P4 to Git Change 1123271 by wchau@wchau_WINDOWS7_OCL on 2015/02/18 16:59:12

ECR #399840 - OpenCL Runtime HW Debug support development - add install_trap API to avoid the expose of runtime trap handler to the DBE.

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_context.cpp#41 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_debugger_amd.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_debugger_amd.h#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudebugmanager.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hwdebug.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hwdebug.hpp#5 edit


[ROCm/clr commit: 11b7e95416]

projects/clr/opencl/api/opencl/amdocl/cl_context.cpp

@@ -419,7 +419,7 @@ RUNTIME_ENTRY(cl_int, clGetContextInfo, (
         }
         return CL_SUCCESS;
     }
-    case CL_CONTEXT_ADAPTER_D3D9EX_KHR: { 
+    case CL_CONTEXT_ADAPTER_D3D9EX_KHR: {
         if (param_value != NULL && param_value_size < sizeof(void*)) {
             return CL_INVALID_VALUE;
         }
@@ -534,7 +534,7 @@ clGetExtensionFunctionAddress(const char* func_name)
         CL_EXTENSION_ENTRYPOINT_CHECK(clEnqueueAcquireD3D10ObjectsKHR);
         CL_EXTENSION_ENTRYPOINT_CHECK(clEnqueueReleaseD3D10ObjectsKHR);
         CL_EXTENSION_ENTRYPOINT_CHECK(clEnqueueAcquireDX9MediaSurfacesKHR);
-        CL_EXTENSION_ENTRYPOINT_CHECK(clEnqueueReleaseDX9MediaSurfacesKHR);        
+        CL_EXTENSION_ENTRYPOINT_CHECK(clEnqueueReleaseDX9MediaSurfacesKHR);
 #endif //_WIN32
 #if cl_amd_open_video
         CL_EXTENSION_ENTRYPOINT_CHECK(clEnqueueRunVideoProgramAMD);
@@ -590,6 +590,7 @@ clGetExtensionFunctionAddress(const char* func_name)
         CL_EXTENSION_ENTRYPOINT_CHECK(clHwDbgUnmapScratchRingAMD);
         CL_EXTENSION_ENTRYPOINT_CHECK(clHwDbgGetKernelParamMemAMD);
         CL_EXTENSION_ENTRYPOINT_CHECK(clHwDbgSetGlobalMemoryAMD);
+        CL_EXTENSION_ENTRYPOINT_CHECK(clHwDbgInstallTrapAMD);
 #endif //_WIN32
         break;
     case 'I':

projects/clr/opencl/api/opencl/amdocl/cl_debugger_amd.cpp

@@ -911,6 +911,46 @@ RUNTIME_ENTRY(cl_int, clHwDbgSetGlobalMemoryAMD, (
 RUNTIME_EXIT
 
 
+/*! \brief Install the trap handler of a given type
+ *
+ *  \param device      specifies the device to be used
+ *
+ *  \param trapType    is the type of trap handler
+ *
+ *  \param trapHandler is the pointer of trap handler (TBA)
+ *
+ *  \param trapBuffer  is the pointer of trap handler buffer (TMA)
+ *
+ *  \return One of the following values:
+ *  - CL_SUCCESS if the event occurs before the timeout
+ *  - CL_INVALID_DEVICE if the device is not valid
+ *  - CL_HWDBG_MANAGER_NOT_AVAILABLE_AMD if there is no HW DEBUG manager
+ */
+RUNTIME_ENTRY(cl_int, clHwDbgInstallTrapAMD, (
+    cl_device_id            device,
+    cl_dbg_trap_type_amd    trapType,
+    cl_mem                  trapHandler,
+    cl_mem                  trapBuffer))
+{
+    if (!is_valid(device)) {
+        return CL_INVALID_DEVICE;
+    }
+
+    amd::HwDebugManager * debugManager = as_amd(device)->hwDebugMgr();
+    if (NULL == debugManager) {
+        return CL_HWDBG_MANAGER_NOT_AVAILABLE_AMD;
+    }
+
+    amd::Memory* pTrapHandler = as_amd(trapHandler);
+    amd::Memory* pTrapBuffer = as_amd(trapBuffer);
+    debugManager->installTrap(trapType, pTrapHandler, pTrapBuffer);
+
+    return CL_SUCCESS;
+}
+
+RUNTIME_EXIT
+
+
 /*! @}
  *  @}
  */

projects/clr/opencl/api/opencl/amdocl/cl_debugger_amd.h

@@ -114,6 +114,16 @@ extern "C" {
 
 typedef uintptr_t     cl_dbg_event_amd;  //! debug event
 
+/*!  \brief  Trap Handler Type
+ *
+ *   The trap handler for each support type.
+ */
+enum cl_dbg_trap_type_amd
+{
+    CL_DBG_DEBUG_TRAP  = 0,    //! HW debug
+    CL_DBG_MAX_TRAP
+};
+
 /*!  \brief  Wave actions used to control the wave execution on the hardware
  *
  *   The wave action enumerations are used to specify the desired
@@ -245,7 +255,7 @@ typedef struct _cl_dispatch_debug_info_amd
  */
 typedef struct _cl_aql_packet_info_amd
 {
-    unsigned int trapReservedVgprIndex;     //! VGPR index reserved for trap
+    cl_uint trapReservedVgprIndex;          //! VGPR index reserved for trap
                                             //!   value is -1 when kernel was not compiled
                                             //!   in debug mode.
     cl_uint scratchBufferWaveOffset;        //! scratch buffer wave offset
@@ -285,7 +295,6 @@ typedef void * (*cl_PreDispatchCallBackFunctionAMD) ( cl_device_id device,
                                                       void *ocl_event_handle,
                                                       const void *aql_packet,
                                                       void *acl_binary,
-                                                      cl_ulong *device_trap_buffer,
                                                       void *user_args);
 
 /*!  \brief Post-dispatch call back function signature
@@ -761,6 +770,36 @@ extern CL_API_ENTRY cl_int CL_API_CALL clHwDbgSetGlobalMemoryAMD(
     cl_uint                             /* size */
 ) CL_API_SUFFIX__VERSION_2_0;
 
+
+/*! \brief Install the trap handler of a given type
+ *
+ *  \param device      specifies the device to be used
+ *
+ *  \param trapType    is the type of trap handler
+ *
+ *  \param trapHandler is the pointer of trap handler (TBA)
+ *
+ *  \param trapBuffer  is the pointer of trap handler buffer (TMA)
+ *
+ *  \param trapHandlerSize   size (in bytes) of the trap handler
+ *
+ *  \param trapBufferSize    size (in bytes) of the trap handler buffer
+ *
+ *  \return One of the following values:
+ *  - CL_SUCCESS if the event occurs before the timeout
+ *  - CL_INVALID_DEVICE if the device is not valid
+ *  - CL_INVALID_VALUE if trapHandler is NULL or trapHandlerSize <= 0
+ *  - CL_HWDBG_MANAGER_NOT_AVAILABLE_AMD if there is no HW DEBUG manager
+ */
+extern CL_API_ENTRY cl_int CL_API_CALL clHwDbgInstallTrapAMD(
+    cl_device_id                        /* device */,
+    cl_dbg_trap_type_amd                /* trapType */,
+    cl_mem                              /* trapHandler */,
+    cl_mem                              /* trapBuffer */
+) CL_API_SUFFIX__VERSION_2_0;
+
+
+
 #ifdef __cplusplus
 } /*extern "C"*/
 #endif /*__cplusplus*/