/* * Copyright (c) 2021, 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 hal_kerneldll_next.c //! \brief Kernel Dynamic Linking/Loading routines for FC //! #ifndef VPHAL_LIB #if IMOLA #include #endif // IMOLA #include //for sin & cos #endif // VPHAL_LIB #if EMUL || VPHAL_LIB #include #include "support.h" #elif LINUX #else // !(EMUL | VPHAL_LIB) && !LINUX #endif // EMUL | VPHAL_LIB #include "hal_kerneldll_next.h" #include "vp_utils.h" // Define _DEBUG symbol for KDLL Release build before loading the "vpkrnheader.h" file // This is necessary for full kernels names in both Release/Debug versions of KDLL app #if EMUL || VPHAL_LIB #ifndef _DEBUG #define _DEBUG 2 #endif // _DEBUG #endif // EMUL || VPHAL_LIB // Kernel IDs and Kernel Names #include "vpkrnheader.h" // IDR_VP_TOTAL_NUM_KERNELS #include "cm_fc_ld.h" // Undefine _DEBUG symbol for the remaining of the KDLL Release build #if _DEBUG == 2 #undef _DEBUG #endif // _DEBUG #ifndef PI #define PI 3.1415926535897932f #endif // PI #ifdef __cplusplus extern "C" { #endif // __cplusplus #define FOLD_HASH(folded_hash, hash) \ { \ folded_hash = (((hash) >> 8) ^ (hash)) & 0x00ff00ff; \ folded_hash = ((folded_hash >> 16) ^ folded_hash) & 0xff; \ } \ const bool g_cIsFormatYUV[Format_Count] = { false, // Format_Any false, // Format_A8R8G8B8 false, // Format_X8R8G8B8 false, // Format_A8B8G8R8 false, // Format_X8B8G8R8 false, // Format_A16B16G16R16 false, // Format_A16R16G16B16 false, // Format_R5G6B5 false, // Format_R32U false, // Format_R32F false, // Format_R8G8B8 false, // Format_RGBP false, // Format_BGRP true, // Format_YUY2 true, // Format_YUYV true, // Format_YVYU true, // Format_UYVY true, // Format_VYUY true, // Format_Y216 true, // Format_Y210 true, // Format_Y416 true, // Format_AYUV true, // Format_AUYV true, // Format_Y410 true, // Format_400P true, // Format_NV12 true, // Format_NV12_UnAligned true, // Format_NV21 true, // Format_NV11 true, // Format_NV11_UnAligned true, // Format_P208 true, // Format_P208_UnAligned true, // Format_IMC1 true, // Format_IMC2 true, // Format_IMC3 true, // Format_IMC4 true, // Format_422H true, // Format_422V true, // Format_444P true, // Format_411P true, // Format_411R true, // Format_I420 true, // Format_IYUV true, // Format_YV12 true, // Format_YVU9 true, // Format_AI44 (YUV originally, palette may be converted to RGB) true, // Format_IA44 (same as above) false, // Format_P8 (using RGB since P8 is uncommon in FC) false, // Format_A8P8 (same as above) false, // Format_A8 false, // Format_L8 false, // Format_A4L4 false, // Format_A8L8 true, // Format_IRW0 true, // Format_IRW1 true, // Format_IRW2 true, // Format_IRW3 true, // Format_IRW4 true, // Format_IRW5 true, // Format_IRW6 true, // Format_IRW7 false, // Format_STMM false, // Format_Buffer false, // Format_Buffer_2D false, // Format_V8U8 false, // Format_R32S false, // Format_R8U false, // Format_R8G8UN false, // Format_R8G8SN false, // Format_G8R8_G8B8 false, // Format_R16U false, // Format_R16S false, // Format_R16UN false, // Format_RAW false, // Format_Y8 false, // Format_Y1 false, // Format_Y16U false, // Format_Y16S false, // Format_L16 false, // Format_D16 false, // Format_R10G10B10A2 false, // Format_B10G10R10A2 true, // Format_P016 true, // Format_P010 true // Format_YV12_Planar }; bool KernelDll_IsYUVFormat(MOS_FORMAT format) { if (format >= Format_Any && format < Format_Count) { return g_cIsFormatYUV[format]; } else { return false; } } /*---------------------------------------------------------------------------- | Purpose : Group common color spaces into one | Returns : Return the representative color space of the group \---------------------------------------------------------------------------*/ VPHAL_CSPACE KernelDll_TranslateCspace(VPHAL_CSPACE cspace) { switch (cspace) { case CSpace_BT709: case CSpace_xvYCC709: return CSpace_BT709; case CSpace_BT601: case CSpace_xvYCC601: return CSpace_BT601; case CSpace_BT601_FullRange: return CSpace_BT601_FullRange; case CSpace_BT709_FullRange: return CSpace_BT709_FullRange; case CSpace_RGB: case CSpace_sRGB: return CSpace_sRGB; case CSpace_stRGB: return CSpace_stRGB; case CSpace_Gray: case CSpace_BT601Gray: return CSpace_BT601Gray; case CSpace_BT601Gray_FullRange: return CSpace_BT601Gray_FullRange; case CSpace_BT2020: return CSpace_BT2020; case CSpace_BT2020_FullRange: return CSpace_BT2020_FullRange; case CSpace_BT2020_RGB: return CSpace_BT2020_RGB; case CSpace_BT2020_stRGB: return CSpace_BT2020_stRGB; default: return CSpace_None; } } void KernelDll_MatrixProduct( float * dest, const float *m1, const float *m2) { bool save; float temp[12]; // setup temp matrix to allow the following operations: // dest = dest * m2 // dest = m1 * dest // dest = dest * dest save = (m1 == dest) || (m2 == dest); m1 = (m1 == dest) ? temp : m1; m2 = (m2 == dest) ? temp : m2; if (save) MOS_SecureMemcpy(temp, sizeof(temp), (void *)dest, sizeof(temp)); // Multiply the matrices dest[0] = m1[0] * m2[0] + m1[1] * m2[4] + m1[2] * m2[8]; dest[1] = m1[0] * m2[1] + m1[1] * m2[5] + m1[2] * m2[9]; dest[2] = m1[0] * m2[2] + m1[1] * m2[6] + m1[2] * m2[10]; dest[3] = m1[0] * m2[3] + m1[1] * m2[7] + m1[2] * m2[11] + m1[3]; dest[4] = m1[4] * m2[0] + m1[5] * m2[4] + m1[6] * m2[8]; dest[5] = m1[4] * m2[1] + m1[5] * m2[5] + m1[6] * m2[9]; dest[6] = m1[4] * m2[2] + m1[5] * m2[6] + m1[6] * m2[10]; dest[7] = m1[4] * m2[3] + m1[5] * m2[7] + m1[6] * m2[11] + m1[7]; dest[8] = m1[8] * m2[0] + m1[9] * m2[4] + m1[10] * m2[8]; dest[9] = m1[8] * m2[1] + m1[9] * m2[5] + m1[10] * m2[9]; dest[10] = m1[8] * m2[2] + m1[9] * m2[6] + m1[10] * m2[10]; dest[11] = m1[8] * m2[3] + m1[9] * m2[7] + m1[10] * m2[11] + m1[11]; } void KernelDll_UpdateCscCoefficients(Kdll_State *pState, Kdll_CSC_Matrix * pMatrix) { float csc[12]; // CSC matrix (YUV->RGB) float icsc[12]; // ICSC matrix (RGB->YUV), (YUV->YUV) float m[12]; // auxiliary matrix float matrix[12]; // final matrix Kdll_CSCType csctype; Kdll_Procamp *pProcamp = nullptr; VPHAL_CSPACE src = pMatrix->SrcSpace; VPHAL_CSPACE dst = pMatrix->DstSpace; bool bCSC, bICSC; bCSC = bICSC = false; MOS_ZeroMemory(m, sizeof(m)); MOS_ZeroMemory(csc, sizeof(csc)); MOS_ZeroMemory(icsc, sizeof(icsc)); // Select procamp parameters if (pMatrix->iProcampID > DL_PROCAMP_DISABLED && pMatrix->iProcampID < pState->iProcampSize && pState->pProcamp != nullptr) { pProcamp = pState->pProcamp + pMatrix->iProcampID; } // Setup CSC matrix if (src != dst) { if ((dst == CSpace_sRGB) && (src != CSpace_stRGB)) { KernelDll_GetCSCMatrix(src, dst, csc); MOS_SecureMemcpy(m, sizeof(csc), (void *)csc, sizeof(csc)); bCSC = true; csctype = CSC_YUV_RGB; } else if ((dst == CSpace_stRGB) && (src != CSpace_sRGB)) { KernelDll_GetCSCMatrix(src, dst, csc); MOS_SecureMemcpy(m, sizeof(csc), (void *)csc, sizeof(csc)); bCSC = true; csctype = CSC_YUV_RGB; } else { KernelDll_GetCSCMatrix(src, dst, icsc); MOS_SecureMemcpy(m, sizeof(icsc), (void *)icsc, sizeof(icsc)); bICSC = true; if (KernelDll_IsCspace(src, CSpace_RGB) && !KernelDll_IsCspace(dst, CSpace_RGB)) { csctype = CSC_RGB_YUV; } else if (KernelDll_IsCspace(src, CSpace_BT2020_RGB) && KernelDll_IsCspace(dst, CSpace_BT2020)) { csctype = CSC_RGB_YUV; } else if (KernelDll_IsCspace(src, CSpace_BT2020) && KernelDll_IsCspace(dst, CSpace_BT2020_RGB)) { csctype = CSC_YUV_RGB; } else if (KernelDll_IsCspace(src, CSpace_BT2020_RGB) && KernelDll_IsCspace(dst, CSpace_BT2020_RGB)) { csctype = CSC_RGB_RGB; // Kernel params didn't support 10bit, it need transformation from 10bit to 8bit. m[3] = ROUND_FLOAT(m[3], 0.25f); // 10bit to 8bit (value/4) m[7] = ROUND_FLOAT(m[7], 0.25f); // 10bit to 8bit (value/4) m[11] = ROUND_FLOAT(m[11], 0.25f); // 10bit to 8bit (value/4) } else { csctype = CSC_YUV_YUV; } } } // Setup CSC matrix for procamp in sRGB space else if ((dst == CSpace_sRGB) && (pProcamp)) { KernelDll_GetCSCMatrix(CSpace_sRGB, CSpace_BT709, icsc); KernelDll_GetCSCMatrix(CSpace_BT709, CSpace_sRGB, csc); bICSC = bCSC = true; csctype = CSC_RGB_RGB; } // Setup CSC matrix for procamp in stRGB space else if ((dst == CSpace_stRGB) && (pProcamp)) { KernelDll_GetCSCMatrix(CSpace_stRGB, CSpace_BT709, icsc); KernelDll_GetCSCMatrix(CSpace_BT709, CSpace_stRGB, csc); bICSC = bCSC = true; csctype = CSC_RGB_RGB; } else { MOS_SecureMemcpy(m, sizeof(g_cCSC_Identity), (void *)g_cCSC_Identity, sizeof(g_cCSC_Identity)); csctype = CSC_YUV_YUV; } // Product only happens if Procamp is present // Otherwise use the original matrix if (pProcamp) { float b, c, h, s; // Calculate procamp parameters b = pProcamp->fBrightness; c = pProcamp->fContrast; h = pProcamp->fHue * (PI / 180.0f); s = pProcamp->fSaturation; // procamp matrix // // [Y'] [ c 0 0 ] [Y] [ 16 - 16 * c + b ] // [U'] = [ 0 c*s*cos(h) c*s*sin(h) ] [U] + [ 128 - 128*c*s*(cos(h)+sin(h)) ] // [V'] [ 0 -c*s*sin(h) c*s*cos(h) ] [V] [ 128 - 128*c*s*(cos(h)-sin(h)) ] matrix[0] = c; matrix[1] = 0.0f; matrix[2] = 0.0f; matrix[3] = 16.0f - 16.0f * c + b; matrix[4] = 0.0f; matrix[5] = (float)cos(h) * c * s; matrix[6] = (float)sin(h) * c * s; matrix[7] = 128.0f * (1.0f - matrix[5] - matrix[6]); matrix[8] = 0.0f; matrix[9] = -matrix[6]; matrix[10] = matrix[5]; matrix[11] = 128.0f * (1.0f - matrix[5] + matrix[6]); // Calculate final CSC matrix (csc * pa * icsc) if (bICSC) { // Calculate [pa] * [icsc] KernelDll_MatrixProduct(matrix, matrix, icsc); } if (bCSC) { // Calculate [csc] * [pa] (if no icsc) // or [csc] * [pa] * [icsc] KernelDll_MatrixProduct(matrix, csc, matrix); } // Update procamp version pMatrix->iProcampVersion = pProcamp->iProcampVersion; // Use the output matrix to generate kernel CSC parameters MOS_SecureMemcpy(m, sizeof(m), (void *)matrix, sizeof(m)); } // normalize for kernel use matrix[0] = ROUND_FLOAT(m[0], 128.0f); // 9.7 matrix[1] = ROUND_FLOAT(m[1], 128.0f); // 9.7 matrix[2] = ROUND_FLOAT(m[2], 128.0f); // 9.7 matrix[3] = ROUND_FLOAT(m[3], 0.5f); // 16.0 (value/2) matrix[4] = ROUND_FLOAT(m[4], 128.0f); // 9.7 matrix[5] = ROUND_FLOAT(m[5], 128.0f); // 9.7 matrix[6] = ROUND_FLOAT(m[6], 128.0f); // 9.7 matrix[7] = ROUND_FLOAT(m[7], 0.5f); // 16.0 (value/2) matrix[8] = ROUND_FLOAT(m[8], 128.0f); // 9.7 matrix[9] = ROUND_FLOAT(m[9], 128.0f); // 9.7 matrix[10] = ROUND_FLOAT(m[10], 128.0f); // 9.7 matrix[11] = ROUND_FLOAT(m[11], 0.5f); // 16.0 (value/2) // Save matrix as kernel CSC coefficients pState->pfnMapCSCMatrix(csctype, matrix, pMatrix->Coeff); } //--------------------------------------------------------------------------------------- // KernelDll_StartKernelSearch_Next - Starts kernel search // // Parameters: // Kdll_State *pState - [in] Dynamic Linking State // Kdll_FilterEntry *pFilter - [in] Search filter (array of search entries) // int iFilterSize - [in] Search filter size // Kdll_SearchState *pSearchState - [in/out] Kernel search state // // Output: none //--------------------------------------------------------------------------------------- void KernelDll_StartKernelSearch_Next( Kdll_State * pState, Kdll_SearchState *pSearchState, Kdll_FilterEntry *pFilter, int32_t iFilterSize, uint32_t uiIs64BInstrEnabled) { int32_t nLayer; VP_RENDER_FUNCTION_ENTER; // Reset all states MOS_ZeroMemory(pSearchState, sizeof(Kdll_SearchState)); // Setup KDLL state pSearchState->pKdllState = pState; // KDLL state // Cleanup kernel table pSearchState->KernelCount = 0; // # of kernels // Cleanup patch data memset(pSearchState->Patches, 0, sizeof(pSearchState->Patches)); memset(pSearchState->PatchID, -1, sizeof(pSearchState->PatchID)); pSearchState->PatchCount = 0; // Copy original filter; filter will be modified as part of the search if (pFilter && iFilterSize > 0) { MOS_SecureMemcpy(pSearchState->Filter, iFilterSize * sizeof(Kdll_FilterEntry), pFilter, iFilterSize * sizeof(Kdll_FilterEntry)); pSearchState->pFilter = pSearchState->Filter; pSearchState->iFilterSize = iFilterSize; // Copy the render target format pSearchState->target_format = pSearchState->pFilter[iFilterSize - 1].format; // Copy the render target tile type pSearchState->target_tiletype = pSearchState->pFilter[iFilterSize - 1].tiletype; // Indicate whether to use 64B save kernel for render target surface if (uiIs64BInstrEnabled && ((pSearchState->target_tiletype == MOS_TILE_X) || (pSearchState->target_tiletype == MOS_TILE_LINEAR))) { pSearchState->b64BSaveEnabled = true; } } } void KernelDll_ModifyFunctionPointers_Next(Kdll_State *pState) { pState->pfnStartKernelSearch = KernelDll_StartKernelSearch_Next; } bool KernelDll_IsCspace(VPHAL_CSPACE cspace, VPHAL_CSPACE match) { switch (match) { case CSpace_RGB: return (cspace == CSpace_sRGB || cspace == CSpace_stRGB); case CSpace_YUV: return (cspace == CSpace_BT709 || cspace == CSpace_BT601 || cspace == CSpace_BT601_FullRange || cspace == CSpace_BT709_FullRange || cspace == CSpace_xvYCC709 || cspace == CSpace_xvYCC601); case CSpace_Gray: return (cspace == CSpace_BT601Gray || cspace == CSpace_BT601Gray_FullRange); case CSpace_Any: return (cspace != CSpace_None); case CSpace_BT2020: return (cspace == CSpace_BT2020 || cspace == CSpace_BT2020_FullRange); case CSpace_BT2020_RGB: return (cspace == CSpace_BT2020_RGB || cspace == CSpace_BT2020_stRGB); default: return (cspace == match); } return false; } /*---------------------------------------------------------------------------- | Name : KernelDll_GetYuvRangeAndOffset | Purpose : Get the YUV offset and excursion for the input color space | Return : true if success else false \---------------------------------------------------------------------------*/ bool KernelDll_GetYuvRangeAndOffset( Kdll_CSpace cspace, float * pLumaOffset, float * pLumaExcursion, float * pChromaZero, float * pChromaExcursion) { bool res = true; switch (cspace) { case CSpace_BT601_FullRange: case CSpace_BT709_FullRange: case CSpace_BT601Gray_FullRange: case CSpace_BT2020_FullRange: *pLumaOffset = 0.0f; *pLumaExcursion = 255.0f; *pChromaZero = 128.0f; *pChromaExcursion = 255.0f; break; case CSpace_BT601: case CSpace_BT709: case CSpace_xvYCC601: // since matrix is the same as 601, use the same range case CSpace_xvYCC709: // since matrix is the same as 709, use the same range case CSpace_BT601Gray: case CSpace_BT2020: *pLumaOffset = 16.0f; *pLumaExcursion = 219.0f; *pChromaZero = 128.0f; *pChromaExcursion = 224.0f; break; default: res = false; break; } return res; } /*---------------------------------------------------------------------------- | Name : KernelDll_GetRgbRangeAndOffset | Purpose : Get the RGB offset and excursion for the input color space | Return : true if success else false \---------------------------------------------------------------------------*/ bool KernelDll_GetRgbRangeAndOffset( Kdll_CSpace cspace, float * pRgbOffset, float * pRgbExcursion) { bool res = true; switch (cspace) { case CSpace_sRGB: case CSpace_BT2020_RGB: *pRgbOffset = 0.0f; *pRgbExcursion = 255.0f; break; case CSpace_stRGB: case CSpace_BT2020_stRGB: *pRgbOffset = 16.0f; *pRgbExcursion = 219.0f; break; default: res = false; break; } return res; } /*---------------------------------------------------------------------------- | Name : KernelDll_CalcYuvToRgbMatrix | Purpose : Given the YUV->RGB transfer matrix, get the final matrix after | applying offsets and excursions. | | [R'] [R_o] [R_e/Y_e 0 0 ] [Y' - Y_o] | [G'] = [R_o] + [YUVtoRGBCoeff (3x3 matrix)]. [ 0 R_e/C_e 0 ]. [Cb' - C_z] | [B'] [R_o] [ 0 0 R_e/C_e]. [Cr' - C_z] | | [R'] = [C0 C1 C2] [Y' ] [C3] {Out pMatrix} | [G'] = [C4 C5 C6].[Cb'] + [C7] | [B'] = [C8 C9 C10] [Cr'] + [C11] | | Return : true if success else false \---------------------------------------------------------------------------*/ bool KernelDll_CalcYuvToRgbMatrix( Kdll_CSpace src, // [in] YUV Color space Kdll_CSpace dst, // [in] RGB Color space float * pTransferMatrix, // [in] Transfer matrix (3x3) float * pOutMatrix) // [out] Conversion matrix (3x4) { bool res; float Y_o, Y_e, C_z, C_e; float R_o, R_e; res = true; res = KernelDll_GetRgbRangeAndOffset(dst, &R_o, &R_e); if (res == false) { goto finish; } res = KernelDll_GetYuvRangeAndOffset(src, &Y_o, &Y_e, &C_z, &C_e); if (res == false) { goto finish; } // after + (3x3)(3x3) pOutMatrix[0] = pTransferMatrix[0] * R_e / Y_e; pOutMatrix[4] = pTransferMatrix[3] * R_e / Y_e; pOutMatrix[8] = pTransferMatrix[6] * R_e / Y_e; pOutMatrix[1] = pTransferMatrix[1] * R_e / C_e; pOutMatrix[5] = pTransferMatrix[4] * R_e / C_e; pOutMatrix[9] = pTransferMatrix[7] * R_e / C_e; pOutMatrix[2] = pTransferMatrix[2] * R_e / C_e; pOutMatrix[6] = pTransferMatrix[5] * R_e / C_e; pOutMatrix[10] = pTransferMatrix[8] * R_e / C_e; // (3x1) - (3x3)(3x3)(3x1) pOutMatrix[3] = R_o - (pOutMatrix[0] * Y_o + pOutMatrix[1] * C_z + pOutMatrix[2] * C_z); pOutMatrix[7] = R_o - (pOutMatrix[4] * Y_o + pOutMatrix[5] * C_z + pOutMatrix[6] * C_z); pOutMatrix[11] = R_o - (pOutMatrix[8] * Y_o + pOutMatrix[9] * C_z + pOutMatrix[10] * C_z); finish: return res; } /*---------------------------------------------------------------------------- | Name : KernelDll_CalcRgbToYuvMatrix | Purpose : Given the RGB->YUV transfer matrix, get the final matrix after | applying offsets and excursions. | | [Y' ] [Y_o - Y_e.R_o/R_e] [Y_e/R_e 0 0 ] [ RGB to YUV ] [R'] | [Cb'] = [C_z] + [ 0 C_e/R_e 0 ]. [Transfer matrix]. [G'] | [Cr'] [C_z] [ 0 0 C_e/R_e] [ 3x3 matrix ] [B'] | | [Y' ] = [C0 C1 C2] [R'] [C3] {Out pMatrix} | [Cb'] = [C4 C5 C6].[G'] + [C7] | [Cr'] = [C8 C9 C10] [B'] + [C11] | | Return : true if success else false \---------------------------------------------------------------------------*/ bool KernelDll_CalcRgbToYuvMatrix( Kdll_CSpace src, // [in] RGB Color space Kdll_CSpace dst, // [in] YUV Color space float * pTransferMatrix, // [in] Transfer matrix (3x3) float * pOutMatrix) // [out] Conversion matrix (3x4) { bool res; float Y_o, Y_e, C_z, C_e; float R_o, R_e; res = true; res = KernelDll_GetRgbRangeAndOffset(src, &R_o, &R_e); if (res == false) { goto finish; } res = KernelDll_GetYuvRangeAndOffset(dst, &Y_o, &Y_e, &C_z, &C_e); if (res == false) { goto finish; } // multiplication of + onwards pOutMatrix[0] = pTransferMatrix[0] * Y_e / R_e; pOutMatrix[1] = pTransferMatrix[1] * Y_e / R_e; pOutMatrix[2] = pTransferMatrix[2] * Y_e / R_e; pOutMatrix[4] = pTransferMatrix[3] * C_e / R_e; pOutMatrix[5] = pTransferMatrix[4] * C_e / R_e; pOutMatrix[6] = pTransferMatrix[5] * C_e / R_e; pOutMatrix[8] = pTransferMatrix[6] * C_e / R_e; pOutMatrix[9] = pTransferMatrix[7] * C_e / R_e; pOutMatrix[10] = pTransferMatrix[8] * C_e / R_e; // before + pOutMatrix[3] = Y_o - Y_e * R_o / R_e; pOutMatrix[7] = C_z; pOutMatrix[11] = C_z; finish: return res; } /*---------------------------------------------------------------------------- | Name : KernelDll_CalcGrayCoeffs | Purpose : Given CSC matrix, calculate the new matrix making Chroma zero. | Chroma will be read from the surface, but we need to factor in C_z | by adjusting this in the constant. | | [R'] = [C0 C1 C2] [Y' ] [C3] {Out pMatrix} | [G'] = [C4 C5 C6].[C_z] + [C7] | [B'] = [C8 C9 C10] [C_z] [C11] | | New C3 = C1 * C_z + C2 * C_z + C3 | | Return : true if success else false \---------------------------------------------------------------------------*/ bool KernelDll_CalcGrayCoeffs( Kdll_CSpace src, // [in] YUV source Color space float * pMatrix) // [in/out] Conversion matrix (3x4) { float Y_o, Y_e, C_z, C_e; bool res; res = true; res = KernelDll_GetYuvRangeAndOffset(src, &Y_o, &Y_e, &C_z, &C_e); if (res == false) { goto finish; } // Calculate the constant offset by factoring in C_z pMatrix[3] = pMatrix[1] * C_z + pMatrix[2] * C_z + pMatrix[3]; pMatrix[7] = pMatrix[5] * C_z + pMatrix[6] * C_z + pMatrix[7]; pMatrix[11] = pMatrix[9] * C_z + pMatrix[10] * C_z + pMatrix[11]; // Nullify the effect of chroma read pMatrix[1] = pMatrix[2] = 0; pMatrix[5] = pMatrix[6] = 0; pMatrix[9] = pMatrix[10] = 0; finish: return res; } /*---------------------------------------------------------------------------- | Name : KernelDll_3x3MatrixProduct | Purpose : Given two [3x4] input matrices, calculate [3x3]x[3x3] ignoring | the last column in both inputs | Return : none \---------------------------------------------------------------------------*/ void KernelDll_3x3MatrixProduct( float * dest, const float *m1, const float *m2) { dest[0] = m1[0] * m2[0] + m1[1] * m2[4] + m1[2] * m2[8]; dest[1] = m1[0] * m2[1] + m1[1] * m2[5] + m1[2] * m2[9]; dest[2] = m1[0] * m2[2] + m1[1] * m2[6] + m1[2] * m2[10]; dest[4] = m1[4] * m2[0] + m1[5] * m2[4] + m1[6] * m2[8]; dest[5] = m1[4] * m2[1] + m1[5] * m2[5] + m1[6] * m2[9]; dest[6] = m1[4] * m2[2] + m1[5] * m2[6] + m1[6] * m2[10]; dest[8] = m1[8] * m2[0] + m1[9] * m2[4] + m1[10] * m2[8]; dest[9] = m1[8] * m2[1] + m1[9] * m2[5] + m1[10] * m2[9]; dest[10] = m1[8] * m2[2] + m1[9] * m2[6] + m1[10] * m2[10]; } /*---------------------------------------------------------------------------- | Name : KernelDll_CalcYuvToYuvMatrix | Purpose : Calculate the matrix equation for converting b/w YUV color spaces. | 1. Get conversion matrix from Source YUV to sRGB | 2. Get conversion matrix from sRGB to Destination YUV | 3. Apply the transformation below to get the final matrix | | [Y'dst] = [C0 C1 C2] [C0 C1 C2][Y'src] [C0 C1 C2] [C3] [C3] | [U'] = [C4 C5 C6].[C4 C5 C6][C_z] + [C4 C5 C6].[C7] + [C7] | [V'] = [C8 C9 C10] [C8 C9 C10][C_z] [C8 C9 C10] [C11] [C11] | dst matrix src matrix dst matrix src dst | | [Y'dst] = [C0 C1 C2] [Y'src] [C3] {Out pMatrix} | [U'] = [C4 C5 C6].[C_z] + [C7] | [V'] = [C8 C9 C10] [C_z] [C11] | | Return : true if success else false \---------------------------------------------------------------------------*/ bool KernelDll_CalcYuvToYuvMatrix( Kdll_CSpace src, // [in] YUV Color space Kdll_CSpace dst, // [in] YUV Color space float * pOutMatrix) // [out] Conversion matrix (3x4) { float fYuvToRgb[12] = {0}; float fRgbToYuv[12] = {0}; bool res; res = true; // 1. Get conversion matrix from Source YUV to sRGB if (IS_BT601_CSPACE(src)) { res = KernelDll_CalcYuvToRgbMatrix(src, CSpace_sRGB, (float *)g_cCSC_BT601_YUV_RGB, fYuvToRgb); } else if(IS_COLOR_SPACE_BT2020_YUV(src)) { switch (src) { case CSpace_BT2020: res = KernelDll_CalcYuvToRgbMatrix(CSpace_BT2020, CSpace_sRGB, (float *)g_cCSC_BT2020_LimitedYUV_RGB, fYuvToRgb); break; case CSpace_BT2020_FullRange: res = KernelDll_CalcYuvToRgbMatrix(CSpace_BT2020_FullRange, CSpace_sRGB, (float *)g_cCSC_BT2020_YUV_RGB, fYuvToRgb); break; default: res = false; break; } } else { res = KernelDll_CalcYuvToRgbMatrix(src, CSpace_sRGB, (float *)g_cCSC_BT709_YUV_RGB, fYuvToRgb); } if (res == false) { goto finish; } // 2. Get conversion matrix from sRGB to Destination YUV if (IS_BT601_CSPACE(dst)) { res = KernelDll_CalcRgbToYuvMatrix(CSpace_sRGB, dst, (float *)g_cCSC_BT601_RGB_YUV, fRgbToYuv); } else if (IS_COLOR_SPACE_BT2020_YUV(dst)) { switch (dst) { case CSpace_BT2020_FullRange: res = KernelDll_CalcRgbToYuvMatrix(CSpace_sRGB, dst, (float *)g_cCSC_BT2020_RGB_YUV, fRgbToYuv); break; case CSpace_BT2020: res = KernelDll_CalcRgbToYuvMatrix(CSpace_sRGB, dst, (float *)g_cCSC_BT2020_RGB_LimitedYUV, fRgbToYuv); break; default: res = false; break; } } else { res = KernelDll_CalcRgbToYuvMatrix(CSpace_sRGB, dst, (float *)g_cCSC_BT709_RGB_YUV, fRgbToYuv); } if (res == false) { goto finish; } // 3. Multiply the 2 matrices above KernelDll_3x3MatrixProduct(pOutMatrix, fRgbToYuv, fYuvToRgb); // Perform [3x3][3x1] matrix multiply + [3x1] matrix pOutMatrix[3] = fRgbToYuv[0] * fYuvToRgb[3] + fRgbToYuv[1] * fYuvToRgb[7] + fRgbToYuv[2] * fYuvToRgb[11] + fRgbToYuv[3]; pOutMatrix[7] = fRgbToYuv[4] * fYuvToRgb[3] + fRgbToYuv[5] * fYuvToRgb[7] + fRgbToYuv[6] * fYuvToRgb[11] + fRgbToYuv[7]; pOutMatrix[11] = fRgbToYuv[8] * fYuvToRgb[3] + fRgbToYuv[9] * fYuvToRgb[7] + fRgbToYuv[10] * fYuvToRgb[11] + fRgbToYuv[11]; finish: return res; } /*---------------------------------------------------------------------------- | Name : KernelDll_GetCSCMatrix | Purpose : Get the required matrix for the given CSC conversion | Return : \---------------------------------------------------------------------------*/ void KernelDll_GetCSCMatrix( Kdll_CSpace src, // [in] Source Color space Kdll_CSpace dst, // [in] Destination Color space float * pCSC_Matrix) // [out] CSC matrix to use { bool bMatrix; bool bSrcGray; Kdll_CSpace temp; int32_t i; bMatrix = false; bSrcGray = KernelDll_IsCspace(src, CSpace_Gray); // convert gray color spaces to its equivalent non-gray cpsace switch (src) { case CSpace_BT601Gray: temp = CSpace_BT601; break; case CSpace_BT601Gray_FullRange: temp = CSpace_BT601_FullRange; break; default: temp = src; break; } // BT601/709 YUV to sRGB/stRGB conversion if (KernelDll_IsCspace(temp, CSpace_YUV) || KernelDll_IsCspace(temp, CSpace_Gray)) { if (KernelDll_IsCspace(dst, CSpace_RGB)) { if (IS_BT601_CSPACE(temp)) { KernelDll_CalcYuvToRgbMatrix(temp, dst, (float *)g_cCSC_BT601_YUV_RGB, pCSC_Matrix); bMatrix = true; } else // if (IS_BT709_CSPACE(temp)) { KernelDll_CalcYuvToRgbMatrix(temp, dst, (float *)g_cCSC_BT709_YUV_RGB, pCSC_Matrix); bMatrix = true; } } } // sRGB/stRGB to BT601/709 YUV conversion else if (KernelDll_IsCspace(temp, CSpace_RGB)) { if (KernelDll_IsCspace(dst, CSpace_YUV)) { if (IS_BT601_CSPACE(dst)) { KernelDll_CalcRgbToYuvMatrix(temp, dst, (float *)g_cCSC_BT601_RGB_YUV, pCSC_Matrix); bMatrix = true; } else // if (IS_BT709_CSPACE(temp)) { KernelDll_CalcRgbToYuvMatrix(temp, dst, (float *)g_cCSC_BT709_RGB_YUV, pCSC_Matrix); bMatrix = true; } } } // BT2020 YUV to RGB conversion else if (KernelDll_IsCspace(temp, CSpace_BT2020)) { if (KernelDll_IsCspace(dst, CSpace_BT2020_RGB)) { KernelDll_CalcYuvToRgbMatrix(temp, dst, (float *)g_cCSC_BT2020_YUV_RGB, pCSC_Matrix); bMatrix = true; } } // BT2020 RGB to YUV conversion else if (KernelDll_IsCspace(temp, CSpace_BT2020_RGB)) { if (KernelDll_IsCspace(dst, CSpace_BT2020)) { KernelDll_CalcRgbToYuvMatrix(temp, dst, (float *)g_cCSC_BT2020_RGB_YUV, pCSC_Matrix); bMatrix = true; } } // If matrix has not been derived yet, its one of the below special cases if (!bMatrix) { if (temp == dst) // Check if its identity matrix { MOS_SecureMemcpy(pCSC_Matrix, sizeof(g_cCSC_Identity), (void *)g_cCSC_Identity, sizeof(g_cCSC_Identity)); } else if (KernelDll_IsCspace(temp, CSpace_RGB)) // sRGB to stRGB inter-conversions { if (temp == CSpace_sRGB) { MOS_SecureMemcpy(pCSC_Matrix, sizeof(g_cCSC_sRGB_stRGB), (void *)g_cCSC_sRGB_stRGB, sizeof(g_cCSC_sRGB_stRGB)); } else //temp == CSpace_stRGB { MOS_SecureMemcpy(pCSC_Matrix, sizeof(g_cCSC_stRGB_sRGB), (void *)g_cCSC_stRGB_sRGB, sizeof(g_cCSC_stRGB_sRGB)); } } else if (KernelDll_IsCspace(temp, CSpace_YUV)) // 601 to 709 inter-conversions { KernelDll_CalcYuvToYuvMatrix(temp, dst, pCSC_Matrix); } else if (KernelDll_IsCspace(temp, CSpace_BT2020_RGB)) { if (temp == CSpace_BT2020_RGB) //BT2020_RGB to BT2020_limited_RGB conversions { MOS_SecureMemcpy(pCSC_Matrix, sizeof(g_cCSC_BT2020RGB_BT2020stRGB), (void *)g_cCSC_BT2020RGB_BT2020stRGB, sizeof(g_cCSC_BT2020RGB_BT2020stRGB)); } else if (temp == CSpace_BT2020_stRGB) //BT2020_limited_RGB to BT2020_RGB conversions { MOS_SecureMemcpy(pCSC_Matrix, sizeof(g_cCSC_BT2020stRGB_BT2020RGB), (void *)g_cCSC_BT2020stRGB_BT2020RGB, sizeof(g_cCSC_BT2020stRGB_BT2020RGB)); } } else if (KernelDll_IsCspace(temp, CSpace_BT2020)) // BT2020 limited_YUV to BT2020_FullRange_YUV conversions { KernelDll_CalcYuvToYuvMatrix(temp, dst, pCSC_Matrix); } else { VP_RENDER_ASSERTMESSAGE("Not supported color space conversion(from %d to %d)", src, dst); MT_ERR2(MT_VP_KERNEL_CSC, MT_VP_KERNEL_CSPACE, src, MT_VP_KERNEL_CSPACE, dst); } } // Calculate the Gray transformation matrix now if (bSrcGray) { KernelDll_CalcGrayCoeffs(src, pCSC_Matrix); } VP_RENDER_NORMALMESSAGE(""); for (i = 0; i < 3; i++) { VP_RENDER_NORMALMESSAGE("%f\t%f\t%f\t%f", pCSC_Matrix[4 * i], pCSC_Matrix[4 * i + 1], pCSC_Matrix[4 * i + 2], pCSC_Matrix[4 * i + 3]); } } bool KernelDll_MapCSCMatrix( Kdll_CSCType csctype, const float *matrix, short * coeff) { // Unified kernel architecture requires that the color space // conversion coefficients programmed in specific orders, depends on the // type of the color space conversion. // // M (matrix) ---> C (coeff) switch (csctype) { case CSC_YUV_RGB: // direct mapping from matrix to coeff coeff[0] = FLOAT_TO_SHORT(matrix[0]); // M0 --> C0 coeff[1] = FLOAT_TO_SHORT(matrix[1]); // M1 --> C1 coeff[2] = FLOAT_TO_SHORT(matrix[2]); // M2 --> C2 coeff[3] = FLOAT_TO_SHORT(matrix[3]); // M3 --> C3 coeff[4] = FLOAT_TO_SHORT(matrix[4]); // M4 --> C4 coeff[5] = FLOAT_TO_SHORT(matrix[5]); // M5 --> C5 coeff[6] = FLOAT_TO_SHORT(matrix[6]); // M6 --> C6 coeff[7] = FLOAT_TO_SHORT(matrix[7]); // M7 --> C7 coeff[8] = FLOAT_TO_SHORT(matrix[8]); // M8 --> C8 coeff[9] = FLOAT_TO_SHORT(matrix[9]); // M9 --> C9 coeff[10] = FLOAT_TO_SHORT(matrix[10]); // M10 --> C10 coeff[11] = FLOAT_TO_SHORT(matrix[11]); // M11 --> C11 break; case CSC_RGB_YUV: coeff[6] = FLOAT_TO_SHORT(matrix[0]); // M0 --> C6 coeff[4] = FLOAT_TO_SHORT(matrix[1]); // M1 --> C4 coeff[5] = FLOAT_TO_SHORT(matrix[2]); // M2 --> C5 coeff[7] = FLOAT_TO_SHORT(matrix[3]); // M3 --> C7 coeff[10] = FLOAT_TO_SHORT(matrix[4]); // M4 --> C10 coeff[8] = FLOAT_TO_SHORT(matrix[5]); // M5 --> C8 coeff[9] = FLOAT_TO_SHORT(matrix[6]); // M6 --> C9 coeff[11] = FLOAT_TO_SHORT(matrix[7]); // M7 --> C11 coeff[2] = FLOAT_TO_SHORT(matrix[8]); // M8 --> C2 coeff[0] = FLOAT_TO_SHORT(matrix[9]); // M9 --> C0 coeff[1] = FLOAT_TO_SHORT(matrix[10]); // M10 --> C1 coeff[3] = FLOAT_TO_SHORT(matrix[11]); // M11 --> C3 break; case CSC_YUV_YUV: coeff[4] = FLOAT_TO_SHORT(matrix[0]); // M0 --> C4 coeff[5] = FLOAT_TO_SHORT(matrix[1]); // M1 --> C5 coeff[6] = FLOAT_TO_SHORT(matrix[2]); // M2 --> C6 coeff[7] = FLOAT_TO_SHORT(matrix[3]); // M3 --> C7 coeff[8] = FLOAT_TO_SHORT(matrix[4]); // M4 --> C8 coeff[9] = FLOAT_TO_SHORT(matrix[5]); // M5 --> C9 coeff[10] = FLOAT_TO_SHORT(matrix[6]); // M6 --> C10 coeff[11] = FLOAT_TO_SHORT(matrix[7]); // M7 --> C11 coeff[0] = FLOAT_TO_SHORT(matrix[8]); // M8 --> C0 coeff[1] = FLOAT_TO_SHORT(matrix[9]); // M9 --> C1 coeff[2] = FLOAT_TO_SHORT(matrix[10]); // M10 --> C2 coeff[3] = FLOAT_TO_SHORT(matrix[11]); // M11 --> C3 break; default: //CSC_RGB_RGB coeff[0] = FLOAT_TO_SHORT(matrix[1]); // M1 --> C0 coeff[1] = FLOAT_TO_SHORT(matrix[2]); // M2 --> C1 coeff[2] = FLOAT_TO_SHORT(matrix[0]); // M0 --> C2 coeff[3] = FLOAT_TO_SHORT(matrix[3]); // M3 --> C3 coeff[4] = FLOAT_TO_SHORT(matrix[5]); // M5 --> C4 coeff[5] = FLOAT_TO_SHORT(matrix[6]); // M6 --> C5 coeff[6] = FLOAT_TO_SHORT(matrix[4]); // M4 --> C6 coeff[7] = FLOAT_TO_SHORT(matrix[7]); // M7 --> C7 coeff[8] = FLOAT_TO_SHORT(matrix[9]); // M9 --> C8 coeff[9] = FLOAT_TO_SHORT(matrix[10]); // M10 --> C9 coeff[10] = FLOAT_TO_SHORT(matrix[8]); // M8 --> C10 coeff[11] = FLOAT_TO_SHORT(matrix[11]); // M11 --> C11 break; } return true; } bool KernelDll_IsFormat( MOS_FORMAT format, VPHAL_CSPACE cspace, MOS_FORMAT match) { switch (match) { case Format_Any: return (format != Format_None); break; case Format_RGB_Swap: return (IS_RGB_SWAP(format)); case Format_RGB_No_Swap: return (IS_RGB_NO_SWAP(format)); case Format_RGB: if (IS_PAL_FORMAT(format)) { return (KernelDll_IsCspace(cspace, CSpace_RGB)); } else { return (IS_RGB_FORMAT(format) && !IS_PL3_RGB_FORMAT(format)); } case Format_RGB32: return (IS_RGB32_FORMAT(format)); case Format_PA: if (IS_PAL_FORMAT(format)) { return (KernelDll_IsCspace(cspace, CSpace_YUV)); } else { return (IS_PA_FORMAT(format) || format == Format_AUYV); } case Format_PL2: return (IS_PL2_FORMAT(format)); case Format_PL2_UnAligned: return (IS_PL2_FORMAT_UnAligned(format)); case Format_PL3: return (IS_PL3_FORMAT(format)); case Format_PL3_RGB: return (IS_PL3_RGB_FORMAT(format)); case Format_AYUV: return (format == Format_AYUV); case Format_PAL: return (IS_PAL_FORMAT(format)); default: return (format == match); } return false; } //--------------------------------------------------------------------------------------- // KernelDll_SetupProcampParameters - Setup Kernel Procamp Parameters // // Parameters: // Kdll_State *pState - [in] Kernel dll State to release // Kdll_Procamp *pProcamp - [in] Pointer to array of Procamp Parameters // int32_t iProcampSize - [in] Size of the array // // Output: Pointer to allocated Kernel dll state // nullptr - Failed to allocate Kernel dll state //----------------------------------------------------------------------------------------- void KernelDll_SetupProcampParameters(Kdll_State *pState, Kdll_Procamp *pProcamp, int32_t iProcampSize) { VP_RENDER_FUNCTION_ENTER; // Setup pointer to procamp parameters pState->pProcamp = pProcamp; pState->iProcampSize = iProcampSize; } //-------------------------------------------------------------- // Fowler/Noll/Vo FNV-1a hash algorithm - public domain //-------------------------------------------------------------- uint32_t KernelDll_SimpleHash(void *pData, int32_t iSize) { static const uint32_t k = 0x1000193; uint32_t hash = 0x811c9dc5; char *p = (char *)pData; for(; iSize > 0; iSize--) { hash ^= (*p++); hash *= k; } return hash; } //-------------------------------------------------------------- // KernelDll_GetCombinedKernel - Search combined kernel //-------------------------------------------------------------- Kdll_CacheEntry *KernelDll_GetCombinedKernel( Kdll_State *pState, Kdll_FilterEntry *pFilter, int32_t iFilterSize, uint32_t dwHash) { Kdll_KernelHashTable *pHashTable; Kdll_KernelHashEntry *entries, *curr, *next; uint32_t folded_hash; uint16_t entry; VP_RENDER_FUNCTION_ENTER; // Get hash table pHashTable = &pState->KernelHashTable; // fold hash from 32 to 8 bit :-) FOLD_HASH(folded_hash, dwHash) // No entries entry = pHashTable->wHashTable[folded_hash]; if (entry == 0 || entry > DL_MAX_COMBINED_KERNELS ) return nullptr; entries = (&pHashTable->HashEntry[0]) - 1; // all indices are 1 based (0 means null) curr = &entries[entry]; for (; (curr != nullptr); curr = next) { // match 32-bit hash, then compare filter if (curr->dwHash == dwHash && curr->iFilter == iFilterSize) { if (memcmp(curr->pFilter, pFilter, iFilterSize * sizeof(Kdll_FilterEntry)) == 0) { break; } } // Next entry with the same 8-bit folded hash next = (curr->next) ? (&entries[curr->next]) : nullptr; } if (curr) { // Kernel already cached curr->pCacheEntry->dwRefresh = pState->dwRefresh++; return (curr->pCacheEntry); } else { // Kernel must be built return nullptr; } } /*---------------------------------------------------------------------------- | Name : KernelDll_FindRule | Purpose : Find a rule that matches the current search/input state | | Input : pState - Kernel Dll state | pSearchState - current DL search state | | Return : \---------------------------------------------------------------------------*/ bool KernelDll_FindRule( Kdll_State * pState, Kdll_SearchState *pSearchState) { uint32_t parser_state = (uint32_t)pSearchState->state; Kdll_RuleEntrySet * pRuleSet; const Kdll_RuleEntry *pRuleEntry; int32_t iRuleCount; int32_t iMatchCount; bool bLayerFormatMatched; bool bSrc0FormatMatched; bool bSrc1FormatMatched; bool bTargetFormatMatched; bool bSrc0SampingMatched; VP_RENDER_FUNCTION_ENTER; // All Custom states are handled as a single group if (parser_state >= Parser_Custom) { parser_state = Parser_Custom; } pRuleSet = pState->pDllRuleTable[parser_state]; iRuleCount = pState->iDllRuleCount[parser_state]; if (pRuleSet == nullptr || iRuleCount == 0) { VP_RENDER_NORMALMESSAGE("Search rules undefined."); pSearchState->pMatchingRuleSet = nullptr; return false; } // Search matching entry for (; iRuleCount > 0; iRuleCount--, pRuleSet++) { // Points to the first rule, get number of matches pRuleEntry = pRuleSet->pRuleEntry; iMatchCount = pRuleSet->iMatchCount; // Initialize for each Ruleset bLayerFormatMatched = false; bSrc0FormatMatched = false; bSrc1FormatMatched = false; bTargetFormatMatched = false; bSrc0SampingMatched = false; // Match all rules within the same RuleSet for (; iMatchCount > 0; iMatchCount--, pRuleEntry++) { switch (pRuleEntry->id) { // Match current Parser State case RID_IsParserState: if (pSearchState->state == (Kdll_ParserState)pRuleEntry->value) { continue; } else { break; } // Match render method case RID_IsRenderMethod: if (pSearchState->pFilter->RenderMethod == (Kdll_RenderMethod)pRuleEntry->value) { continue; } else { break; } // Match target color space case RID_IsTargetCspace: if (KernelDll_IsCspace(pSearchState->cspace, (VPHAL_CSPACE)pRuleEntry->value)) { continue; } else { break; } // Match current layer ID case RID_IsLayerID: if (pSearchState->pFilter->layer == (Kdll_Layer)pRuleEntry->value) { continue; } else { break; } // Match current layer format case RID_IsLayerFormat: if (pRuleEntry->logic == Kdll_Or && bLayerFormatMatched) { // Already found matching format in the ruleset continue; } else { // Check if the layer format matches the rule if (KernelDll_IsFormat(pSearchState->pFilter->format, pSearchState->pFilter->cspace, (MOS_FORMAT)pRuleEntry->value)) { bLayerFormatMatched = true; } if (pRuleEntry->logic == Kdll_None && !bLayerFormatMatched) { // Last entry and No matching format was found break; } else { continue; } } // Match shuffling requirement case RID_IsShuffling: if (pSearchState->ShuffleSamplerData == (Kdll_Shuffling)pRuleEntry->value) { continue; } else { break; } // Check if RT rotates case RID_IsRTRotate: if (pSearchState->bRTRotate == (pRuleEntry->value ? true : false)) { continue; } else { break; } // Match current layer rotation case RID_IsLayerRotation: if (pSearchState->pFilter->rotation == (MEDIA_ROTATION)pRuleEntry->value) { continue; } else { break; } // Match Src0 source format (surface) case RID_IsSrc0Format: if (pRuleEntry->logic == Kdll_Or && bSrc0FormatMatched) { // Already found matching format in the ruleset continue; } else { // Check if the source 0 format matches the rule // The intermediate colorspace is used to determine // if palettized input is given in RGB or YUV format. if (KernelDll_IsFormat(pSearchState->src0_format, pSearchState->cspace, (MOS_FORMAT)pRuleEntry->value)) { bSrc0FormatMatched = true; } if (pRuleEntry->logic == Kdll_None && !bSrc0FormatMatched) { // Last entry and No matching format was found break; } else { continue; } } // Match Src0 sampling mode case RID_IsSrc0Sampling: // Check if the layer format matches the rule if (pSearchState->src0_sampling == (Kdll_Sampling)pRuleEntry->value) { bSrc0SampingMatched = true; continue; } else if (bSrc0SampingMatched || pRuleEntry->logic == Kdll_Or) { continue; } else if ((Kdll_Sampling)pRuleEntry->value == Sample_Any && pSearchState->src0_sampling != Sample_None) { continue; } else { break; } // Match Src0 rotation case RID_IsSrc0Rotation: if (pSearchState->src0_rotation == (MEDIA_ROTATION)pRuleEntry->value) { continue; } else { break; } // Match Src0 Colorfill case RID_IsSrc0ColorFill: if (pSearchState->src0_colorfill == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match Src0 Luma Key case RID_IsSrc0LumaKey: if (pSearchState->src0_lumakey == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match Src0 Procamp case RID_IsSrc0Procamp: if (pSearchState->pFilter->procamp == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match Src0 CSC coefficients case RID_IsSrc0Coeff: if (pSearchState->src0_coeff == (Kdll_CoeffID)pRuleEntry->value) { continue; } else if ((Kdll_CoeffID)pRuleEntry->value == CoeffID_Any && pSearchState->src0_coeff != CoeffID_None) { continue; } else { break; } // Match Src0 CSC coefficients setting mode case RID_IsSetCoeffMode: if (pSearchState->pFilter->SetCSCCoeffMode == (Kdll_SetCSCCoeffMethod)pRuleEntry->value) { continue; } else { break; } // Match Src0 processing mode case RID_IsSrc0Processing: if (pSearchState->src0_process == (Kdll_Processing)pRuleEntry->value) { continue; } if ((Kdll_Processing)pRuleEntry->value == Process_Any && pSearchState->src0_process != Process_None) { continue; } else { break; } // Match Src0 chromasiting mode case RID_IsSrc0Chromasiting: if (pSearchState->Filter->chromasiting == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match Src1 source format (surface) case RID_IsSrc1Format: if (pRuleEntry->logic == Kdll_Or && bSrc1FormatMatched) { // Already found matching format in the ruleset continue; } else { // Check if the source 1 format matches the rule // The intermediate colorspace is used to determine // if palettized input is given in RGB or YUV format. if (KernelDll_IsFormat(pSearchState->src1_format, pSearchState->cspace, (MOS_FORMAT)pRuleEntry->value)) { bSrc1FormatMatched = true; } if (pRuleEntry->logic == Kdll_None && !bSrc1FormatMatched) { // Last entry and No matching format was found break; } else { continue; } } // Match Src1 sampling mode case RID_IsSrc1Sampling: if (pSearchState->src1_sampling == (Kdll_Sampling)pRuleEntry->value) { continue; } else if ((Kdll_Sampling)pRuleEntry->value == Sample_Any && pSearchState->src1_sampling != Sample_None) { continue; } else { break; } // Match Src1 Luma Key case RID_IsSrc1LumaKey: if (pSearchState->src1_lumakey == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match Src1 Sampler LumaKey case RID_IsSrc1SamplerLumaKey: if (pSearchState->src1_samplerlumakey == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match Src1 Procamp case RID_IsSrc1Procamp: if (pSearchState->pFilter->procamp == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match Src1 CSC coefficients case RID_IsSrc1Coeff: if (pSearchState->src1_coeff == (Kdll_CoeffID)pRuleEntry->value) { continue; } else if ((Kdll_CoeffID)pRuleEntry->value == CoeffID_Any && pSearchState->src1_coeff != CoeffID_None) { continue; } else { break; } // Match Src1 processing mode case RID_IsSrc1Processing: if (pSearchState->src1_process == (Kdll_Processing)pRuleEntry->value) { continue; } if ((Kdll_Processing)pRuleEntry->value == Process_Any && pSearchState->src1_process != Process_None) { continue; } else { break; } // Match Src1 chromasiting mode case RID_IsSrc1Chromasiting: //pSearchState->pFilter is pointed to the real sub layer if (pSearchState->pFilter->chromasiting == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match Layer number case RID_IsLayerNumber: if (pSearchState->layer_number == (int32_t)pRuleEntry->value) { continue; } else { break; } // Match quadrant case RID_IsQuadrant: if (pSearchState->quadrant == (int32_t)pRuleEntry->value) { continue; } else { break; } // Set CSC flag before Mix case RID_IsCSCBeforeMix: if (pSearchState->bCscBeforeMix == (pRuleEntry->value ? true : false)) { continue; } else { break; } case RID_IsDualOutput: if (pSearchState->pFilter->dualout == (pRuleEntry->value ? true : false)) { continue; } else { break; } case RID_IsTargetFormat: if (pRuleEntry->logic == Kdll_Or && bTargetFormatMatched) { // Already found matching format in the ruleset continue; } else { if (pSearchState->target_format == (MOS_FORMAT)pRuleEntry->value) { bTargetFormatMatched = true; } if (pRuleEntry->logic == Kdll_None && !bTargetFormatMatched) { // Last entry and No matching format was found break; } else { continue; } } case RID_Is64BSaveEnabled: if (pSearchState->b64BSaveEnabled == (pRuleEntry->value ? true : false)) { continue; } else { break; } case RID_IsTargetTileType: if (pRuleEntry->logic == Kdll_None && pSearchState->target_tiletype == (MOS_TILE_TYPE)pRuleEntry->value) { continue; } else if (pRuleEntry->logic == Kdll_Not && pSearchState->target_tiletype != (MOS_TILE_TYPE)pRuleEntry->value) { continue; } else { break; } case RID_IsProcampEnabled: if (pSearchState->bProcamp == (pRuleEntry->value ? true : false)) { continue; } else { break; } case RID_IsConstOutAlpha: if (pSearchState->pFilter->bFillOutputAlphaWithConstant == (pRuleEntry->value ? true : false)) { continue; } else { break; } case RID_IsDitherNeeded: if (pSearchState->pFilter->bIsDitherNeeded == (pRuleEntry->value ? true : false)) { continue; } else { break; } // Undefined search rule will fail default: VP_RENDER_ASSERTMESSAGE("Invalid rule %d @ layer %d, state %d.", pRuleEntry->id, pSearchState->layer_number, pSearchState->state); MT_ERR3(MT_VP_KERNEL_RULE, MT_VP_KERNEL_RULE_ID, pRuleEntry->id, MT_VP_KERNEL_RULE_LAYERNUM, pSearchState->layer_number, MT_VP_KERNEL_RULE_SEARCH_STATE, pSearchState->state); break; } // End of switch to deal with all matching rule IDs // Rule didn't match - try another RuleSet break; } // End of file loop to test all rules for the current RuleSet // Match if (iMatchCount == 0) { pSearchState->pMatchingRuleSet = pRuleSet; return true; } } // End of for loop to test all RuleSets for the current parser state // Failed to find a matching rule -> kernel search will fail VP_RENDER_NORMALMESSAGE("Fail to find a matching rule @ layer %d, state %d.", pSearchState->layer_number, pSearchState->state); MT_ERR2(MT_VP_KERNEL_RULE, MT_VP_KERNEL_RULE_LAYERNUM, pSearchState->layer_number, MT_VP_KERNEL_RULE_SEARCH_STATE, pSearchState->state); // No match -> return pSearchState->pMatchingRuleSet = nullptr; return false; } //-------------------------------------------------------------- // Append kernel, include symbols to resolve //-------------------------------------------------------------- bool Kdll_AppendKernel(Kdll_KernelCache *pKernelCache, Kdll_SearchState * pSearchState, int32_t iKUID, Kdll_PatchData * pKernelPatch) { Kdll_State * pState; Kdll_Symbol * pSymbols; Kdll_CacheEntry *kernels; Kdll_LinkData * link; Kdll_LinkData * liSearch_reloc; uint8_t * kernel; int * size; int * left; int dwSize; int i; int base; bool bInline; bool res; VP_RENDER_FUNCTION_ENTER; res = false; // Check if Kernel ID is valid if (iKUID >= pKernelCache->iCacheEntries) { VP_RENDER_NORMALMESSAGE("invalid Kernel ID %d.", iKUID); goto cleanup; } // Get KDLL state pState = pSearchState->pKdllState; // Get current combined kernel kernel = pSearchState->Kernel; size = &pSearchState->KernelSize; left = &pSearchState->KernelLeft; pSymbols = &pSearchState->KernelLink; base = (*size) >> 2; // Find selected kernel and kernel size; check if there is enough space kernels = &pKernelCache->pCacheEntries[iKUID]; dwSize = kernels->iSize; if (*left < dwSize) { VP_RENDER_NORMALMESSAGE("exceeded maximum kernel size."); goto cleanup; } // Check if there is enough space for symbols if (pSymbols->dwCount + kernels->nLink >= pSymbols->dwSize) { VP_RENDER_NORMALMESSAGE("exceeded maximum numbers of symbols to resolve."); goto cleanup; } #if EMUL || VPHAL_LIB VP_RENDER_NORMALMESSAGE("%s.", kernels->szName); if (pState->pfnCbListKernel) { pState->pfnCbListKernel(pState->pToken, kernels->szName); } #elif _DEBUG // EMUL || VPHAL_LIB VP_RENDER_NORMALMESSAGE("%s.", kernels->szName); #endif // _DEBUG // Append symbols to resolve, relocate symbols link = kernels->pLink; liSearch_reloc = pSymbols->pLink + pSymbols->dwCount; bInline = false; if (link) { for (i = kernels->nLink; i > 0; i--, link++) { if (link->bInline) { // Inline code included if (!link->bExport) { bInline = true; } } else { *liSearch_reloc = *link; liSearch_reloc->dwOffset += base; liSearch_reloc++; pSymbols->dwCount++; } } } // Append kernel MOS_SecureMemcpy(&kernel[*size], dwSize, (void *)kernels->pBinary, dwSize); // Patch kernel if (pKernelPatch) { uint8_t *pSource = pKernelPatch->Data; uint8_t *pDestination = kernel + (*size); int32_t i; Kdll_PatchBlock *pBlock = pKernelPatch->Patch; for (i = pKernelPatch->nPatches; i > 0; i--, pBlock++) { MOS_SecureMemcpy(pDestination + pBlock->DstOffset, pBlock->BlockSize, (void *)(pSource + pBlock->SrcOffset), pBlock->BlockSize); } } res = true; *size += dwSize; *left -= dwSize; // Insert inline code if (bInline) { for (link = kernels->pLink, i = kernels->nLink; (i > 0) && (res); i--, link++) { if (link->bInline && (!link->bExport)) { iKUID = pKernelCache->pExports[link->iLabelID].iKUID; res &= Kdll_AppendKernel(pKernelCache, pSearchState, iKUID, pKernelPatch); } } } cleanup: return res; } //-------------------------------------------------------------- // Resolve kernel dependencies and perform patching //-------------------------------------------------------------- bool Kdll_ResolveKernelDependencies( Kdll_State * pState, Kdll_SearchState *pSearchState) { Kdll_KernelCache *cache = &pState->ComponentKernelCache; uint8_t * kernel = pSearchState->Kernel; Kdll_Symbol * pSymbols = &pSearchState->KernelLink; uint32_t nExports = cache->nExports; Kdll_LinkData * pExports = cache->pExports; Kdll_LinkData * pLink; int32_t iKUID; int32_t iOffset; uint32_t dwResolveOffset[DL_MAX_EXPORT_COUNT]; bool bResolveDone; int32_t i; uint32_t * d; VP_RENDER_FUNCTION_ENTER; MOS_ZeroMemory(dwResolveOffset, sizeof(dwResolveOffset)); do { // Update exports for (pLink = pSymbols->pLink, i = pSymbols->dwCount; i > 0; i--, pLink++) { if (pLink->bExport) { dwResolveOffset[pLink->iLabelID] = pLink->dwOffset; } } bResolveDone = true; for (pLink = pSymbols->pLink, i = pSymbols->dwCount; i > 0; i--, pLink++) { // validate label if (pLink->iLabelID > nExports || // invalid label pExports[pLink->iLabelID].bExport == 0) // label not in the export table { VP_RENDER_ASSERTMESSAGE("Invalid/unresolved label %d.", pLink->iLabelID); return false; } // load dependencies if (!pLink->bExport && !dwResolveOffset[pLink->iLabelID]) { // set flag for another pass as newly loaded // kernels may contain dependencies of their own bResolveDone = false; // Load dependencies iKUID = pExports[pLink->iLabelID].iKUID; Kdll_AppendKernel(cache, pSearchState, iKUID, nullptr); // Restart break; } } // for } while (!bResolveDone); // All modules must be loaded by now, start patching for (pLink = pSymbols->pLink, i = pSymbols->dwCount; i > 0; i--, pLink++) { iOffset = (int32_t)dwResolveOffset[pLink->iLabelID] - 4; iOffset -= pLink->dwOffset; d = ((uint32_t *)kernel) + pLink->dwOffset; // Patch offset if (!pLink->bExport && !pLink->bInline) { d[3] = iOffset << 2; // jmpi - index * 8 bits } } return true; } //--------------------------------------------------------------------------------------- // Kdll_SearchKernel - Performs full kernel search, including selection of best match // Search state must be initialized by KernelDll_StartKernelSearch // // Parameters: // Kdll_State *pState - [in] Dynamic Linking state // Kdll_SearchState *pSearchState - [in/out] Kernel search state // // Output: true if suceeded, false otherwise //--------------------------------------------------------------------------------------- bool KernelDll_SearchKernel(Kdll_State *pState, Kdll_SearchState * pSearchState) { VP_RENDER_FUNCTION_ENTER; // Check parameters if ((!pSearchState) || pSearchState->iFilterSize < 1) { VP_RENDER_NORMALMESSAGE("Search is empty, must contain 2 or more layers."); return false; } // Setup CSC; allocate and calculate CSC matrices if (!pState->pfnSetupCSC(pState, pSearchState)) { VP_RENDER_NORMALMESSAGE("CSC setup failed."); return false; } // Initial search states pSearchState->bCscBeforeMix = false; pSearchState->state = Parser_Begin; pSearchState->cspace = pSearchState->CscParams.ColorSpace; pSearchState->quadrant = 0; pSearchState->layer_number = 0; pSearchState->pMatchingRuleSet = nullptr; // Reset Src0 state pSearchState->src0_format = Format_None; pSearchState->src0_sampling = Sample_None; pSearchState->src0_colorfill = false; pSearchState->src0_lumakey = LumaKey_False; pSearchState->src0_coeff = CoeffID_None; // Reset Src1 state pSearchState->src1_format = Format_None; pSearchState->src1_sampling = Sample_None; pSearchState->src1_lumakey = LumaKey_False; pSearchState->src1_samplerlumakey = LumaKey_False; pSearchState->src1_coeff = CoeffID_None; pSearchState->src1_process = Process_None; // Search loop while (pSearchState->state != Parser_End) { #if EMUL || VPHAL_LIB if (pState->pfnCbSearchSate) { pState->pfnCbSearchSate(pState->pToken, CB_REASON_BEGIN_SEARCH, pSearchState); } #endif // Find rule that matches if (!pState->pfnFindRule(pState, pSearchState)) { #if EMUL || VPHAL_LIB if (pState->pfnCbSearchSate) { pState->pfnCbSearchSate(pState->pToken, CB_REASON_SEARCH_FAILED, pSearchState); } #endif return false; } #if EMUL || VPHAL_LIB if (pState->pfnCbSearchSate) { pState->pfnCbSearchSate(pState->pToken, CB_REASON_BEGIN_UPDATE, pSearchState); } #endif // Update state if (!pState->pfnUpdateState(pState, pSearchState)) { #if EMUL || VPHAL_LIB if (pState->pfnCbSearchSate) { pState->pfnCbSearchSate(pState->pToken, CB_REASON_UPDATE_FAILED, pSearchState); } #endif return false; } } #if EMUL || VPHAL_LIB if (pState->pfnCbSearchSate) { pState->pfnCbSearchSate(pState->pToken, CB_REASON_END_SEARCH, pSearchState); } #endif VP_RENDER_VERBOSEMESSAGE("Search completed successfully."); return true; } //-------------------------------------------------------------- // KernelDll_BuildKernel - build kernel //-------------------------------------------------------------- bool KernelDll_BuildKernel(Kdll_State *pState, Kdll_SearchState *pSearchState) { Kdll_KernelCache *pKernelCache = &pState->ComponentKernelCache; Kdll_KernelCache *pCustomCache = pState->pCustomKernelCache; Kdll_PatchData * pKernelPatch; bool res; int32_t offset = 0; int32_t * pKernelID, *pGroupID, *pPatchID; VP_RENDER_FUNCTION_ENTER; pSearchState->KernelLink.dwSize = DL_MAX_SYMBOLS; pSearchState->KernelLink.dwCount = 0; pSearchState->KernelLink.pLink = pSearchState->LinkArray; pSearchState->KernelSize = 0; pSearchState->KernelLeft = sizeof(pSearchState->Kernel); pSearchState->KernelLink.dwCount = 0; #if EMUL || VPHAL_LIB || _DEBUG || _RELEASE_INTERNAL VP_RENDER_NORMALMESSAGE("Component Kernels:"); #endif // EMUL || VPHAL_LIB || _DEBUG pKernelID = pSearchState->KernelID; pGroupID = pSearchState->KernelGrp; pPatchID = pSearchState->PatchID; for (offset = 0; offset < pSearchState->KernelCount; offset++, pKernelID++, pGroupID++, pPatchID++) { // Get patch information associated with the kernel pKernelPatch = (*pPatchID >= 0) ? &(pSearchState->Patches[*pPatchID]) : nullptr; // Append/Patch kernel from custom cache if (*pGroupID == GROUP_CUSTOM) { res = Kdll_AppendKernel(pCustomCache, pSearchState, *pKernelID, pKernelPatch); } // Append/Patch kernel from internal cache else { res = Kdll_AppendKernel(pKernelCache, pSearchState, *pKernelID, pKernelPatch); } if (!res) { VP_RENDER_ASSERTMESSAGE("Failed to build kernel ID %d.", pSearchState->KernelID[offset]); return false; } else { Kdll_CacheEntry *kernels = (*pGroupID == GROUP_CUSTOM) ? &pCustomCache->pCacheEntries[*pKernelID] : &pKernelCache->pCacheEntries[*pKernelID]; VP_RENDER_NORMALMESSAGE("Component kernels [%d]: %s", *pKernelID, kernels->szName); } } // Resolve kernel dependencies res = Kdll_ResolveKernelDependencies(pState, pSearchState); if (!res) { VP_RENDER_ASSERTMESSAGE("Failed to resolve symbols."); return false; } return true; } //--------------------------------------------------------------------------------------- // KernelDll_StartKernelSearch - Starts kernel search // // Parameters: // Kdll_State *pState - [in] Dynamic Linking State // Kdll_FilterEntry *pFilter - [in] Search filter (array of search entries) // int iFilterSize - [in] Search filter size // Kdll_SearchState *pSearchState - [in/out] Kernel search state // // Output: none //--------------------------------------------------------------------------------------- void KernelDll_StartKernelSearch( Kdll_State * pState, Kdll_SearchState *pSearchState, Kdll_FilterEntry *pFilter, int32_t iFilterSize, uint32_t uiIs64BInstrEnabled) { int32_t nLayer; VP_RENDER_FUNCTION_ENTER; // Reset all states MOS_ZeroMemory(pSearchState, sizeof(Kdll_SearchState)); // Setup KDLL state pSearchState->pKdllState = pState; // KDLL state // Cleanup kernel table pSearchState->KernelCount = 0; // # of kernels // Cleanup patch data memset(pSearchState->Patches, 0, sizeof(pSearchState->Patches)); memset(pSearchState->PatchID, -1, sizeof(pSearchState->PatchID)); pSearchState->PatchCount = 0; // Copy original filter; filter will be modified as part of the search if (pFilter && iFilterSize > 0) { MOS_SecureMemcpy(pSearchState->Filter, iFilterSize * sizeof(Kdll_FilterEntry), pFilter, iFilterSize * sizeof(Kdll_FilterEntry)); pSearchState->pFilter = pSearchState->Filter; pSearchState->iFilterSize = iFilterSize; for (nLayer = 0; nLayer < iFilterSize; nLayer++) { // DScale Kernels are enabled for all gen9 stepping. //For Gen9+, kernel don't support sublayer DScale+rotation //Sampler_unorm does not support Y410/RGB10, we need to use sampler_16 to support Y410/RGB10 if (!pFilter[nLayer].bEnableDscale && (!pFilter[nLayer].bWaEnableDscale || (pFilter[nLayer].layer == Layer_SubVideo && pFilter[nLayer].rotation != MEDIA_ROTATION(VPHAL_ROTATION_IDENTITY)))) { if (pFilter[nLayer].sampler == Sample_Scaling_034x) { pSearchState->pFilter[nLayer].sampler = Sample_Scaling; } else if (pFilter[nLayer].sampler == Sample_iScaling_034x) { pSearchState->pFilter[nLayer].sampler = Sample_iScaling; } else if (pFilter[nLayer].sampler == Sample_iScaling_AVS) { pSearchState->pFilter[nLayer].sampler = Sample_iScaling_AVS; } } } // Copy the render target format pSearchState->target_format = pSearchState->pFilter[iFilterSize - 1].format; // Copy the render target tile type pSearchState->target_tiletype = pSearchState->pFilter[iFilterSize - 1].tiletype; // Indicate whether to use 64B save kernel for render target surface if (uiIs64BInstrEnabled && ((pSearchState->target_tiletype == MOS_TILE_X) || (pSearchState->target_tiletype == MOS_TILE_LINEAR))) { pSearchState->b64BSaveEnabled = true; } } } /*---------------------------------------------------------------------------- | Name : KernelDll_SetupCSC | Purpose : Defines CSC conversions necessary for a given filter | | Input : pState - Kernel Dll state | pSearchState - current DL search state | | Return : \---------------------------------------------------------------------------*/ bool KernelDll_SetupCSC( Kdll_State * pState, Kdll_SearchState *pSearchState) { int i, m; // Integer iterators bool bCoeffID_0_Used = false; VPHAL_CSPACE cspace = CSpace_None; // Current ColorSpace VPHAL_CSPACE out_cspace = CSpace_None; // Render Target CS VPHAL_CSPACE main_cspace = CSpace_None; // Main video CS VPHAL_CSPACE sel_cspace = CSpace_Any; // Selected CS Kdll_FilterEntry *pFilter; // Current Filter information int iFilterSize = pSearchState->iFilterSize; Kdll_CSC_Params * pCSC = &pSearchState->CscParams; int csc_count; // Number of CSC operations int matrix_count; // Number of Matrices in use int procamp_count = 0; // Number of PA operations int sel_csc_count = -1; // Minimum number of CSC operations int iCoeffID; // coeffID for layers other than main video uint8_t cspace_in_use[CSpace_Count]; // Color Spaces in use Kdll_CSC_Matrix curr_matrix; Kdll_CSC_Matrix *matrix = pCSC->Matrix; // Color Space conversion matrix uint8_t * matrixID = pCSC->MatrixID; // CSC coefficient allocation table bool forceToTargetColorSpace = false; // Clear all CSC matrices MOS_ZeroMemory(matrix, sizeof(pCSC->Matrix)); memset(matrixID, DL_CSC_DISABLED, sizeof(pCSC->MatrixID)); memset(pCSC->PatchMatrixID, DL_CSC_DISABLED, sizeof(pCSC->PatchMatrixID)); pCSC->PatchMatrixNum = 0; // Clear array of color spaces in use MOS_ZeroMemory(cspace_in_use, sizeof(cspace_in_use)); //---------------------------------------------------------------// // Collect information about Color Spaces in use // Get Primary Video and Render Target Color Spaces // Force xvYCC passthrough if enabled //---------------------------------------------------------------// for (i = iFilterSize, pFilter = pSearchState->Filter; i > 0; i--, pFilter++) { if (pFilter->forceToTargetColorSpace) { forceToTargetColorSpace = true; } // Disable Procamp for all layers except Main Video // Disable Procamp if source is RGB if (pFilter->layer != Layer_MainVideo || pFilter->cspace == CSpace_sRGB || pFilter->cspace == CSpace_stRGB) { pFilter->procamp = DL_PROCAMP_DISABLED; } // Count number of procamp operations (limited by number of independent coefficients) // Ignore layers with palletized/constant colors if (pFilter->procamp != DL_PROCAMP_DISABLED && pFilter->cspace != CSpace_Any) { procamp_count++; } // Set xvYCC passthrough mode if (pFilter->cspace == CSpace_xvYCC709 || pFilter->cspace == CSpace_xvYCC601) { sel_cspace = pFilter->cspace; } // Get Main Video color space if (pFilter->layer == Layer_MainVideo) { main_cspace = pFilter->cspace; } // Get Render Target color space if (pFilter->layer == Layer_RenderTarget) { // Target is sRGB/stRGB if (!KernelDll_IsYUVFormat(pFilter->format)) { // Disable xvYCC passthrough (sRGB cannot have extended gamut) sel_cspace = CSpace_Any; } out_cspace = pFilter->cspace; } // Mark color spaces in use for search that follows if (pFilter->cspace > CSpace_Any && pFilter->cspace < CSpace_Count) { cspace_in_use[pFilter->cspace] = 1; } } // Check max number of procamp operations if (procamp_count > DL_PROCAMP_MAX) { return false; } //---------------------------------------------------------------// // Search Color Space that provides minimum number of CSC conversions // If there are multiple solutions, select main video cspace (quality) //---------------------------------------------------------------// if (sel_cspace == CSpace_Any) { if (forceToTargetColorSpace) { sel_cspace = out_cspace; } else { int cs; for (cs = (CSpace_Any + 1); cs < CSpace_Count; cs++) { // Skip color spaces not in use cspace = (VPHAL_CSPACE)cs; if (!cspace_in_use[cspace]) { continue; } // xvYCC and BT are treated as same for CSC considerations (BT.x to xvYCC.x matrix is I) cspace = KernelDll_TranslateCspace(cspace); // Count # of CS conversions and matrices csc_count = 0; for (i = iFilterSize, pFilter = pSearchState->Filter; i > 0; i--, pFilter++) { // Ignore layers where the Color Space may be set in software (colorfill, palletized) if (pFilter->cspace == CSpace_Any) { continue; } // Check if CSC/PA is required if (KernelDll_TranslateCspace(pFilter->cspace) != cspace || pFilter->procamp != DL_PROCAMP_DISABLED) { csc_count++; } } // Save best choice as requiring minimum number of CSC operations if ((sel_csc_count < 0) || // Initial value (csc_count < sel_csc_count) || // Minimum number of CSC operations (csc_count == sel_csc_count && cs == main_cspace)) // Use main cspace as default if same CSC count { sel_cspace = cspace; sel_csc_count = csc_count; } } } } // Due to put the colorfill behind CSC, so Src0 cspace needs to change // to selspace in order to fill colorfill values correctly. pState->colorfill_cspace = sel_cspace; // color space is selected by now... setup CSC matrices matrix_count = 0; iCoeffID = 1; for (i = iFilterSize, pFilter = pSearchState->Filter; i > 0; i--, pFilter++) { // Setup CSC for palettized/colorfill layers if (pFilter->cspace == CSpace_Any) { // Set Color Space and format (for software) if (pFilter->format == Format_Any) { pFilter->format = KernelDll_IsCspace(sel_cspace, CSpace_YUV) ? Format_AYUV : Format_A8R8G8B8; } pFilter->cspace = sel_cspace; pFilter->matrix = DL_CSC_DISABLED; } else { // Setup CSC parameters: SrcSpace is the layer color space, // DstSpace is the internal color space selected curr_matrix.SrcSpace = KernelDll_TranslateCspace(pFilter->cspace); curr_matrix.DstSpace = KernelDll_TranslateCspace(sel_cspace); curr_matrix.iProcampID = pFilter->procamp; // Check if CSC is necessary if (curr_matrix.SrcSpace == curr_matrix.DstSpace && curr_matrix.iProcampID == DL_PROCAMP_DISABLED) { pFilter->matrix = DL_CSC_DISABLED; curr_matrix.iCoeffID = CoeffID_None; continue; } // Reserve CoeffID_0 for main video - CoeffID_0 gets CSC coeff from static parameters // If main video doesn't use CoeffID_0, assign to RT if ((pFilter->layer == Layer_MainVideo) || (pFilter->layer == Layer_RenderTarget)) { if (bCoeffID_0_Used) { curr_matrix.iCoeffID = (Kdll_CoeffID)iCoeffID++; } else { curr_matrix.iCoeffID = CoeffID_0; bCoeffID_0_Used = true; } } else { curr_matrix.iCoeffID = (Kdll_CoeffID)iCoeffID++; } // CSC at the target layer is from internal cspace (SrcSpace) // to external cspace (DstCspace) if (pFilter->layer == Layer_RenderTarget) { VPHAL_CSPACE aux = curr_matrix.SrcSpace; curr_matrix.SrcSpace = curr_matrix.DstSpace; curr_matrix.DstSpace = aux; } // Search CSC matrix - avoid duplicated CSC matrices for (m = 0; m < matrix_count; m++) { if (curr_matrix.SrcSpace == matrix[m].SrcSpace && curr_matrix.DstSpace == matrix[m].DstSpace && curr_matrix.iProcampID == matrix[m].iProcampID) { break; } } // Check limit if (m == matrix_count) { // Exceeded number of CSC matrices allowed if (matrix_count == DL_CSC_MAX) { VP_RENDER_ASSERTMESSAGE("CSC matrix count %d exceeded number of CSC matrices allowed!", matrix_count); return false; } matrix[m].bInUse = true; matrix[m].SrcSpace = curr_matrix.SrcSpace; matrix[m].DstSpace = curr_matrix.DstSpace; matrix[m].iProcampID = curr_matrix.iProcampID; matrix[m].iCoeffID = curr_matrix.iCoeffID; // Calculate coefficients for the first time KernelDll_UpdateCscCoefficients(pState, &matrix[m]); // Next matrix matrix_count++; } // point to the matrix pFilter->matrix = m; } } // Link matrices to kernel coefficients (and vice-versa) matrix = pCSC->Matrix; for (m = 0; m < matrix_count; m++, matrix++) { // Coefficient table points to matrix index matrixID[matrix->iCoeffID] = (uint8_t)m; } // Save selected color space pCSC->ColorSpace = sel_cspace; return true; } /*---------------------------------------------------------------------------- | Name : KernelDll_GetPatchData | Purpose : Get binary data block to be used for kernel patching | | Input : pState - [in] Current DL state | pSearchState - [in] Current DL search state | iPatchKind - [in] Patch kind | pSize - [out] Data block Size | | Return : nullptr - Unsupported patch data kind | <>nullptr - Pointer to data block \---------------------------------------------------------------------------*/ static uint8_t *KernelDll_GetPatchData( Kdll_State * pState, Kdll_SearchState *pSearchState, int32_t iPatchKind, int32_t * pSize) { MOS_UNUSED(pState); VP_RENDER_FUNCTION_ENTER; if (iPatchKind == PatchKind_CSC_Coeff_Src0 || iPatchKind == PatchKind_CSC_Coeff_Src1) { Kdll_CoeffID coeffID = CoeffID_None; uint8_t matrixID = DL_CSC_DISABLED; // Get matrix id if (iPatchKind == PatchKind_CSC_Coeff_Src0) { coeffID = pSearchState->src0_coeff; } else { coeffID = pSearchState->src1_coeff; } // Get matrix associated with the coefficient ID if (coeffID > CoeffID_None) { matrixID = pSearchState->CscParams.MatrixID[coeffID]; } // Found matrix if (matrixID < DL_CSC_MAX) { Kdll_CSC_Matrix *pMatrix = &(pSearchState->CscParams.Matrix[matrixID]); *pSize = 12 * sizeof(uint16_t); if (pState->bEnableCMFC) { if (pSearchState->CscParams.PatchMatrixNum < DL_CSC_MAX) { pSearchState->CscParams.PatchMatrixID[pSearchState->CscParams.PatchMatrixNum] = matrixID; pSearchState->CscParams.PatchMatrixNum++; } else { VP_RENDER_ASSERTMESSAGE("Patch CSC coefficient number %d exceed limitation %d!", pSearchState->CscParams.PatchMatrixNum, DL_CSC_MAX); return nullptr; } } return ((uint8_t *)pMatrix->Coeff); } } else { VP_RENDER_ASSERTMESSAGE("Invalid patch kind %d.", iPatchKind); } return nullptr; } /*---------------------------------------------------------------------------- | Name : KernelDll_UpdateState | Purpose : Update search state using current matching rule | | Input : pState - Kernel Dll state | pSearchState - current DL search state | | Return : \---------------------------------------------------------------------------*/ bool KernelDll_UpdateState( Kdll_State * pState, Kdll_SearchState *pSearchState) { Kdll_RuleEntrySet * pRuleSet = pSearchState->pMatchingRuleSet; const Kdll_RuleEntry *pRuleEntry; int32_t iSetCount; VP_RENDER_FUNCTION_ENTER; // Ensures that we have a matching rule if (pRuleSet == nullptr) { return false; } // Get rule entry and number of state update ("Set") rules; validate pRuleEntry = pRuleSet->pRuleEntry; iSetCount = pRuleSet->iSetCount; if (pRuleEntry == nullptr || iSetCount < 1) { VP_RENDER_NORMALMESSAGE("Invalid rule set."); return false; } // Jump to set rules (skip match rules) pRuleEntry += pRuleSet->iMatchCount; // Apply state update rules for (; iSetCount > 0; iSetCount--, pRuleEntry++) { switch (pRuleEntry->id) { // Add kernel to the Dynamic Linking array case RID_SetKernel: if (pSearchState->KernelCount < DL_MAX_KERNELS) { int32_t i = pSearchState->KernelCount++; pSearchState->KernelID[i] = pRuleEntry->value; pSearchState->KernelGrp[i] = pRuleSet->iGroup; // Group associated with the kernel ID } else { VP_RENDER_ASSERTMESSAGE("reached maximum number of component kernels."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } break; // Set Parser State case RID_SetParserState: pSearchState->state = (Kdll_ParserState)pRuleEntry->value; break; // Move to Next/Prev Layer case RID_SetNextLayer: if (pRuleEntry->value == -1) { pSearchState->layer_number--; pSearchState->pFilter--; } else if (pRuleEntry->value == -2) // jump to layer main video { do { pSearchState->layer_number--; pSearchState->pFilter--; if (pSearchState->pFilter == nullptr || pSearchState->layer_number < 0) { return false; } } while (pSearchState->pFilter->layer != Layer_MainVideo); } else if (pRuleEntry->value == 2) // jump to target layer { while (pSearchState->pFilter->layer < Layer_RenderTarget) { pSearchState->layer_number++; pSearchState->pFilter++; } } else { pSearchState->layer_number++; pSearchState->pFilter++; } break; // Set patch data case RID_SetPatchData: { uint8_t * pData = nullptr; int32_t iSize = 0; int32_t iKernelIndex = pSearchState->KernelCount - 1; int32_t iPatchIndex; Kdll_PatchData *pPatch; // Get block of data for patching pData = KernelDll_GetPatchData(pState, pSearchState, (Kdll_PatchKind)pRuleEntry->value, &iSize); if (pData == nullptr || iSize == 0) { VP_RENDER_ASSERTMESSAGE("invalid patch."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } // Append to the existing patch data block iPatchIndex = pSearchState->PatchID[iKernelIndex]; // Allocate new patch structure if (iPatchIndex < 0) { // Fail to allocate if (pSearchState->PatchCount >= DL_MAX_PATCHES) { VP_RENDER_ASSERTMESSAGE("reached maximum number of patches."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } // Get new patch block iPatchIndex = pSearchState->PatchCount++; pSearchState->PatchID[iKernelIndex] = iPatchIndex; // Reset new patch entry pPatch = &(pSearchState->Patches[iPatchIndex]); MOS_ZeroMemory(pPatch, sizeof(Kdll_PatchData)); } else { // Get Patch entry already in use pPatch = &(pSearchState->Patches[iPatchIndex]); } // Check if data can be appended if (pPatch->iPatchDataSize + iSize > DL_MAX_PATCH_DATA_SIZE) { VP_RENDER_ASSERTMESSAGE("exceeded maximum patch size."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } // Append patch data MOS_SecureMemcpy(pPatch->Data + pPatch->iPatchDataSize, iSize, (void *)pData, iSize); pPatch->iPatchDataSize += iSize; } break; // Set patch operation case RID_SetPatch: { int32_t iKernelIndex = pSearchState->KernelCount - 1; int32_t iPatchIndex = pSearchState->PatchID[iKernelIndex]; Kdll_PatchData * pPatch; uint8_t * pPatchRule; Kdll_PatchBlock *pPatchBlock; int32_t nPatches; // No patch associated with the current kernel if (iPatchIndex < 0) { return false; } // Get Patch entry pPatch = &(pSearchState->Patches[iPatchIndex]); // Get number of patches and pointer to first rule extension (patch rule) nPatches = pRuleEntry->value; pPatchRule = (uint8_t *)(pRuleEntry + 1); // Check if rules can be applied if (nPatches + pPatch->nPatches > DL_MAX_PATCH_BLOCKS) { VP_RENDER_ASSERTMESSAGE("exceeded number of patch blocks."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } // Set Patches pPatchBlock = &(pPatch->Patch[pPatch->nPatches]); for (; nPatches > 0; nPatches--, pPatchBlock++, pPatch->nPatches++) { pPatchBlock->BlockSize = ((Kdll_PatchRuleEntry *)pPatchRule)->Size; pPatchBlock->SrcOffset = ((Kdll_PatchRuleEntry *)pPatchRule)->Source; pPatchBlock->DstOffset = ((Kdll_PatchRuleEntry *)pPatchRule)->Dest; pPatchRule += sizeof(Kdll_RuleEntry); } // Skip rule extensions iSetCount -= pRuleEntry->value; pRuleEntry += pRuleEntry->value; } break; // Set destination colorspace case RID_SetTargetCspace: if ((VPHAL_CSPACE)pRuleEntry->value == CSpace_Source) { pSearchState->cspace = pSearchState->pFilter->cspace; } else { pSearchState->cspace = (VPHAL_CSPACE)pRuleEntry->value; } break; // Set Src0 source format case RID_SetSrc0Format: if ((MOS_FORMAT)pRuleEntry->value == Format_Source) { pSearchState->src0_format = pSearchState->pFilter->format; } else { pSearchState->src0_format = (MOS_FORMAT)pRuleEntry->value; } break; // Set Src0 sampling mode case RID_SetSrc0Sampling: if ((Kdll_Sampling)pRuleEntry->value == Sample_Source) { pSearchState->src0_sampling = pSearchState->pFilter->sampler; } else { pSearchState->src0_sampling = (Kdll_Sampling)pRuleEntry->value; } break; // Set Src0 Rotation case RID_SetSrc0Rotation: pSearchState->src0_rotation = pSearchState->pFilter->rotation; break; // Set Src0 Colorfill case RID_SetSrc0ColorFill: if ((int32_t)pRuleEntry->value == ColorFill_Source) { pSearchState->src0_colorfill = pSearchState->pFilter->colorfill; } else { pSearchState->src0_colorfill = (int32_t)pRuleEntry->value; } break; // Set Src0 luma key case RID_SetSrc0LumaKey: if (pRuleEntry->value == LumaKey_Source) { pSearchState->src0_lumakey = pSearchState->pFilter->lumakey; } else { pSearchState->src0_lumakey = (int32_t)pRuleEntry->value; } break; // Set Src0 Procamp case RID_SetSrc0Procamp: if (pRuleEntry->value == Procamp_Source) { pSearchState->src0_procamp = pSearchState->pFilter->procamp; } else { pSearchState->src0_procamp = (int32_t)pRuleEntry->value; } break; // Set Src0 CSC coefficients case RID_SetSrc0Coeff: if ((Kdll_CoeffID)pRuleEntry->value == CoeffID_Source) { if (pSearchState->pFilter->matrix == DL_CSC_DISABLED) { pSearchState->src0_coeff = CoeffID_None; } else { Kdll_CSC_Matrix *matrix = pSearchState->CscParams.Matrix; matrix += pSearchState->pFilter->matrix; pSearchState->src0_coeff = matrix->iCoeffID; } } else { pSearchState->src0_coeff = (Kdll_CoeffID)pRuleEntry->value; } break; case RID_SetSrc0Processing: if ((Kdll_Processing)pRuleEntry->value == Process_Source) { pSearchState->src0_process = pSearchState->pFilter->process; } else { pSearchState->src0_process = (Kdll_Processing)pRuleEntry->value; } break; // Set Src1 source format case RID_SetSrc1Format: if ((MOS_FORMAT)pRuleEntry->value == Format_Source) { pSearchState->src1_format = pSearchState->pFilter->format; } else { pSearchState->src1_format = (MOS_FORMAT)pRuleEntry->value; } break; // Set Src1 sampling mode case RID_SetSrc1Sampling: if ((Kdll_Sampling)pRuleEntry->value == Sample_Source) { pSearchState->src1_sampling = pSearchState->pFilter->sampler; } else { pSearchState->src1_sampling = (Kdll_Sampling)pRuleEntry->value; } break; // Set Src1 Rotation case RID_SetSrc1Rotation: pSearchState->src1_rotation = pSearchState->pFilter->rotation; break; // Set Src1 luma key case RID_SetSrc1LumaKey: if (pRuleEntry->value == LumaKey_Source) { pSearchState->src1_lumakey = pSearchState->pFilter->lumakey; } else { pSearchState->src1_lumakey = (int32_t)pRuleEntry->value; } break; // Set Src1 Sampler LumaKey case RID_SetSrc1SamplerLumaKey: if (pRuleEntry->value == LumaKey_Source) { pSearchState->src1_samplerlumakey = pSearchState->pFilter->samplerlumakey; } else { pSearchState->src1_samplerlumakey = (int32_t)pRuleEntry->value; } break; // Set Src1 Procamp case RID_SetSrc1Procamp: if (pRuleEntry->value == Procamp_Source) { pSearchState->src1_procamp = pSearchState->pFilter->procamp; } else { pSearchState->src1_procamp = (int32_t)pRuleEntry->value; } break; // Set Src1 CSC coefficients case RID_SetSrc1Coeff: if ((Kdll_CoeffID)pRuleEntry->value == CoeffID_Source) { if (pSearchState->pFilter->matrix == DL_CSC_DISABLED) { pSearchState->src1_coeff = CoeffID_None; } else { Kdll_CSC_Matrix *matrix = pSearchState->CscParams.Matrix; matrix += pSearchState->pFilter->matrix; pSearchState->src1_coeff = matrix->iCoeffID; } } else { pSearchState->src1_coeff = (Kdll_CoeffID)pRuleEntry->value; } break; // Set Src1 processing mode case RID_SetSrc1Processing: if ((Kdll_Processing)pRuleEntry->value == Process_Source) { pSearchState->src1_process = pSearchState->pFilter->process; } else { pSearchState->src1_process = (Kdll_Processing)pRuleEntry->value; } break; // Set current quadrant case RID_SetQuadrant: pSearchState->quadrant = (int32_t)pRuleEntry->value; break; // Set CSC flag before Mix case RID_SetCSCBeforeMix: pSearchState->bCscBeforeMix = pRuleEntry->value ? true : false; break; // Unsupported "Set" rule default: // Failed to find a matching rule -> kernel search will fail VP_RENDER_ASSERTMESSAGE("Invalid rule %d @ layer %d, state %d.", pRuleEntry->id, pSearchState->layer_number, pSearchState->state); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } } // Reset matching rule pSearchState->pMatchingRuleSet = nullptr; return true; } //----------------------------------------------------------------------------------------- // KernelDll_SortRuleTable - Sort master dynamic linking rule table // // Parameters: // char *pState - [in] Kernel Dll state // // Output: true - Master rule table (and acceleration table) successfully created // false - Failed to setup master rule table //----------------------------------------------------------------------------------------- bool KernelDll_SortRuleTable(Kdll_State *pState) { uint8_t group; int32_t state; const Kdll_RuleEntry *pRule = nullptr; Kdll_RuleEntrySet * pRuleSet; int32_t i, j; int32_t iTotal = 0; int32_t iNoOverr[Parser_Count]; // Non-overridable (enforced) rules int32_t iDefault[Parser_Count]; // Default rules int32_t iCustom[Parser_Count]; // Custom rules VP_RENDER_FUNCTION_ENTER; // Release previous table (rule table update) if (pState->pSortedRules) { MOS_FreeMemory(pState->pSortedRules); pState->pSortedRules = nullptr; MOS_ZeroMemory(pState->pDllRuleTable, sizeof(pState->pDllRuleTable)); MOS_ZeroMemory(pState->iDllRuleCount, sizeof(pState->iDllRuleCount)); } // Zero counters MOS_ZeroMemory(iNoOverr, sizeof(iNoOverr)); MOS_ZeroMemory(iDefault, sizeof(iDefault)); MOS_ZeroMemory(iCustom, sizeof(iCustom)); // Count number of entries for each state for (i = 0; i < 2; i++) { if (i == 0) { pRule = pState->pRuleTableDefault; } else if (i == 1) { pRule = pState->pRuleTableCustom; } // Table not set - continue if (!pRule) continue; for (; pRule->id != RID_Op_EOF; pRule++) { // Skip extended rules (variable lenght) if (RID_IS_EXTENDED(pRule->id)) { // value contains number of entries pRule += pRule->value; } else if (pRule->id == RID_Op_NewEntry) { // Save Rule Group if (i == 0) { group = pRule->value; } else { group = RULE_CUSTOM; } // Second rule must always be RID_IsParserState pRule++; if (pRule->id != RID_IsParserState) { VP_RENDER_ASSERTMESSAGE("Rule does not start with State."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } // Get Parser State -> validate value state = pRule->value; if (state < Parser_Begin) { VP_RENDER_ASSERTMESSAGE("Invalid State %d.", state); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } else if (state >= Parser_Custom) { // Custom states are set together in the same entry state = Parser_Custom; } if (group == RULE_NO_OVERRIDE) { iNoOverr[state]++; } else if (group == RULE_DEFAULT) { iDefault[state]++; } else { iCustom[state]++; } iTotal++; } } } // Allocate rules pState->pSortedRules = (Kdll_RuleEntrySet *)MOS_AllocAndZeroMemory(iTotal * sizeof(Kdll_RuleEntrySet)); if (!pState->pSortedRules) { VP_RENDER_ASSERTMESSAGE("Failed to allocate rule table."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } // Setup pointers to sorted rules pState->pDllRuleTable[0] = pState->pSortedRules; for (j = 0, i = 0; i < Parser_Count; i++) { // Setup start pointer and number of entries to search for each state pState->pDllRuleTable[i] = pState->pDllRuleTable[j] + pState->iDllRuleCount[j]; pState->iDllRuleCount[i] = iNoOverr[i] + iCustom[i] + iDefault[i]; // Setup offsets to rules for sorting iDefault[i] = iNoOverr[i] + iCustom[i]; // Last set of rules iCustom[i] = iNoOverr[i]; // 2nd set of rules iNoOverr[i] = 0; // 1st set of rules j = i; } // Sort rules for fast access // Integrate enforced, custom, default rules into one single access table for (i = 0; i < 2; i++) { if (i == 0) { pRule = pState->pRuleTableDefault; } else if (i == 1) { pRule = pState->pRuleTableCustom; } // Table not set - continue if (!pRule) continue; while (pRule->id != RID_Op_EOF) { if (pRule->id != RID_Op_NewEntry) { VP_RENDER_ASSERTMESSAGE("New rule entry expected."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } // Save Rule Group if (i == 0) { group = pRule->value; } else { group = RULE_CUSTOM; } // Get Parser State -> validate value pRule++; state = pRule->value; if (state >= Parser_Custom) { // Custom states are set together in the same entry state = Parser_Custom; } else { // Skip state check - already handled by acceleration table pRule++; } // Point to sorted rule set entry if (group == RULE_NO_OVERRIDE) { j = iNoOverr[state]++; } else if (group == RULE_DEFAULT) { j = iDefault[state]++; } else { j = iCustom[state]++; } // Point to sorted ruleset for the current parser state pRuleSet = pState->pDllRuleTable[state] + j; // Fill RuleSet pRuleSet->pRuleEntry = pRule; pRuleSet->iGroup = group; // Count number of match rules, including extended rules while (RID_IS_MATCH(pRule->id)) { if (RID_IS_EXTENDED(pRule->id)) { pRuleSet->iMatchCount += pRule->value; pRule += pRule->value; } pRuleSet->iMatchCount++; pRule++; } // Count number of set rules, including extended rules while (RID_IS_SET(pRule->id)) { if (RID_IS_EXTENDED(pRule->id)) { pRuleSet->iSetCount += pRule->value; pRule += pRule->value; } pRuleSet->iSetCount++; pRule++; } // Rule must have at least one "Set" rule if (pRuleSet->iSetCount < 1) { VP_RENDER_ASSERTMESSAGE("Ruleset must have at least one set rule."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); return false; } } } // Rule table is now sorted and integrated with custom rules return true; } //--------------------------------------------------------------------------------------- // KernelDll_AllocateStates - Allocate Kernel Dynamic Linking/Loading (Dll) States // // - Allocate DL states // - Setup export/import list for linking // - Prepare pool of search nodes // - Load component kernels from binary file // - Setup kernel cache // - Setup kernel dynamic linking rules // // Parameters: [in] pKernelBin - Pointer to Kernel binary file loaded in sys memory // [in] uKernelSize - Kernel file size // [in] pFcPatchBin - Pointer to FC patch binary file loaded in sys memory // [in] uFcPatchCacheSize - FC patch binary file size // [in] platform - Gfx platform // [in] pDefaultRules - Dynamic Linking Rules Table // // Output: Pointer to allocated Kernel dll state // nullptr - Failed to allocate Kernel dll state //----------------------------------------------------------------------------------------- Kdll_State *KernelDll_AllocateStates( void * pKernelBin, uint32_t uKernelSize, void * pFcPatchCache, uint32_t uFcPatchCacheSize, const Kdll_RuleEntry *pDefaultRules, void (*ModifyFunctionPointers)(PKdll_State)) { Kdll_State * pState; Kdll_CacheEntry * pCacheEntry; Kdll_KernelCache * pKernelCache; Kdll_KernelHashTable *pHashTable; Kdll_KernelHashEntry *pHashEntries; int32_t iSize; int32_t nExports = 0; int32_t nImports = 0; uint32_t * pLinkOffset = nullptr; Kdll_LinkData * pLinkSort = nullptr; Kdll_LinkData * pLinkData; Kdll_LinkData * pExports; Kdll_LinkFileHeader *pLinkHeader; int32_t i, j; uint32_t *pOffsets; uint8_t * pBase; VP_RENDER_FUNCTION_ENTER; // Allocate dynamic linking states i = sizeof(Kdll_State); // Dynamic linking states i += sizeof(Kdll_CacheEntry) * IDR_VP_TOTAL_NUM_KERNELS; // Component kernel cache entries i += sizeof(Kdll_CacheEntry) * IDR_VP_TOTAL_NUM_KERNELS; // CMFC kernel patch cache entries i += sizeof(Kdll_CacheEntry) * DL_DEFAULT_COMBINED_KERNELS; // Combined kernel cache entries i += DL_COMBINED_KERNEL_CACHE_SIZE; // Combined kernel buffer i += sizeof(Kdll_LinkData) * DL_MAX_EXPORT_COUNT; // Kernel Export table pState = (Kdll_State *)MOS_AllocAndZeroMemory(i); if (!pState) { VP_RENDER_ASSERTMESSAGE("Failed to allocate kernel dll states."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); goto cleanup; } pState->iSize = i; pState->dwRefresh = 0; pState->pProcamp = nullptr; pState->iProcampSize = 0; pState->pSortedRules = nullptr; if ((pFcPatchCache != nullptr) && (uFcPatchCacheSize != 0)) { pState->bEnableCMFC = true; } // Initialize platform specific function pointers if (!KernelDll_SetupFunctionPointers(pState, ModifyFunctionPointers)) { VP_RENDER_ASSERTMESSAGE("Failed to setup function pointers."); MT_ERR1(MT_VP_KERNEL_RULE, MT_CODE_LINE, __LINE__); goto cleanup; } pKernelCache = &pState->ComponentKernelCache; // No custom kernels/rules pState->pRuleTableCustom = nullptr; pState->pCustomKernelCache = nullptr; // Set Kernel DLL Rules pState->pRuleTableDefault = pDefaultRules; // Integrate and sort rule tables KernelDll_SortRuleTable(pState); // Setup component kernel cache pKernelCache->pCache = (uint8_t *)pKernelBin; pKernelCache->iCacheSize = (int32_t)uKernelSize; pKernelCache->iCacheFree = 0; pKernelCache->iCacheMaxEntries = IDR_VP_TOTAL_NUM_KERNELS; pKernelCache->iCacheEntries = IDR_VP_TOTAL_NUM_KERNELS; pKernelCache->pCacheEntries = (Kdll_CacheEntry *)(pState + 1); pOffsets = (uint32_t *)pKernelCache->pCache; pBase = (uint8_t *)(pOffsets + IDR_VP_TOTAL_NUM_KERNELS + 1); pCacheEntry = pKernelCache->pCacheEntries; for (i = 0; i < IDR_VP_TOTAL_NUM_KERNELS; i++, pCacheEntry++) { pCacheEntry->iKUID = i; pCacheEntry->iKCID = -1; pCacheEntry->dwLoaded = 0; pCacheEntry->dwRefresh = 0; pCacheEntry->wHashEntry = 0; pCacheEntry->szName = g_cInit_ComponentNames[i]; pCacheEntry->iSize = pOffsets[i + 1] - pOffsets[i]; pCacheEntry->pBinary = (pCacheEntry->iSize > 0) ? (pBase + pOffsets[i]) : nullptr; } // Setup CMFC kernel patch cache pKernelCache = &pState->CmFcPatchCache; if (pState->bEnableCMFC && pFcPatchCache) { pKernelCache->pCache = (uint8_t *)pFcPatchCache; pKernelCache->iCacheSize = (int32_t)uFcPatchCacheSize; pKernelCache->iCacheFree = 0; pKernelCache->iCacheMaxEntries = IDR_VP_TOTAL_NUM_KERNELS; pKernelCache->iCacheEntries = IDR_VP_TOTAL_NUM_KERNELS; pKernelCache->pCacheEntries = pCacheEntry; pOffsets = (uint32_t *)pKernelCache->pCache; pBase = (uint8_t *)(pOffsets + IDR_VP_TOTAL_NUM_KERNELS + 1); for (i = 0; i < IDR_VP_TOTAL_NUM_KERNELS; i++, pCacheEntry++) { pCacheEntry->iKUID = i; pCacheEntry->iKCID = -1; pCacheEntry->dwLoaded = 0; pCacheEntry->dwRefresh = 0; pCacheEntry->wHashEntry = 0; pCacheEntry->szName = g_cInit_ComponentNames[i]; pCacheEntry->iSize = pOffsets[i + 1] - pOffsets[i]; pCacheEntry->pBinary = (pCacheEntry->iSize > 0) ? (pBase + pOffsets[i]) : nullptr; } } else { pCacheEntry += IDR_VP_TOTAL_NUM_KERNELS; } // Setup combined kernel cache pKernelCache = &pState->KernelCache; pKernelCache->iCacheMaxEntries = DL_DEFAULT_COMBINED_KERNELS; pKernelCache->iCacheEntries = 0; pKernelCache->iCacheSize = DL_COMBINED_KERNEL_CACHE_SIZE; // Size of kernel cache pKernelCache->iCacheFree = DL_COMBINED_KERNEL_CACHE_SIZE; // Free cache size pKernelCache->iCacheID = 0x00010000; // Cache ID pKernelCache->pCacheEntries = pCacheEntry; // Cached kernel entries pKernelCache->pCache = (uint8_t *)(pCacheEntry + DL_DEFAULT_COMBINED_KERNELS); // kernels // reset cache entries for (i = 0; i < DL_DEFAULT_COMBINED_KERNELS; i++, pCacheEntry++) { pCacheEntry->iKUID = -1; pCacheEntry->iKCID = -1; pCacheEntry->pBinary = pKernelCache->pCache + i * DL_CACHE_BLOCK_SIZE; if (i != DL_DEFAULT_COMBINED_KERNELS - 1) { pCacheEntry->pNextEntry = pCacheEntry + 1; } else { pCacheEntry->pNextEntry = nullptr; } } //------------------------------------ // Setup hash table //------------------------------------ pHashTable = &pState->KernelHashTable; pHashEntries = pState->KernelHashTable.HashEntry - 1; pHashTable->pool = 1; // first in pool (1 based index) pHashTable->last = DL_MAX_COMBINED_KERNELS; // last in pool (for releasing) for (i = 1; i <= DL_MAX_COMBINED_KERNELS; i++) { pHashEntries[i].next = i + 1; } pHashEntries[i - 1].next = 0; // last entry //------------------------------------ // Setup dynamic linking import/export array //------------------------------------ pCacheEntry = pState->ComponentKernelCache.pCacheEntries; iSize = pCacheEntry[IDR_VP_LinkFile].iSize; if (iSize == 0) { VP_RENDER_NORMALMESSAGE("Link file is missing."); goto cleanup; } // Get link file binary data pLinkHeader = (Kdll_LinkFileHeader *)pCacheEntry[IDR_VP_LinkFile].pBinary; if (pLinkHeader == nullptr || pLinkHeader->dwVersion != IDR_VP_LINKFILE_VERSION || sizeof(Kdll_LinkFileHeader) != IDR_VP_LINKFILE_HEADER) { VP_RENDER_ASSERTMESSAGE("Invalid link file version."); goto cleanup; } iSize = (iSize - IDR_VP_LINKFILE_HEADER) / sizeof(Kdll_LinkData); // Create temporary list of sorted link data and offsets pLinkSort = (Kdll_LinkData *)MOS_AllocAndZeroMemory(iSize * sizeof(Kdll_LinkData)); pLinkOffset = (uint32_t *)MOS_AllocAndZeroMemory((IDR_VP_TOTAL_NUM_KERNELS + 1) * sizeof(uint32_t)); if (!pLinkSort || !pLinkOffset) { VP_RENDER_ASSERTMESSAGE("Failed to allocate temporary buffers."); goto cleanup; } // Count number of imports for each component kernel pCacheEntry[0].pLink = pLinkData = (Kdll_LinkData *)(pLinkHeader + 1); for (i = iSize; i > 0; i--, pLinkData++) { if (pLinkData->iKUID < IDR_VP_TOTAL_NUM_KERNELS) { pCacheEntry[pLinkData->iKUID].nLink++; } nExports += pLinkData->bExport; nImports += !pLinkData->bExport; } // Sanity check if (nExports != (int32_t)pLinkHeader->dwExports || nImports != (int32_t)pLinkHeader->dwImports) { VP_RENDER_ASSERTMESSAGE("Inconsistent header data."); goto cleanup; } if (nExports > DL_MAX_EXPORT_COUNT) { VP_RENDER_ASSERTMESSAGE("Unsupported number of exports %d > %d.", nExports, DL_MAX_EXPORT_COUNT); goto cleanup; } pState->ComponentKernelCache.pExports = pExports = (Kdll_LinkData *)(pKernelCache->pCache + pKernelCache->iCacheSize); pState->ComponentKernelCache.nExports = nExports; // Calculate offsets for sorting pLinkOffset[0] = 0; pLinkData = pCacheEntry[0].pLink; for (i = 1; i < IDR_VP_TOTAL_NUM_KERNELS; i++) { pLinkOffset[i] = pLinkOffset[i - 1] + pCacheEntry[i - 1].nLink; pCacheEntry[i].pLink = (pCacheEntry[i].nLink) ? (pLinkData + pLinkOffset[i]) : nullptr; } pLinkOffset[i] = pLinkOffset[i - 1] + pCacheEntry[i - 1].nLink; // Sort link data for (i = iSize; i > 0; i--, pLinkData++) { j = pLinkOffset[MOS_MIN(pLinkData->iKUID, IDR_VP_TOTAL_NUM_KERNELS)]++; pLinkSort[j] = *pLinkData; // Add to export table if (pLinkData->bExport && pLinkData->iLabelID < DL_MAX_EXPORT_COUNT) { pExports[pLinkData->iLabelID] = *pLinkData; } } // Copy sort data pLinkData = pCacheEntry[0].pLink; MOS_SecureMemcpy(pLinkData, iSize * sizeof(Kdll_LinkData), (void *)pLinkSort, iSize * sizeof(Kdll_LinkData)); // Release sort buffers MOS_FreeMemory(pLinkOffset); MOS_FreeMemory(pLinkSort); // Return return pState; cleanup: if (pState) { MOS_FreeMemory(pState->pSortedRules); pState->pSortedRules = nullptr; } // Free DL States and temporary sort buffers MOS_FreeMemory(pState); MOS_FreeMemory(pLinkSort); MOS_FreeMemory(pLinkOffset); return nullptr; } //-------------------------------------------------------------- // KernelDll_ReleaseAdditionalCacheEntries - Release the additional kernel cache entries //-------------------------------------------------------------- void KernelDll_ReleaseAdditionalCacheEntries(Kdll_KernelCache *pCache) { VP_RENDER_FUNCTION_ENTER; if (pCache->iCacheMaxEntries > DL_DEFAULT_COMBINED_KERNELS) { Kdll_CacheEntry *pNewEntries, *pEntries; pNewEntries = (pCache->pCacheEntries + DL_DEFAULT_COMBINED_KERNELS - 1)->pNextEntry; for (int i = 0; i < (pCache->iCacheMaxEntries - DL_DEFAULT_COMBINED_KERNELS) / DL_NEW_COMBINED_KERNELS; i++) { pEntries = (pNewEntries + DL_NEW_COMBINED_KERNELS - 1)->pNextEntry; MOS_FreeMemory(pNewEntries); pNewEntries = pEntries; } } } //--------------------------------------------------------------------------------------- // KernelDll_ReleaseStates - Release Kernel Dynamic Linking/Loading (Dll) States // // Parameters: // Kdll_State *pState - [in] Kernel dll State to release // // Output: Pointer to allocated Kernel dll state // nullptr - Failed to allocate Kernel dll state //----------------------------------------------------------------------------------------- void KernelDll_ReleaseStates(Kdll_State *pState) { VP_RENDER_FUNCTION_ENTER; if (!pState) return; KernelDll_ReleaseAdditionalCacheEntries(&pState->KernelCache); MOS_FreeMemory(pState->ComponentKernelCache.pCache); MOS_FreeMemory(pState->CmFcPatchCache.pCache); MOS_FreeMemory(pState->pSortedRules); MOS_FreeMemory(pState); } //--------------------------------------------------------------------------------------- // KernelDll_SetupFunctionPointers - Setup Function pointers based on platform // // Parameters: // char *pState - [in/out] Kernel Dll state // platform - [in] platform // // Output: true - Function pointers are set // false - Failed to setup function pointers (invalid platform) //----------------------------------------------------------------------------------------- static bool KernelDll_SetupFunctionPointers( Kdll_State *pState, void (*ModifyFunctionPointers)(PKdll_State)) { VP_RENDER_FUNCTION_ENTER; pState->pfnSetupCSC = KernelDll_SetupCSC; pState->pfnMapCSCMatrix = KernelDll_MapCSCMatrix; pState->pfnFindRule = KernelDll_FindRule; pState->pfnUpdateState = KernelDll_UpdateState; pState->pfnSearchKernel = KernelDll_SearchKernel; pState->pfnBuildKernel = KernelDll_BuildKernel; pState->pfnStartKernelSearch = KernelDll_StartKernelSearch; if (ModifyFunctionPointers != nullptr) { (*ModifyFunctionPointers)(pState); } #if EMUL || VPHAL_LIB // Disable callbacks pState->pToken = nullptr; pState->pfnCbListKernel = nullptr; pState->pfnCbSearchSate = nullptr; #endif // EMUL || VPHAL_LIB return true; } //--------------------------------------------------------------------------------------- // Kdll_AddKernelList - Add kernel to CM FC kernel list // // Parameters: // Kdll_KernelCache *pKernelCache - [in] Component kernel cache // Kdll_KernelCache *pCmFcPatchCache - [in] Component kernel patch data cache // Kdll_SearchState *pSearchState - [in/out] Kernel search state // Kdll_PatchData *pKernelPatch - [in] Kernel Patch data // void *pPatchDst - [in] Patch data Dst address // int32_t iKUID - [in] Kernel Unique ID // cm_fc_kernel_t *Cm_Fc_Kernels - [in/out] CM FC Kernels // // Output: true if suceeded, false otherwise //--------------------------------------------------------------------------------------- bool Kdll_AddKernelList(Kdll_KernelCache *pKernelCache, Kdll_KernelCache * pCmFcPatchCache, Kdll_SearchState * pSearchState, int32_t iKUID, Kdll_PatchData * pKernelPatch, void * pPatchDst, cm_fc_kernel_t * Cm_Fc_Kernels) { Kdll_State * pState; Kdll_Symbol * pSymbols; Kdll_CacheEntry *kernels; Kdll_CacheEntry *pPatch; Kdll_LinkData * link; Kdll_LinkData * liSearch_reloc; int * size; int * left; int dwSize; int i; int base; bool bInline; bool res; VP_RENDER_FUNCTION_ENTER; res = false; // Check if Kernel ID is valid if (iKUID >= pKernelCache->iCacheEntries) { VP_RENDER_NORMALMESSAGE("invalid Kernel ID %d.", iKUID); goto finish; } // Get KDLL state pState = pSearchState->pKdllState; // Get current combined kernel size = &pSearchState->KernelSize; left = &pSearchState->KernelLeft; pSymbols = &pSearchState->KernelLink; base = (*size) >> 2; // Find selected kernel/patch and kernel size; check if there is enough space kernels = &pKernelCache->pCacheEntries[iKUID]; pPatch = &pCmFcPatchCache->pCacheEntries[iKUID]; dwSize = kernels->iSize; if (*left < dwSize) { VP_RENDER_NORMALMESSAGE("exceeded maximum kernel size."); goto finish; } // Check if there is enough space for symbols if (pSymbols->dwCount + kernels->nLink >= pSymbols->dwSize) { VP_RENDER_NORMALMESSAGE("exceeded maximum numbers of symbols to resolve."); goto finish; } #if EMUL || VPHAL_LIB VP_RENDER_NORMALMESSAGE("%s.", kernels->szName); if (pState->pfnCbListKernel) { pState->pfnCbListKernel(pState->pToken, kernels->szName); } #elif _DEBUG || _RELEASE_INTERNAL // EMUL || VPHAL_LIB VP_RENDER_NORMALMESSAGE("%s.", kernels->szName); #endif // _DEBUG MT_LOG1(MT_VP_KERNEL_LIST_ADD, MT_NORMAL, MT_VP_KERNEL_ID, iKUID); // Append symbols to resolve, relocate symbols link = kernels->pLink; liSearch_reloc = pSymbols->pLink + pSymbols->dwCount; bInline = false; if (link) { for (i = kernels->nLink; i > 0; i--, link++) { if (link->bInline) { // Inline code included if (!link->bExport) { bInline = true; } } else { *liSearch_reloc = *link; liSearch_reloc->dwOffset += base; liSearch_reloc++; pSymbols->dwCount++; } } } *size += dwSize; *left -= dwSize; Cm_Fc_Kernels->binary_buf = (const char *)kernels->pBinary; Cm_Fc_Kernels->binary_size = kernels->iSize; Cm_Fc_Kernels->patch_buf = (const char *)pPatch->pBinary; Cm_Fc_Kernels->patch_size = pPatch->iSize; res = true; finish: return res; } //--------------------------------------------------------------------------------------- // KernelDll_BuildKernel_CmFc - Build CM based FC combine Kernel // // Parameters: [in/out] pState - Pointer to Kernel binary file loaded in sys memory // [in/out] pSearchState - Kernel file size // // Output: bool // TRUE - Successful FALSE - Failed //----------------------------------------------------------------------------------------- bool KernelDll_BuildKernel_CmFc(Kdll_State *pState, Kdll_SearchState *pSearchState) { Kdll_KernelCache *pKernelCache = &pState->ComponentKernelCache; Kdll_KernelCache *pPatchCache = &pState->CmFcPatchCache; Kdll_KernelCache *pCustomCache = pState->pCustomKernelCache; bool res; int32_t offset = 0; int32_t * pKernelID, *pPatchID; uint8_t * pPatchData; Kdll_PatchData * pKernelPatch; uint8_t * kernel = pSearchState->Kernel; Kdll_Symbol * pSymbols = &pSearchState->KernelLink; uint32_t nExports = pKernelCache->nExports; Kdll_LinkData * pExports = pKernelCache->pExports; Kdll_LinkData * pLink; int32_t iOffset; uint32_t dwResolveOffset[DL_MAX_EXPORT_COUNT]; uint32_t dwTotalKernelCount; size_t stEstimatedKernelSize; int32_t iKUID; bool bResolveDone; int32_t i; cm_fc_kernel_t Cm_Fc_kernels[DL_MAX_KERNELS]; VP_RENDER_FUNCTION_ENTER; // Disable pop-up box window for STL assertion to avoid VM hang in auto test. #if (!LINUX && !ANDROID) uint32_t prevErrorMode = ::SetErrorMode(SEM_FAILCRITICALERRORS | SEM_NOGPFAULTERRORBOX); #if defined(_MSC_VER) ::_set_error_mode(_OUT_TO_STDERR); _CrtSetReportMode(_CRT_WARN, _CRTDBG_MODE_FILE | _CRTDBG_MODE_DEBUG); _CrtSetReportFile(_CRT_WARN, _CRTDBG_FILE_STDERR); _CrtSetReportMode(_CRT_ERROR, _CRTDBG_MODE_FILE | _CRTDBG_MODE_DEBUG); _CrtSetReportFile(_CRT_ERROR, _CRTDBG_FILE_STDERR); _CrtSetReportMode(_CRT_ASSERT, _CRTDBG_MODE_FILE | _CRTDBG_MODE_DEBUG); _CrtSetReportFile(_CRT_ASSERT, _CRTDBG_FILE_STDERR); #endif #endif pSearchState->KernelLink.dwSize = DL_MAX_SYMBOLS; pSearchState->KernelLink.dwCount = 0; pSearchState->KernelLink.pLink = pSearchState->LinkArray; pSearchState->KernelSize = 0; pSearchState->KernelLeft = sizeof(pSearchState->Kernel); pSearchState->KernelLink.dwCount = 0; MOS_ZeroMemory(Cm_Fc_kernels, sizeof(Cm_Fc_kernels)); dwTotalKernelCount = 0; stEstimatedKernelSize = 0; #if EMUL || VPHAL_LIB || _DEBUG VP_RENDER_NORMALMESSAGE("Component Kernels:"); #endif // EMUL || VPHAL_LIB || _DEBUG pKernelID = pSearchState->KernelID; pPatchID = pSearchState->PatchID; pPatchData = nullptr; for (offset = 0; offset < pSearchState->KernelCount; offset++, pKernelID++, pPatchID++, dwTotalKernelCount++) { // Get patch information associated with the kernel pKernelPatch = (*pPatchID >= 0) ? &(pSearchState->Patches[*pPatchID]) : nullptr; // Append/Patch kernel from internal cache res = Kdll_AddKernelList(pKernelCache, pPatchCache, pSearchState, *pKernelID, pKernelPatch, pPatchData, &Cm_Fc_kernels[dwTotalKernelCount]); stEstimatedKernelSize += Cm_Fc_kernels[dwTotalKernelCount].binary_size; if (*pKernelID == IDR_VP_EOT) { dwTotalKernelCount--; } if (!res) { VP_RENDER_NORMALMESSAGE("Failed to build kernel ID %d.", pSearchState->KernelID[offset]); res = false; goto finish; } } // Resolve kernel dependencies MOS_ZeroMemory(dwResolveOffset, sizeof(dwResolveOffset)); do { // Update exports for (pLink = pSymbols->pLink, i = pSymbols->dwCount; i > 0; i--, pLink++) { if (pLink->bExport) { dwResolveOffset[pLink->iLabelID] = pLink->dwOffset; } } bResolveDone = true; for (pLink = pSymbols->pLink, i = pSymbols->dwCount; i > 0; i--, pLink++) { // validate label if (pLink->iLabelID > nExports || // invalid label pExports[pLink->iLabelID].bExport == 0) // label not in the export table { VP_RENDER_NORMALMESSAGE("Invalid/unresolved label %d.", pLink->iLabelID); res = false; goto finish; } // load dependencies if (!pLink->bExport && !dwResolveOffset[pLink->iLabelID]) { // set flag for another pass as newly loaded // kernels may contain dependencies of their own bResolveDone = false; // Add dependencies to kernel list iKUID = pExports[pLink->iLabelID].iKUID; res = Kdll_AddKernelList(pKernelCache, pPatchCache, pSearchState, iKUID, nullptr, nullptr, &Cm_Fc_kernels[dwTotalKernelCount]); if (!res) { VP_RENDER_NORMALMESSAGE("Failed to build kernel ID %d.", pSearchState->KernelID[offset]); res = false; goto finish; } dwTotalKernelCount++; // Restart break; } } // for } while (!bResolveDone); if (stEstimatedKernelSize > DL_MAX_KERNEL_SIZE) { res = false; VP_RENDER_NORMALMESSAGE("Kernel size exceeded kdll limitatin."); goto finish; } stEstimatedKernelSize = DL_MAX_KERNEL_SIZE; // Get combine kernel binary from CMFC lib if (CM_FC_OK != cm_fc_combine_kernels(dwTotalKernelCount, Cm_Fc_kernels, (char *)pSearchState->Kernel, &stEstimatedKernelSize, nullptr)) { res = false; VP_RENDER_NORMALMESSAGE("cm_fc_combine_kernels() function call failed."); goto finish; } // Get combine kernel binary size from CMFC lib pSearchState->KernelSize = (int)stEstimatedKernelSize; res = true; finish: #if (!LINUX && !ANDROID) ::SetErrorMode(prevErrorMode); #endif return res; } //-------------------------------------------------------------- // KernelDll_AllocateHashEntry - Allocate hash entry //-------------------------------------------------------------- uint16_t KernelDll_AllocateHashEntry(Kdll_KernelHashTable *pHashTable, uint32_t hash) { Kdll_KernelHashEntry *pHashEntry = &pHashTable->HashEntry[0] - 1; Kdll_KernelHashEntry *pNewEntry; uint32_t folded_hash; uint16_t entry; VP_RENDER_FUNCTION_ENTER; entry = pHashTable->pool; if (!entry) { return 0; } // Get entry from pool pNewEntry = &pHashEntry[entry]; pHashTable->pool = pNewEntry->next; if (pHashTable->last == entry) { pHashTable->last = 0; } // Initialize entry, attach to the hash table FOLD_HASH(folded_hash, hash); pNewEntry->dwHash = hash; pNewEntry->next = pHashTable->wHashTable[folded_hash]; pNewEntry->iFilter = 0; pNewEntry->pFilter = nullptr; pNewEntry->pCacheEntry = nullptr; pHashTable->wHashTable[folded_hash] = entry; return entry; } //-------------------------------------------------------------- // KernelDll_CacheGarbageCollection - performs garbage collection //-------------------------------------------------------------- bool KernelDll_GarbageCollection(Kdll_State *pState, int32_t size) { Kdll_KernelCache *pCache = &pState->KernelCache; Kdll_CacheEntry *pEntry = pCache->pCacheEntries; Kdll_CacheEntry *pOldest = nullptr; Kdll_KernelHashTable *pHashTable = &pState->KernelHashTable; Kdll_KernelHashEntry *pHashEntry = &pHashTable->HashEntry[0] - 1; uint32_t dwOldest = (uint32_t)-1; uint16_t wEntry = 0; int32_t i; MOS_UNUSED(size); VP_RENDER_FUNCTION_ENTER; // Adjust refresh values to avoid overflow if (pState->dwRefresh > 0xffff0000) { pState->dwRefresh -= 0x80000000; for (i = pCache->iCacheMaxEntries; i > 0; i--) { if (pEntry->dwRefresh < 0x80000000) pEntry->dwRefresh = 0; else pEntry->dwRefresh -= 0x80000000; pEntry = pEntry->pNextEntry; } } // No need to deallocate old entries if (pCache->iCacheEntries < DL_MAX_COMBINED_KERNELS) { return true; } for (i = pCache->iCacheMaxEntries; i > 0; i--) { // deallocate old unreferenced entries if (pEntry->iKCID != -1 && pEntry->dwLoaded == 0) { if (pEntry->dwRefresh < dwOldest) { pOldest = pEntry; dwOldest = pEntry->dwRefresh; wEntry = pEntry->wHashEntry; } } pEntry = pEntry->pNextEntry; } // No entry to release, sanity checks pHashEntry += wEntry; if (!pOldest || wEntry == 0 || pHashEntry->pCacheEntry != pOldest) { VP_RENDER_ASSERT(false); return false; } // Release hash and cache entries KernelDll_ReleaseHashEntry(pHashTable, wEntry); KernelDll_ReleaseCacheEntry(pCache, pOldest); return true; } //-------------------------------------------------------------- // KernelDll_AllocateCacheEntry - Allocate cache entry for a given size //-------------------------------------------------------------- Kdll_CacheEntry * KernelDll_AllocateCacheEntry(Kdll_KernelCache *pCache, int32_t iSize) { Kdll_CacheEntry *pEntry = pCache->pCacheEntries; uint8_t *pCacheBinary = nullptr; Kdll_CacheEntry *pCacheNextEntry = nullptr; int32_t i, j; VP_RENDER_FUNCTION_ENTER; // Check size if (iSize > DL_CACHE_BLOCK_SIZE) { return nullptr; } // Search empty entry j = pCache->iCacheMaxEntries; for (i = 0; i < j; i++) { if (pEntry->iKCID == -1) { break; } pEntry = pEntry->pNextEntry; } if (i == j) { // Try to allocate more cache entries pEntry = KernelDll_AllocateAdditionalCacheEntries(pCache); if(! pEntry) { return nullptr; } } // Reset entry pCacheBinary = pEntry->pBinary; pCacheNextEntry = pEntry->pNextEntry; MOS_ZeroMemory(pEntry, sizeof(Kdll_CacheEntry)); pEntry->iSize = iSize; pEntry->pBinary = pCacheBinary; pEntry->pNextEntry = pCacheNextEntry; // Increment entries pCache->iCacheEntries++; return pEntry; } //-------------------------------------------------------------- // KernelDll_AllocateAdditionalCacheEntries - Allocate more kernel cache entries //-------------------------------------------------------------- Kdll_CacheEntry * KernelDll_AllocateAdditionalCacheEntries(Kdll_KernelCache *pCache) { Kdll_CacheEntry *pNewEntry = nullptr; Kdll_CacheEntry *pChcheEntry; int i, j; VP_RENDER_FUNCTION_ENTER; // Check num if (pCache->iCacheEntries + DL_NEW_COMBINED_KERNELS > DL_MAX_COMBINED_KERNELS) { VP_RENDER_ASSERTMESSAGE("KernelDll_AllocateAdditionalCacheEntries: Can't allocate more kernel cache entries\n"); return nullptr; } // Allocate the new entires i = (sizeof(Kdll_CacheEntry) + DL_CACHE_BLOCK_SIZE) * DL_NEW_COMBINED_KERNELS; pNewEntry = (Kdll_CacheEntry *)MOS_AllocAndZeroMemory(i); if (!pNewEntry) { VP_RENDER_ASSERTMESSAGE("KernelDll_AllocateAdditionalCacheEntries: Failed to allocate kernel cache entries\n"); return nullptr; } // Update the cache entires pChcheEntry = pCache->pCacheEntries; for(j = 0; j < pCache->iCacheMaxEntries - 1; j++) { pChcheEntry = pChcheEntry->pNextEntry; } pChcheEntry->pNextEntry = pNewEntry; for(j = 0; j < DL_NEW_COMBINED_KERNELS; j++, pNewEntry++) { pNewEntry->iKUID = -1; pNewEntry->iKCID = -1; pNewEntry->pBinary = (uint8_t *)(pNewEntry + DL_NEW_COMBINED_KERNELS - j) + j * DL_CACHE_BLOCK_SIZE; if(j != DL_NEW_COMBINED_KERNELS - 1) { pNewEntry->pNextEntry = pNewEntry + 1; } else { pNewEntry->pNextEntry = nullptr; } } pCache->iCacheMaxEntries += DL_NEW_COMBINED_KERNELS; pCache->iCacheSize += DL_NEW_COMBINED_KERNELS * DL_CACHE_BLOCK_SIZE; pCache->iCacheFree += DL_NEW_COMBINED_KERNELS * DL_CACHE_BLOCK_SIZE; return (Kdll_CacheEntry *)(pNewEntry - DL_NEW_COMBINED_KERNELS); } //-------------------------------------------------------------- // KernelDll_AddKernel - Add kernel into hash table and kernel cache //-------------------------------------------------------------- Kdll_CacheEntry * KernelDll_AddKernel(Kdll_State *pState, // Kernel Dll state Kdll_SearchState *pSearchState, // Search state Kdll_FilterEntry *pFilter, // Original filter int32_t iFilterSize, // Original filter size uint32_t dwHash) { Kdll_CacheEntry *pCacheEntry; Kdll_KernelHashTable *pHashTable; Kdll_KernelHashEntry *pHashEntry; uint16_t entry; int32_t size; uint8_t *ptr; VP_RENDER_FUNCTION_ENTER; // Check kernel if (pSearchState->KernelSize <= 0) { return nullptr; } // Get hash table pHashTable = &pState->KernelHashTable; pHashEntry = &pHashTable->HashEntry[0] - 1; // all indices are 1 based (0 = null) // allocate space in kernel cache to store the kernel, filter, CSC parameters size = pSearchState->KernelSize + // Kernel pSearchState->iFilterSize * sizeof(Kdll_FilterEntry) * 2 + // Original + Modified Filter sizeof(Kdll_CSC_Params) + // CSC parameters sizeof(VPHAL_CSPACE); // Intermediate Color Space for colorfill // Run garbage collection, create space for new kernel and metadata KernelDll_GarbageCollection(pState, size); // Get new kernel cache entry pCacheEntry = KernelDll_AllocateCacheEntry(&pState->KernelCache, size); if (!pCacheEntry) { VP_RENDER_ASSERTMESSAGE("Failed to allocate cache space for new kernel."); return nullptr; } // Get hash entry entry = KernelDll_AllocateHashEntry(pHashTable, dwHash); if (!entry) { VP_RENDER_ASSERTMESSAGE("Failed to allocate hash entry for new kernel."); KernelDll_ReleaseCacheEntry(&pState->KernelCache, pCacheEntry); return nullptr; } // Setup cache entry, copy kernel pCacheEntry->iKUID = -1; pCacheEntry->iKCID = pState->KernelCache.iCacheID; // Create new kernel cache id (KCID) pCacheEntry->dwRefresh = pState->dwRefresh++; pCacheEntry->wHashEntry = entry; // Save kernel pCacheEntry->iSize = pSearchState->KernelSize; MOS_SecureMemcpy(pCacheEntry->pBinary, pSearchState->KernelSize, (void *)pSearchState->Kernel, pSearchState->KernelSize); ptr = pCacheEntry->pBinary + pSearchState->KernelSize; // Save modified filter pCacheEntry->iFilterSize = pSearchState->iFilterSize; pCacheEntry->pFilter = (Kdll_FilterEntry *) (ptr); MOS_SecureMemcpy(ptr, pSearchState->iFilterSize * sizeof(Kdll_FilterEntry), (void *)pSearchState->Filter, pSearchState->iFilterSize * sizeof(Kdll_FilterEntry)); ptr += pSearchState->iFilterSize * sizeof(Kdll_FilterEntry); // Save CSC parameters associated with the kernel pCacheEntry->pCscParams = (Kdll_CSC_Params *) (ptr); MOS_SecureMemcpy(ptr, sizeof(Kdll_CSC_Params), (void *)&pSearchState->CscParams, sizeof(Kdll_CSC_Params)); ptr += sizeof(Kdll_CSC_Params); // Save intermediate color space for colorfill pCacheEntry->colorfill_cspace = pState->colorfill_cspace; ptr += sizeof(VPHAL_CSPACE); // increment KCID (Range = 0x00010000 - 0x7fffffff) pState->KernelCache.iCacheID = 0x00010000 + (pState->KernelCache.iCacheID - 0x0000ffff) % 0x7fff0000; // Setup hash entry, copy filter pHashEntry += entry; pHashEntry->pCacheEntry = pCacheEntry; // Save original filter for search purposes - modified filter is used for rendering pHashEntry->iFilter = iFilterSize; pHashEntry->pFilter = (Kdll_FilterEntry *) (ptr); MOS_SecureMemcpy(ptr, iFilterSize * sizeof(Kdll_FilterEntry), (void *)pFilter, iFilterSize * sizeof(Kdll_FilterEntry)); return pCacheEntry; } //-------------------------------------------------------------- // KernelDll_ReleaseHashEntry - Release hash table entry //-------------------------------------------------------------- void KernelDll_ReleaseHashEntry(Kdll_KernelHashTable *pHashTable, uint16_t entry) { Kdll_KernelHashEntry *pHashEntry = &pHashTable->HashEntry[0] - 1; uint32_t folded_hash; uint16_t next; VP_RENDER_FUNCTION_ENTER; if (entry == 0) { return; } // unlink entry next = pHashEntry[entry].next; pHashEntry[entry].next = 0; // remove references to entry from hash table FOLD_HASH(folded_hash, pHashEntry[entry].dwHash); if (pHashTable->wHashTable[folded_hash] == entry) { pHashTable->wHashTable[folded_hash] = next; } else { uint16_t prev = pHashTable->wHashTable[folded_hash]; while (prev != 0 && pHashEntry[prev].next != entry) { prev = pHashEntry[prev].next; } if (prev) { pHashEntry[prev].next = next; } } // return entry to pool if (pHashTable->pool == 0) { pHashTable->pool = entry; } else { pHashEntry[pHashTable->last].next = entry; } pHashTable->last = entry; } //-------------------------------------------------------------- // KernelDll_ReleaseCacheEntry - Release cache entry //-------------------------------------------------------------- void KernelDll_ReleaseCacheEntry(Kdll_KernelCache *pCache, Kdll_CacheEntry *pEntry) { pEntry->iKUID = -1; pEntry->iKCID = -1; pCache->iCacheEntries--; } //--------------------------------------------------------------------------------------- // KernelDll_SetupFunctionPointers - Setup Function pointers based on platform // // Parameters: // KdllState *pState - [in/out] Kernel Dll state // // Output: true - Function pointers are set // false - Failed to setup function pointers (invalid platform) //----------------------------------------------------------------------------------------- bool KernelDll_SetupFunctionPointers_Ext( Kdll_State *pState) { VP_RENDER_FUNCTION_ENTER; if (pState && pState->bEnableCMFC) { pState->pfnBuildKernel = KernelDll_BuildKernel_CmFc; } return true; } #ifdef __cplusplus } #endif // __cplusplus