/* * Copyright (c) 2021-2022, 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. */ //! //! \file media_render_cmd_packet.cpp //! \brief Defines the interface for media render workload cmd packet //! \details The media cmd packet is dedicated for command buffer sequenece submit //! #include "media_render_cmd_packet.h" #include "renderhal_platform_interface.h" #include "hal_oca_interface_next.h" #include "mos_os_cp_interface_specific.h" #define COMPUTE_WALKER_THREAD_SPACE_WIDTH 1 #define COMPUTE_WALKER_THREAD_SPACE_HEIGHT 1 #define COMPUTE_WALKER_THREAD_SPACE_DEPTH 1 RenderCmdPacket::RenderCmdPacket(MediaTask* task, PMOS_INTERFACE pOsInterface, RENDERHAL_INTERFACE* renderHal) : CmdPacket(task), m_renderHal(renderHal), m_cpInterface(nullptr) { m_osInterface = pOsInterface; } RenderCmdPacket::~RenderCmdPacket() { Destroy(); } MOS_STATUS RenderCmdPacket::Init() { bool mediaWalkerUsed = false; bool computeWalkerUsed = false; if (nullptr == m_renderHal) { // Could not allocate renderhal in packet class // Because we need to share renderhal for many Packets to reduce memory RENDER_PACKET_ASSERTMESSAGE("renderHal should be passed When doing construction"); return MOS_STATUS_INVALID_PARAMETER; } mediaWalkerUsed = m_renderHal->pfnGetMediaWalkerStatus(m_renderHal) ? true : false; computeWalkerUsed = m_renderHal->pRenderHalPltInterface->IsComputeContextInUse(m_renderHal); if (mediaWalkerUsed && !computeWalkerUsed) { m_walkerType = WALKER_TYPE_MEDIA; } else if (computeWalkerUsed) { m_walkerType = WALKER_TYPE_COMPUTE; } else { m_walkerType = WALKER_TYPE_DISABLED; } if (m_renderHal->pRenderHalPltInterface) { m_miItf = m_renderHal->pRenderHalPltInterface->GetMhwMiItf(); } return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::Destroy() { MOS_STATUS eStatus = MOS_STATUS_SUCCESS; return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::Submit(MOS_COMMAND_BUFFER *commandBuffer, uint8_t packetPhase) { PMOS_INTERFACE pOsInterface = nullptr; MOS_STATUS eStatus = MOS_STATUS_SUCCESS; uint32_t dwSyncTag = 0; int32_t i = 0, iRemaining = 0; MHW_MEDIA_STATE_FLUSH_PARAM FlushParam = {}; bool bEnableSLM = false; RENDERHAL_GENERIC_PROLOG_PARAMS GenericPrologParams = {}; MOS_RESOURCE GpuStatusBuffer = {}; MOS_CONTEXT* pOsContext = nullptr; PMHW_MI_MMIOREGISTERS pMmioRegisters = nullptr; RENDER_PACKET_CHK_NULL_RETURN(m_renderHal); RENDER_PACKET_CHK_NULL_RETURN(m_renderHal->pRenderHalPltInterface); RENDER_PACKET_CHK_NULL_RETURN(m_renderHal->pRenderHalPltInterface->GetMmioRegisters(m_renderHal)); RENDER_PACKET_CHK_NULL_RETURN(m_renderHal->pOsInterface); RENDER_PACKET_CHK_NULL_RETURN(m_renderHal->pOsInterface->pOsContext); RENDER_PACKET_CHK_NULL_RETURN(commandBuffer); eStatus = MOS_STATUS_UNKNOWN; pOsInterface = m_renderHal->pOsInterface; iRemaining = 0; FlushParam = g_cRenderHal_InitMediaStateFlushParams; pOsContext = pOsInterface->pOsContext; pMmioRegisters = m_renderHal->pRenderHalPltInterface->GetMmioRegisters(m_renderHal); RENDER_PACKET_CHK_STATUS_RETURN(SetPowerMode(0)); m_renderHal->pRenderHalPltInterface->On1stLevelBBStart(m_renderHal, commandBuffer, pOsContext, pOsInterface->CurrentGpuContextHandle, pMmioRegisters); OcaDumpDbgInfo(*commandBuffer, *pOsContext); RENDER_PACKET_CHK_STATUS_RETURN(SetMediaFrameTracking(GenericPrologParams)); // Initialize command buffer and insert prolog RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnInitCommandBuffer(m_renderHal, commandBuffer, &GenericPrologParams)); RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddPerfCollectStartCmd(m_renderHal, pOsInterface, commandBuffer)); RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->StartPredicate(m_renderHal, commandBuffer)); // Write timing data for 3P budget RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSendTimingData(m_renderHal, commandBuffer, true)); bEnableSLM = false; // Media walker first RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetCacheOverrideParams( m_renderHal, &m_renderHal->L3CacheSettings, bEnableSLM)); if (m_renderHal->bCmfcCoeffUpdate) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSendCscCoeffSurface(m_renderHal, commandBuffer, m_renderHal->pCmfcCoeffSurface, m_renderHal->pStateHeap->pKernelAllocation[m_renderHal->iKernelAllocationID].pKernelEntry)); } // Flush media states if (m_isMultiKernelOneMediaState) { RENDER_PACKET_CHK_STATUS_RETURN(SendMultiKernelMediaStates( m_renderHal, commandBuffer)); } else { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSendMediaStates( m_renderHal, commandBuffer, (m_walkerType == WALKER_TYPE_MEDIA) ? &m_mediaWalkerParams : nullptr, &m_gpgpuWalkerParams)); } // Write back GPU Status tag if (!pOsInterface->bEnableKmdMediaFrameTracking) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSendRcsStatusTag(m_renderHal, commandBuffer)); } RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->StopPredicate(m_renderHal, commandBuffer)); RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddPerfCollectEndCmd(m_renderHal, pOsInterface, commandBuffer)); // Write timing data for 3P budget RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSendTimingData(m_renderHal, commandBuffer, false)); MHW_PIPE_CONTROL_PARAMS PipeControlParams; MOS_ZeroMemory(&PipeControlParams, sizeof(PipeControlParams)); PipeControlParams.dwFlushMode = MHW_FLUSH_WRITE_CACHE; PipeControlParams.bGenericMediaStateClear = true; PipeControlParams.bIndirectStatePointersDisable = true; PipeControlParams.bDisableCSStall = false; RENDER_PACKET_CHK_NULL_RETURN(pOsInterface->pfnGetSkuTable); auto *skuTable = pOsInterface->pfnGetSkuTable(pOsInterface); if (skuTable && MEDIA_IS_SKU(skuTable, FtrEnablePPCFlush)) { // Add PPC fulsh PipeControlParams.bPPCFlush = true; } RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddMiPipeControl(m_renderHal, commandBuffer, &PipeControlParams)); if (MEDIA_IS_WA(m_renderHal->pWaTable, WaSendDummyVFEafterPipelineSelect)) { MHW_VFE_PARAMS VfeStateParams = {}; VfeStateParams.dwNumberofURBEntries = 1; RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddMediaVfeCmd(m_renderHal, commandBuffer, &VfeStateParams)); } // Add media flush command in case HW not cleaning the media state if (MEDIA_IS_WA(m_renderHal->pWaTable, WaMSFWithNoWatermarkTSGHang)) { FlushParam.bFlushToGo = true; if (m_walkerType == WALKER_TYPE_MEDIA) { FlushParam.ui8InterfaceDescriptorOffset = m_mediaWalkerParams.InterfaceDescriptorOffset; } else { RENDER_PACKET_ASSERTMESSAGE("ERROR, pWalkerParams is nullptr and cannot get InterfaceDescriptorOffset."); } RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddMediaStateFlush(m_renderHal, commandBuffer, &FlushParam)); } else if (MEDIA_IS_WA(m_renderHal->pWaTable, WaAddMediaStateFlushCmd)) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddMediaStateFlush(m_renderHal, commandBuffer, &FlushParam)); } #if (_DEBUG || _RELEASE_INTERNAL) RENDER_PACKET_CHK_STATUS_RETURN(StallBatchBuffer(commandBuffer)); #endif HalOcaInterfaceNext::On1stLevelBBEnd(*commandBuffer, *pOsInterface); if (pBatchBuffer) { // Send Batch Buffer end command (HW/OS dependent) RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddMiBatchBufferEnd(m_renderHal, commandBuffer, nullptr)); } else if (IsMiBBEndNeeded(pOsInterface)) { // Send Batch Buffer end command for 1st level Batch Buffer RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddMiBatchBufferEnd(m_renderHal, commandBuffer, nullptr)); } else if (m_renderHal->pOsInterface->bNoParsingAssistanceInKmd) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pRenderHalPltInterface->AddMiBatchBufferEnd(m_renderHal, commandBuffer, nullptr)); } // No need return command buffer here, which will be done in CmdTask::Submit. MOS_NULL_RENDERING_FLAGS NullRenderingFlags; NullRenderingFlags = pOsInterface->pfnGetNullHWRenderFlags(pOsInterface); if ((NullRenderingFlags.VPLgca || NullRenderingFlags.VPGobal) == false) { dwSyncTag = m_renderHal->pStateHeap->dwNextTag++; // Set media state and batch buffer as busy m_renderHal->pStateHeap->pCurMediaState->bBusy = true; if (pBatchBuffer) { pBatchBuffer->bBusy = true; pBatchBuffer->dwSyncTag = dwSyncTag; } } return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::RenderEngineSetup() { // pls make sure the context already switched to render/compute engine before submit RENDER_PACKET_CHK_NULL_RETURN(m_renderHal); // Register the resource of GSH RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnReset(m_renderHal)); // Assign media state m_renderData.mediaState = m_renderHal->pfnAssignMediaState(m_renderHal, RENDERHAL_COMPONENT_PACKET); RENDER_PACKET_CHK_NULL_RETURN(m_renderData.mediaState); RENDER_PACKET_CHK_NULL_RETURN(m_renderHal->pStateHeap); // Allocate and reset SSH instance if ((m_isMultiBindingTables == false) || (m_renderHal->pStateHeap->iCurrentBindingTable >= m_renderHal->StateHeapSettings.iBindingTables) || (m_renderHal->pStateHeap->iCurrentSurfaceState >= m_renderHal->StateHeapSettings.iSurfaceStates) || m_isMultiKernelOneMediaState || m_renderHal->isBindlessHeapInUse) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnAssignSshInstance(m_renderHal)); } // Assign and Reset binding table RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnAssignBindingTable( m_renderHal, &m_renderData.bindingTable)); // Reset a new Binding index from start m_renderData.bindingTableEntry = 0; // load kernels before packet submit PipeLine to load kernel return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::InitKernelEntry() { if (m_kernelCount == 0) { RENDER_PACKET_NORMALMESSAGE("no kernel set up"); return MOS_STATUS_LOAD_LIBRARY_FAILED; } return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::SetPowerMode(uint32_t KernelID) { MOS_STATUS eStatus = MOS_STATUS_SUCCESS; uint16_t wNumRequestedEUSlices = 1; // Default to 1 slice uint16_t wNumRequestedSubSlices = 3; // Default to 3 subslice uint16_t wNumRequestedEUs = 8; // Default to 8 EUs RENDERHAL_POWEROPTION PowerOption; bool bSetRequestedSlices = false; const SseuSetting * pcSSEUTable = nullptr; MediaUserSettingSharedPtr userSettingPtr = nullptr; RENDER_PACKET_CHK_NULL_RETURN(m_renderHal); if ((m_renderHal->bRequestSingleSlice) || (m_renderHal->bEUSaturationNoSSD)) { bSetRequestedSlices = true; // bEUSaturationNoSSD: No slice shutdown, must request 2 slices [CM EU saturation on]. // bRequestSingleSlice: Always single slice. wNumRequestedEUSlices = (m_renderHal->bEUSaturationNoSSD) ? 2 : 1; } else { bSetRequestedSlices = false; } if (m_renderHal->sseuTable) { pcSSEUTable = (const SseuSetting *)m_renderHal->sseuTable; } else { RENDER_PACKET_ASSERTMESSAGE("SSEU Table not valid."); return MOS_STATUS_UNKNOWN; } RENDER_PACKET_CHK_NULL_RETURN(pcSSEUTable); pcSSEUTable += KernelID; if (!bSetRequestedSlices) // If num Slices is already programmed, then don't change it { if (wNumRequestedEUSlices < pcSSEUTable->numSlices) { wNumRequestedEUSlices = pcSSEUTable->numSlices; } } wNumRequestedSubSlices = pcSSEUTable->numSubSlices; wNumRequestedEUs = pcSSEUTable->numEUs; #if (_DEBUG || _RELEASE_INTERNAL) // User feature key reads userSettingPtr = m_osInterface->pfnGetUserSettingInstance(m_osInterface); uint32_t value = 0; ReadUserSettingForDebug( userSettingPtr, value, __MEDIA_USER_FEATURE_VALUE_SSEU_SETTING_OVERRIDE, MediaUserSetting::Group::Device); if (value != 0xDEADC0DE) { wNumRequestedEUSlices = value & 0xFF; // Bits 0-7 wNumRequestedSubSlices = (value >> 8) & 0xFF; // Bits 8-15 wNumRequestedEUs = (value >> 16) & 0xFFFF; // Bits 16-31 } #endif PowerOption.nSlice = wNumRequestedEUSlices; PowerOption.nSubSlice = wNumRequestedSubSlices; PowerOption.nEU = wNumRequestedEUs; m_renderHal->pfnSetPowerOptionMode(m_renderHal, &PowerOption); return eStatus; } uint32_t RenderCmdPacket::SetSurfaceForHwAccess( PMOS_SURFACE surface, PRENDERHAL_SURFACE_NEXT pRenderSurface, PRENDERHAL_SURFACE_STATE_PARAMS pSurfaceParams, bool bWrite) { PMOS_INTERFACE pOsInterface; PRENDERHAL_SURFACE_STATE_ENTRY pSurfaceEntries[MHW_MAX_SURFACE_PLANES]; int32_t iSurfaceEntries; int32_t i; MOS_STATUS eStatus; RENDERHAL_SURFACE_STATE_PARAMS surfaceParams; // Initialize Variables eStatus = MOS_STATUS_SUCCESS; pOsInterface = m_osInterface; RENDER_PACKET_CHK_NULL_RETURN(pRenderSurface); RENDER_PACKET_CHK_NULL_RETURN(pOsInterface); // Register surfaces for rendering (GfxAddress/Allocation index) // Register resource RENDER_PACKET_CHK_STATUS_RETURN(m_osInterface->pfnRegisterResource( m_osInterface, &surface->OsResource, bWrite, true)); if (!pSurfaceParams) { MOS_ZeroMemory(&surfaceParams, sizeof(RENDERHAL_SURFACE_STATE_PARAMS)); //set mem object control for cache surfaceParams.MemObjCtl = (m_renderHal->pOsInterface->pfnCachePolicyGetMemoryObject( MOS_HW_RESOURCE_USAGE_VP_INTERNAL_READ_WRITE_RENDER, m_renderHal->pOsInterface->pfnGetGmmClientContext(m_renderHal->pOsInterface))).DwordValue; pSurfaceParams = &surfaceParams; } if (pSurfaceParams->bAVS) { pSurfaceParams->Type = m_renderHal->SurfaceTypeAdvanced; } else { pSurfaceParams->Type = m_renderHal->SurfaceTypeDefault; } RENDER_PACKET_CHK_STATUS_RETURN(InitRenderHalSurface( *surface, pRenderSurface)); if (bWrite) { pRenderSurface->SurfType = RENDERHAL_SURF_OUT_RENDERTARGET; } // Setup surface states----------------------------------------------------- RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetupSurfaceState( m_renderHal, pRenderSurface, pSurfaceParams, &iSurfaceEntries, // for most cases, surface entry should only take 1 entry, need align with kerenl design pSurfaceEntries, nullptr)); if (!m_isLargeSurfaceStateNeeded) { if (m_renderData.bindingTableEntry > 15) { RENDER_PACKET_ASSERTMESSAGE("input surface support up to 16 RSS"); m_renderData.bindingTableEntry = 0; } } else { if (m_renderData.bindingTableEntry > 255) { RENDER_PACKET_ASSERTMESSAGE("input surface support up to 256 RSS"); m_renderData.bindingTableEntry = 0; } } uint32_t iBTEntry = m_renderData.bindingTableEntry; // Bind surface states------------------------------------------------------ for (i = 0; i < iSurfaceEntries; i++, m_renderData.bindingTableEntry++) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnBindSurfaceState( m_renderHal, m_renderData.bindingTable, m_renderData.bindingTableEntry, pSurfaceEntries[i])); pRenderSurface->Index = m_renderData.bindingTableEntry; } return iBTEntry; } uint32_t RenderCmdPacket::SetSurfaceForHwAccess( PMOS_SURFACE surface, PRENDERHAL_SURFACE_NEXT pRenderSurface, PRENDERHAL_SURFACE_STATE_PARAMS pSurfaceParams, bool bWrite, std::vector &stateLocations) { PMOS_INTERFACE pOsInterface; PRENDERHAL_SURFACE_STATE_ENTRY pSurfaceEntries[MHW_MAX_SURFACE_PLANES]; int32_t iSurfaceEntries; int32_t i; MOS_STATUS eStatus; RENDERHAL_SURFACE_STATE_PARAMS surfaceParams; // Initialize Variables eStatus = MOS_STATUS_SUCCESS; pOsInterface = m_osInterface; RENDER_PACKET_CHK_NULL_RETURN(pRenderSurface); RENDER_PACKET_CHK_NULL_RETURN(pOsInterface); // Register surfaces for rendering (GfxAddress/Allocation index) // Register resource RENDER_PACKET_CHK_STATUS_RETURN(m_osInterface->pfnRegisterResource( m_osInterface, &surface->OsResource, bWrite, true)); if (!pSurfaceParams) { MOS_ZeroMemory(&surfaceParams, sizeof(RENDERHAL_SURFACE_STATE_PARAMS)); //set mem object control for cache surfaceParams.MemObjCtl = (m_renderHal->pOsInterface->pfnCachePolicyGetMemoryObject( MOS_HW_RESOURCE_USAGE_VP_INTERNAL_READ_WRITE_RENDER, m_renderHal->pOsInterface->pfnGetGmmClientContext(m_renderHal->pOsInterface))).DwordValue; pSurfaceParams = &surfaceParams; } if (pSurfaceParams->bAVS) { pSurfaceParams->Type = m_renderHal->SurfaceTypeAdvanced; } else { pSurfaceParams->Type = m_renderHal->SurfaceTypeDefault; } RENDER_PACKET_CHK_STATUS_RETURN(InitRenderHalSurface( *surface, pRenderSurface)); if (bWrite) { pRenderSurface->SurfType = RENDERHAL_SURF_OUT_RENDERTARGET; } // Setup surface states----------------------------------------------------- RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetupSurfaceState( m_renderHal, pRenderSurface, pSurfaceParams, &iSurfaceEntries, // for most cases, surface entry should only take 1 entry, need align with kerenl design pSurfaceEntries, nullptr)); if (!m_isLargeSurfaceStateNeeded) { if (m_renderData.bindingTableEntry > 15) { RENDER_PACKET_ASSERTMESSAGE("input surface support up to 16 RSS"); m_renderData.bindingTableEntry = 0; } } else { if (m_renderData.bindingTableEntry > 255) { RENDER_PACKET_ASSERTMESSAGE("input surface support up to 256 RSS"); m_renderData.bindingTableEntry = 0; } } uint32_t iBTEntry = m_renderData.bindingTableEntry; if (m_renderHal->isBindlessHeapInUse == false) { // Bind surface states------------------------------------------------------ for (i = 0; i < iSurfaceEntries; i++, m_renderData.bindingTableEntry++) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnBindSurfaceState( m_renderHal, m_renderData.bindingTable, m_renderData.bindingTableEntry, pSurfaceEntries[i])); pRenderSurface->Index = m_renderData.bindingTableEntry; } } else { for (i = 0; i < iSurfaceEntries; i++) { stateLocations.push_back(pSurfaceEntries[i]->stateLocation); } } return iBTEntry; } MOS_STATUS RenderCmdPacket::SetSurfaceForHwAccess( PMOS_SURFACE surface, PRENDERHAL_SURFACE_NEXT pRenderSurface, PRENDERHAL_SURFACE_STATE_PARAMS pSurfaceParams, uint32_t &bindingIndex, bool bWrite, PRENDERHAL_SURFACE_STATE_ENTRY *surfaceEntries, uint32_t * numOfSurfaceEntries) { PMOS_INTERFACE pOsInterface = nullptr; PRENDERHAL_SURFACE_STATE_ENTRY surfaceEntriesTmp[MHW_MAX_SURFACE_PLANES] = {}; PRENDERHAL_SURFACE_STATE_ENTRY *pSurfaceEntries = nullptr; int32_t iSurfaceEntries = 0; int32_t i = 0; MOS_STATUS eStatus = MOS_STATUS_SUCCESS; RENDERHAL_SURFACE_STATE_PARAMS surfaceParams = {}; if (nullptr == surfaceEntries || nullptr == numOfSurfaceEntries) { pSurfaceEntries = surfaceEntriesTmp; } else { pSurfaceEntries = surfaceEntries; } // Initialize Variables eStatus = MOS_STATUS_SUCCESS; pOsInterface = m_osInterface; RENDER_PACKET_CHK_NULL_RETURN(pRenderSurface); RENDER_PACKET_CHK_NULL_RETURN(pOsInterface); // Register surfaces for rendering (GfxAddress/Allocation index) // Register resource RENDER_PACKET_CHK_STATUS_RETURN(m_osInterface->pfnRegisterResource( m_osInterface, &surface->OsResource, bWrite, true)); if (!pSurfaceParams) { MOS_ZeroMemory(&surfaceParams, sizeof(RENDERHAL_SURFACE_STATE_PARAMS)); //set mem object control for cache surfaceParams.MemObjCtl = (m_renderHal->pOsInterface->pfnCachePolicyGetMemoryObject( MOS_HW_RESOURCE_USAGE_VP_INTERNAL_READ_WRITE_RENDER, m_renderHal->pOsInterface->pfnGetGmmClientContext(m_renderHal->pOsInterface))).DwordValue; pSurfaceParams = &surfaceParams; } if (pSurfaceParams->bAVS) { pSurfaceParams->Type = m_renderHal->SurfaceTypeAdvanced; } else { pSurfaceParams->Type = m_renderHal->SurfaceTypeDefault; } RENDER_PACKET_CHK_STATUS_RETURN(InitRenderHalSurface( *surface, pRenderSurface)); if (bWrite) { pRenderSurface->SurfType = RENDERHAL_SURF_OUT_RENDERTARGET; } // Setup surface states----------------------------------------------------- RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetupSurfaceState( m_renderHal, pRenderSurface, pSurfaceParams, &iSurfaceEntries, // for most cases, surface entry should only take 1 entry, need align with kerenl design pSurfaceEntries, nullptr)); uint32_t iBTEntry = bindingIndex; // Bind surface states------------------------------------------------------ for (i = 0; i < iSurfaceEntries; i++, iBTEntry++) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnBindSurfaceState( m_renderHal, m_renderData.bindingTable, iBTEntry, pSurfaceEntries[i])); pRenderSurface->Index = iBTEntry; } if (numOfSurfaceEntries) { *numOfSurfaceEntries = iSurfaceEntries; } return eStatus; } MOS_STATUS RenderCmdPacket::SetSurfaceForHwAccess( PMOS_SURFACE surface, PRENDERHAL_SURFACE_NEXT pRenderSurface, PRENDERHAL_SURFACE_STATE_PARAMS pSurfaceParams, std::set &bindingIndexes, bool bWrite, std::vector &stateLocations, uint32_t capcityOfSurfaceEntries, PRENDERHAL_SURFACE_STATE_ENTRY *surfaceEntries, uint32_t *numOfSurfaceEntries) { PMOS_INTERFACE pOsInterface = nullptr; PRENDERHAL_SURFACE_STATE_ENTRY pSurfaceEntries[MHW_MAX_SURFACE_PLANES] = {}; int32_t iSurfaceEntries = 0; int32_t i = 0; MOS_STATUS eStatus = MOS_STATUS_SUCCESS; RENDERHAL_SURFACE_STATE_PARAMS surfaceParams = {}; // Initialize Variables eStatus = MOS_STATUS_SUCCESS; pOsInterface = m_osInterface; RENDER_PACKET_CHK_NULL_RETURN(pRenderSurface); RENDER_PACKET_CHK_NULL_RETURN(pOsInterface); // Register surfaces for rendering (GfxAddress/Allocation index) // Register resource RENDER_PACKET_CHK_STATUS_RETURN(m_osInterface->pfnRegisterResource( m_osInterface, &surface->OsResource, bWrite, true)); if (!pSurfaceParams) { MOS_ZeroMemory(&surfaceParams, sizeof(RENDERHAL_SURFACE_STATE_PARAMS)); //set mem object control for cache surfaceParams.MemObjCtl = (m_renderHal->pOsInterface->pfnCachePolicyGetMemoryObject( MOS_HW_RESOURCE_USAGE_VP_INTERNAL_READ_WRITE_RENDER, m_renderHal->pOsInterface->pfnGetGmmClientContext(m_renderHal->pOsInterface))) .DwordValue; pSurfaceParams = &surfaceParams; } if (pSurfaceParams->bAVS) { pSurfaceParams->Type = m_renderHal->SurfaceTypeAdvanced; } else { pSurfaceParams->Type = m_renderHal->SurfaceTypeDefault; } RENDER_PACKET_CHK_STATUS_RETURN(InitRenderHalSurface( *surface, pRenderSurface)); if (bWrite) { pRenderSurface->SurfType = RENDERHAL_SURF_OUT_RENDERTARGET; } // Setup surface states----------------------------------------------------- RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetupSurfaceState( m_renderHal, pRenderSurface, pSurfaceParams, &iSurfaceEntries, // for most cases, surface entry should only take 1 entry, need align with kerenl design pSurfaceEntries, nullptr)); if (iSurfaceEntries > MHW_MAX_SURFACE_PLANES) { RENDER_PACKET_CHK_STATUS_RETURN(MOS_STATUS_INVALID_PARAMETER); } if (static_cast(iSurfaceEntries) <= capcityOfSurfaceEntries && surfaceEntries != nullptr) { for (int32_t i = 0; i < iSurfaceEntries; ++i) { surfaceEntries[i] = pSurfaceEntries[i]; } } if (m_renderHal->isBindlessHeapInUse == false) { for (uint32_t const &bindingIndex : bindingIndexes) { uint32_t iBTEntry = bindingIndex; // Bind surface states------------------------------------------------------ for (i = 0; i < iSurfaceEntries; i++, iBTEntry++) { RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnBindSurfaceState( m_renderHal, m_renderData.bindingTable, iBTEntry, pSurfaceEntries[i])); pRenderSurface->Index = iBTEntry; } } } else { for (i = 0; i < iSurfaceEntries; i++) { stateLocations.push_back(pSurfaceEntries[i]->stateLocation); } } if (numOfSurfaceEntries) { *numOfSurfaceEntries = iSurfaceEntries; } return eStatus; } uint32_t RenderCmdPacket::SetBufferForHwAccess( PMOS_SURFACE buffer, PRENDERHAL_SURFACE_NEXT pRenderSurface, PRENDERHAL_SURFACE_STATE_PARAMS pSurfaceParams, bool bWrite, std::vector &stateLocations) { RENDERHAL_SURFACE RenderHalSurface; RENDERHAL_SURFACE_STATE_PARAMS SurfaceParam; PRENDERHAL_SURFACE_STATE_ENTRY pSurfaceEntry; RENDER_PACKET_CHK_NULL_RETURN(m_osInterface); RENDER_PACKET_CHK_NULL_RETURN(m_osInterface->osCpInterface); RENDER_PACKET_CHK_NULL_RETURN(buffer); MOS_ZeroMemory(&RenderHalSurface, sizeof(RenderHalSurface)); // Register surfaces for rendering (GfxAddress/Allocation index) // Register resource RENDER_PACKET_CHK_STATUS_RETURN(m_osInterface->pfnRegisterResource( m_osInterface, &buffer->OsResource, bWrite, true)); // Setup Buffer surface----------------------------------------------------- if (pSurfaceParams == nullptr) { MOS_ZeroMemory(&SurfaceParam, sizeof(SurfaceParam)); auto memObjCtrlState = m_osInterface->pfnGetResourceCachePolicyMemoryObject(m_renderHal->pOsInterface, &buffer->OsResource); SurfaceParam.MemObjCtl = memObjCtrlState.DwordValue; pSurfaceParams = &SurfaceParam; } RENDER_PACKET_CHK_STATUS_RETURN(InitRenderHalSurface( *buffer, &RenderHalSurface)); RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetupBufferSurfaceState( m_renderHal, &RenderHalSurface, pSurfaceParams, &pSurfaceEntry)); if (m_renderHal->isBindlessHeapInUse == false) { // Bind surface state------------------------------------------------------- RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnBindSurfaceState( m_renderHal, m_renderData.bindingTable, m_renderData.bindingTableEntry, pSurfaceEntry)); pRenderSurface->Index = m_renderData.bindingTableEntry; m_renderData.bindingTableEntry++; } else { stateLocations.push_back(pSurfaceEntry->stateLocation); } return pRenderSurface->Index; } uint32_t RenderCmdPacket::SetBufferForHwAccess(PMOS_SURFACE buffer, PRENDERHAL_SURFACE_NEXT pRenderSurface, PRENDERHAL_SURFACE_STATE_PARAMS pSurfaceParams, uint32_t bindingIndex, bool bWrite) { RENDERHAL_SURFACE RenderHalSurface; RENDERHAL_SURFACE_STATE_PARAMS SurfaceParam; PRENDERHAL_SURFACE_STATE_ENTRY pSurfaceEntry; RENDER_PACKET_CHK_NULL_RETURN(m_osInterface); RENDER_PACKET_CHK_NULL_RETURN(m_osInterface->osCpInterface); RENDER_PACKET_CHK_NULL_RETURN(buffer); MOS_ZeroMemory(&RenderHalSurface, sizeof(RenderHalSurface)); // Register surfaces for rendering (GfxAddress/Allocation index) // Register resource RENDER_PACKET_CHK_STATUS_RETURN(m_osInterface->pfnRegisterResource( m_osInterface, &buffer->OsResource, bWrite, true)); // Setup Buffer surface----------------------------------------------------- if (pSurfaceParams == nullptr) { MOS_ZeroMemory(&SurfaceParam, sizeof(SurfaceParam)); //set mem object control for cache SurfaceParam.MemObjCtl = (m_renderHal->pOsInterface->pfnCachePolicyGetMemoryObject( MOS_HW_RESOURCE_USAGE_VP_INTERNAL_READ_WRITE_RENDER, m_renderHal->pOsInterface->pfnGetGmmClientContext(m_renderHal->pOsInterface))).DwordValue; pSurfaceParams = &SurfaceParam; } RENDER_PACKET_CHK_STATUS_RETURN(InitRenderHalSurface( *buffer, &RenderHalSurface)); RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetupBufferSurfaceState( m_renderHal, &RenderHalSurface, pSurfaceParams, &pSurfaceEntry)); uint32_t iBTEntry = bindingIndex; // Bind surface state------------------------------------------------------- RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnBindSurfaceState( m_renderHal, m_renderData.bindingTable, iBTEntry, pSurfaceEntry)); pRenderSurface->Index = bindingIndex; return bindingIndex; } MOS_STATUS RenderCmdPacket::SetBufferForHwAccess( PMOS_SURFACE buffer, PRENDERHAL_SURFACE_NEXT pRenderSurface, PRENDERHAL_SURFACE_STATE_PARAMS pSurfaceParams, std::set &bindingIndexes, bool bWrite, std::vector &stateLocations) { RENDERHAL_SURFACE RenderHalSurface = {}; RENDERHAL_SURFACE_STATE_PARAMS SurfaceParam = {}; PRENDERHAL_SURFACE_STATE_ENTRY pSurfaceEntry = {}; MOS_STATUS eStatus = MOS_STATUS_SUCCESS; RENDER_PACKET_CHK_NULL_RETURN(m_osInterface); RENDER_PACKET_CHK_NULL_RETURN(m_osInterface->osCpInterface); RENDER_PACKET_CHK_NULL_RETURN(buffer); MOS_ZeroMemory(&RenderHalSurface, sizeof(RenderHalSurface)); // Register surfaces for rendering (GfxAddress/Allocation index) // Register resource RENDER_PACKET_CHK_STATUS_RETURN(m_osInterface->pfnRegisterResource( m_osInterface, &buffer->OsResource, bWrite, true)); // Setup Buffer surface----------------------------------------------------- if (pSurfaceParams == nullptr) { MOS_ZeroMemory(&SurfaceParam, sizeof(SurfaceParam)); //set mem object control for cache SurfaceParam.MemObjCtl = (m_renderHal->pOsInterface->pfnCachePolicyGetMemoryObject( MOS_HW_RESOURCE_USAGE_VP_INTERNAL_READ_WRITE_RENDER, m_renderHal->pOsInterface->pfnGetGmmClientContext(m_renderHal->pOsInterface))) .DwordValue; pSurfaceParams = &SurfaceParam; } RENDER_PACKET_CHK_STATUS_RETURN(InitRenderHalSurface( *buffer, &RenderHalSurface)); RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetupBufferSurfaceState( m_renderHal, &RenderHalSurface, pSurfaceParams, &pSurfaceEntry)); if (m_renderHal->isBindlessHeapInUse == false) { for (uint32_t const &bindingIndex : bindingIndexes) { uint32_t iBTEntry = bindingIndex; // Bind surface state------------------------------------------------------- RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnBindSurfaceState( m_renderHal, m_renderData.bindingTable, iBTEntry, pSurfaceEntry)); pRenderSurface->Index = bindingIndex; } } else { stateLocations.push_back(pSurfaceEntry->stateLocation); } return eStatus; } uint32_t RenderCmdPacket::SetBufferForHwAccess(MOS_BUFFER buffer, PRENDERHAL_SURFACE_NEXT pRenderSurface, PRENDERHAL_SURFACE_STATE_PARAMS pSurfaceParams, bool bWrite) { RENDERHAL_SURFACE RenderHalSurface; RENDERHAL_SURFACE_STATE_PARAMS SurfaceParam; PRENDERHAL_SURFACE_STATE_ENTRY pSurfaceEntry; RENDER_PACKET_CHK_NULL_RETURN(m_osInterface); RENDER_PACKET_CHK_NULL_RETURN(m_osInterface->osCpInterface); MOS_ZeroMemory(&RenderHalSurface, sizeof(RenderHalSurface)); // not support CP yet if (m_osInterface->osCpInterface->IsHMEnabled()) { RENDER_PACKET_ASSERTMESSAGE("ERROR, need to use VpHal_CommonSetBufferSurfaceForHwAccess if under CP HM."); } // Register surfaces for rendering (GfxAddress/Allocation index) // Register resource RENDER_PACKET_CHK_STATUS_RETURN(m_osInterface->pfnRegisterResource( m_osInterface, &buffer.OsResource, bWrite, true)); // Setup Buffer surface----------------------------------------------------- if (pSurfaceParams == nullptr) { MOS_ZeroMemory(&SurfaceParam, sizeof(SurfaceParam)); //set mem object control for cache SurfaceParam.MemObjCtl = (m_renderHal->pOsInterface->pfnCachePolicyGetMemoryObject( MOS_HW_RESOURCE_USAGE_VP_INTERNAL_READ_WRITE_RENDER, m_renderHal->pOsInterface->pfnGetGmmClientContext(m_renderHal->pOsInterface))).DwordValue; pSurfaceParams = &SurfaceParam; } RENDER_PACKET_CHK_STATUS_RETURN(InitRenderHalBuffer( buffer, &RenderHalSurface)); RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetupBufferSurfaceState( m_renderHal, &RenderHalSurface, pSurfaceParams, &pSurfaceEntry)); // Bind surface state------------------------------------------------------- RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnBindSurfaceState( m_renderHal, m_renderData.bindingTable, m_renderData.bindingTableEntry, pSurfaceEntry)); pRenderSurface->Index = m_renderData.bindingTableEntry; m_renderData.bindingTableEntry++; return pRenderSurface->Index; } MOS_STATUS RenderCmdPacket::SetupCurbe(void *pData, uint32_t curbeLength, uint32_t maximumNumberofThreads) { m_renderData.iCurbeOffset = m_renderHal->pfnLoadCurbeData( m_renderHal, m_renderData.mediaState, pData, curbeLength); if (m_renderData.iCurbeOffset < 0) { RENDER_PACKET_ASSERTMESSAGE("Curbe Set Fail, return error"); return MOS_STATUS_UNKNOWN; } m_renderData.iCurbeLength = curbeLength; RENDER_PACKET_CHK_STATUS_RETURN(m_renderHal->pfnSetVfeStateParams( m_renderHal, MEDIASTATE_DEBUG_COUNTER_FREE_RUNNING, maximumNumberofThreads, m_renderData.iCurbeLength, m_renderData.iInlineLength, nullptr)); return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::PrepareMediaWalkerParams(KERNEL_WALKER_PARAMS params, MHW_WALKER_PARAMS &mediaWalker) { uint32_t uiMediaWalkerBlockSize; RECT alignedRect = {}; bool bVerticalPattern = false; uiMediaWalkerBlockSize = m_renderHal->pHwSizes->dwSizeMediaWalkerBlock; alignedRect = params.alignedRect; bVerticalPattern = params.isVerticalPattern; // Calculate aligned output area in order to determine the total # blocks // to process in case of non-16x16 aligned target. alignedRect.right += uiMediaWalkerBlockSize - 1; alignedRect.bottom += uiMediaWalkerBlockSize - 1; alignedRect.left -= alignedRect.left % uiMediaWalkerBlockSize; alignedRect.top -= alignedRect.top % uiMediaWalkerBlockSize; alignedRect.right -= alignedRect.right % uiMediaWalkerBlockSize; alignedRect.bottom -= alignedRect.bottom % uiMediaWalkerBlockSize; if (params.calculateBlockXYByAlignedRect) { // Set number of blocks params.iBlocksX = (alignedRect.right - alignedRect.left) / uiMediaWalkerBlockSize; params.iBlocksY = (alignedRect.bottom - alignedRect.top) / uiMediaWalkerBlockSize; } // Set walker cmd params - Rasterscan mediaWalker.InterfaceDescriptorOffset = params.iMediaID; mediaWalker.dwGlobalLoopExecCount = 1; if (uiMediaWalkerBlockSize == 32) { mediaWalker.ColorCountMinusOne = 3; } else { mediaWalker.ColorCountMinusOne = 0; } if (alignedRect.left != 0 || alignedRect.top != 0) { // if the rect starts from any other macro block other than the first // then the global resolution should be the whole frame and the global // start should be the rect start. mediaWalker.GlobalResolution.x = (alignedRect.right / uiMediaWalkerBlockSize); mediaWalker.GlobalResolution.y = (alignedRect.bottom / uiMediaWalkerBlockSize); } else { mediaWalker.GlobalResolution.x = params.iBlocksX; mediaWalker.GlobalResolution.y = params.iBlocksY; } mediaWalker.GlobalStart.x = (alignedRect.left / uiMediaWalkerBlockSize); mediaWalker.GlobalStart.y = (alignedRect.top / uiMediaWalkerBlockSize); mediaWalker.GlobalOutlerLoopStride.x = params.iBlocksX; mediaWalker.GlobalOutlerLoopStride.y = 0; mediaWalker.GlobalInnerLoopUnit.x = 0; mediaWalker.GlobalInnerLoopUnit.y = params.iBlocksY; mediaWalker.BlockResolution.x = params.iBlocksX; mediaWalker.BlockResolution.y = params.iBlocksY; mediaWalker.LocalStart.x = 0; mediaWalker.LocalStart.y = 0; if (bVerticalPattern) { mediaWalker.LocalOutLoopStride.x = 1; mediaWalker.LocalOutLoopStride.y = 0; mediaWalker.LocalInnerLoopUnit.x = 0; mediaWalker.LocalInnerLoopUnit.y = 1; mediaWalker.dwLocalLoopExecCount = params.iBlocksX - 1; mediaWalker.LocalEnd.x = 0; mediaWalker.LocalEnd.y = params.iBlocksY - 1; } else { mediaWalker.LocalOutLoopStride.x = 0; mediaWalker.LocalOutLoopStride.y = 1; mediaWalker.LocalInnerLoopUnit.x = 1; mediaWalker.LocalInnerLoopUnit.y = 0; mediaWalker.dwLocalLoopExecCount = params.iBlocksY - 1; mediaWalker.LocalEnd.x = params.iBlocksX - 1; mediaWalker.LocalEnd.y = 0; } mediaWalker.UseScoreboard = m_renderHal->VfeScoreboard.ScoreboardEnable; mediaWalker.ScoreboardMask = m_renderHal->VfeScoreboard.ScoreboardMask; return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::PrepareComputeWalkerParams(KERNEL_WALKER_PARAMS params, MHW_GPGPU_WALKER_PARAMS &gpgpuWalker) { uint32_t uiMediaWalkerBlockSize; RECT alignedRect = {}; // Get media walker kernel block size uiMediaWalkerBlockSize = m_renderHal->pHwSizes->dwSizeMediaWalkerBlock; alignedRect = params.alignedRect; // Calculate aligned output area in order to determine the total # blocks // to process in case of non-16x16 aligned target. alignedRect.right += uiMediaWalkerBlockSize - 1; alignedRect.bottom += uiMediaWalkerBlockSize - 1; alignedRect.left -= alignedRect.left % uiMediaWalkerBlockSize; alignedRect.top -= alignedRect.top % uiMediaWalkerBlockSize; alignedRect.right -= alignedRect.right % uiMediaWalkerBlockSize; alignedRect.bottom -= alignedRect.bottom % uiMediaWalkerBlockSize; if (params.calculateBlockXYByAlignedRect) { // Set number of blocks params.iBlocksX = (alignedRect.right - alignedRect.left) / uiMediaWalkerBlockSize; params.iBlocksY = (alignedRect.bottom - alignedRect.top) / uiMediaWalkerBlockSize; } // Set walker cmd params - Rasterscan gpgpuWalker.InterfaceDescriptorOffset = params.iMediaID; // Specifies the initial value of the X component of the thread group when walker is started. // During the walker operation, when X is incremented to the X Dimension limit, on the next step // it is re-loaded with theStarting Xvalue. Same as GroupStartingY. gpgpuWalker.GroupStartingX = (alignedRect.left / uiMediaWalkerBlockSize); gpgpuWalker.GroupStartingY = (alignedRect.top / uiMediaWalkerBlockSize); // The X dimension of the thread group (maximum X is dimension -1), same as GroupHeight. gpgpuWalker.GroupWidth = params.iBlocksX; gpgpuWalker.GroupHeight = params.iBlocksY; gpgpuWalker.GroupDepth = params.iBlocksZ; if (params.isGroupStartInvolvedInGroupSize) { gpgpuWalker.GroupWidth += gpgpuWalker.GroupStartingX; gpgpuWalker.GroupHeight += gpgpuWalker.GroupStartingY; } if (params.threadDepth && params.threadWidth && params.threadHeight && ((params.isGenerateLocalID && params.emitLocal != MHW_EMIT_LOCAL_NONE) || params.slmSize > 0)) { gpgpuWalker.ThreadWidth = params.threadWidth; gpgpuWalker.ThreadHeight = params.threadHeight; gpgpuWalker.ThreadDepth = params.threadDepth; } else { gpgpuWalker.ThreadWidth = COMPUTE_WALKER_THREAD_SPACE_WIDTH; gpgpuWalker.ThreadHeight = COMPUTE_WALKER_THREAD_SPACE_HEIGHT; gpgpuWalker.ThreadDepth = COMPUTE_WALKER_THREAD_SPACE_DEPTH; } gpgpuWalker.simdSize = params.simdSize; gpgpuWalker.IndirectDataStartAddress = params.iCurbeOffset; // Indirect Data Length is a multiple of 64 bytes (size of L3 cacheline). Bits [5:0] are zero. gpgpuWalker.IndirectDataLength = MOS_ALIGN_CEIL(params.iCurbeLength, 1 << MHW_COMPUTE_INDIRECT_SHIFT); gpgpuWalker.BindingTableID = params.iBindingTable; gpgpuWalker.ForcePreferredSLMZero = params.forcePreferredSLMZero; gpgpuWalker.isEmitInlineParameter = params.isEmitInlineParameter; gpgpuWalker.inlineDataLength = params.inlineDataLength; gpgpuWalker.inlineData = params.inlineData; gpgpuWalker.isGenerateLocalID = params.isGenerateLocalID; gpgpuWalker.emitLocal = params.emitLocal; gpgpuWalker.SLMSize = params.slmSize; gpgpuWalker.hasBarrier = params.hasBarrier; gpgpuWalker.registersPerThread = params.registersPerThread; gpgpuWalker.curbeResourceList = params.curbeResourceList; gpgpuWalker.curbeResourceListSize = params.curbeResourceListSize; gpgpuWalker.inlineResourceList = params.inlineResourceList; gpgpuWalker.inlineResourceListSize = params.inlineResourceListSize; return MOS_STATUS_SUCCESS; } void RenderCmdPacket::UpdateKernelConfigParam(RENDERHAL_KERNEL_PARAM &kernelParam) { // CURBE_Length is set as the size of curbe buffer. 32 alignment with 5 bits right shift need be done before it being used. kernelParam.CURBE_Length = (kernelParam.CURBE_Length + 31) >> 5; } MOS_STATUS RenderCmdPacket::LoadKernel() { int32_t iKrnAllocation = 0; MHW_KERNEL_PARAM MhwKernelParam = {}; RENDERHAL_KERNEL_PARAM KernelParam = m_renderData.KernelParam; // Load kernel to GSH INIT_MHW_KERNEL_PARAM(MhwKernelParam, &m_renderData.KernelEntry); UpdateKernelConfigParam(KernelParam); iKrnAllocation = m_renderHal->pfnLoadKernel( m_renderHal, &KernelParam, &MhwKernelParam, nullptr); if (iKrnAllocation < 0) { RENDER_PACKET_ASSERTMESSAGE("kernel load failed"); return MOS_STATUS_UNKNOWN; } m_renderData.kernelAllocationID = iKrnAllocation; if (m_renderData.iCurbeOffset < 0) { RENDER_PACKET_ASSERTMESSAGE("Curbe Set Fail, return error"); return MOS_STATUS_UNKNOWN; } // Allocate Media ID, link to kernel m_renderData.mediaID = m_renderHal->pfnAllocateMediaID( m_renderHal, iKrnAllocation, m_renderData.bindingTable, m_renderData.iCurbeOffset, (m_renderData.iCurbeLength), 0, nullptr); if (m_renderData.mediaID < 0) { RENDER_PACKET_ASSERTMESSAGE("Allocate Media ID failed, return error"); return MOS_STATUS_UNKNOWN; } return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::InitRenderHalSurface(MOS_SURFACE surface, PRENDERHAL_SURFACE pRenderSurface) { RENDER_PACKET_CHK_NULL_RETURN(pRenderSurface); RENDERHAL_GET_SURFACE_INFO info; MOS_ZeroMemory(&info, sizeof(info)); RENDER_PACKET_CHK_STATUS_RETURN(RenderHal_GetSurfaceInfo( m_renderHal->pOsInterface, &info, &surface)); if (Mos_ResourceIsNull(&pRenderSurface->OsSurface.OsResource)) { pRenderSurface->OsSurface = surface; } return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::InitRenderHalBuffer(MOS_BUFFER surface, PRENDERHAL_SURFACE pRenderSurface) { RENDER_PACKET_CHK_NULL_RETURN(pRenderSurface); pRenderSurface->OsSurface.OsResource = surface.OsResource; pRenderSurface->OsSurface.dwWidth = surface.size; pRenderSurface->OsSurface.dwHeight = 1; pRenderSurface->OsSurface.dwPitch = surface.size; pRenderSurface->OsSurface.Format = Format_RAW; return MOS_STATUS_SUCCESS; } MOS_STATUS RenderCmdPacket::SendMultiKernelMediaStates( PRENDERHAL_INTERFACE pRenderHal, PMOS_COMMAND_BUFFER pCmdBuffer) { PMOS_INTERFACE pOsInterface = nullptr; PRENDERHAL_STATE_HEAP pStateHeap = nullptr; MOS_STATUS eStatus = MOS_STATUS_SUCCESS; MHW_VFE_PARAMS *pVfeStateParams = nullptr; MOS_CONTEXT *pOsContext = nullptr; MHW_MI_LOAD_REGISTER_IMM_PARAMS loadRegisterImmParams = {}; PMHW_MI_MMIOREGISTERS pMmioRegisters = nullptr; MOS_OCA_BUFFER_HANDLE hOcaBuf = 0; bool flushL1 = false; //--------------------------------------- MHW_RENDERHAL_CHK_NULL(pRenderHal); MHW_RENDERHAL_CHK_NULL(pRenderHal->pStateHeap); MHW_RENDERHAL_CHK_NULL(pRenderHal->pRenderHalPltInterface); MHW_RENDERHAL_ASSERT(pRenderHal->pStateHeap->bGshLocked); RENDER_PACKET_CHK_NULL_RETURN(pRenderHal->pRenderHalPltInterface); RENDER_PACKET_CHK_NULL_RETURN(pRenderHal->pRenderHalPltInterface->GetMmioRegisters(pRenderHal)); //--------------------------------------- pOsInterface = pRenderHal->pOsInterface; pStateHeap = pRenderHal->pStateHeap; pOsContext = pOsInterface->pOsContext; pMmioRegisters = pRenderHal->pRenderHalPltInterface->GetMmioRegisters(pRenderHal); // Setup L3$ Config, LRI commands used here & hence must be launched from a secure bb pRenderHal->L3CacheSettings.bEnableSLM = (m_walkerType == WALKER_TYPE_COMPUTE && m_slmSize > 0) ? true : false; MHW_RENDERHAL_CHK_STATUS(pRenderHal->pfnEnableL3Caching(pRenderHal, &pRenderHal->L3CacheSettings)); // Send L3 Cache Configuration MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->SetL3Cache(pRenderHal, pCmdBuffer)); MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->EnablePreemption(pRenderHal, pCmdBuffer)); // Send Pipeline Select command MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->AddPipelineSelectCmd(pRenderHal, pCmdBuffer, (m_walkerType == WALKER_TYPE_COMPUTE) ? true : false)); // The binding table for surface states is at end of command buffer. No need to add it to indirect state heap. HalOcaInterfaceNext::OnIndirectState(*pCmdBuffer, (MOS_CONTEXT_HANDLE)pOsContext, pRenderHal->StateBaseAddressParams.presInstructionBuffer, pStateHeap->CurIDEntryParams.dwKernelOffset, false, pStateHeap->iKernelUsedForDump); // Send State Base Address command MHW_RENDERHAL_CHK_STATUS(pRenderHal->pfnSendStateBaseAddress(pRenderHal, pCmdBuffer)); if (pRenderHal->bComputeContextInUse && !pRenderHal->isBindlessHeapInUse) { pRenderHal->pRenderHalPltInterface->SendTo3DStateBindingTablePoolAlloc(pRenderHal, pCmdBuffer); } // Send Surface States MHW_RENDERHAL_CHK_STATUS(pRenderHal->pfnSendSurfaces(pRenderHal, pCmdBuffer)); // Send SIP State if ASM debug enabled if (pRenderHal->bIsaAsmDebugEnable) { MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->AddSipStateCmd(pRenderHal, pCmdBuffer)); } pVfeStateParams = pRenderHal->pRenderHalPltInterface->GetVfeStateParameters(); if (!pRenderHal->bComputeContextInUse) { // set VFE State MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->AddMediaVfeCmd(pRenderHal, pCmdBuffer, pVfeStateParams)); } else { if (!pRenderHal->isBindlessHeapInUse) { // set CFE State MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->AddCfeStateCmd(pRenderHal, pCmdBuffer, pVfeStateParams)); } } // Send CURBE Load if (!pRenderHal->bComputeContextInUse) { MHW_RENDERHAL_CHK_STATUS(pRenderHal->pfnSendCurbeLoad(pRenderHal, pCmdBuffer)); } // Send Interface Desc Load if (!pRenderHal->bComputeContextInUse) { MHW_RENDERHAL_CHK_STATUS(pRenderHal->pfnSendMediaIdLoad(pRenderHal, pCmdBuffer)); } // Send Chroma Keys MHW_RENDERHAL_CHK_STATUS(pRenderHal->pfnSendChromaKey(pRenderHal, pCmdBuffer)); // Send Palettes in use MHW_RENDERHAL_CHK_STATUS(pRenderHal->pfnSendPalette(pRenderHal, pCmdBuffer)); pRenderHal->pRenderHalPltInterface->OnDispatch(pRenderHal, pCmdBuffer, pOsInterface, pMmioRegisters); for (uint32_t kernelIndex = 0; kernelIndex < m_kernelRenderData.size(); kernelIndex++) { auto it = m_kernelRenderData.find(kernelIndex); if (it == m_kernelRenderData.end()) { eStatus = MOS_STATUS_INVALID_PARAMETER; goto finish; } if (kernelIndex > 0 && it->second.walkerParam.bSyncFlag) { MHW_PIPE_CONTROL_PARAMS pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_NOWRITE; pipeCtlParams.dwFlushMode = MHW_FLUSH_CUSTOM; pipeCtlParams.bInvalidateTextureCache = true; pipeCtlParams.bFlushRenderTargetCache = true; if (it->second.walkerParam.pipeControlParams.bUpdateNeeded) { pipeCtlParams.bHdcPipelineFlush = it->second.walkerParam.pipeControlParams.bEnableDataPortFlush; pipeCtlParams.bUnTypedDataPortCacheFlush = it->second.walkerParam.pipeControlParams.bUnTypedDataPortCacheFlush; pipeCtlParams.bFlushRenderTargetCache = it->second.walkerParam.pipeControlParams.bFlushRenderTargetCache; pipeCtlParams.bInvalidateTextureCache = it->second.walkerParam.pipeControlParams.bInvalidateTextureCache; } if (flushL1) { //Flush L1 cache after consumer walker when there is a producer-consumer relationship walker. pipeCtlParams.bUnTypedDataPortCacheFlush = true; pipeCtlParams.bHdcPipelineFlush = true; } MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->AddMiPipeControl(pRenderHal, pCmdBuffer, &pipeCtlParams)); } if (m_walkerType == WALKER_TYPE_MEDIA) { MOS_ZeroMemory(&m_mediaWalkerParams, sizeof(m_mediaWalkerParams)); MHW_RENDERHAL_CHK_STATUS(PrepareMediaWalkerParams(it->second.walkerParam, m_mediaWalkerParams)); MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->AddMediaObjectWalkerCmd(pRenderHal, pCmdBuffer, &m_mediaWalkerParams)); } else if (m_walkerType == WALKER_TYPE_COMPUTE) { MOS_ZeroMemory(&m_gpgpuWalkerParams, sizeof(m_gpgpuWalkerParams)); MHW_RENDERHAL_CHK_STATUS(PrepareComputeWalkerParams(it->second.walkerParam, m_gpgpuWalkerParams)); pRenderHal->iKernelAllocationID = it->second.kernelAllocationID; MHW_RENDERHAL_CHK_STATUS(pRenderHal->pRenderHalPltInterface->SendComputeWalker( pRenderHal, pCmdBuffer, &m_gpgpuWalkerParams)); flushL1 = it->second.walkerParam.bFlushL1; } else { eStatus = MOS_STATUS_UNIMPLEMENTED; goto finish; } } // This need not be secure, since PPGTT will be used here. But moving this after // L3 cache configuration will delay UMD from fetching another media state. // Send Sync Tag MHW_RENDERHAL_CHK_STATUS(pRenderHal->pfnSendSyncTag(pRenderHal, pCmdBuffer)); m_kernelRenderData.clear(); finish: return eStatus; }