, int l3_workspace_size = 0xfffc00,

diff --git a/fastdeploy/runtime/backends/lite/option.h b/fastdeploy/runtime/backends/lite/option.h
@@ -75,12 +75,12 @@ struct LiteBackendOption {
   std::string kunlunxin_autotune_file = "";
   /// kunlunxin_precision
   std::string kunlunxin_precision = "int16";
-  /// kunlunxin_gm_default_size 
-  int kunlunxin_gm_default_size = 0;
   /// kunlunxin_adaptive_seqlen
   bool kunlunxin_adaptive_seqlen = false;
   /// kunlunxin_enable_multi_stream
   bool kunlunxin_enable_multi_stream = false;
+  /// kunlunxin_gm_default_size 
+  int64_t kunlunxin_gm_default_size = 0;
 
   /// Optimized model dir for CxxConfig
   std::string optimized_model_dir = "";

diff --git a/fastdeploy/runtime/runtime_option.h b/fastdeploy/runtime/runtime_option.h
@@ -106,6 +106,7 @@ struct FASTDEPLOY_DECL RuntimeOption {
   /// \param adaptive_seqlen Is the input of multi_encoder variable length
   /// \param enable_multi_stream Whether to enable the multi stream of
   ///        KunlunXin XPU.
+  /// \param gm_default_size The default size of global memory of KunlunXin XPU.
   ///
   void UseKunlunXin(int kunlunxin_id = 0, int l3_workspace_size = 0xfffc00,
                     bool locked = false, bool autotune = true,

diff --git a/python/fastdeploy/runtime.py b/python/fastdeploy/runtime.py
@@ -242,6 +242,7 @@ def use_kunlunxin(self,
         :param precision: (str)Calculation accuracy of multi_encoder
         :param adaptive_seqlen: (bool)adaptive_seqlen Is the input of multi_encoder variable length
         :param enable_multi_stream: (bool)Whether to enable the multi stream of KunlunXin XPU.
+        :param gm_default_size The default size of context global memory of KunlunXin XPU.
         """
         return self._option.use_kunlunxin(device_id, l3_workspace_size, locked,
                                           autotune, autotune_file, precision,