/* * Copyright (c) 2017, Intel Corporation * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be included * in all copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR * OTHER DEALINGS IN THE SOFTWARE. */ #ifndef CMRTLIB_AGNOSTIC_SHARE_CM_EVENT_BASE_H_ #define CMRTLIB_AGNOSTIC_SHARE_CM_EVENT_BASE_H_ #include "cm_def_os.h" #include "cm_include.h" enum CM_EVENT_PROFILING_INFO { CM_EVENT_PROFILING_HWSTART, CM_EVENT_PROFILING_HWEND, CM_EVENT_PROFILING_SUBMIT, CM_EVENT_PROFILING_COMPLETE, CM_EVENT_PROFILING_ENQUEUE, CM_EVENT_PROFILING_KERNELCOUNT, CM_EVENT_PROFILING_KERNELNAMES, CM_EVENT_PROFILING_THREADSPACE, CM_EVENT_PROFILING_CALLBACK }; enum CM_STATUS { CM_STATUS_QUEUED = 0, CM_STATUS_FLUSHED = 1, CM_STATUS_FINISHED = 2, CM_STATUS_STARTED = 3, CM_STATUS_RESET = 4 }; //Time in seconds before kernel should timeout #define CM_MAX_TIMEOUT 2 //Time in milliseconds before kernel should timeout #define CM_MAX_TIMEOUT_MS CM_MAX_TIMEOUT*1000 //|GT-PIN struct CM_SURFACE_DETAILS { uint32_t width; //width of surface uint32_t height; //height of surface, 0 if surface is CmBuffer uint32_t depth; //depth of surface, 0 if surface is CmBuffer or CmSurface2D CM_SURFACE_FORMAT format; //format of surface, if surface is CmBuffer uint32_t planeIndex; //plane Index for this BTI, 0 if surface is not planar surface uint32_t pitch; //pitch of surface, 0 if surface is CmBuffer uint32_t slicePitch; // pitch of a slice in CmSurface3D, 0 if surface is CmBuffer or CmSurface2D uint32_t surfaceBaseAddress; uint8_t tiledSurface; uint8_t tileWalk; uint32_t xOffset; uint32_t yOffset; }; //! \brief A CmEvent object is genereated after a task is enqueued. //! \details A CmEvent object is generated after a task(one kernel or multiples //! kernels running concurrently) is enqueued for execution.This //! event object can subsequently be queried to determine the status //! of execution and other execution related information. class CmEvent { public: //! //! \brief Query the status of the task associated with the event. //! \details An event is generated when a task (one kernel or multiples //! kernels running concurrently) is enqueued. This is a //! non-blocking call. //! \param [out] status //! The reference to the execution status. Four status values, //! CM_STATUS_QUEUED, CM_STATUS_FLUSHED, CM_STATUS_STARTED, and //! CM_STATUS_FINISHED are supported. CM_STATUS_FLUSHED means //! that the task is already sent to driver/hardware. //! CM_STATUS_STARTED means hardware starts to execute the task. //! CM_STATUS_FINISHED means hardware has finished the //! execution of all kernels in the task. In Emulation/Simulation //! mode since the Enqueue operation is a blocking call, this //! function always returns CM_STATUS_FINISHED. //! \retval CM_SUCCESS if the status is successfully returned. //! \retval CM_FAILURE if otherwise. //! CM_RT_API virtual int32_t GetStatus( CM_STATUS& status) = 0 ; //! //! \brief Query the execution time of a task (one kernel or multiples //! kernels running concurrently) in the unit of nanoseconds. //! \details The execution time is measured from the time the task //! started execution in GPU to the time when the task finished //! execution.We recommend pair this API with //! CmEvent::WaitForTaskFinished in practice when you try to //! get GPU HW execution time. This is a non-blocking call. //! In Emulation/Simulation mode this function always returns 100. //! \param [out] time //! Reference to time. //! \retval CM_SUCCESS if the execution time is successfully returned. //! \retval CM_FAILURE if not, e.g. the task hasn't finished. //! CM_RT_API virtual int32_t GetExecutionTime(uint64_t& time) = 0; //! //! \brief Wait for the task associated with the event finishing //! execution on GPU (task status became CM_STATUS_FINISHED). //! \details It applies event driven synchronization between GPU and CPU //! to reduce CPU usage and power consumption in the waiting: //! current process goes to sleep and waits for the //! notification from OS until task finishes. We recommend use //! this API followed by CmEvent::GetExecutionTime when you try //! to get GPU HW execution time. //! \param [in] timeOutMs //! Timeout in milliseconds for the waiting, 2000 milliseconds //! by-default. //! \retval CM_SUCCESS if successfully wait and get notification from //! OS. //! \retval CM_NOT_IMPLEMENTED if DDI version is less than 2.4. //! \retval CM_EVENT_DRIVEN_FAILURE otherwise. //! \note This API is unnecessary to be added before the ReadSurface //! method of class Surface2D/3D etc. We already apply //! event-driven synchronization optimization inside //! ReadSurface if the dependent event is given. //! CM_RT_API virtual int32_t WaitForTaskFinished(uint32_t timeOutMs = CM_MAX_TIMEOUT_MS) = 0; //! //! \brief Gets surface details for GT-Pin. //! \param [in] kernIndex //! Index of the kernel which owns the surface. //! \param [in] surfBTI //! Index in Binding table for queried surface. //! \param [out] outDetails //! Detail info about this surface, which includes width, //! height, depth, format, pitch, slice Pitch, surface base //! address, tiled type, vertical offset, and horizontal offset. //! \retval CM_SUCCESS if query is successfully finished. //! \retval CM_NOT_IMPLEMENTED if DDI version is less than 2.3. //! \retval CM_INVALID_ARG_VALUE if input parameter is invalid. //! \retval CM_FAILURE otherwise. //! CM_RT_API virtual int32_t GetSurfaceDetails( uint32_t kernIndex, uint32_t surfBTI,CM_SURFACE_DETAILS& outDetails )=0; //! //! \brief This function can be used to get more profiling //! information for vTune. //! \details It can provided 9 profiling values, for profiling //! information,including //! CM_EVENT_PROFILING_HWSTART,CM_EVENT_PROFILING_HWEND, //! CM_EVENT_PROFILING_SUBMIT,CM_EVENT_PROFILING_COMPLETE, //! CM_EVENT_PROFILING_ENQUEUE,CM_EVENT_PROFILING_KERNELCOUNT, //! CM_EVENT_PROFILING_KERNELNAMES, //! CM_EVENT_PROFILING_THREADSPACE, //! CM_EVENT_PROFILING_CALLBACK. //! \param [in] infoType //! One value of CM_EVENT_PROFILING_INFO, specify which //! information to get. //! \param [in] paramSize //! Size of the parameter. //! \param [in] inputValue //! Pointer pointing to memory where to get kernel index or //! callback function. //! \param [out] value //! Pointer pointing to memory where the cap information should //! be returned. //! \retval CM_SUCCESS if get information successfully. //! \retval CM_FAILURE otherwise. //! CM_RT_API virtual int32_t GetProfilingInfo(CM_EVENT_PROFILING_INFO infoType, size_t paramSize, void *inputValue, void *value) = 0; //! //! \brief Query the raw tick time of a task(one kernel or multiples //! kernels running concurrently). //! \details We recommend pair this API with //! CmEvent::WaitForTaskFinished in practice when you try to //! get HW execution tick time. //! \param [out] tick //! Reference to time. //! \retval CM_SUCCESS if the raw tick time is successfully returned. //! \retval CM_FAILURE if not, e.g. the task hasn't finished. //! CM_RT_API virtual int32_t GetExecutionTickTime(uint64_t& tick) = 0; }; #endif // #ifndef CMRTLIB_AGNOSTIC_SHARE_CM_EVENT_BASE_H_