P4 to Git Change 1702079 by jujiang@JJ-OCL-w8 on 2018/11/02 16:01:11

SWDEV-155310 - Request for OpenCL extension function to set stable pstate

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_context.cpp#60 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_profile_amd.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_profile_amd.h#3 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#324 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#600 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.hpp#170 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#115 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.hpp#35 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#102 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#32 edit

opencl/api/opencl/amdocl/cl_context.cpp

@@ -574,6 +574,7 @@ CL_API_ENTRY void* CL_API_CALL clGetExtensionFunctionAddress(const char* func_na
       break;
     case 'S':
       CL_EXTENSION_ENTRYPOINT_CHECK(clSetThreadTraceParamAMD);
+      CL_EXTENSION_ENTRYPOINT_CHECK(clSetDeviceClockModeAMD);
       break;
     case 'U':
       CL_EXTENSION_ENTRYPOINT_CHECK(clUnloadPlatformAMD);

opencl/api/opencl/amdocl/cl_profile_amd.cpp

@@ -6,6 +6,7 @@
 #include "platform/context.hpp"
 #include "platform/command.hpp"
 #include "platform/perfctr.hpp"
+#include "device/device.hpp"
 #include <cstring>
 
 /*! \addtogroup API
@@ -347,6 +348,22 @@ RUNTIME_ENTRY(cl_int, clGetPerfCounterInfoAMD,
 }
 RUNTIME_EXIT
 
+RUNTIME_ENTRY(cl_int, clSetDeviceClockModeAMD,
+              (cl_device_id device, cl_set_device_clock_mode_input_amd set_clock_mode_input,
+               cl_set_device_clock_mode_output_amd* set_clock_mode_output)) {
+  // Make sure we have a valid device object
+  if (!is_valid(device)) {
+    return CL_INVALID_DEVICE;
+  }
+  if (set_clock_mode_input.clock_mode >= CL_DEVICE_CLOCK_MODE_COUNT_AMD) {
+    return CL_INVALID_VALUE;
+  }
+  amd::Device* amdDevice = as_amd(device);
+  bool ret = amdDevice->SetClockMode(set_clock_mode_input, set_clock_mode_output);
+  return (ret == true)? CL_SUCCESS : CL_INVALID_OPERATION;
+}
+RUNTIME_EXIT
+
 /*! @}
  *  @}
  */

opencl/api/opencl/amdocl/cl_profile_amd.h

@@ -114,6 +114,35 @@ enum PerfcounterInfo {
   CL_PERFCOUNTER_LAST
 };
 
+/*********************************
+* Set device clock mode data
+*********************************/
+enum cl_DeviceClockMode_AMD {
+  CL_DEVICE_CLOCK_MODE_DEFAULT_AMD = 0x0, /*Device clocks and other power settings are restored to default*/
+  CL_DEVICE_CLOCK_MODE_QUERY_AMD = 0x1, /*Queries the current device clock ratios. Leaves the clock mode of the device unchanged*/
+  CL_DEVICE_CLOCK_MODE_PROFILING_AMD = 0x2, /*Scale down from peak ratio*/
+  CL_DEVICE_CLOCK_MODE_MINIMUMMEMORY_AMD = 0x3, /* Memory clock is set to the lowest available level*/
+  CL_DEVICE_CLOCK_MODE_MINIMUMENGINE_AMD = 0x4, /*Engine clock is set to the lowest available level*/
+  CL_DEVICE_CLOCK_MODE_PEAK_AMD = 0x5, /*Clocks set to maximum when possible. Fan set to maximum.*/
+  CL_DEVICE_CLOCK_MODE_QUERYPROFILING_AMD = 0x6, /*Queries the profiling device clock ratios. Leaves the clock mode of the device unchanged*/
+  CL_DEVICE_CLOCK_MODE_QUERYPEAK_AMD = 0x7, /*Queries the peak device clock ratios.Leaves the clock mode of the device unchanged*/
+  CL_DEVICE_CLOCK_MODE_COUNT_AMD = 0x8, /*Maxmium count of device clock mode*/
+};
+
+typedef struct _cl_set_device_clock_mode_input_amd
+{
+  /* specify the clock mode for AMD GPU device*/
+  cl_DeviceClockMode_AMD clock_mode;
+} cl_set_device_clock_mode_input_amd;
+
+typedef struct _cl_set_device_clock_mode_output_amd
+{
+  /*Ratio of current mem clock to peak clock as obtained from DeviceProperties::maxGpuClock*/
+  cl_float memory_clock_ratio_to_peak;
+  /*Ratio of current gpu core clock to peak clock as obtained from DeviceProperties::maxGpuClock*/
+  cl_float engine_clock_ratio_to_peak;
+} cl_set_device_clock_mode_output_amd;
+
 /*! \brief Creates a new HW performance counter
  *   for the specified OpenCL context.
  *
@@ -220,6 +249,11 @@ extern CL_API_ENTRY cl_int CL_API_CALL clGetPerfCounterInfoAMD(
     size_t /* param_value_size */, void* /* param_value */, size_t* /* param_value_size_ret */
     ) CL_API_SUFFIX__VERSION_1_0;
 
+extern CL_API_ENTRY cl_int CL_API_CALL clSetDeviceClockModeAMD(
+    cl_device_id /* device*/, cl_set_device_clock_mode_input_amd /* Clock_Mode_Input */,
+    cl_set_device_clock_mode_output_amd* /* Clock_Mode_Output */
+    ) CL_API_SUFFIX__VERSION_1_0;
+
 #ifdef __cplusplus
 } /*extern "C"*/
 #endif /*__cplusplus*/