Fix to add missing zeKernelGetExp API and header updates

Signed-off-by: Neil R. Spruit <[email protected]>
oneapi-src · Nov 13, 2024 · 5f9d031 · 5f9d031
1 parent 0a681d7
commit 5f9d031
Show file tree

Hide file tree

Showing 23 changed files with 390 additions and 34 deletions.
diff --git a/include/layers/zel_tracing_register_cb.h b/include/layers/zel_tracing_register_cb.h
@@ -1410,6 +1410,33 @@ typedef void (ZE_APICALL *ze_pfnKernelSetGlobalOffsetExpCb_t)(
     void** ppTracerInstanceUserData
     );
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Callback function parameters for zeKernelGetBinaryExp
+/// @details Each entry is a pointer to the parameter passed to the function;
+///     allowing the callback the ability to modify the parameter's value
+
+typedef struct _ze_kernel_get_binary_exp_params_t
+{
+    ze_kernel_handle_t* phKernel;
+    size_t** ppSize;
+    uint8_t** ppKernelBinary;
+} ze_kernel_get_binary_exp_params_t;
+
+
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Callback function-pointer for zeKernelGetBinaryExp
+/// @param[in] params Parameters passed to this instance
+/// @param[in] result Return value
+/// @param[in] pTracerUserData Per-Tracer user data
+/// @param[in,out] ppTracerInstanceUserData Per-Tracer, Per-Instance user data
+
+typedef void (ZE_APICALL *ze_pfnKernelGetBinaryExpCb_t)(
+    ze_kernel_get_binary_exp_params_t* params,
+    ze_result_t result,
+    void* pTracerUserData,
+    void** ppTracerInstanceUserData
+    );
+
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Callback function parameters for zeKernelSchedulingHintExp
 /// @details Each entry is a pointer to the parameter passed to the function;
@@ -3023,6 +3050,14 @@ zelTracerKernelSetGlobalOffsetExpRegisterCallback(
     );
 
 
+ZE_APIEXPORT ze_result_t ZE_APICALL
+zelTracerKernelGetBinaryExpRegisterCallback(
+    zel_tracer_handle_t hTracer,
+    zel_tracer_reg_t callback_type,
+    ze_pfnKernelGetBinaryExpCb_t pfnGetBinaryExpCb
+    );
+
+
 ZE_APIEXPORT ze_result_t ZE_APICALL
 zelTracerDeviceReserveCacheExtRegisterCallback(
     zel_tracer_handle_t hTracer,

diff --git a/include/ze.py b/include/ze.py
@@ -4,7 +4,7 @@
  SPDX-License-Identifier: MIT
 
  @file ze.py
- @version v1.11-r1.11.5
+ @version v1.11-r1.11.9
 
  """
 import platform
@@ -5520,13 +5520,21 @@ class _ze_kernel_dditable_t(Structure):
 else:
     _zeKernelSchedulingHintExp_t = CFUNCTYPE( ze_result_t, ze_kernel_handle_t, POINTER(ze_scheduling_hint_exp_desc_t) )
 
+###############################################################################
+## @brief Function-pointer for zeKernelGetBinaryExp
+if __use_win_types:
+    _zeKernelGetBinaryExp_t = WINFUNCTYPE( ze_result_t, ze_kernel_handle_t, POINTER(c_size_t), POINTER(c_ubyte) )
+else:
+    _zeKernelGetBinaryExp_t = CFUNCTYPE( ze_result_t, ze_kernel_handle_t, POINTER(c_size_t), POINTER(c_ubyte) )
+
 
 ###############################################################################
 ## @brief Table of KernelExp functions pointers
 class _ze_kernel_exp_dditable_t(Structure):
     _fields_ = [
         ("pfnSetGlobalOffsetExp", c_void_p),                            ## _zeKernelSetGlobalOffsetExp_t
-        ("pfnSchedulingHintExp", c_void_p)                              ## _zeKernelSchedulingHintExp_t
+        ("pfnSchedulingHintExp", c_void_p),                             ## _zeKernelSchedulingHintExp_t
+        ("pfnGetBinaryExp", c_void_p)                                   ## _zeKernelGetBinaryExp_t
     ]
 
 ###############################################################################
@@ -6129,6 +6137,7 @@ def __init__(self, version : ze_api_version_t):
         # attach function interface to function address
         self.zeKernelSetGlobalOffsetExp = _zeKernelSetGlobalOffsetExp_t(self.__dditable.KernelExp.pfnSetGlobalOffsetExp)
         self.zeKernelSchedulingHintExp = _zeKernelSchedulingHintExp_t(self.__dditable.KernelExp.pfnSchedulingHintExp)
+        self.zeKernelGetBinaryExp = _zeKernelGetBinaryExp_t(self.__dditable.KernelExp.pfnGetBinaryExp)
 
         # call driver to get function pointers
         _Sampler = _ze_sampler_dditable_t()

diff --git a/include/ze_api.h b/include/ze_api.h
@@ -5,7 +5,7 @@
  * SPDX-License-Identifier: MIT
  *
  * @file ze_api.h
- * @version v1.11-r1.11.5
+ * @version v1.11-r1.11.9
  *
  */
 #ifndef _ZE_API_H
@@ -968,8 +968,9 @@ typedef enum _ze_init_flag_t
 /// @brief Initialize the 'oneAPI' driver(s)
 /// 
 /// @details
-///     - The application must call this function before calling any other
-///       function.
+///     - @deprecated since 1.10. Please use zeInitDrivers()
+///     - The application must call this function or zeInitDrivers before
+///       calling any other function.
 ///     - If this function is not called then all other functions will return
 ///       ::ZE_RESULT_ERROR_UNINITIALIZED.
 ///     - Only one instance of each driver will be initialized per process.
@@ -999,6 +1000,10 @@ zeInit(
 /// @brief Retrieves driver instances
 /// 
 /// @details
+///     - @deprecated since 1.10. Please use zeInitDrivers()
+///     - Usage of zeInitDrivers and zeDriverGet is mutually exclusive and
+///       should not be used together. Usage of them together will result in
+///       undefined behavior.
 ///     - A driver represents a collection of physical devices.
 ///     - Multiple calls to this function will return identical driver handles,
 ///       in the same order.
@@ -1075,20 +1080,20 @@ typedef struct _ze_init_driver_type_desc_t
 /// 
 /// @details
 ///     - The application must call this function or zeInit before calling any
-///       other function.
-///     - The application can call InitDrivers or zeInit to init the drivers on
-///       the system.
-///     - Calls to zeInit or InitDrivers will not alter the drivers retrieved
-///       thru either api.
-///     - Drivers init thru zeInit or InitDrivers will not be reInitialized once
-///       init in an application. The Loader will determine if the already init
-///       driver needs to be delivered to the user thru the init type flags.
+///       other function. (zeInit is [Deprecated] and is replaced by
+///       zeInitDrivers)
+///     - Calls to zeInit[Deprecated] or InitDrivers will not alter the drivers
+///       retrieved thru either api.
+///     - Drivers init thru zeInit[Deprecated] or InitDrivers will not be
+///       reInitialized once init in an application. The Loader will determine
+///       if the already init driver needs to be delivered to the user thru the
+///       init type flags.
 ///     - Already init Drivers will not be uninitialized if the call to
 ///       InitDrivers does not include that driver's type. Those init drivers
 ///       which don't match the init flags will not have their driver handles
 ///       returned to the user in that InitDrivers call.
-///     - If this function or zeInit is not called, then all other functions
-///       will return ::ZE_RESULT_ERROR_UNINITIALIZED.
+///     - If this function or zeInit[Deprecated] is not called, then all other
+///       functions will return ::ZE_RESULT_ERROR_UNINITIALIZED.
 ///     - Only one instance of each driver will be initialized per process.
 ///     - A driver represents a collection of physical devices.
 ///     - Multiple calls to this function will return identical driver handles,
@@ -7659,6 +7664,34 @@ typedef struct _ze_relaxed_allocation_limits_exp_desc_t
 #define ZE_GET_KERNEL_BINARY_EXP_NAME  "ZE_extension_kernel_binary_exp"
 #endif // ZE_GET_KERNEL_BINARY_EXP_NAME
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Retrieves kernel binary program data (ISA GEN format).
+/// 
+/// @details
+///     - A valid kernel handle must be created with ::zeKernelCreate.
+///     - Returns Intel Graphics Assembly (GEN ISA) format binary program data
+///       for kernel handle.
+///     - The application may call this function from simultaneous threads.
+///     - The implementation of this function must be thread-safe.
+/// 
+/// @returns
+///     - ::ZE_RESULT_SUCCESS
+///     - ::ZE_RESULT_ERROR_UNINITIALIZED
+///     - ::ZE_RESULT_ERROR_DEVICE_LOST
+///     - ::ZE_RESULT_ERROR_OUT_OF_HOST_MEMORY
+///     - ::ZE_RESULT_ERROR_OUT_OF_DEVICE_MEMORY
+///     - ::ZE_RESULT_ERROR_INVALID_NULL_HANDLE
+///         + `nullptr == hKernel`
+///     - ::ZE_RESULT_ERROR_INVALID_NULL_POINTER
+///         + `nullptr == pSize`
+///         + `nullptr == pKernelBinary`
+ZE_APIEXPORT ze_result_t ZE_APICALL
+zeKernelGetBinaryExp(
+    ze_kernel_handle_t hKernel,                                             ///< [in] Kernel handle.
+    size_t* pSize,                                                          ///< [in,out] pointer to variable with size of GEN ISA binary.
+    uint8_t* pKernelBinary                                                  ///< [in,out] pointer to storage area for GEN ISA binary function.
+    );
+
 #if !defined(__GNUC__)
 #pragma endregion
 #endif

diff --git a/include/ze_ddi.h b/include/ze_ddi.h
@@ -5,7 +5,7 @@
  * SPDX-License-Identifier: MIT
  *
  * @file ze_ddi.h
- * @version v1.11-r1.11.5
+ * @version v1.11-r1.11.9
  *
  */
 #ifndef _ZE_DDI_H
@@ -2196,12 +2196,21 @@ typedef ze_result_t (ZE_APICALL *ze_pfnKernelSchedulingHintExp_t)(
     ze_scheduling_hint_exp_desc_t*
     );
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Function-pointer for zeKernelGetBinaryExp 
+typedef ze_result_t (ZE_APICALL *ze_pfnKernelGetBinaryExp_t)(
+    ze_kernel_handle_t,
+    size_t*,
+    uint8_t*
+    );
+
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Table of KernelExp functions pointers
 typedef struct _ze_kernel_exp_dditable_t
 {
     ze_pfnKernelSetGlobalOffsetExp_t                            pfnSetGlobalOffsetExp;
     ze_pfnKernelSchedulingHintExp_t                             pfnSchedulingHintExp;
+    ze_pfnKernelGetBinaryExp_t                                  pfnGetBinaryExp;
 } ze_kernel_exp_dditable_t;
 
 ///////////////////////////////////////////////////////////////////////////////

diff --git a/include/zes.py b/include/zes.py
@@ -4,7 +4,7 @@
  SPDX-License-Identifier: MIT
 
  @file zes.py
- @version v1.11-r1.11.5
+ @version v1.11-r1.11.9
 
  """
 import platform

diff --git a/include/zes_api.h b/include/zes_api.h
@@ -5,7 +5,7 @@
  * SPDX-License-Identifier: MIT
  *
  * @file zes_api.h
- * @version v1.11-r1.11.5
+ * @version v1.11-r1.11.9
  *
  */
 #ifndef _ZES_API_H

diff --git a/include/zes_ddi.h b/include/zes_ddi.h
@@ -5,7 +5,7 @@
  * SPDX-License-Identifier: MIT
  *
  * @file zes_ddi.h
- * @version v1.11-r1.11.5
+ * @version v1.11-r1.11.9
  *
  */
 #ifndef _ZES_DDI_H

diff --git a/include/zet.py b/include/zet.py
@@ -4,7 +4,7 @@
  SPDX-License-Identifier: MIT
 
  @file zet.py
- @version v1.11-r1.11.5
+ @version v1.11-r1.11.9
 
  """
 import platform

diff --git a/include/zet_api.h b/include/zet_api.h
@@ -5,7 +5,7 @@
  * SPDX-License-Identifier: MIT
  *
  * @file zet_api.h
- * @version v1.11-r1.11.5
+ * @version v1.11-r1.11.9
  *
  */
 #ifndef _ZET_API_H

diff --git a/include/zet_ddi.h b/include/zet_ddi.h
@@ -5,7 +5,7 @@
  * SPDX-License-Identifier: MIT
  *
  * @file zet_ddi.h
- * @version v1.11-r1.11.5
+ * @version v1.11-r1.11.9
  *
  */
 #ifndef _ZET_DDI_H

diff --git a/source/drivers/null/ze_nullddi.cpp b/source/drivers/null/ze_nullddi.cpp
@@ -3969,6 +3969,31 @@ namespace driver
         return result;
     }
 
+    ///////////////////////////////////////////////////////////////////////////////
+    /// @brief Intercept function for zeKernelGetBinaryExp
+    __zedlllocal ze_result_t ZE_APICALL
+    zeKernelGetBinaryExp(
+        ze_kernel_handle_t hKernel,                     ///< [in] Kernel handle.
+        size_t* pSize,                                  ///< [in,out] pointer to variable with size of GEN ISA binary.
+        uint8_t* pKernelBinary                          ///< [in,out] pointer to storage area for GEN ISA binary function.
+        )
+    {
+        ze_result_t result = ZE_RESULT_SUCCESS;
+
+        // if the driver has created a custom function, then call it instead of using the generic path
+        auto pfnGetBinaryExp = context.zeDdiTable.KernelExp.pfnGetBinaryExp;
+        if( nullptr != pfnGetBinaryExp )
+        {
+            result = pfnGetBinaryExp( hKernel, pSize, pKernelBinary );
+        }
+        else
+        {
+            // generic implementation
+        }
+
+        return result;
+    }
+
     ///////////////////////////////////////////////////////////////////////////////
     /// @brief Intercept function for zeDeviceReserveCacheExt
     __zedlllocal ze_result_t ZE_APICALL
@@ -5876,6 +5901,8 @@ zeGetKernelExpProcAddrTable(
 
     pDdiTable->pfnSetGlobalOffsetExp                     = driver::zeKernelSetGlobalOffsetExp;
 
+    pDdiTable->pfnGetBinaryExp                           = driver::zeKernelGetBinaryExp;
+
     pDdiTable->pfnSchedulingHintExp                      = driver::zeKernelSchedulingHintExp;
 
     return result;

diff --git a/source/layers/tracing/ze_tracing_cb_structs.h b/source/layers/tracing/ze_tracing_cb_structs.h
@@ -235,6 +235,7 @@ typedef struct _zel_kernel_callbacks_t
     ze_pfnKernelGetPropertiesCb_t                                   pfnGetPropertiesCb;
     ze_pfnKernelGetNameCb_t                                         pfnGetNameCb;
     ze_pfnKernelSetGlobalOffsetExpCb_t                              pfnSetGlobalOffsetExpCb;
+    ze_pfnKernelGetBinaryExpCb_t                                    pfnGetBinaryExpCb;
     ze_pfnKernelSchedulingHintExpCb_t                               pfnSchedulingHintExpCb;
 } zel_kernel_callbacks_t;
 

diff --git a/source/layers/tracing/ze_tracing_register_cb.cpp b/source/layers/tracing/ze_tracing_register_cb.cpp
@@ -2355,6 +2355,22 @@ zelTracerKernelSetGlobalOffsetExpRegisterCallback(
 }
 
 
+ZE_DLLEXPORT ze_result_t ZE_APICALL
+zelTracerKernelGetBinaryExpRegisterCallback(
+    zel_tracer_handle_t hTracer,
+    zel_tracer_reg_t callback_type,
+    ze_pfnKernelGetBinaryExpCb_t pfnGetBinaryExpCb
+    ) {
+
+    ze_result_t result;
+    auto& cbs = tracing_layer::APITracer::fromHandle(hTracer)->getProEpilogues(callback_type, result);
+    if (result == ZE_RESULT_SUCCESS)
+        cbs.Kernel.pfnGetBinaryExpCb = pfnGetBinaryExpCb;
+
+    return result;
+}
+
+
 ZE_DLLEXPORT ze_result_t ZE_APICALL
 zelTracerDeviceReserveCacheExtRegisterCallback(
     zel_tracer_handle_t hTracer,

diff --git a/source/layers/tracing/ze_trcddi.cpp b/source/layers/tracing/ze_trcddi.cpp
@@ -5827,6 +5827,44 @@ namespace tracing_layer
                                                   *tracerParams.poffsetZ);
     }
 
+    ///////////////////////////////////////////////////////////////////////////////
+    /// @brief Intercept function for zeKernelGetBinaryExp
+    __zedlllocal ze_result_t ZE_APICALL
+    zeKernelGetBinaryExp(
+        ze_kernel_handle_t hKernel,                     ///< [in] Kernel handle.
+        size_t* pSize,                                  ///< [in,out] pointer to variable with size of GEN ISA binary.
+        uint8_t* pKernelBinary                          ///< [in,out] pointer to storage area for GEN ISA binary function.
+        )
+    {
+        auto pfnGetBinaryExp = context.zeDdiTable.KernelExp.pfnGetBinaryExp;
+
+        if( nullptr == pfnGetBinaryExp)
+            return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
+
+        ZE_HANDLE_TRACER_RECURSION(context.zeDdiTable.KernelExp.pfnGetBinaryExp, hKernel, pSize, pKernelBinary);
+
+        // capture parameters
+        ze_kernel_get_binary_exp_params_t tracerParams = {
+            &hKernel,
+            &pSize,
+            &pKernelBinary
+        };
+
+        tracing_layer::APITracerCallbackDataImp<ze_pfnKernelGetBinaryExpCb_t> apiCallbackData;
+
+        ZE_GEN_PER_API_CALLBACK_STATE(apiCallbackData, ze_pfnKernelGetBinaryExpCb_t, Kernel, pfnGetBinaryExpCb);
+
+
+        return tracing_layer::APITracerWrapperImp(context.zeDdiTable.KernelExp.pfnGetBinaryExp,
+                                                  &tracerParams,
+                                                  apiCallbackData.apiOrdinal,
+                                                  apiCallbackData.prologCallbacks,
+                                                  apiCallbackData.epilogCallbacks,
+                                                  *tracerParams.phKernel,
+                                                  *tracerParams.ppSize,
+                                                  *tracerParams.ppKernelBinary);
+    }
+
     ///////////////////////////////////////////////////////////////////////////////
     /// @brief Intercept function for zeDeviceReserveCacheExt
     __zedlllocal ze_result_t ZE_APICALL
@@ -8470,6 +8508,9 @@ zeGetKernelExpProcAddrTable(
     dditable.pfnSetGlobalOffsetExp                       = pDdiTable->pfnSetGlobalOffsetExp;
     pDdiTable->pfnSetGlobalOffsetExp                     = tracing_layer::zeKernelSetGlobalOffsetExp;
 
+    dditable.pfnGetBinaryExp                             = pDdiTable->pfnGetBinaryExp;
+    pDdiTable->pfnGetBinaryExp                           = tracing_layer::zeKernelGetBinaryExp;
+
     dditable.pfnSchedulingHintExp                        = pDdiTable->pfnSchedulingHintExp;
     pDdiTable->pfnSchedulingHintExp                      = tracing_layer::zeKernelSchedulingHintExp;