/* * Copyright (c) 2017-2020, 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 cm_hal_g9.cpp //! \brief Common HAL CM Gen9 functions //! #include "cm_hal_g9.h" #include "mhw_render_hwcmd_g9_X.h" #include "renderhal_platform_interface.h" #include "mhw_render_legacy.h" #include "hal_oca_interface.h" #if defined(ENABLE_KERNELS) && (!defined(_FULL_OPEN_SOURCE)) #include "cm_gpucopy_kernel_g9.h" #include "cm_gpuinit_kernel_g9.h" #else unsigned int iGPUCopy_kernel_isa_size_gen9 = 0; unsigned int iGPUInit_kernel_isa_size_Gen9 = 0; unsigned char *pGPUCopy_kernel_isa_gen9 = nullptr; unsigned char *pGPUInit_kernel_isa_Gen9 = nullptr; #endif #define CM_NS_PER_TICK_RENDER_G9 (83.333) // For SKL, 83.333 nano seconds per tick in render engine #define CM_NS_PER_TICK_RENDER_G9LP (52.083) //For BXT, 52.083 nano seconds per tick in render engine #define PLATFORM_INTEL_BXT 8 #define PLATFORM_INTEL_GLK 16 // Gen9 Surface state tokenized commands - a SURFACE_STATE_G9 command and // a surface state command, either SURFACE_STATE_G9 or SURFACE_STATE_ADV_G9 struct PACKET_SURFACE_STATE { SURFACE_STATE_TOKEN_COMMON token; union { mhw_state_heap_g9_X::RENDER_SURFACE_STATE_CMD cmdSurfaceState; mhw_state_heap_g9_X::MEDIA_SURFACE_STATE_CMD cmdSurfaceStateAdv; }; }; //------------------------------------------------------------------------------ //| Purpose: Sets Media Walker Parameters from engineering API for GEN9 //| Returns: Result of the operation //------------------------------------------------------------------------------ MOS_STATUS CM_HAL_G9_X::SetMediaWalkerParams( CM_WALKING_PARAMETERS engineeringParams, PCM_HAL_WALKER_PARAMS walkerParams) { mhw_render_g9_X::MEDIA_OBJECT_WALKER_CMD mediaWalkerCmd; mediaWalkerCmd.DW5.Value = engineeringParams.Value[0]; walkerParams->scoreboardMask = mediaWalkerCmd.DW5.ScoreboardMask; mediaWalkerCmd.DW6.Value = engineeringParams.Value[1]; walkerParams->colorCountMinusOne = mediaWalkerCmd.DW6.ColorCountMinusOne; walkerParams->midLoopUnitX = mediaWalkerCmd.DW6.MidLoopUnitX; walkerParams->midLoopUnitY = mediaWalkerCmd.DW6.LocalMidLoopUnitY; walkerParams->middleLoopExtraSteps = mediaWalkerCmd.DW6.MiddleLoopExtraSteps; mediaWalkerCmd.DW7.Value = engineeringParams.Value[2]; walkerParams->localLoopExecCount = mediaWalkerCmd.DW7.LocalLoopExecCount; walkerParams->globalLoopExecCount = mediaWalkerCmd.DW7.GlobalLoopExecCount; mediaWalkerCmd.DW8.Value = engineeringParams.Value[3]; walkerParams->blockResolution.x = mediaWalkerCmd.DW8.BlockResolutionX; walkerParams->blockResolution.y = mediaWalkerCmd.DW8.BlockResolutionY; mediaWalkerCmd.DW9.Value = engineeringParams.Value[4]; walkerParams->localStart.x = mediaWalkerCmd.DW9.LocalStartX; walkerParams->localStart.y = mediaWalkerCmd.DW9.LocalStartY; mediaWalkerCmd.DW11.Value = engineeringParams.Value[6]; walkerParams->localOutLoopStride.x = mediaWalkerCmd.DW11.LocalOuterLoopStrideX; walkerParams->localOutLoopStride.y = mediaWalkerCmd.DW11.LocalOuterLoopStrideY; mediaWalkerCmd.DW12.Value = engineeringParams.Value[7]; walkerParams->localInnerLoopUnit.x = mediaWalkerCmd.DW12.LocalInnerLoopUnitX; walkerParams->localInnerLoopUnit.y = mediaWalkerCmd.DW12.LocalInnerLoopUnitY; mediaWalkerCmd.DW13.Value = engineeringParams.Value[8]; walkerParams->globalResolution.x = mediaWalkerCmd.DW13.GlobalResolutionX; walkerParams->globalResolution.y = mediaWalkerCmd.DW13.GlobalResolutionY; mediaWalkerCmd.DW14.Value = engineeringParams.Value[9]; walkerParams->globalStart.x = mediaWalkerCmd.DW14.GlobalStartX; walkerParams->globalStart.y = mediaWalkerCmd.DW14.GlobalStartY; mediaWalkerCmd.DW15.Value = engineeringParams.Value[10]; walkerParams->globalOutlerLoopStride.x = mediaWalkerCmd.DW15.GlobalOuterLoopStrideX; walkerParams->globalOutlerLoopStride.y = mediaWalkerCmd.DW15.GlobalOuterLoopStrideY; mediaWalkerCmd.DW16.Value = engineeringParams.Value[11]; walkerParams->globalInnerLoopUnit.x = mediaWalkerCmd.DW16.GlobalInnerLoopUnitX; walkerParams->globalInnerLoopUnit.y = mediaWalkerCmd.DW16.GlobalInnerLoopUnitY; walkerParams->localEnd.x = 0; walkerParams->localEnd.y = 0; return MOS_STATUS_SUCCESS; } MOS_STATUS CM_HAL_G9_X::SetupHwDebugControl( PRENDERHAL_INTERFACE renderHal, PMOS_COMMAND_BUFFER cmdBuffer) { MOS_STATUS eStatus = MOS_STATUS_SUCCESS; MHW_MI_LOAD_REGISTER_IMM_PARAMS loadRegImm; //--------------------------------------- CM_CHK_NULL_GOTOFINISH_MOSERROR(renderHal); CM_CHK_NULL_GOTOFINISH_MOSERROR(renderHal->pMhwMiInterface); CM_CHK_NULL_GOTOFINISH_MOSERROR(cmdBuffer); //--------------------------------------- MOS_ZeroMemory(&loadRegImm, sizeof(MHW_MI_LOAD_REGISTER_IMM_PARAMS)); // CS_DEBUG_MODE1, global debug enable loadRegImm.dwRegister = CS_DEBUG_MODE1; loadRegImm.dwData = (CS_DEBUG_MODE1_GLOBAL_DEBUG << 16) | CS_DEBUG_MODE1_GLOBAL_DEBUG; CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pMhwMiInterface->AddMiLoadRegisterImmCmd(cmdBuffer, &loadRegImm)); // TD_CTL, force thread breakpoint enable // Also enable external exception, because the source-level debugger has to // be able to interrupt runing EU threads. loadRegImm.dwRegister = TD_CTL; loadRegImm.dwData = TD_CTL_FORCE_THREAD_BKPT_ENABLE | TD_CTL_FORCE_EXT_EXCEPTION_ENABLE; CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pMhwMiInterface->AddMiLoadRegisterImmCmd(cmdBuffer, &loadRegImm)); finish: return eStatus; } //------------------------------------------------------------------------------ //| Purpose: Registers the sampler 8x8 AVS table: DWORDS 152, 153 and coefficients //| Returns: Result of the operation //------------------------------------------------------------------------------ MOS_STATUS CM_HAL_G9_X::RegisterSampler8x8AVSTable( PCM_HAL_SAMPLER_8X8_TABLE sampler8x8AvsTable, PCM_AVS_TABLE_STATE_PARAMS avsTable ) { MOS_ZeroMemory( &sampler8x8AvsTable->mhwSamplerAvsTableParam, sizeof( sampler8x8AvsTable->mhwSamplerAvsTableParam ) ); sampler8x8AvsTable->mhwSamplerAvsTableParam.byteTransitionArea8Pixels = MEDIASTATE_AVS_TRANSITION_AREA_8_PIXELS; sampler8x8AvsTable->mhwSamplerAvsTableParam.byteTransitionArea4Pixels = MEDIASTATE_AVS_TRANSITION_AREA_4_PIXELS; sampler8x8AvsTable->mhwSamplerAvsTableParam.byteMaxDerivative8Pixels = MEDIASTATE_AVS_MAX_DERIVATIVE_8_PIXELS; sampler8x8AvsTable->mhwSamplerAvsTableParam.byteMaxDerivative4Pixels = MEDIASTATE_AVS_MAX_DERIVATIVE_4_PIXELS; sampler8x8AvsTable->mhwSamplerAvsTableParam.bEnableRGBAdaptive = false; sampler8x8AvsTable->mhwSamplerAvsTableParam.bAdaptiveFilterAllChannels = avsTable->adaptiveFilterAllChannels; // Assign the coefficient table; for ( uint32_t i = 0; i < CM_NUM_HW_POLYPHASE_TABLES_G9; i++ ) { sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroXFilterCoefficient[0] = ( uint8_t )avsTable->tbl0X[ i ].FilterCoeff_0_0; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroXFilterCoefficient[1] = ( uint8_t )avsTable->tbl0X[ i ].FilterCoeff_0_1; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroXFilterCoefficient[2] = ( uint8_t )avsTable->tbl0X[ i ].FilterCoeff_0_2; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroXFilterCoefficient[3] = ( uint8_t )avsTable->tbl0X[ i ].FilterCoeff_0_3; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroXFilterCoefficient[4] = ( uint8_t )avsTable->tbl0X[ i ].FilterCoeff_0_4; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroXFilterCoefficient[5] = ( uint8_t )avsTable->tbl0X[ i ].FilterCoeff_0_5; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroXFilterCoefficient[6] = ( uint8_t )avsTable->tbl0X[ i ].FilterCoeff_0_6; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroXFilterCoefficient[7] = ( uint8_t )avsTable->tbl0X[ i ].FilterCoeff_0_7; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroYFilterCoefficient[0] = ( uint8_t )avsTable->tbl0Y[ i ].FilterCoeff_0_0; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroYFilterCoefficient[1] = ( uint8_t )avsTable->tbl0Y[ i ].FilterCoeff_0_1; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroYFilterCoefficient[2] = ( uint8_t )avsTable->tbl0Y[ i ].FilterCoeff_0_2; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroYFilterCoefficient[3] = ( uint8_t )avsTable->tbl0Y[ i ].FilterCoeff_0_3; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroYFilterCoefficient[4] = ( uint8_t )avsTable->tbl0Y[ i ].FilterCoeff_0_4; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroYFilterCoefficient[5] = ( uint8_t )avsTable->tbl0Y[ i ].FilterCoeff_0_5; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroYFilterCoefficient[6] = ( uint8_t )avsTable->tbl0Y[ i ].FilterCoeff_0_6; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].ZeroYFilterCoefficient[7] = ( uint8_t )avsTable->tbl0Y[ i ].FilterCoeff_0_7; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].OneXFilterCoefficient[0] = ( uint8_t )avsTable->tbl1X[ i ].FilterCoeff_0_2; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].OneXFilterCoefficient[1] = ( uint8_t )avsTable->tbl1X[ i ].FilterCoeff_0_3; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].OneXFilterCoefficient[2] = ( uint8_t )avsTable->tbl1X[ i ].FilterCoeff_0_4; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].OneXFilterCoefficient[3] = ( uint8_t )avsTable->tbl1X[ i ].FilterCoeff_0_5; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].OneYFilterCoefficient[0] = ( uint8_t )avsTable->tbl1Y[ i ].FilterCoeff_0_2; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].OneYFilterCoefficient[1] = ( uint8_t )avsTable->tbl1Y[ i ].FilterCoeff_0_3; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].OneYFilterCoefficient[2] = ( uint8_t )avsTable->tbl1Y[ i ].FilterCoeff_0_4; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParam[ i ].OneYFilterCoefficient[3] = ( uint8_t )avsTable->tbl1Y[ i ].FilterCoeff_0_5; } sampler8x8AvsTable->mhwSamplerAvsTableParam.byteDefaultSharpnessLevel = avsTable->defaultSharpLevel; sampler8x8AvsTable->mhwSamplerAvsTableParam.bBypassXAdaptiveFiltering = avsTable->bypassXAF; sampler8x8AvsTable->mhwSamplerAvsTableParam.bBypassYAdaptiveFiltering = avsTable->bypassYAF; if ( !avsTable->bypassXAF && !avsTable->bypassYAF ) { sampler8x8AvsTable->mhwSamplerAvsTableParam.byteMaxDerivative8Pixels = avsTable->maxDerivative8Pixels; sampler8x8AvsTable->mhwSamplerAvsTableParam.byteMaxDerivative4Pixels = avsTable->maxDerivative4Pixels; sampler8x8AvsTable->mhwSamplerAvsTableParam.byteTransitionArea8Pixels = avsTable->transitionArea8Pixels; sampler8x8AvsTable->mhwSamplerAvsTableParam.byteTransitionArea4Pixels = avsTable->transitionArea4Pixels; } for ( int i = 0; i < CM_NUM_HW_POLYPHASE_EXTRA_TABLES_G9; i++ ) { int src = i + CM_NUM_HW_POLYPHASE_TABLES_G9; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroXFilterCoefficient[0] = ( uint8_t )avsTable->tbl0X[ src ].FilterCoeff_0_0; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroXFilterCoefficient[1] = ( uint8_t )avsTable->tbl0X[ src ].FilterCoeff_0_1; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroXFilterCoefficient[2] = ( uint8_t )avsTable->tbl0X[ src ].FilterCoeff_0_2; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroXFilterCoefficient[3] = ( uint8_t )avsTable->tbl0X[ src ].FilterCoeff_0_3; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroXFilterCoefficient[4] = ( uint8_t )avsTable->tbl0X[ src ].FilterCoeff_0_4; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroXFilterCoefficient[5] = ( uint8_t )avsTable->tbl0X[ src ].FilterCoeff_0_5; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroXFilterCoefficient[6] = ( uint8_t )avsTable->tbl0X[ src ].FilterCoeff_0_6; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroXFilterCoefficient[7] = ( uint8_t )avsTable->tbl0X[ src ].FilterCoeff_0_7; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroYFilterCoefficient[0] = ( uint8_t )avsTable->tbl0Y[ src ].FilterCoeff_0_0; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroYFilterCoefficient[1] = ( uint8_t )avsTable->tbl0Y[ src ].FilterCoeff_0_1; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroYFilterCoefficient[2] = ( uint8_t )avsTable->tbl0Y[ src ].FilterCoeff_0_2; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroYFilterCoefficient[3] = ( uint8_t )avsTable->tbl0Y[ src ].FilterCoeff_0_3; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroYFilterCoefficient[4] = ( uint8_t )avsTable->tbl0Y[ src ].FilterCoeff_0_4; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroYFilterCoefficient[5] = ( uint8_t )avsTable->tbl0Y[ src ].FilterCoeff_0_5; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroYFilterCoefficient[6] = ( uint8_t )avsTable->tbl0Y[ src ].FilterCoeff_0_6; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].ZeroYFilterCoefficient[7] = ( uint8_t )avsTable->tbl0Y[ src ].FilterCoeff_0_7; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].OneXFilterCoefficient[0] = ( uint8_t )avsTable->tbl1X[ src ].FilterCoeff_0_2; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].OneXFilterCoefficient[1] = ( uint8_t )avsTable->tbl1X[ src ].FilterCoeff_0_3; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].OneXFilterCoefficient[2] = ( uint8_t )avsTable->tbl1X[ src ].FilterCoeff_0_4; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].OneXFilterCoefficient[3] = ( uint8_t )avsTable->tbl1X[ src ].FilterCoeff_0_5; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].OneYFilterCoefficient[0] = ( uint8_t )avsTable->tbl1Y[ src ].FilterCoeff_0_2; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].OneYFilterCoefficient[1] = ( uint8_t )avsTable->tbl1Y[ src ].FilterCoeff_0_3; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].OneYFilterCoefficient[2] = ( uint8_t )avsTable->tbl1Y[ src ].FilterCoeff_0_4; sampler8x8AvsTable->mhwSamplerAvsTableParam.paMhwAvsCoeffParamExtra[ i ].OneYFilterCoefficient[3] = ( uint8_t )avsTable->tbl1Y[ src ].FilterCoeff_0_5; } return MOS_STATUS_SUCCESS; } MOS_STATUS CM_HAL_G9_X::RegisterSampler8x8( PCM_HAL_SAMPLER_8X8_PARAM param) { PCM_HAL_STATE state = m_cmState; MOS_STATUS eStatus = MOS_STATUS_SUCCESS; int16_t samplerIndex = 0; PMHW_SAMPLER_STATE_PARAM samplerEntry = nullptr; PCM_HAL_SAMPLER_8X8_ENTRY sampler8x8Entry = nullptr; if (param->sampler8x8State.stateType == CM_SAMPLER8X8_AVS) { for (uint32_t i = 0; i < state->cmDeviceParam.maxSamplerTableSize; i++) { if (!state->samplerTable[i].bInUse) { samplerEntry = &state->samplerTable[i]; param->handle = (uint32_t)i << 16; samplerEntry->bInUse = true; break; } } for (uint32_t i = 0; i < state->cmDeviceParam.maxSampler8x8TableSize; i++) { if (!state->sampler8x8Table[i].inUse) { sampler8x8Entry = &state->sampler8x8Table[i]; samplerIndex = (int16_t)i; param->handle |= (uint32_t)(i & 0xffff); sampler8x8Entry->inUse = true; break; } } if (!samplerEntry || !sampler8x8Entry) { eStatus = MOS_STATUS_INVALID_PARAMETER; CM_ASSERTMESSAGE("Sampler or AVS table is full"); goto finish; } //State data from application samplerEntry->SamplerType = MHW_SAMPLER_TYPE_AVS; samplerEntry->ElementType = MHW_Sampler128Elements; samplerEntry->Avs = param->sampler8x8State.avsParam.avsState; samplerEntry->Avs.stateID = samplerIndex; samplerEntry->Avs.iTable8x8_Index = samplerIndex; // Used for calculating the Media offset of 8x8 table samplerEntry->Avs.pMhwSamplerAvsTableParam = &sampler8x8Entry->sampler8x8State.mhwSamplerAvsTableParam; if (samplerEntry->Avs.EightTapAFEnable) param->sampler8x8State.avsParam.avsTable.adaptiveFilterAllChannels = true; else param->sampler8x8State.avsParam.avsTable.adaptiveFilterAllChannels = false; CM_CHK_MOSSTATUS_GOTOFINISH(RegisterSampler8x8AVSTable(&sampler8x8Entry->sampler8x8State, ¶m->sampler8x8State.avsParam.avsTable)); sampler8x8Entry->sampler8x8State.stateType = CM_SAMPLER8X8_AVS; } else if (param->sampler8x8State.stateType == CM_SAMPLER8X8_MISC) { for (uint32_t i = 0; i < state->cmDeviceParam.maxSamplerTableSize; i++) { if (!state->samplerTable[i].bInUse) { samplerEntry = &state->samplerTable[i]; param->handle = (uint32_t)i << 16; samplerEntry->bInUse = true; break; } } if ( samplerEntry == nullptr ) { return MOS_STATUS_INVALID_HANDLE; } samplerEntry->SamplerType = MHW_SAMPLER_TYPE_MISC; samplerEntry->ElementType = MHW_Sampler2Elements; samplerEntry->Misc.byteHeight = param->sampler8x8State.miscState.DW0.Height; samplerEntry->Misc.byteWidth = param->sampler8x8State.miscState.DW0.Width; samplerEntry->Misc.wRow[0] = param->sampler8x8State.miscState.DW0.Row0; samplerEntry->Misc.wRow[1] = param->sampler8x8State.miscState.DW1.Row1; samplerEntry->Misc.wRow[2] = param->sampler8x8State.miscState.DW1.Row2; samplerEntry->Misc.wRow[3] = param->sampler8x8State.miscState.DW2.Row3; samplerEntry->Misc.wRow[4] = param->sampler8x8State.miscState.DW2.Row4; samplerEntry->Misc.wRow[5] = param->sampler8x8State.miscState.DW3.Row5; samplerEntry->Misc.wRow[6] = param->sampler8x8State.miscState.DW3.Row6; samplerEntry->Misc.wRow[7] = param->sampler8x8State.miscState.DW4.Row7; samplerEntry->Misc.wRow[8] = param->sampler8x8State.miscState.DW4.Row8; samplerEntry->Misc.wRow[9] = param->sampler8x8State.miscState.DW5.Row9; samplerEntry->Misc.wRow[10] = param->sampler8x8State.miscState.DW5.Row10; samplerEntry->Misc.wRow[11] = param->sampler8x8State.miscState.DW6.Row11; samplerEntry->Misc.wRow[12] = param->sampler8x8State.miscState.DW6.Row12; samplerEntry->Misc.wRow[13] = param->sampler8x8State.miscState.DW7.Row13; samplerEntry->Misc.wRow[14] = param->sampler8x8State.miscState.DW7.Row14; } else if (param->sampler8x8State.stateType == CM_SAMPLER8X8_CONV) { for (uint32_t i = 0; i < state->cmDeviceParam.maxSamplerTableSize; i++) { if (!state->samplerTable[i].bInUse) { samplerEntry = &state->samplerTable[i]; param->handle = (uint32_t)i << 16; samplerEntry->bInUse = true; break; } } if ( samplerEntry == nullptr ) { return MOS_STATUS_INVALID_HANDLE; } MOS_ZeroMemory(&samplerEntry->Convolve, sizeof(samplerEntry->Convolve)); samplerEntry->SamplerType = MHW_SAMPLER_TYPE_CONV; samplerEntry->Convolve.ui8Height = param->sampler8x8State.convolveState.height; samplerEntry->Convolve.ui8Width = param->sampler8x8State.convolveState.width; samplerEntry->Convolve.ui8ScaledDownValue = param->sampler8x8State.convolveState.scaleDownValue; samplerEntry->Convolve.ui8SizeOfTheCoefficient = param->sampler8x8State.convolveState.coeffSize; samplerEntry->Convolve.ui8MSBWidth = param->sampler8x8State.convolveState.isHorizontal32Mode; samplerEntry->Convolve.ui8MSBHeight = param->sampler8x8State.convolveState.isVertical32Mode; samplerEntry->Convolve.skl_mode = param->sampler8x8State.convolveState.sklMode; // Currently use DW0.Reserved0 to save the detailed Convolve Type, the DW0.Reserved0 will be cleared when copy to sampelr heap samplerEntry->Convolve.ui8ConvolveType = param->sampler8x8State.convolveState.nConvolveType; if (samplerEntry->Convolve.skl_mode && samplerEntry->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_2D) { samplerEntry->ElementType = MHW_Sampler128Elements; } else if ((!samplerEntry->Convolve.skl_mode && samplerEntry->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_2D) || samplerEntry->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_1P) { samplerEntry->ElementType = MHW_Sampler64Elements; } else { samplerEntry->ElementType = MHW_Sampler8Elements; } for ( int i = 0; i < CM_NUM_CONVOLVE_ROWS_SKL; i++ ) { MHW_SAMPLER_CONVOLVE_COEFF_TABLE *coeffTable = &(samplerEntry->Convolve.CoeffTable[i]); CM_HAL_CONVOLVE_COEFF_TABLE *sourceTable = &(param->sampler8x8State.convolveState.table[i]); if ( samplerEntry->Convolve.ui8SizeOfTheCoefficient == 1 ) { coeffTable->wFilterCoeff[0] = FloatToS3_12( sourceTable->FilterCoeff_0_0 ); coeffTable->wFilterCoeff[1] = FloatToS3_12( sourceTable->FilterCoeff_0_1 ); coeffTable->wFilterCoeff[2] = FloatToS3_12( sourceTable->FilterCoeff_0_2 ); coeffTable->wFilterCoeff[3] = FloatToS3_12( sourceTable->FilterCoeff_0_3 ); coeffTable->wFilterCoeff[4] = FloatToS3_12( sourceTable->FilterCoeff_0_4 ); coeffTable->wFilterCoeff[5] = FloatToS3_12( sourceTable->FilterCoeff_0_5 ); coeffTable->wFilterCoeff[6] = FloatToS3_12( sourceTable->FilterCoeff_0_6 ); coeffTable->wFilterCoeff[7] = FloatToS3_12( sourceTable->FilterCoeff_0_7 ); coeffTable->wFilterCoeff[8] = FloatToS3_12( sourceTable->FilterCoeff_0_8 ); coeffTable->wFilterCoeff[9] = FloatToS3_12( sourceTable->FilterCoeff_0_9 ); coeffTable->wFilterCoeff[10] = FloatToS3_12( sourceTable->FilterCoeff_0_10 ); coeffTable->wFilterCoeff[11] = FloatToS3_12( sourceTable->FilterCoeff_0_11 ); coeffTable->wFilterCoeff[12] = FloatToS3_12( sourceTable->FilterCoeff_0_12 ); coeffTable->wFilterCoeff[13] = FloatToS3_12( sourceTable->FilterCoeff_0_13 ); coeffTable->wFilterCoeff[14] = FloatToS3_12( sourceTable->FilterCoeff_0_14 ); coeffTable->wFilterCoeff[15] = FloatToS3_12( sourceTable->FilterCoeff_0_15 ); } else { coeffTable->wFilterCoeff[0] = FloatToS3_4( sourceTable->FilterCoeff_0_0 ); coeffTable->wFilterCoeff[1] = FloatToS3_4( sourceTable->FilterCoeff_0_1 ); coeffTable->wFilterCoeff[2] = FloatToS3_4( sourceTable->FilterCoeff_0_2 ); coeffTable->wFilterCoeff[3] = FloatToS3_4( sourceTable->FilterCoeff_0_3 ); coeffTable->wFilterCoeff[4] = FloatToS3_4( sourceTable->FilterCoeff_0_4 ); coeffTable->wFilterCoeff[5] = FloatToS3_4( sourceTable->FilterCoeff_0_5 ); coeffTable->wFilterCoeff[6] = FloatToS3_4( sourceTable->FilterCoeff_0_6 ); coeffTable->wFilterCoeff[7] = FloatToS3_4( sourceTable->FilterCoeff_0_7 ); coeffTable->wFilterCoeff[8] = FloatToS3_4( sourceTable->FilterCoeff_0_8 ); coeffTable->wFilterCoeff[9] = FloatToS3_4( sourceTable->FilterCoeff_0_9 ); coeffTable->wFilterCoeff[10] = FloatToS3_4( sourceTable->FilterCoeff_0_10 ); coeffTable->wFilterCoeff[11] = FloatToS3_4( sourceTable->FilterCoeff_0_11 ); coeffTable->wFilterCoeff[12] = FloatToS3_4( sourceTable->FilterCoeff_0_12 ); coeffTable->wFilterCoeff[13] = FloatToS3_4( sourceTable->FilterCoeff_0_13 ); coeffTable->wFilterCoeff[14] = FloatToS3_4( sourceTable->FilterCoeff_0_14 ); coeffTable->wFilterCoeff[15] = FloatToS3_4( sourceTable->FilterCoeff_0_15 ); } } for ( int i = CM_NUM_CONVOLVE_ROWS_SKL; i < CM_NUM_CONVOLVE_ROWS_SKL * 2; i++ ) { MHW_SAMPLER_CONVOLVE_COEFF_TABLE *coeffTable = &(samplerEntry->Convolve.CoeffTable[i]); CM_HAL_CONVOLVE_COEFF_TABLE *sourceTable = &(param->sampler8x8State.convolveState.table[i - CM_NUM_CONVOLVE_ROWS_SKL]); if ( samplerEntry->Convolve.ui8SizeOfTheCoefficient == 1 ) { coeffTable->wFilterCoeff[0] = FloatToS3_12( sourceTable->FilterCoeff_0_16 ); coeffTable->wFilterCoeff[1] = FloatToS3_12( sourceTable->FilterCoeff_0_17 ); coeffTable->wFilterCoeff[2] = FloatToS3_12( sourceTable->FilterCoeff_0_18 ); coeffTable->wFilterCoeff[3] = FloatToS3_12( sourceTable->FilterCoeff_0_19 ); coeffTable->wFilterCoeff[4] = FloatToS3_12( sourceTable->FilterCoeff_0_20 ); coeffTable->wFilterCoeff[5] = FloatToS3_12( sourceTable->FilterCoeff_0_21 ); coeffTable->wFilterCoeff[6] = FloatToS3_12( sourceTable->FilterCoeff_0_22 ); coeffTable->wFilterCoeff[7] = FloatToS3_12( sourceTable->FilterCoeff_0_23 ); coeffTable->wFilterCoeff[8] = FloatToS3_12( sourceTable->FilterCoeff_0_24 ); coeffTable->wFilterCoeff[9] = FloatToS3_12( sourceTable->FilterCoeff_0_25 ); coeffTable->wFilterCoeff[10] = FloatToS3_12( sourceTable->FilterCoeff_0_26 ); coeffTable->wFilterCoeff[11] = FloatToS3_12( sourceTable->FilterCoeff_0_27 ); coeffTable->wFilterCoeff[12] = FloatToS3_12( sourceTable->FilterCoeff_0_28 ); coeffTable->wFilterCoeff[13] = FloatToS3_12( sourceTable->FilterCoeff_0_29 ); coeffTable->wFilterCoeff[14] = FloatToS3_12( sourceTable->FilterCoeff_0_30 ); coeffTable->wFilterCoeff[15] = FloatToS3_12( sourceTable->FilterCoeff_0_31 ); } else { coeffTable->wFilterCoeff[0] = FloatToS3_4( sourceTable->FilterCoeff_0_16 ); coeffTable->wFilterCoeff[1] = FloatToS3_4( sourceTable->FilterCoeff_0_17 ); coeffTable->wFilterCoeff[2] = FloatToS3_4( sourceTable->FilterCoeff_0_18 ); coeffTable->wFilterCoeff[3] = FloatToS3_4( sourceTable->FilterCoeff_0_19 ); coeffTable->wFilterCoeff[4] = FloatToS3_4( sourceTable->FilterCoeff_0_20 ); coeffTable->wFilterCoeff[5] = FloatToS3_4( sourceTable->FilterCoeff_0_21 ); coeffTable->wFilterCoeff[6] = FloatToS3_4( sourceTable->FilterCoeff_0_22 ); coeffTable->wFilterCoeff[7] = FloatToS3_4( sourceTable->FilterCoeff_0_23 ); coeffTable->wFilterCoeff[8] = FloatToS3_4( sourceTable->FilterCoeff_0_24 ); coeffTable->wFilterCoeff[9] = FloatToS3_4( sourceTable->FilterCoeff_0_25 ); coeffTable->wFilterCoeff[10] = FloatToS3_4( sourceTable->FilterCoeff_0_26 ); coeffTable->wFilterCoeff[11] = FloatToS3_4( sourceTable->FilterCoeff_0_27 ); coeffTable->wFilterCoeff[12] = FloatToS3_4( sourceTable->FilterCoeff_0_28 ); coeffTable->wFilterCoeff[13] = FloatToS3_4( sourceTable->FilterCoeff_0_29 ); coeffTable->wFilterCoeff[14] = FloatToS3_4( sourceTable->FilterCoeff_0_30 ); coeffTable->wFilterCoeff[15] = FloatToS3_4( sourceTable->FilterCoeff_0_31 ); } } } finish: return eStatus; } /*---------------------------------------------------------------------------- | Purpose : Set's surface state memory object control settings | Returns : dword value \---------------------------------------------------------------------------*/ MOS_STATUS CM_HAL_G9_X::HwSetSurfaceMemoryObjectControl( uint16_t memObjCtl, PRENDERHAL_SURFACE_STATE_PARAMS surfStateParams) { PRENDERHAL_INTERFACE renderHal = m_cmState->renderHal; MOS_STATUS eStatus = MOS_STATUS_SUCCESS; MOS_HW_RESOURCE_DEF mosUsage; // The memory object control uint16_t is composed with cache type(8:15), memory type(4:7), ages(0:3) mosUsage = (MOS_HW_RESOURCE_DEF)((memObjCtl & CM_MEMOBJCTL_CACHE_MASK) >> 8); if (mosUsage >= MOS_HW_RESOURCE_DEF_MAX) mosUsage = GetDefaultMOCS(); surfStateParams->MemObjCtl = renderHal->pOsInterface->pfnCachePolicyGetMemoryObject(mosUsage, renderHal->pOsInterface->pfnGetGmmClientContext(renderHal->pOsInterface)).DwordValue; return eStatus; } #if (_RELEASE_INTERNAL || _DEBUG) #if defined (CM_DIRECT_GUC_SUPPORT) MOS_STATUS CM_HAL_G9_X::SubmitDummyCommands( PMHW_BATCH_BUFFER batchBuffer, int32_t taskId, PCM_HAL_KERNEL_PARAM *kernelParam, void **cmdBuffer) { MOS_STATUS eStatus = MOS_STATUS_SUCCESS; PCM_HAL_STATE state = m_cmState; PRENDERHAL_INTERFACE_LEGACY renderHal = state->renderHal; MhwRenderInterface *mhwRender = renderHal->pMhwRenderInterface; PRENDERHAL_STATE_HEAP stateHeap = renderHal->pStateHeap; PMOS_INTERFACE osInterface = renderHal->pOsInterface; PMHW_MI_INTERFACE mhwMiInterface = renderHal->pMhwMiInterface; MHW_PIPE_CONTROL_PARAMS pipeCtlParams; MHW_ID_LOAD_PARAMS idLoadParams; int32_t remaining = 0; bool enableWalker = state->walkerParams.CmWalkerEnable; bool enableGpGpu = state->taskParam->blGpGpuWalkerEnabled; MOS_COMMAND_BUFFER mosCmdBuffer; int64_t *taskSyncLocation; int32_t syncOffset; int32_t tmp; RENDERHAL_GENERIC_PROLOG_PARAMS genericPrologParams = {}; MOS_ZeroMemory(&mosCmdBuffer, sizeof(MOS_COMMAND_BUFFER)); // Get the task sync offset syncOffset = state->pfnGetTaskSyncLocation(taskId); // Initialize the location taskSyncLocation = (int64_t*)(state->renderTimeStampResource.data + syncOffset); *taskSyncLocation = CM_INVALID_INDEX; *(taskSyncLocation + 1) = CM_INVALID_INDEX; if (state->cbbEnabled) { *(taskSyncLocation + 2) = CM_INVALID_TAG; } // Register batch buffer for rendering if (!enableWalker && !enableGpGpu) { CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnRegisterResource( osInterface, &batchBuffer->OsResource, true, true)); } // Register Timestamp Buffer CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnRegisterResource( osInterface, &state->renderTimeStampResource.osResource, true, true)); // Allocate all available space, unused buffer will be returned later CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnGetCommandBuffer(osInterface, &mosCmdBuffer, 0)); remaining = mosCmdBuffer.iRemaining; // Linux will just return next sync tag here since currently no frame tracking support //dwFrameId = pRenderHal->pfnEnableFrameTracking(pRenderHal, pOsInterface->CurrentGpuContextOrdinal, &genericPrologParams, &OsResource); //pStateHeap->pCurMediaState->dwSyncTag = dwFrameId; // Initialize command buffer and insert prolog CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pfnInitCommandBuffer(renderHal, &mosCmdBuffer, &genericPrologParams)); //Send the First PipeControl Command to indicate the beginning of execution pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwResourceOffset = syncOffset; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_WRITE_TIMESTAMP_REG; pipeCtlParams.dwFlushMode = MHW_FLUSH_WRITE_CACHE; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); // Send Pipeline Select command CM_CHK_MOSSTATUS_GOTOFINISH(mhwRender->AddPipelineSelectCmd(&mosCmdBuffer, enableGpGpu)); // issue a PIPE_CONTROL to flush all caches and the stall the CS before // issuing a PIPE_CONTROL to write the timestamp pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_NOWRITE; pipeCtlParams.dwFlushMode = MHW_FLUSH_WRITE_CACHE; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); // issue a PIPE_CONTROL to write timestamp syncOffset += sizeof(uint64_t); pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwResourceOffset = syncOffset; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_WRITE_TIMESTAMP_REG; pipeCtlParams.dwFlushMode = MHW_FLUSH_READ_CACHE; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); // Add PipeControl to invalidate ISP and MediaState to avoid PageFault issue 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; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeControlParams)); //Couple to the BB_START , otherwise GPU Hang without it in Linux KMD CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddMiBatchBufferEnd(&mosCmdBuffer, nullptr)); // Return unused command buffer space to OS osInterface->pfnReturnCommandBuffer(osInterface, &mosCmdBuffer, 0); CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnSubmitCommandBuffer(osInterface, &mosCmdBuffer, state->nullHwRenderCm)) if (state->nullHwRenderCm == false) { stateHeap->pCurMediaState->bBusy = true; if (!enableWalker && !enableGpGpu) { batchBuffer->bBusy = true; } } // reset API call number of HW threads state->maxHWThreadValues.apiValue = 0; state->pfnReferenceCommandBuffer(&mosCmdBuffer.OsResource, cmdBuffer); eStatus = MOS_STATUS_SUCCESS; finish: // Failed -> discard all changes in Command Buffer if (eStatus != MOS_STATUS_SUCCESS) { // Buffer overflow - display overflow size if (mosCmdBuffer.iRemaining < 0) { CM_ASSERTMESSAGE("Command Buffer overflow by %d bytes.", -mosCmdBuffer.iRemaining); } // Move command buffer back to beginning tmp = remaining - mosCmdBuffer.iRemaining; mosCmdBuffer.iRemaining = remaining; mosCmdBuffer.iOffset -= tmp; mosCmdBuffer.pCmdPtr = mosCmdBuffer.pCmdBase + mosCmdBuffer.iOffset / sizeof(uint32_t); // Return unused command buffer space to OS osInterface->pfnReturnCommandBuffer(osInterface, &mosCmdBuffer, 0); } return eStatus; } #endif #endif MOS_STATUS CM_HAL_G9_X::SubmitCommands( PMHW_BATCH_BUFFER batchBuffer, int32_t taskId, PCM_HAL_KERNEL_PARAM *kernelParam, void **cmdBuffer) { MOS_STATUS eStatus = MOS_STATUS_SUCCESS; PCM_HAL_STATE state = m_cmState; PRENDERHAL_INTERFACE_LEGACY renderHal = state->renderHal; MhwRenderInterface *mhwRender = renderHal->pMhwRenderInterface; PRENDERHAL_STATE_HEAP stateHeap = renderHal->pStateHeap; PMOS_INTERFACE osInterface = renderHal->pOsInterface; PMHW_MI_INTERFACE mhwMiInterface = renderHal->pMhwMiInterface; MHW_PIPE_CONTROL_PARAMS pipeCtlParams; MHW_ID_LOAD_PARAMS idLoadParams; int32_t remaining = 0; bool enableWalker = state->walkerParams.CmWalkerEnable; bool enableGpGpu = state->taskParam->blGpGpuWalkerEnabled; MOS_COMMAND_BUFFER mosCmdBuffer; uint32_t syncTag; int64_t *taskSyncLocation; int32_t syncOffset; int32_t tmp; PCM_HAL_TASK_PARAM taskParam = state->taskParam; bool sipEnable = renderHal->bSIPKernel? true: false; bool csrEnable = renderHal->bCSRKernel? true: false; PCM_HAL_BB_ARGS bbCmArgs; RENDERHAL_GENERIC_PROLOG_PARAMS genericPrologParams = {}; MOS_RESOURCE *osResource; uint32_t tag; uint32_t tagOffset = 0; CM_HAL_MI_REG_OFFSETS miRegG9 = { REG_TIMESTAMP_BASE_G9, REG_GPR_BASE_G9 }; #if (_RELEASE_INTERNAL || _DEBUG) #if defined (CM_DIRECT_GUC_SUPPORT) uint64_t batchbufferaddress; #endif #endif MOS_CONTEXT *pOsContext = renderHal->pOsInterface->pOsContext; PMHW_MI_MMIOREGISTERS pMmioRegisters = renderHal->pMhwRenderInterface->GetMmioRegisters(); MOS_ZeroMemory(&mosCmdBuffer, sizeof(MOS_COMMAND_BUFFER)); // get the tag tag = renderHal->trackerProducer.GetNextTracker(renderHal->currentTrackerIndex); // Get the task sync offset syncOffset = state->pfnGetTaskSyncLocation(state, taskId); // Initialize the location taskSyncLocation = (int64_t*)(state->renderTimeStampResource.data + syncOffset); *taskSyncLocation = CM_INVALID_INDEX; *(taskSyncLocation + 1) = CM_INVALID_INDEX; if(state->cbbEnabled) { *(taskSyncLocation + 2) = tag; *(taskSyncLocation + 3) = state->renderHal->currentTrackerIndex; } // Register batch buffer for rendering if (!enableWalker && !enableGpGpu) { CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnRegisterResource( osInterface, &batchBuffer->OsResource, true, true)); } #if (_RELEASE_INTERNAL || _DEBUG) #if !defined(CM_DIRECT_GUC_SUPPORT) // Register Timestamp Buffer CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnRegisterResource( osInterface, &state->renderTimeStampResource.osResource, true, true)); #endif #endif // Allocate all available space, unused buffer will be returned later CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnGetCommandBuffer(osInterface, &mosCmdBuffer, 0)); remaining = mosCmdBuffer.iRemaining; #if (_RELEASE_INTERNAL || _DEBUG) #if defined(CM_DIRECT_GUC_SUPPORT) batchbufferaddress = osInterface->pfnGetResourceGfxAddress( osInterface, &mosCmdBuffer.OsResource); batchbufferaddress += mosCmdBuffer.iOffset; #endif #endif // Update power option of this command; CM_CHK_MOSSTATUS_GOTOFINISH( state->pfnUpdatePowerOption( state, &state->powerOption ) ); // use frame tracking to write the tracker ID to CM tracker resource renderHal->trackerProducer.GetLatestTrackerResource(renderHal->currentTrackerIndex, &osResource, &tagOffset); renderHal->pfnSetupPrologParams(renderHal, &genericPrologParams, osResource, tagOffset, tag); FrameTrackerTokenFlat_SetProducer(&stateHeap->pCurMediaState->trackerToken, &renderHal->trackerProducer); FrameTrackerTokenFlat_Merge(&stateHeap->pCurMediaState->trackerToken, renderHal->currentTrackerIndex, tag); // Record registers by unified media profiler in the beginning if (state->perfProfiler != nullptr) { CM_CHK_MOSSTATUS_GOTOFINISH(state->perfProfiler->AddPerfCollectStartCmd((void *)state, state->osInterface, mhwMiInterface, &mosCmdBuffer)); } //Send the First PipeControl Command to indicate the beginning of execution pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwResourceOffset = syncOffset; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_WRITE_TIMESTAMP_REG; pipeCtlParams.dwFlushMode = MHW_FLUSH_WRITE_CACHE; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); // Initialize command buffer and insert prolog CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pfnInitCommandBuffer(renderHal, &mosCmdBuffer, &genericPrologParams)); HalOcaInterface::On1stLevelBBStart(mosCmdBuffer, *pOsContext, osInterface->CurrentGpuContextHandle, *renderHal->pMhwMiInterface, *pMmioRegisters); // update tracker tag used with CM tracker resource renderHal->trackerProducer.StepForward(renderHal->currentTrackerIndex); // Increment sync tag syncTag = stateHeap->dwNextTag++; //enable SKL L3 config HalCm_GetLegacyRenderHalL3Setting( &state->l3Settings, &renderHal->L3CacheSettings ); renderHal->pfnEnableL3Caching(renderHal, &renderHal->L3CacheSettings); mhwRender->SetL3Cache(&mosCmdBuffer); if (sipEnable) { CM_CHK_MOSSTATUS_GOTOFINISH(SetupHwDebugControl(renderHal, &mosCmdBuffer)); } // Adds granularity control for preemption for Gen9. // Supporting Preemption granularity control reg for 3D and GPGPU mode for per ctx and with non-privileged access if ( MEDIA_IS_SKU(state->skuTable, FtrPerCtxtPreemptionGranularityControl )) { MHW_MI_LOAD_REGISTER_IMM_PARAMS loadRegImm; MOS_ZeroMemory( &loadRegImm, sizeof( MHW_MI_LOAD_REGISTER_IMM_PARAMS ) ); loadRegImm.dwRegister = MHW_RENDER_ENGINE_PREEMPTION_CONTROL_OFFSET; // Same reg offset and value for gpgpu pipe and media pipe if ( enableGpGpu ) { if ( MEDIA_IS_SKU(state->skuTable, FtrGpGpuThreadGroupLevelPreempt ) || MEDIA_IS_SKU(state->skuTable, FtrGpGpuMidThreadLevelPreempt)) { //if FtrGpGpuThreadGroupLevelPreempt is true, still program the //it to MID_THREAD_GROUP.Gen9 doesn't support MID_THREAD level loadRegImm.dwData = MHW_RENDER_ENGINE_THREAD_GROUP_PREEMPT_VALUE; state->renderHal->pfnEnableGpgpuMiddleBatchBufferPreemption( state->renderHal ); } else if ( MEDIA_IS_SKU(state->skuTable, FtrGpGpuMidBatchPreempt )) { loadRegImm.dwData = MHW_RENDER_ENGINE_MID_BATCH_PREEMPT_VALUE; state->renderHal->pfnEnableGpgpuMiddleBatchBufferPreemption( state->renderHal ); } else { // if hit this branch then platform does not support any media preemption in render engine. Still program the register to avoid GPU hang loadRegImm.dwData = MHW_RENDER_ENGINE_MID_BATCH_PREEMPT_VALUE; } } else { if ( MEDIA_IS_SKU(state->skuTable, FtrMediaThreadGroupLevelPreempt) || MEDIA_IS_SKU(state->skuTable, FtrMediaMidThreadLevelPreempt)) { //if FtrMediaMidThreadLevelPreempt is true, still program the //it to MID_THREAD_GROUP.Gen9 doesn't support MID_THREAD. loadRegImm.dwData = MHW_RENDER_ENGINE_THREAD_GROUP_PREEMPT_VALUE; } else if ( MEDIA_IS_SKU(state->skuTable, FtrMediaMidBatchPreempt)) { loadRegImm.dwData = MHW_RENDER_ENGINE_MID_BATCH_PREEMPT_VALUE; } else { // if hit this branch then platform does not support any media preemption in render engine. Still program the register to avoid GPU hang loadRegImm.dwData = MHW_RENDER_ENGINE_MID_BATCH_PREEMPT_VALUE; } } CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddMiLoadRegisterImmCmd(&mosCmdBuffer, &loadRegImm ) ); } // Send Pipeline Select command CM_CHK_MOSSTATUS_GOTOFINISH(mhwRender->AddPipelineSelectCmd(&mosCmdBuffer, enableGpGpu)); // Send State Base Address command CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pfnSendStateBaseAddress(renderHal, &mosCmdBuffer)); // Send Surface States CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pfnSendSurfaces(renderHal, &mosCmdBuffer)); if (enableGpGpu) { if (csrEnable) { // Send CS_STALL pipe control //Insert a pipe control as synchronization pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_NOWRITE; pipeCtlParams.dwFlushMode = MHW_FLUSH_WRITE_CACHE; pipeCtlParams.bDisableCSStall = 0; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); } if (sipEnable || csrEnable) { // Send SIP State CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pfnSendSipStateCmd(renderHal, &mosCmdBuffer)); CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnRegisterResource( osInterface, &state->csrResource, true, true)); // Send csr base addr command CM_CHK_MOSSTATUS_GOTOFINISH(mhwRender->AddGpgpuCsrBaseAddrCmd(&mosCmdBuffer, &state->csrResource)); } } // Setup VFE State params. Each Renderer MUST call pfnSetVfeStateParams(). // See comment in pfnSetVfeStateParams() for details. tmp = RENDERHAL_USE_MEDIA_THREADS_MAX; if (state->maxHWThreadValues.userFeatureValue != 0) { if( state->maxHWThreadValues.userFeatureValue < renderHal->pHwCaps->dwMaxThreads) { tmp = state->maxHWThreadValues.userFeatureValue; } } else if (state->maxHWThreadValues.apiValue != 0) { if( state->maxHWThreadValues.apiValue < renderHal->pHwCaps->dwMaxThreads) { tmp = state->maxHWThreadValues.apiValue; } } renderHal->pfnSetVfeStateParams( renderHal, MEDIASTATE_DEBUG_COUNTER_FREE_RUNNING, tmp, state->taskParam->vfeCurbeSize, state->taskParam->urbEntrySize, &state->scoreboardParams); // Send VFE State CM_CHK_MOSSTATUS_GOTOFINISH(mhwRender->AddMediaVfeCmd(&mosCmdBuffer, renderHal->pRenderHalPltInterface->GetVfeStateParameters())); // Send CURBE Load if (state->taskParam->vfeCurbeSize > 0) { CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pfnSendCurbeLoad(renderHal, &mosCmdBuffer)); } // Send Interface Descriptor Load if (state->dshEnabled) { PRENDERHAL_DYNAMIC_STATE dynamicState = ((PRENDERHAL_MEDIA_STATE_LEGACY)stateHeap->pCurMediaState)->pDynamicState; idLoadParams.dwInterfaceDescriptorStartOffset = dynamicState->memoryBlock.GetOffset() + dynamicState->MediaID.dwOffset; idLoadParams.dwInterfaceDescriptorLength = dynamicState->MediaID.iCount * stateHeap->dwSizeMediaID; } else { idLoadParams.dwInterfaceDescriptorStartOffset = stateHeap->pCurMediaState->dwOffset + stateHeap->dwOffsetMediaID; idLoadParams.dwInterfaceDescriptorLength = renderHal->StateHeapSettings.iMediaIDs * stateHeap->dwSizeMediaID; } idLoadParams.pKernelState = nullptr; CM_CHK_MOSSTATUS_GOTOFINISH(mhwRender->AddMediaIDLoadCmd(&mosCmdBuffer, &idLoadParams)); HalOcaInterface::OnDispatch(mosCmdBuffer, *osInterface, *renderHal->pMhwMiInterface, *pMmioRegisters); if (enableWalker) { // send media walker command, if required for (uint32_t i = 0; i < state->taskParam->numKernels; i ++) { // Insert CONDITIONAL_BATCH_BUFFER_END if ( taskParam->conditionalEndBitmap & ((uint64_t)1 << (i))) { // this could be batch buffer end so need to update sync tag, media state flush, write end timestamp CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pfnSendSyncTag(renderHal, &mosCmdBuffer)); // conditionally write timestamp CM_CHK_MOSSTATUS_GOTOFINISH(HalCm_OsAddArtifactConditionalPipeControl(&miRegG9, state, &mosCmdBuffer, syncOffset, &taskParam->conditionalBBEndParams[i], tag)); // Insert conditional batch buffer end mhwMiInterface->AddMiConditionalBatchBufferEndCmd(&mosCmdBuffer, &taskParam->conditionalBBEndParams[i]); } //Insert PIPE_CONTROL at two cases: // 1. synchronization is set // 2. the next kernel has dependency pattern if((i > 0) && ((taskParam->syncBitmap & ((uint64_t)1 << (i-1))) || (kernelParam[i]->kernelThreadSpaceParam.patternType != CM_NONE_DEPENDENCY))) { //Insert a pipe control as synchronization pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_NOWRITE; pipeCtlParams.dwFlushMode = MHW_FLUSH_CUSTOM; pipeCtlParams.bInvalidateTextureCache = true; pipeCtlParams.bFlushRenderTargetCache = true; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); } CM_CHK_MOSSTATUS_GOTOFINISH(state->pfnSendMediaWalkerState(state, kernelParam[i], &mosCmdBuffer)); } } else if (enableGpGpu) { // send GPGPU walker command, if required for (uint32_t i = 0; i < state->taskParam->numKernels; i ++) { //Insert PIPE_CONTROL as synchronization if synchronization is set if((i > 0) && (taskParam->syncBitmap & ((uint64_t)1 << (i-1)))) { //Insert a pipe control as synchronization pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_NOWRITE; pipeCtlParams.dwFlushMode = MHW_FLUSH_CUSTOM; pipeCtlParams.bInvalidateTextureCache = true; pipeCtlParams.bFlushRenderTargetCache = true; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); } CM_CHK_MOSSTATUS_GOTOFINISH(state->pfnSendGpGpuWalkerState(state, kernelParam[i], &mosCmdBuffer)); } } else { // Send Start batch buffer command CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddMiBatchBufferStartCmd( &mosCmdBuffer, batchBuffer)); CM_CHK_NULL_GOTOFINISH_MOSERROR(batchBuffer->pPrivateData); bbCmArgs = (PCM_HAL_BB_ARGS) batchBuffer->pPrivateData; if ( (bbCmArgs->refCount == 1) || (state->taskParam->reuseBBUpdateMask == 1) ) { // Add BB end command CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddMiBatchBufferEnd(nullptr, batchBuffer)); } else //reuse BB { // Skip BB end command CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->SkipMiBatchBufferEndBb(batchBuffer)); } // UnLock the batch buffer if ( (bbCmArgs->refCount == 1) || (state->taskParam->reuseBBUpdateMask == 1) ) { CM_CHK_MOSSTATUS_GOTOFINISH(renderHal->pfnUnlockBB(renderHal, batchBuffer)); } } // issue a PIPE_CONTROL to flush all caches and the stall the CS before // issuing a PIPE_CONTROL to write the timestamp pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_NOWRITE; pipeCtlParams.dwFlushMode = MHW_FLUSH_WRITE_CACHE; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); if (state->svmBufferUsed || state->statelessBufferUsed) { // Find the SVM/statelessBuffer slot, patch it into this dummy pipe_control for (uint32_t i = 0; i < state->cmDeviceParam.maxBufferTableSize; i++) { //register resource here if (state->bufferTable[i].address) { CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnRegisterResource( osInterface, &state->bufferTable[i].osResource, true, false)); // sync resource MOS_SURFACE mosSurface; MOS_ZeroMemory(&mosSurface, sizeof(mosSurface)); CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnGetResourceInfo( osInterface, &state->bufferTable[i].osResource, &mosSurface)); mosSurface.OsResource = state->bufferTable[i].osResource; CM_CHK_HRESULT_GOTOFINISH_MOSERROR(state->pfnSurfaceSync(state, &mosSurface, false)); } } } // Send Sync Tag CM_CHK_MOSSTATUS_GOTOFINISH( renderHal->pfnSendSyncTag( renderHal, &mosCmdBuffer ) ); // Update tracker resource CM_CHK_MOSSTATUS_GOTOFINISH(state->pfnUpdateTrackerResource(state, &mosCmdBuffer, tag)); // issue a PIPE_CONTROL to write timestamp syncOffset += sizeof(uint64_t); pipeCtlParams = g_cRenderHal_InitPipeControlParams; pipeCtlParams.presDest = &state->renderTimeStampResource.osResource; pipeCtlParams.dwResourceOffset = syncOffset; pipeCtlParams.dwPostSyncOp = MHW_FLUSH_WRITE_TIMESTAMP_REG; pipeCtlParams.dwFlushMode = MHW_FLUSH_READ_CACHE; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeCtlParams)); // Record registers by unified media profiler in the end if (state->perfProfiler != nullptr) { CM_CHK_MOSSTATUS_GOTOFINISH(state->perfProfiler->AddPerfCollectEndCmd((void *)state, state->osInterface, mhwMiInterface, &mosCmdBuffer)); } // Add PipeControl to invalidate ISP and MediaState to avoid PageFault issue 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; CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddPipeControl(&mosCmdBuffer, nullptr, &pipeControlParams)); if (MEDIA_IS_WA(renderHal->pWaTable, WaSendDummyVFEafterPipelineSelect)) { MHW_VFE_PARAMS vfeStateParams = {}; vfeStateParams.dwNumberofURBEntries = 1; CM_CHK_MOSSTATUS_GOTOFINISH(mhwRender->AddMediaVfeCmd(&mosCmdBuffer, &vfeStateParams)); } HalOcaInterface::On1stLevelBBEnd(mosCmdBuffer, *osInterface); //Couple to the BB_START , otherwise GPU Hang without it in KMD. CM_CHK_MOSSTATUS_GOTOFINISH(mhwMiInterface->AddMiBatchBufferEnd(&mosCmdBuffer, nullptr)); // Return unused command buffer space to OS osInterface->pfnReturnCommandBuffer(osInterface, &mosCmdBuffer, 0); #if MDF_COMMAND_BUFFER_DUMP if (state->dumpCommandBuffer) { state->pfnDumpCommadBuffer( state, &mosCmdBuffer, offsetof(PACKET_SURFACE_STATE, cmdSurfaceState), mhw_state_heap_g9_X::RENDER_SURFACE_STATE_CMD::byteSize); } #endif #if MDF_SURFACE_STATE_DUMP if (state->dumpSurfaceState) { state->pfnDumpSurfaceState( state, offsetof(PACKET_SURFACE_STATE, cmdSurfaceState), mhw_state_heap_g9_X::RENDER_SURFACE_STATE_CMD::byteSize); } #endif CM_CHK_MOSSTATUS_GOTOFINISH(state->pfnGetGlobalTime(&state->taskTimeStamp->submitTimeInCpu[taskId])); CM_CHK_MOSSTATUS_GOTOFINISH(state->pfnGetGpuTime(state, &state->taskTimeStamp->submitTimeInGpu[taskId])); // Submit command buffer #if (_RELEASE_INTERNAL || _DEBUG) #if defined (CM_DIRECT_GUC_SUPPORT) CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnSubmitWorkQueue(osInterface, MOS_GPU_NODE_3D, batchbufferaddress)); #endif #endif #if !defined (CM_DIRECT_GUC_SUPPORT) CM_CHK_HRESULT_GOTOFINISH_MOSERROR(osInterface->pfnSubmitCommandBuffer(osInterface, &mosCmdBuffer, state->nullHwRenderCm)); #endif if (state->nullHwRenderCm == false) { stateHeap->pCurMediaState->bBusy = true; if ( !enableWalker && !enableGpGpu ) { batchBuffer->bBusy = true; batchBuffer->dwSyncTag = syncTag; } } // reset API call number of HW threads state->maxHWThreadValues.apiValue = 0; state->pfnReferenceCommandBuffer(&mosCmdBuffer.OsResource, cmdBuffer); eStatus = MOS_STATUS_SUCCESS; finish: // Failed -> discard all changes in Command Buffer if (eStatus != MOS_STATUS_SUCCESS) { // Buffer overflow - display overflow size if (mosCmdBuffer.iRemaining < 0) { CM_ASSERTMESSAGE("Command Buffer overflow by %d bytes.", -mosCmdBuffer.iRemaining); } // Move command buffer back to beginning tmp = remaining - mosCmdBuffer.iRemaining; mosCmdBuffer.iRemaining = remaining; mosCmdBuffer.iOffset -= tmp; mosCmdBuffer.pCmdPtr = mosCmdBuffer.pCmdBase + mosCmdBuffer.iOffset/sizeof(uint32_t); // Return unused command buffer space to OS osInterface->pfnReturnCommandBuffer(osInterface, &mosCmdBuffer, 0); } return eStatus; } MOS_STATUS CM_HAL_G9_X::GetCopyKernelIsa(void *&isa, uint32_t &isaSize) { isa = (void *)pGPUCopy_kernel_isa_gen9; isaSize = iGPUCopy_kernel_isa_size_gen9; return MOS_STATUS_SUCCESS; } MOS_STATUS CM_HAL_G9_X::GetInitKernelIsa(void *&isa, uint32_t &isaSize) { isa = (void *)pGPUInit_kernel_isa_Gen9; isaSize = iGPUInit_kernel_isa_size_Gen9; return MOS_STATUS_SUCCESS; } MOS_STATUS CM_HAL_G9_X::UpdatePlatformInfoFromPower( PCM_PLATFORM_INFO platformInfo, bool euSaturated) { PCM_HAL_STATE state = m_cmState; PRENDERHAL_INTERFACE renderHal = state->renderHal; CM_POWER_OPTION cmPower; if ( state->requestSingleSlice || renderHal->bRequestSingleSlice || (state->powerOption.nSlice != 0 && state->powerOption.nSlice < platformInfo->numSlices)) { platformInfo->numSubSlices = platformInfo->numSubSlices / platformInfo->numSlices; if (state->powerOption.nSlice > 1) { platformInfo->numSubSlices *= state->powerOption.nSlice; platformInfo->numSlices = state->powerOption.nSlice; } else { platformInfo->numSlices = 1; } } else if (euSaturated) { // No SSD and EU Saturation, request maximum number of slices/subslices/EUs cmPower.nSlice = (uint16_t)platformInfo->numSlices; cmPower.nSubSlice = (uint16_t)platformInfo->numSubSlices; cmPower.nEU = (uint16_t)(platformInfo->numEUsPerSubSlice * platformInfo->numSubSlices); state->pfnSetPowerOption(state, &cmPower); } return MOS_STATUS_SUCCESS; } uint32_t CM_HAL_G9_X::GetMediaWalkerMaxThreadWidth() { return CM_MAX_THREADSPACE_WIDTH_SKLUP_FOR_MW; } uint32_t CM_HAL_G9_X::GetMediaWalkerMaxThreadHeight() { return CM_MAX_THREADSPACE_HEIGHT_SKLUP_FOR_MW; } MOS_STATUS CM_HAL_G9_X::GetHwSurfaceBTIInfo( PCM_SURFACE_BTI_INFO btiInfo) { if (btiInfo == nullptr) { return MOS_STATUS_NULL_POINTER; } btiInfo->normalSurfaceStart = CM_GLOBAL_SURFACE_INDEX_START_GEN9_PLUS + \ CM_GLOBAL_SURFACE_NUMBER + CM_GTPIN_SURFACE_NUMBER ; btiInfo->normalSurfaceEnd = GT_RESERVED_INDEX_START_GEN9_PLUS - 1; btiInfo->reservedSurfaceStart = CM_GLOBAL_SURFACE_INDEX_START_GEN9_PLUS; btiInfo->reservedSurfaceEnd = CM_GLOBAL_SURFACE_NUMBER + CM_GTPIN_SURFACE_NUMBER; return MOS_STATUS_SUCCESS; } MOS_STATUS CM_HAL_G9_X::SetSuggestedL3Conf( L3_SUGGEST_CONFIG l3Config) { if (l3Config >= sizeof(SKL_L3_PLANE)/sizeof(L3ConfigRegisterValues)) { return MOS_STATUS_INVALID_PARAMETER; } return HalCm_SetL3Cache((L3ConfigRegisterValues *)&SKL_L3_PLANE[l3Config], &m_cmState->l3Settings); } MOS_STATUS CM_HAL_G9_X::AllocateSIPCSRResource() { MOS_STATUS eStatus = MOS_STATUS_SUCCESS; if (Mos_ResourceIsNull(&m_cmState->sipResource.osResource)) { CM_CHK_MOSSTATUS_RETURN(HalCm_AllocateSipResource(m_cmState)); // create sip resource if it does not exist CM_CHK_MOSSTATUS_RETURN(HalCm_AllocateCSRResource(m_cmState)); } return eStatus; } MOS_STATUS CM_HAL_G9_X::GetGenStepInfo(char*& stepInfoStr) { const char *cmSteppingInfo[] = { "A", "B", "C", "D", "E", "F", "G", "H", "I", "J" }; uint32_t genStepId = m_cmState->platform.usRevId; if (m_steppingTable.size() != 0) //check if the stepping table been overwritten { if (genStepId < m_steppingTable.size()) { stepInfoStr = (char *)m_steppingTable[genStepId]; } else { stepInfoStr = nullptr; } } else { if (genStepId < sizeof(cmSteppingInfo)/sizeof(const char *)) { stepInfoStr = (char *)cmSteppingInfo[genStepId]; } else { stepInfoStr = nullptr; } } return MOS_STATUS_SUCCESS; } int32_t CM_HAL_G9_X::ColorCountSanityCheck(uint32_t colorCount) { if (colorCount == CM_INVALID_COLOR_COUNT || colorCount > CM_THREADSPACE_MAX_COLOR_COUNT) { CM_ASSERTMESSAGE("Error: Invalid color count."); return CM_INVALID_ARG_VALUE; } return CM_SUCCESS; } bool CM_HAL_G9_X::MemoryObjectCtrlPolicyCheck(uint32_t memCtrl) { if ( memCtrl > MEMORY_OBJECT_CONTROL_SKL_NO_CACHE ) { return false; } return true; } int32_t CM_HAL_G9_X::GetConvSamplerIndex( PMHW_SAMPLER_STATE_PARAM samplerParam, char *samplerIndexTable, int32_t nSamp8X8Num, int32_t nSampConvNum) { int32_t samplerIndex = 0; if ((samplerParam->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_2D) && (samplerParam->Convolve.skl_mode)) { // 2D convolve & SKL+ samplerIndex = 1 + nSampConvNum + nSamp8X8Num; } else if (samplerParam->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_1D) { // 1D convolve & SKL+ samplerIndex = nSampConvNum; } else { // 1P convolve SKL+ samplerIndex = 1 + (nSamp8X8Num + nSampConvNum) * 2; while (samplerIndexTable[samplerIndex] != CM_INVALID_INDEX) { samplerIndex += 2; } } return samplerIndex; } MOS_STATUS CM_HAL_G9_X::SetL3CacheConfig( const L3ConfigRegisterValues *values, PCmHalL3Settings cmHalL3Setting) { return HalCm_SetL3Cache( values, cmHalL3Setting ); } MOS_STATUS CM_HAL_G9_X::GetSamplerParamInfoForSamplerType( PMHW_SAMPLER_STATE_PARAM mhwSamplerParam, SamplerParam &samplerParam) { const unsigned int samplerElementSize[MAX_ELEMENT_TYPE_COUNT] = {16, 32, 64, 128, 1024, 2048}; // gets element_type switch (mhwSamplerParam->SamplerType) { case MHW_SAMPLER_TYPE_3D: samplerParam.elementType = MHW_Sampler1Element; break; case MHW_SAMPLER_TYPE_MISC: samplerParam.elementType = MHW_Sampler2Elements; break; case MHW_SAMPLER_TYPE_CONV: if ((!mhwSamplerParam->Convolve.skl_mode && mhwSamplerParam->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_2D) || mhwSamplerParam->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_1P) { samplerParam.elementType = MHW_Sampler64Elements; } else if (mhwSamplerParam->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_1D) { samplerParam.elementType = MHW_Sampler8Elements; } else { samplerParam.elementType = MHW_Sampler128Elements; } break; case MHW_SAMPLER_TYPE_AVS: samplerParam.elementType = MHW_Sampler128Elements; break; default: break; } // bti_stepping for BDW mode convolve or 1P convolve is 2, other cases are 1. if ((mhwSamplerParam->SamplerType == MHW_SAMPLER_TYPE_CONV) && ((!mhwSamplerParam->Convolve.skl_mode && mhwSamplerParam->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_2D) || mhwSamplerParam->Convolve.ui8ConvolveType == CM_CONVOLVE_SKL_TYPE_1P)) { samplerParam.btiStepping = 2; } else { samplerParam.btiStepping = 1; } // gets multiplier samplerParam.btiMultiplier = samplerElementSize[samplerParam.elementType] / samplerParam.btiStepping; // gets size samplerParam.size = samplerElementSize[samplerParam.elementType]; // Temporary solution for conv because MHW use 2048 bytes for all of the convolve samplers. // size should always be equal to bti_stepping * bti_multiplier except for this one. if (mhwSamplerParam->SamplerType == MHW_SAMPLER_TYPE_CONV) { samplerParam.size = 2048; } return MOS_STATUS_SUCCESS; } MOS_STATUS CM_HAL_G9_X::GetExpectedGtSystemConfig( PCM_EXPECTED_GT_SYSTEM_INFO expectedConfig) { if (m_genGT == PLATFORM_INTEL_GT1) { expectedConfig->numSlices = SKL_GT1_MAX_NUM_SLICES; expectedConfig->numSubSlices = SKL_GT1_MAX_NUM_SUBSLICES; } else if (m_genGT == PLATFORM_INTEL_GT1_5) { expectedConfig->numSlices = SKL_GT1_5_MAX_NUM_SLICES; expectedConfig->numSubSlices = SKL_GT1_5_MAX_NUM_SUBSLICES; } else if (m_genGT == PLATFORM_INTEL_GT2) { expectedConfig->numSlices = SKL_GT2_MAX_NUM_SLICES; expectedConfig->numSubSlices = SKL_GT2_MAX_NUM_SUBSLICES; } else if (m_genGT == PLATFORM_INTEL_GT3) { expectedConfig->numSlices = SKL_GT3_MAX_NUM_SLICES; expectedConfig->numSubSlices = SKL_GT3_MAX_NUM_SUBSLICES; } else if (m_genGT == PLATFORM_INTEL_GT4) { expectedConfig->numSlices = SKL_GT4_MAX_NUM_SLICES; expectedConfig->numSubSlices = SKL_GT4_MAX_NUM_SUBSLICES; } else { expectedConfig->numSlices = 0; expectedConfig->numSubSlices = 0; } return MOS_STATUS_SUCCESS; } uint64_t CM_HAL_G9_X::ConverTicksToNanoSecondsDefault(uint64_t ticks) { if (m_platformID == PLATFORM_INTEL_BXT || m_platformID == PLATFORM_INTEL_GLK) { return (uint64_t)(ticks * CM_NS_PER_TICK_RENDER_G9LP); } else { return (uint64_t)(ticks * CM_NS_PER_TICK_RENDER_G9); } }