/* * Copyright (C) 2008 The Android Open Source Project * Copyright (c) 2010-2013, The Linux Foundation. All rights reserved. * * Not a Contribution, Apache license notifications and license are retained * for attribution purposes only. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "c2d2.h" #include "software_converter.h" #include using gralloc::IMemAlloc; using gralloc::IonController; using gralloc::alloc_data; C2D_STATUS (*LINK_c2dCreateSurface)( uint32 *surface_id, uint32 surface_bits, C2D_SURFACE_TYPE surface_type, void *surface_definition ); C2D_STATUS (*LINK_c2dUpdateSurface)( uint32 surface_id, uint32 surface_bits, C2D_SURFACE_TYPE surface_type, void *surface_definition ); C2D_STATUS (*LINK_c2dReadSurface)( uint32 surface_id, C2D_SURFACE_TYPE surface_type, void *surface_definition, int32 x, int32 y ); C2D_STATUS (*LINK_c2dDraw)( uint32 target_id, uint32 target_config, C2D_RECT *target_scissor, uint32 target_mask_id, uint32 target_color_key, C2D_OBJECT *objects_list, uint32 num_objects ); C2D_STATUS (*LINK_c2dFinish)( uint32 target_id); C2D_STATUS (*LINK_c2dFlush)( uint32 target_id, c2d_ts_handle *timestamp); C2D_STATUS (*LINK_c2dWaitTimestamp)( c2d_ts_handle timestamp ); C2D_STATUS (*LINK_c2dDestroySurface)( uint32 surface_id ); C2D_STATUS (*LINK_c2dMapAddr) ( int mem_fd, void * hostptr, uint32 len, uint32 offset, uint32 flags, void ** gpuaddr); C2D_STATUS (*LINK_c2dUnMapAddr) ( void * gpuaddr); C2D_STATUS (*LINK_c2dGetDriverCapabilities) ( C2D_DRIVER_INFO * driver_info); /* create a fence fd for the timestamp */ C2D_STATUS (*LINK_c2dCreateFenceFD) ( uint32 target_id, c2d_ts_handle timestamp, int32 *fd); C2D_STATUS (*LINK_c2dFillSurface) ( uint32 surface_id, uint32 fill_color, C2D_RECT * fill_rect); /******************************************************************************/ #if defined(COPYBIT_Z180) #define MAX_SCALE_FACTOR (4096) #define MAX_DIMENSION (4096) #else #error "Unsupported HW version" #endif // The following defines can be changed as required i.e. as we encounter // complex use cases. #define MAX_RGB_SURFACES 8 // Max. RGB layers currently supported per draw #define MAX_YUV_2_PLANE_SURFACES 4// Max. 2-plane YUV layers currently supported per draw #define MAX_YUV_3_PLANE_SURFACES 1// Max. 3-plane YUV layers currently supported per draw // +1 for the destination surface. We cannot have multiple destination surfaces. #define MAX_SURFACES (MAX_RGB_SURFACES + MAX_YUV_2_PLANE_SURFACES + MAX_YUV_3_PLANE_SURFACES + 1) #define NUM_SURFACE_TYPES 3 // RGB_SURFACE + YUV_SURFACE_2_PLANES + YUV_SURFACE_3_PLANES #define MAX_BLIT_OBJECT_COUNT 50 // Max. blit objects that can be passed per draw enum { RGB_SURFACE, YUV_SURFACE_2_PLANES, YUV_SURFACE_3_PLANES }; enum eConversionType { CONVERT_TO_ANDROID_FORMAT, CONVERT_TO_C2D_FORMAT }; enum eC2DFlags { FLAGS_PREMULTIPLIED_ALPHA = 1<<0, FLAGS_YUV_DESTINATION = 1<<1, FLAGS_TEMP_SRC_DST = 1<<2 }; static gralloc::IAllocController* sAlloc = 0; /******************************************************************************/ /** State information for each device instance */ struct copybit_context_t { struct copybit_device_t device; // Templates for the various source surfaces. These templates are created // to avoid the expensive create/destroy C2D Surfaces C2D_OBJECT_STR blit_rgb_object[MAX_RGB_SURFACES]; C2D_OBJECT_STR blit_yuv_2_plane_object[MAX_YUV_2_PLANE_SURFACES]; C2D_OBJECT_STR blit_yuv_3_plane_object[MAX_YUV_3_PLANE_SURFACES]; C2D_OBJECT_STR blit_list[MAX_BLIT_OBJECT_COUNT]; // Z-ordered list of blit objects C2D_DRIVER_INFO c2d_driver_info; void *libc2d2; alloc_data temp_src_buffer; alloc_data temp_dst_buffer; unsigned int dst[NUM_SURFACE_TYPES]; // dst surfaces unsigned int mapped_gpu_addr[MAX_SURFACES]; // GPU addresses mapped inside copybit int blit_rgb_count; // Total RGB surfaces being blit int blit_yuv_2_plane_count; // Total 2 plane YUV surfaces being int blit_yuv_3_plane_count; // Total 3 plane YUV surfaces being blit int blit_count; // Total blit objects. unsigned int trg_transform; /* target transform */ int fb_width; int fb_height; int src_global_alpha; int config_mask; int dst_surface_type; bool is_premultiplied_alpha; void* time_stamp; // used for signaling the wait thread bool wait_timestamp; pthread_t wait_thread_id; bool stop_thread; pthread_mutex_t wait_cleanup_lock; pthread_cond_t wait_cleanup_cond; }; struct bufferInfo { int width; int height; int format; }; struct yuvPlaneInfo { int yStride; //luma stride int plane1_stride; int plane2_stride; int plane1_offset; int plane2_offset; }; /** * Common hardware methods */ static int open_copybit(const struct hw_module_t* module, const char* name, struct hw_device_t** device); static struct hw_module_methods_t copybit_module_methods = { .open = open_copybit }; /* * The COPYBIT Module */ struct copybit_module_t HAL_MODULE_INFO_SYM = { .common = { .tag = HARDWARE_MODULE_TAG, .version_major = 1, .version_minor = 0, .id = COPYBIT_HARDWARE_MODULE_ID, .name = "QCT COPYBIT C2D 2.0 Module", .author = "Qualcomm", .methods = ©bit_module_methods } }; /* thread function which waits on the timeStamp and cleans up the surfaces */ static void* c2d_wait_loop(void* ptr) { copybit_context_t* ctx = (copybit_context_t*)(ptr); char thread_name[64] = "copybitWaitThr"; prctl(PR_SET_NAME, (unsigned long) &thread_name, 0, 0, 0); setpriority(PRIO_PROCESS, 0, HAL_PRIORITY_URGENT_DISPLAY); while(ctx->stop_thread == false) { pthread_mutex_lock(&ctx->wait_cleanup_lock); while(ctx->wait_timestamp == false && !ctx->stop_thread) { pthread_cond_wait(&(ctx->wait_cleanup_cond), &(ctx->wait_cleanup_lock)); } if(ctx->wait_timestamp) { if(LINK_c2dWaitTimestamp(ctx->time_stamp)) { ALOGE("%s: LINK_c2dWaitTimeStamp ERROR!!", __FUNCTION__); } ctx->wait_timestamp = false; // Unmap any mapped addresses. for (int i = 0; i < MAX_SURFACES; i++) { if (ctx->mapped_gpu_addr[i]) { LINK_c2dUnMapAddr( (void*)ctx->mapped_gpu_addr[i]); ctx->mapped_gpu_addr[i] = 0; } } // Reset the counts after the draw. ctx->blit_rgb_count = 0; ctx->blit_yuv_2_plane_count = 0; ctx->blit_yuv_3_plane_count = 0; ctx->blit_count = 0; } pthread_mutex_unlock(&ctx->wait_cleanup_lock); if(ctx->stop_thread) break; } pthread_exit(NULL); return NULL; } /* convert COPYBIT_FORMAT to C2D format */ static int get_format(int format) { switch (format) { case HAL_PIXEL_FORMAT_RGB_565: return C2D_COLOR_FORMAT_565_RGB; case HAL_PIXEL_FORMAT_RGBX_8888: return C2D_COLOR_FORMAT_8888_ARGB | C2D_FORMAT_SWAP_RB | C2D_FORMAT_DISABLE_ALPHA; case HAL_PIXEL_FORMAT_RGBA_8888: return C2D_COLOR_FORMAT_8888_ARGB | C2D_FORMAT_SWAP_RB; case HAL_PIXEL_FORMAT_BGRA_8888: return C2D_COLOR_FORMAT_8888_ARGB; case HAL_PIXEL_FORMAT_YCbCr_420_SP: return C2D_COLOR_FORMAT_420_NV12; case HAL_PIXEL_FORMAT_NV12_ENCODEABLE:return C2D_COLOR_FORMAT_420_NV12; case HAL_PIXEL_FORMAT_YCrCb_420_SP: return C2D_COLOR_FORMAT_420_NV21; case HAL_PIXEL_FORMAT_YCbCr_420_SP_TILED: return C2D_COLOR_FORMAT_420_NV12 | C2D_FORMAT_MACROTILED; default: ALOGE("%s: invalid format (0x%x", __FUNCTION__, format); return -EINVAL; } return -EINVAL; } /* Get the C2D formats needed for conversion to YUV */ static int get_c2d_format_for_yuv_destination(int halFormat) { switch (halFormat) { // We do not swap the RB when the target is YUV case HAL_PIXEL_FORMAT_RGBX_8888: return C2D_COLOR_FORMAT_8888_ARGB | C2D_FORMAT_DISABLE_ALPHA; case HAL_PIXEL_FORMAT_RGBA_8888: return C2D_COLOR_FORMAT_8888_ARGB; // The U and V need to be interchanged when the target is YUV case HAL_PIXEL_FORMAT_YCbCr_420_SP: return C2D_COLOR_FORMAT_420_NV21; case HAL_PIXEL_FORMAT_NV12_ENCODEABLE:return C2D_COLOR_FORMAT_420_NV21; case HAL_PIXEL_FORMAT_YCrCb_420_SP: return C2D_COLOR_FORMAT_420_NV12; default: return get_format(halFormat); } return -EINVAL; } /* ------------------------------------------------------------------- *//*! * \internal * \brief Get the bpp for a particular color format * \param color format * \return bits per pixel *//* ------------------------------------------------------------------- */ int c2diGetBpp(int32 colorformat) { int c2dBpp = 0; switch(colorformat&0xFF) { case C2D_COLOR_FORMAT_4444_RGBA: case C2D_COLOR_FORMAT_4444_ARGB: case C2D_COLOR_FORMAT_1555_ARGB: case C2D_COLOR_FORMAT_565_RGB: case C2D_COLOR_FORMAT_5551_RGBA: c2dBpp = 16; break; case C2D_COLOR_FORMAT_8888_RGBA: case C2D_COLOR_FORMAT_8888_ARGB: c2dBpp = 32; break; case C2D_COLOR_FORMAT_8_L: case C2D_COLOR_FORMAT_8_A: c2dBpp = 8; break; case C2D_COLOR_FORMAT_4_A: c2dBpp = 4; break; case C2D_COLOR_FORMAT_1: c2dBpp = 1; break; default: ALOGE("%s ERROR", __func__); break; } return c2dBpp; } static uint32 c2d_get_gpuaddr(copybit_context_t* ctx, struct private_handle_t *handle, int &mapped_idx) { uint32 memtype, *gpuaddr; C2D_STATUS rc; if(!handle) return 0; if (handle->flags & (private_handle_t::PRIV_FLAGS_USES_PMEM | private_handle_t::PRIV_FLAGS_USES_PMEM_ADSP)) memtype = KGSL_USER_MEM_TYPE_PMEM; else if (handle->flags & private_handle_t::PRIV_FLAGS_USES_ASHMEM) memtype = KGSL_USER_MEM_TYPE_ASHMEM; else if (handle->flags & private_handle_t::PRIV_FLAGS_USES_ION) memtype = KGSL_USER_MEM_TYPE_ION; else { ALOGE("Invalid handle flags: 0x%x", handle->flags); return 0; } rc = LINK_c2dMapAddr(handle->fd, (void*)handle->base, handle->size, handle->offset, memtype, (void**)&gpuaddr); if (rc == C2D_STATUS_OK) { // We have mapped the GPU address inside copybit. We need to unmap this // address after the blit. Store this address for (int i = 0; i < MAX_SURFACES; i++) { if (ctx->mapped_gpu_addr[i] == 0) { ctx->mapped_gpu_addr[i] = (uint32) gpuaddr; mapped_idx = i; break; } } return (uint32) gpuaddr; } return 0; } static void unmap_gpuaddr(copybit_context_t* ctx, int mapped_idx) { if (!ctx || (mapped_idx == -1)) return; if (ctx->mapped_gpu_addr[mapped_idx]) { LINK_c2dUnMapAddr( (void*)ctx->mapped_gpu_addr[mapped_idx]); ctx->mapped_gpu_addr[mapped_idx] = 0; } } static int is_supported_rgb_format(int format) { switch(format) { case HAL_PIXEL_FORMAT_RGBA_8888: case HAL_PIXEL_FORMAT_RGBX_8888: case HAL_PIXEL_FORMAT_RGB_565: case HAL_PIXEL_FORMAT_BGRA_8888: { return COPYBIT_SUCCESS; } default: return COPYBIT_FAILURE; } } static int get_num_planes(int format) { switch(format) { case HAL_PIXEL_FORMAT_YCbCr_420_SP: case HAL_PIXEL_FORMAT_YCrCb_420_SP: case HAL_PIXEL_FORMAT_NV12_ENCODEABLE: case HAL_PIXEL_FORMAT_YCbCr_420_SP_TILED: { return 2; } case HAL_PIXEL_FORMAT_YV12: { return 3; } default: return COPYBIT_FAILURE; } } static int is_supported_yuv_format(int format) { switch(format) { case HAL_PIXEL_FORMAT_YCbCr_420_SP: case HAL_PIXEL_FORMAT_YCrCb_420_SP: case HAL_PIXEL_FORMAT_NV12_ENCODEABLE: case HAL_PIXEL_FORMAT_YCbCr_420_SP_TILED: { return COPYBIT_SUCCESS; } default: return COPYBIT_FAILURE; } } static int is_valid_destination_format(int format) { if (format == HAL_PIXEL_FORMAT_YCbCr_420_SP_TILED) { // C2D does not support NV12Tile as a destination format. return COPYBIT_FAILURE; } return COPYBIT_SUCCESS; } static int calculate_yuv_offset_and_stride(const bufferInfo& info, yuvPlaneInfo& yuvInfo) { int width = info.width; int height = info.height; int format = info.format; int aligned_height = 0; int aligned_width = 0, size = 0; switch (format) { case HAL_PIXEL_FORMAT_YCbCr_420_SP_TILED: { /* NV12 Tile buffers have their luma height aligned to 32bytes and width * aligned to 128 bytes. The chroma offset starts at an 8K boundary */ aligned_height = ALIGN(height, 32); aligned_width = ALIGN(width, 128); size = aligned_width * aligned_height; yuvInfo.plane1_offset = ALIGN(size,8192); yuvInfo.yStride = aligned_width; yuvInfo.plane1_stride = aligned_width; break; } case HAL_PIXEL_FORMAT_YCbCr_420_SP: case HAL_PIXEL_FORMAT_NV12_ENCODEABLE: case HAL_PIXEL_FORMAT_YCrCb_420_SP: { aligned_width = ALIGN(width, 32); yuvInfo.yStride = aligned_width; yuvInfo.plane1_stride = aligned_width; if (HAL_PIXEL_FORMAT_NV12_ENCODEABLE == format) { // The encoder requires a 2K aligned chroma offset yuvInfo.plane1_offset = ALIGN(aligned_width * height, 2048); } else yuvInfo.plane1_offset = aligned_width * height; break; } default: { return COPYBIT_FAILURE; } } return COPYBIT_SUCCESS; } /** create C2D surface from copybit image */ static int set_image(copybit_context_t* ctx, uint32 surfaceId, const struct copybit_image_t *rhs, const eC2DFlags flags, int &mapped_idx) { struct private_handle_t* handle = (struct private_handle_t*)rhs->handle; C2D_SURFACE_TYPE surfaceType; int status = COPYBIT_SUCCESS; uint32 gpuaddr = 0; int c2d_format; mapped_idx = -1; if (flags & FLAGS_YUV_DESTINATION) { c2d_format = get_c2d_format_for_yuv_destination(rhs->format); } else { c2d_format = get_format(rhs->format); } if(c2d_format == -EINVAL) { ALOGE("%s: invalid format", __FUNCTION__); return -EINVAL; } if(handle == NULL) { ALOGE("%s: invalid handle", __func__); return -EINVAL; } if (handle->gpuaddr == 0) { gpuaddr = c2d_get_gpuaddr(ctx, handle, mapped_idx); if(!gpuaddr) { ALOGE("%s: c2d_get_gpuaddr failed", __FUNCTION__); return COPYBIT_FAILURE; } } else { gpuaddr = handle->gpuaddr; } /* create C2D surface */ if(is_supported_rgb_format(rhs->format) == COPYBIT_SUCCESS) { /* RGB */ C2D_RGB_SURFACE_DEF surfaceDef; surfaceType = (C2D_SURFACE_TYPE) (C2D_SURFACE_RGB_HOST | C2D_SURFACE_WITH_PHYS); surfaceDef.phys = (void*) gpuaddr; surfaceDef.buffer = (void*) (handle->base); surfaceDef.format = c2d_format | ((flags & FLAGS_PREMULTIPLIED_ALPHA) ? C2D_FORMAT_PREMULTIPLIED : 0); surfaceDef.width = rhs->w; surfaceDef.height = rhs->h; int aligned_width = ALIGN(surfaceDef.width,32); surfaceDef.stride = (aligned_width * c2diGetBpp(surfaceDef.format))>>3; if(LINK_c2dUpdateSurface( surfaceId,C2D_TARGET | C2D_SOURCE, surfaceType, &surfaceDef)) { ALOGE("%s: RGB Surface c2dUpdateSurface ERROR", __FUNCTION__); unmap_gpuaddr(ctx, mapped_idx); status = COPYBIT_FAILURE; } } else if (is_supported_yuv_format(rhs->format) == COPYBIT_SUCCESS) { C2D_YUV_SURFACE_DEF surfaceDef; memset(&surfaceDef, 0, sizeof(surfaceDef)); surfaceType = (C2D_SURFACE_TYPE)(C2D_SURFACE_YUV_HOST | C2D_SURFACE_WITH_PHYS); surfaceDef.format = c2d_format; bufferInfo info; info.width = rhs->w; info.height = rhs->h; info.format = rhs->format; yuvPlaneInfo yuvInfo = {0}; status = calculate_yuv_offset_and_stride(info, yuvInfo); if(status != COPYBIT_SUCCESS) { ALOGE("%s: calculate_yuv_offset_and_stride error", __FUNCTION__); unmap_gpuaddr(ctx, mapped_idx); } surfaceDef.width = rhs->w; surfaceDef.height = rhs->h; surfaceDef.plane0 = (void*) (handle->base); surfaceDef.phys0 = (void*) (gpuaddr); surfaceDef.stride0 = yuvInfo.yStride; surfaceDef.plane1 = (void*) (handle->base + yuvInfo.plane1_offset); surfaceDef.phys1 = (void*) (gpuaddr + yuvInfo.plane1_offset); surfaceDef.stride1 = yuvInfo.plane1_stride; if (3 == get_num_planes(rhs->format)) { surfaceDef.plane2 = (void*) (handle->base + yuvInfo.plane2_offset); surfaceDef.phys2 = (void*) (gpuaddr + yuvInfo.plane2_offset); surfaceDef.stride2 = yuvInfo.plane2_stride; } if(LINK_c2dUpdateSurface( surfaceId,C2D_TARGET | C2D_SOURCE, surfaceType, &surfaceDef)) { ALOGE("%s: YUV Surface c2dUpdateSurface ERROR", __FUNCTION__); unmap_gpuaddr(ctx, mapped_idx); status = COPYBIT_FAILURE; } } else { ALOGE("%s: invalid format 0x%x", __FUNCTION__, rhs->format); unmap_gpuaddr(ctx, mapped_idx); status = COPYBIT_FAILURE; } return status; } /** copy the bits */ static int msm_copybit(struct copybit_context_t *ctx, unsigned int target) { if (ctx->blit_count == 0) { return COPYBIT_SUCCESS; } for (int i = 0; i < ctx->blit_count; i++) { ctx->blit_list[i].next = &(ctx->blit_list[i+1]); } ctx->blit_list[ctx->blit_count-1].next = NULL; uint32_t target_transform = ctx->trg_transform; if (ctx->c2d_driver_info.capabilities_mask & C2D_DRIVER_SUPPORTS_OVERRIDE_TARGET_ROTATE_OP) { // For A3xx - set 0x0 as the transform is set in the config_mask target_transform = 0x0; } if(LINK_c2dDraw(target, target_transform, 0x0, 0, 0, ctx->blit_list, ctx->blit_count)) { ALOGE("%s: LINK_c2dDraw ERROR", __FUNCTION__); return COPYBIT_FAILURE; } return COPYBIT_SUCCESS; } static int flush_get_fence_copybit (struct copybit_device_t *dev, int* fd) { struct copybit_context_t* ctx = (struct copybit_context_t*)dev; int status = COPYBIT_FAILURE; if (!ctx) return COPYBIT_FAILURE; pthread_mutex_lock(&ctx->wait_cleanup_lock); status = msm_copybit(ctx, ctx->dst[ctx->dst_surface_type]); if(LINK_c2dFlush(ctx->dst[ctx->dst_surface_type], &ctx->time_stamp)) { ALOGE("%s: LINK_c2dFlush ERROR", __FUNCTION__); // unlock the mutex and return failure pthread_mutex_unlock(&ctx->wait_cleanup_lock); return COPYBIT_FAILURE; } if(LINK_c2dCreateFenceFD(ctx->dst[ctx->dst_surface_type], ctx->time_stamp, fd)) { ALOGE("%s: LINK_c2dCreateFenceFD ERROR", __FUNCTION__); status = COPYBIT_FAILURE; } if(status == COPYBIT_SUCCESS) { //signal the wait_thread ctx->wait_timestamp = true; pthread_cond_signal(&ctx->wait_cleanup_cond); } pthread_mutex_unlock(&ctx->wait_cleanup_lock); return status; } static int finish_copybit(struct copybit_device_t *dev) { struct copybit_context_t* ctx = (struct copybit_context_t*)dev; if (!ctx) return COPYBIT_FAILURE; int status = msm_copybit(ctx, ctx->dst[ctx->dst_surface_type]); if(LINK_c2dFinish(ctx->dst[ctx->dst_surface_type])) { ALOGE("%s: LINK_c2dFinish ERROR", __FUNCTION__); return COPYBIT_FAILURE; } // Unmap any mapped addresses. for (int i = 0; i < MAX_SURFACES; i++) { if (ctx->mapped_gpu_addr[i]) { LINK_c2dUnMapAddr( (void*)ctx->mapped_gpu_addr[i]); ctx->mapped_gpu_addr[i] = 0; } } // Reset the counts after the draw. ctx->blit_rgb_count = 0; ctx->blit_yuv_2_plane_count = 0; ctx->blit_yuv_3_plane_count = 0; ctx->blit_count = 0; return status; } static int clear_copybit(struct copybit_device_t *dev, struct copybit_image_t const *buf, struct copybit_rect_t *rect) { int ret = COPYBIT_SUCCESS; int flags = FLAGS_PREMULTIPLIED_ALPHA; int mapped_dst_idx = -1; struct copybit_context_t* ctx = (struct copybit_context_t*)dev; C2D_RECT c2drect = {rect->l, rect->t, rect->r - rect->l, rect->b - rect->t}; pthread_mutex_lock(&ctx->wait_cleanup_lock); ret = set_image(ctx, ctx->dst[RGB_SURFACE], buf, (eC2DFlags)flags, mapped_dst_idx); if(ret) { ALOGE("%s: set_image error", __FUNCTION__); unmap_gpuaddr(ctx, mapped_dst_idx); pthread_mutex_unlock(&ctx->wait_cleanup_lock); return COPYBIT_FAILURE; } ret = LINK_c2dFillSurface(ctx->dst[RGB_SURFACE], 0x0, &c2drect); pthread_mutex_unlock(&ctx->wait_cleanup_lock); return ret; } /** setup rectangles */ static void set_rects(struct copybit_context_t *ctx, C2D_OBJECT *c2dObject, const struct copybit_rect_t *dst, const struct copybit_rect_t *src, const struct copybit_rect_t *scissor) { // Set the target rect. if((ctx->trg_transform & C2D_TARGET_ROTATE_90) && (ctx->trg_transform & C2D_TARGET_ROTATE_180)) { /* target rotation is 270 */ c2dObject->target_rect.x = (dst->t)<<16; c2dObject->target_rect.y = ctx->fb_width?(ALIGN(ctx->fb_width,32)- dst->r):dst->r; c2dObject->target_rect.y = c2dObject->target_rect.y<<16; c2dObject->target_rect.height = ((dst->r) - (dst->l))<<16; c2dObject->target_rect.width = ((dst->b) - (dst->t))<<16; } else if(ctx->trg_transform & C2D_TARGET_ROTATE_90) { c2dObject->target_rect.x = ctx->fb_height?(ctx->fb_height - dst->b):dst->b; c2dObject->target_rect.x = c2dObject->target_rect.x<<16; c2dObject->target_rect.y = (dst->l)<<16; c2dObject->target_rect.height = ((dst->r) - (dst->l))<<16; c2dObject->target_rect.width = ((dst->b) - (dst->t))<<16; } else if(ctx->trg_transform & C2D_TARGET_ROTATE_180) { c2dObject->target_rect.y = ctx->fb_height?(ctx->fb_height - dst->b):dst->b; c2dObject->target_rect.y = c2dObject->target_rect.y<<16; c2dObject->target_rect.x = ctx->fb_width?(ALIGN(ctx->fb_width,32) - dst->r):dst->r; c2dObject->target_rect.x = c2dObject->target_rect.x<<16; c2dObject->target_rect.height = ((dst->b) - (dst->t))<<16; c2dObject->target_rect.width = ((dst->r) - (dst->l))<<16; } else { c2dObject->target_rect.x = (dst->l)<<16; c2dObject->target_rect.y = (dst->t)<<16; c2dObject->target_rect.height = ((dst->b) - (dst->t))<<16; c2dObject->target_rect.width = ((dst->r) - (dst->l))<<16; } c2dObject->config_mask |= C2D_TARGET_RECT_BIT; // Set the source rect c2dObject->source_rect.x = (src->l)<<16; c2dObject->source_rect.y = (src->t)<<16; c2dObject->source_rect.height = ((src->b) - (src->t))<<16; c2dObject->source_rect.width = ((src->r) - (src->l))<<16; c2dObject->config_mask |= C2D_SOURCE_RECT_BIT; // Set the scissor rect c2dObject->scissor_rect.x = scissor->l; c2dObject->scissor_rect.y = scissor->t; c2dObject->scissor_rect.height = (scissor->b) - (scissor->t); c2dObject->scissor_rect.width = (scissor->r) - (scissor->l); c2dObject->config_mask |= C2D_SCISSOR_RECT_BIT; } /*****************************************************************************/ /** Set a parameter to value */ static int set_parameter_copybit( struct copybit_device_t *dev, int name, int value) { struct copybit_context_t* ctx = (struct copybit_context_t*)dev; int status = COPYBIT_SUCCESS; if (!ctx) { ALOGE("%s: null context", __FUNCTION__); return -EINVAL; } pthread_mutex_lock(&ctx->wait_cleanup_lock); switch(name) { case COPYBIT_PLANE_ALPHA: { if (value < 0) value = 0; if (value >= 256) value = 255; ctx->src_global_alpha = value; if (value < 255) ctx->config_mask |= C2D_GLOBAL_ALPHA_BIT; else ctx->config_mask &= ~C2D_GLOBAL_ALPHA_BIT; } break; case COPYBIT_BLEND_MODE: { if (value == COPYBIT_BLENDING_NONE) { ctx->config_mask |= C2D_ALPHA_BLEND_NONE; ctx->is_premultiplied_alpha = true; } else if (value == COPYBIT_BLENDING_PREMULT) { ctx->is_premultiplied_alpha = true; } else { ctx->config_mask &= ~C2D_ALPHA_BLEND_NONE; } } break; case COPYBIT_TRANSFORM: { unsigned int transform = 0; uint32 config_mask = 0; config_mask |= C2D_OVERRIDE_GLOBAL_TARGET_ROTATE_CONFIG; if((value & 0x7) == COPYBIT_TRANSFORM_ROT_180) { transform = C2D_TARGET_ROTATE_180; config_mask |= C2D_OVERRIDE_TARGET_ROTATE_180; } else if((value & 0x7) == COPYBIT_TRANSFORM_ROT_270) { transform = C2D_TARGET_ROTATE_90; config_mask |= C2D_OVERRIDE_TARGET_ROTATE_90; } else if(value == COPYBIT_TRANSFORM_ROT_90) { transform = C2D_TARGET_ROTATE_270; config_mask |= C2D_OVERRIDE_TARGET_ROTATE_270; } else { config_mask |= C2D_OVERRIDE_TARGET_ROTATE_0; if(value & COPYBIT_TRANSFORM_FLIP_H) { config_mask |= C2D_MIRROR_H_BIT; } else if(value & COPYBIT_TRANSFORM_FLIP_V) { config_mask |= C2D_MIRROR_V_BIT; } } if (ctx->c2d_driver_info.capabilities_mask & C2D_DRIVER_SUPPORTS_OVERRIDE_TARGET_ROTATE_OP) { ctx->config_mask |= config_mask; } else { // The transform for this surface does not match the current // target transform. Draw all previous surfaces. This will be // changed once we have a new mechanism to send different // target rotations to c2d. finish_copybit(dev); } ctx->trg_transform = transform; } break; case COPYBIT_FRAMEBUFFER_WIDTH: ctx->fb_width = value; break; case COPYBIT_FRAMEBUFFER_HEIGHT: ctx->fb_height = value; break; case COPYBIT_ROTATION_DEG: case COPYBIT_DITHER: case COPYBIT_BLUR: case COPYBIT_BLIT_TO_FRAMEBUFFER: // Do nothing break; default: ALOGE("%s: default case param=0x%x", __FUNCTION__, name); status = -EINVAL; break; } pthread_mutex_unlock(&ctx->wait_cleanup_lock); return status; } /** Get a static info value */ static int get(struct copybit_device_t *dev, int name) { struct copybit_context_t* ctx = (struct copybit_context_t*)dev; int value; if (!ctx) { ALOGE("%s: null context error", __FUNCTION__); return -EINVAL; } switch(name) { case COPYBIT_MINIFICATION_LIMIT: value = MAX_SCALE_FACTOR; break; case COPYBIT_MAGNIFICATION_LIMIT: value = MAX_SCALE_FACTOR; break; case COPYBIT_SCALING_FRAC_BITS: value = 32; break; case COPYBIT_ROTATION_STEP_DEG: value = 1; break; default: ALOGE("%s: default case param=0x%x", __FUNCTION__, name); value = -EINVAL; } return value; } static int is_alpha(int cformat) { int alpha = 0; switch (cformat & 0xFF) { case C2D_COLOR_FORMAT_8888_ARGB: case C2D_COLOR_FORMAT_8888_RGBA: case C2D_COLOR_FORMAT_5551_RGBA: case C2D_COLOR_FORMAT_4444_ARGB: alpha = 1; break; default: alpha = 0; break; } if(alpha && (cformat&C2D_FORMAT_DISABLE_ALPHA)) alpha = 0; return alpha; } /* Function to check if we need a temporary buffer for the blit. * This would happen if the requested destination stride and the * C2D stride do not match. We ignore RGB buffers, since their * stride is always aligned to 32. */ static bool need_temp_buffer(struct copybit_image_t const *img) { if (COPYBIT_SUCCESS == is_supported_rgb_format(img->format)) return false; struct private_handle_t* handle = (struct private_handle_t*)img->handle; // The width parameter in the handle contains the aligned_w. We check if we // need to convert based on this param. YUV formats have bpp=1, so checking // if the requested stride is aligned should suffice. if (0 == (handle->width)%32) { return false; } return true; } /* Function to extract the information from the copybit image and set the corresponding * values in the bufferInfo struct. */ static void populate_buffer_info(struct copybit_image_t const *img, bufferInfo& info) { info.width = img->w; info.height = img->h; info.format = img->format; } /* Function to get the required size for a particular format, inorder for C2D to perform * the blit operation. */ static size_t get_size(const bufferInfo& info) { size_t size = 0; int w = info.width; int h = info.height; int aligned_w = ALIGN(w, 32); switch(info.format) { case HAL_PIXEL_FORMAT_NV12_ENCODEABLE: { // Chroma for this format is aligned to 2K. size = ALIGN((aligned_w*h), 2048) + ALIGN(aligned_w/2, 32) * (h/2) *2; size = ALIGN(size, 4096); } break; case HAL_PIXEL_FORMAT_YCbCr_420_SP: case HAL_PIXEL_FORMAT_YCrCb_420_SP: { size = aligned_w * h + ALIGN(aligned_w/2, 32) * (h/2) * 2; size = ALIGN(size, 4096); } break; default: break; } return size; } /* Function to allocate memory for the temporary buffer. This memory is * allocated from Ashmem. It is the caller's responsibility to free this * memory. */ static int get_temp_buffer(const bufferInfo& info, alloc_data& data) { ALOGD("%s E", __FUNCTION__); // Alloc memory from system heap data.base = 0; data.fd = -1; data.offset = 0; data.size = get_size(info); data.align = getpagesize(); data.uncached = true; int allocFlags = GRALLOC_USAGE_PRIVATE_SYSTEM_HEAP; if (sAlloc == 0) { sAlloc = gralloc::IAllocController::getInstance(); } if (sAlloc == 0) { ALOGE("%s: sAlloc is still NULL", __FUNCTION__); return COPYBIT_FAILURE; } int err = sAlloc->allocate(data, allocFlags); if (0 != err) { ALOGE("%s: allocate failed", __FUNCTION__); return COPYBIT_FAILURE; } ALOGD("%s X", __FUNCTION__); return err; } /* Function to free the temporary allocated memory.*/ static void free_temp_buffer(alloc_data &data) { if (-1 != data.fd) { IMemAlloc* memalloc = sAlloc->getAllocator(data.allocType); memalloc->free_buffer(data.base, data.size, 0, data.fd); } } /* Function to perform the software color conversion. Convert the * C2D compatible format to the Android compatible format */ static int copy_image(private_handle_t *src_handle, struct copybit_image_t const *rhs, eConversionType conversionType) { if (src_handle->fd == -1) { ALOGE("%s: src_handle fd is invalid", __FUNCTION__); return COPYBIT_FAILURE; } // Copy the info. int ret = COPYBIT_SUCCESS; switch(rhs->format) { case HAL_PIXEL_FORMAT_NV12_ENCODEABLE: case HAL_PIXEL_FORMAT_YCbCr_420_SP: case HAL_PIXEL_FORMAT_YCrCb_420_SP: { if (CONVERT_TO_ANDROID_FORMAT == conversionType) { return convert_yuv_c2d_to_yuv_android(src_handle, rhs); } else { return convert_yuv_android_to_yuv_c2d(src_handle, rhs); } } break; default: { ALOGE("%s: invalid format 0x%x", __FUNCTION__, rhs->format); ret = COPYBIT_FAILURE; } break; } return ret; } static void delete_handle(private_handle_t *handle) { if (handle) { delete handle; handle = 0; } } static bool need_to_execute_draw(struct copybit_context_t* ctx, eC2DFlags flags) { if (flags & FLAGS_TEMP_SRC_DST) { return true; } if (flags & FLAGS_YUV_DESTINATION) { return true; } return false; } /** do a stretch blit type operation */ static int stretch_copybit_internal( struct copybit_device_t *dev, struct copybit_image_t const *dst, struct copybit_image_t const *src, struct copybit_rect_t const *dst_rect, struct copybit_rect_t const *src_rect, struct copybit_region_t const *region, bool enableBlend) { struct copybit_context_t* ctx = (struct copybit_context_t*)dev; int status = COPYBIT_SUCCESS; int flags = 0; int src_surface_type; int mapped_src_idx = -1, mapped_dst_idx = -1; C2D_OBJECT_STR src_surface; if (!ctx) { ALOGE("%s: null context error", __FUNCTION__); return -EINVAL; } if (src->w > MAX_DIMENSION || src->h > MAX_DIMENSION) { ALOGE("%s: src dimension error", __FUNCTION__); return -EINVAL; } if (dst->w > MAX_DIMENSION || dst->h > MAX_DIMENSION) { ALOGE("%s : dst dimension error dst w %d h %d", __FUNCTION__, dst->w, dst->h); return -EINVAL; } if (is_valid_destination_format(dst->format) == COPYBIT_FAILURE) { ALOGE("%s: Invalid destination format format = 0x%x", __FUNCTION__, dst->format); return COPYBIT_FAILURE; } int dst_surface_type; if (is_supported_rgb_format(dst->format) == COPYBIT_SUCCESS) { dst_surface_type = RGB_SURFACE; flags |= FLAGS_PREMULTIPLIED_ALPHA; } else if (is_supported_yuv_format(dst->format) == COPYBIT_SUCCESS) { int num_planes = get_num_planes(dst->format); flags |= FLAGS_YUV_DESTINATION; if (num_planes == 2) { dst_surface_type = YUV_SURFACE_2_PLANES; } else if (num_planes == 3) { dst_surface_type = YUV_SURFACE_3_PLANES; } else { ALOGE("%s: dst number of YUV planes is invalid dst format = 0x%x", __FUNCTION__, dst->format); return COPYBIT_FAILURE; } } else { ALOGE("%s: Invalid dst surface format 0x%x", __FUNCTION__, dst->format); return COPYBIT_FAILURE; } if (ctx->blit_rgb_count == MAX_RGB_SURFACES || ctx->blit_yuv_2_plane_count == MAX_YUV_2_PLANE_SURFACES || ctx->blit_yuv_3_plane_count == MAX_YUV_2_PLANE_SURFACES || ctx->blit_count == MAX_BLIT_OBJECT_COUNT || ctx->dst_surface_type != dst_surface_type) { // we have reached the max. limits of our internal structures or // changed the target. // Draw the remaining surfaces. We need to do the finish here since // we need to free up the surface templates. finish_copybit(dev); } ctx->dst_surface_type = dst_surface_type; // Update the destination copybit_image_t dst_image; dst_image.w = dst->w; dst_image.h = dst->h; dst_image.format = dst->format; dst_image.handle = dst->handle; // Check if we need a temp. copy for the destination. We'd need this the destination // width is not aligned to 32. This case occurs for YUV formats. RGB formats are // aligned to 32. bool need_temp_dst = need_temp_buffer(dst); bufferInfo dst_info; populate_buffer_info(dst, dst_info); private_handle_t* dst_hnd = new private_handle_t(-1, 0, 0, 0, dst_info.format, dst_info.width, dst_info.height); if (dst_hnd == NULL) { ALOGE("%s: dst_hnd is null", __FUNCTION__); return COPYBIT_FAILURE; } if (need_temp_dst) { if (get_size(dst_info) != ctx->temp_dst_buffer.size) { free_temp_buffer(ctx->temp_dst_buffer); // Create a temp buffer and set that as the destination. if (COPYBIT_FAILURE == get_temp_buffer(dst_info, ctx->temp_dst_buffer)) { ALOGE("%s: get_temp_buffer(dst) failed", __FUNCTION__); delete_handle(dst_hnd); return COPYBIT_FAILURE; } } dst_hnd->fd = ctx->temp_dst_buffer.fd; dst_hnd->size = ctx->temp_dst_buffer.size; dst_hnd->flags = ctx->temp_dst_buffer.allocType; dst_hnd->base = (int)(ctx->temp_dst_buffer.base); dst_hnd->offset = ctx->temp_dst_buffer.offset; dst_hnd->gpuaddr = 0; dst_image.handle = dst_hnd; } status = set_image(ctx, ctx->dst[ctx->dst_surface_type], &dst_image, (eC2DFlags)flags, mapped_dst_idx); if(status) { ALOGE("%s: dst: set_image error", __FUNCTION__); delete_handle(dst_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); return COPYBIT_FAILURE; } // Update the source flags = 0; if(is_supported_rgb_format(src->format) == COPYBIT_SUCCESS) { src_surface_type = RGB_SURFACE; src_surface = ctx->blit_rgb_object[ctx->blit_rgb_count]; } else if (is_supported_yuv_format(src->format) == COPYBIT_SUCCESS) { int num_planes = get_num_planes(src->format); if (num_planes == 2) { src_surface_type = YUV_SURFACE_2_PLANES; src_surface = ctx->blit_yuv_2_plane_object[ctx->blit_yuv_2_plane_count]; } else if (num_planes == 3) { src_surface_type = YUV_SURFACE_3_PLANES; src_surface = ctx->blit_yuv_3_plane_object[ctx->blit_yuv_2_plane_count]; } else { ALOGE("%s: src number of YUV planes is invalid src format = 0x%x", __FUNCTION__, src->format); delete_handle(dst_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); return -EINVAL; } } else { ALOGE("%s: Invalid source surface format 0x%x", __FUNCTION__, src->format); delete_handle(dst_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); return -EINVAL; } copybit_image_t src_image; src_image.w = src->w; src_image.h = src->h; src_image.format = src->format; src_image.handle = src->handle; bool need_temp_src = need_temp_buffer(src); bufferInfo src_info; populate_buffer_info(src, src_info); private_handle_t* src_hnd = new private_handle_t(-1, 0, 0, 0, src_info.format, src_info.width, src_info.height); if (NULL == src_hnd) { ALOGE("%s: src_hnd is null", __FUNCTION__); delete_handle(dst_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); return COPYBIT_FAILURE; } if (need_temp_src) { if (get_size(src_info) != ctx->temp_src_buffer.size) { free_temp_buffer(ctx->temp_src_buffer); // Create a temp buffer and set that as the destination. if (COPYBIT_SUCCESS != get_temp_buffer(src_info, ctx->temp_src_buffer)) { ALOGE("%s: get_temp_buffer(src) failed", __FUNCTION__); delete_handle(dst_hnd); delete_handle(src_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); return COPYBIT_FAILURE; } } src_hnd->fd = ctx->temp_src_buffer.fd; src_hnd->size = ctx->temp_src_buffer.size; src_hnd->flags = ctx->temp_src_buffer.allocType; src_hnd->base = (int)(ctx->temp_src_buffer.base); src_hnd->offset = ctx->temp_src_buffer.offset; src_hnd->gpuaddr = 0; src_image.handle = src_hnd; // Copy the source. status = copy_image((private_handle_t *)src->handle, &src_image, CONVERT_TO_C2D_FORMAT); if (status == COPYBIT_FAILURE) { ALOGE("%s:copy_image failed in temp source",__FUNCTION__); delete_handle(dst_hnd); delete_handle(src_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); return status; } // Clean the cache IMemAlloc* memalloc = sAlloc->getAllocator(src_hnd->flags); if (memalloc->clean_buffer((void *)(src_hnd->base), src_hnd->size, src_hnd->offset, src_hnd->fd, gralloc::CACHE_CLEAN)) { ALOGE("%s: clean_buffer failed", __FUNCTION__); delete_handle(dst_hnd); delete_handle(src_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); return COPYBIT_FAILURE; } } flags |= (ctx->is_premultiplied_alpha) ? FLAGS_PREMULTIPLIED_ALPHA : 0; flags |= (ctx->dst_surface_type != RGB_SURFACE) ? FLAGS_YUV_DESTINATION : 0; status = set_image(ctx, src_surface.surface_id, &src_image, (eC2DFlags)flags, mapped_src_idx); if(status) { ALOGE("%s: set_image (src) error", __FUNCTION__); delete_handle(dst_hnd); delete_handle(src_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); unmap_gpuaddr(ctx, mapped_src_idx); return COPYBIT_FAILURE; } src_surface.config_mask = C2D_NO_ANTIALIASING_BIT | ctx->config_mask; src_surface.global_alpha = ctx->src_global_alpha; if (enableBlend) { if(src_surface.config_mask & C2D_GLOBAL_ALPHA_BIT) { src_surface.config_mask &= ~C2D_ALPHA_BLEND_NONE; if(!(src_surface.global_alpha)) { // src alpha is zero delete_handle(dst_hnd); delete_handle(src_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); unmap_gpuaddr(ctx, mapped_src_idx); return COPYBIT_FAILURE; } } } else { src_surface.config_mask |= C2D_ALPHA_BLEND_NONE; } if (src_surface_type == RGB_SURFACE) { ctx->blit_rgb_object[ctx->blit_rgb_count] = src_surface; ctx->blit_rgb_count++; } else if (src_surface_type == YUV_SURFACE_2_PLANES) { ctx->blit_yuv_2_plane_object[ctx->blit_yuv_2_plane_count] = src_surface; ctx->blit_yuv_2_plane_count++; } else { ctx->blit_yuv_3_plane_object[ctx->blit_yuv_3_plane_count] = src_surface; ctx->blit_yuv_3_plane_count++; } struct copybit_rect_t clip; while ((status == 0) && region->next(region, &clip)) { set_rects(ctx, &(src_surface), dst_rect, src_rect, &clip); if (ctx->blit_count == MAX_BLIT_OBJECT_COUNT) { ALOGW("Reached end of blit count"); finish_copybit(dev); } ctx->blit_list[ctx->blit_count] = src_surface; ctx->blit_count++; } // Check if we need to perform an early draw-finish. flags |= (need_temp_dst || need_temp_src) ? FLAGS_TEMP_SRC_DST : 0; if (need_to_execute_draw(ctx, (eC2DFlags)flags)) { finish_copybit(dev); } if (need_temp_dst) { // copy the temp. destination without the alignment to the actual // destination. status = copy_image(dst_hnd, dst, CONVERT_TO_ANDROID_FORMAT); if (status == COPYBIT_FAILURE) { ALOGE("%s:copy_image failed in temp Dest",__FUNCTION__); delete_handle(dst_hnd); delete_handle(src_hnd); unmap_gpuaddr(ctx, mapped_dst_idx); unmap_gpuaddr(ctx, mapped_src_idx); return status; } // Clean the cache. IMemAlloc* memalloc = sAlloc->getAllocator(dst_hnd->flags); memalloc->clean_buffer((void *)(dst_hnd->base), dst_hnd->size, dst_hnd->offset, dst_hnd->fd, gralloc::CACHE_CLEAN); } delete_handle(dst_hnd); delete_handle(src_hnd); ctx->is_premultiplied_alpha = false; ctx->fb_width = 0; ctx->fb_height = 0; ctx->config_mask = 0; return status; } static int stretch_copybit( struct copybit_device_t *dev, struct copybit_image_t const *dst, struct copybit_image_t const *src, struct copybit_rect_t const *dst_rect, struct copybit_rect_t const *src_rect, struct copybit_region_t const *region) { struct copybit_context_t* ctx = (struct copybit_context_t*)dev; int status = COPYBIT_SUCCESS; bool needsBlending = (ctx->src_global_alpha != 0); pthread_mutex_lock(&ctx->wait_cleanup_lock); status = stretch_copybit_internal(dev, dst, src, dst_rect, src_rect, region, needsBlending); pthread_mutex_unlock(&ctx->wait_cleanup_lock); return status; } /** Perform a blit type operation */ static int blit_copybit( struct copybit_device_t *dev, struct copybit_image_t const *dst, struct copybit_image_t const *src, struct copybit_region_t const *region) { int status = COPYBIT_SUCCESS; struct copybit_context_t* ctx = (struct copybit_context_t*)dev; struct copybit_rect_t dr = { 0, 0, dst->w, dst->h }; struct copybit_rect_t sr = { 0, 0, src->w, src->h }; pthread_mutex_lock(&ctx->wait_cleanup_lock); status = stretch_copybit_internal(dev, dst, src, &dr, &sr, region, false); pthread_mutex_unlock(&ctx->wait_cleanup_lock); return status; } /*****************************************************************************/ static void clean_up(copybit_context_t* ctx) { void* ret; if (!ctx) return; // stop the wait_cleanup_thread pthread_mutex_lock(&ctx->wait_cleanup_lock); ctx->stop_thread = true; // Signal waiting thread pthread_cond_signal(&ctx->wait_cleanup_cond); pthread_mutex_unlock(&ctx->wait_cleanup_lock); // waits for the cleanup thread to exit pthread_join(ctx->wait_thread_id, &ret); pthread_mutex_destroy(&ctx->wait_cleanup_lock); pthread_cond_destroy (&ctx->wait_cleanup_cond); for (int i = 0; i < NUM_SURFACE_TYPES; i++) { if (ctx->dst[i]) LINK_c2dDestroySurface(ctx->dst[i]); } for (int i = 0; i < MAX_RGB_SURFACES; i++) { if (ctx->blit_rgb_object[i].surface_id) LINK_c2dDestroySurface(ctx->blit_rgb_object[i].surface_id); } for (int i = 0; i < MAX_YUV_2_PLANE_SURFACES; i++) { if (ctx->blit_yuv_2_plane_object[i].surface_id) LINK_c2dDestroySurface(ctx->blit_yuv_2_plane_object[i].surface_id); } for (int i = 0; i < MAX_YUV_3_PLANE_SURFACES; i++) { if (ctx->blit_yuv_3_plane_object[i].surface_id) LINK_c2dDestroySurface(ctx->blit_yuv_3_plane_object[i].surface_id); } if (ctx->libc2d2) { ::dlclose(ctx->libc2d2); ALOGV("dlclose(libc2d2)"); } free(ctx); } /** Close the copybit device */ static int close_copybit(struct hw_device_t *dev) { struct copybit_context_t* ctx = (struct copybit_context_t*)dev; if (ctx) { free_temp_buffer(ctx->temp_src_buffer); free_temp_buffer(ctx->temp_dst_buffer); } clean_up(ctx); return 0; } /** Open a new instance of a copybit device using name */ static int open_copybit(const struct hw_module_t* module, const char* name, struct hw_device_t** device) { int status = COPYBIT_SUCCESS; C2D_RGB_SURFACE_DEF surfDefinition = {0}; C2D_YUV_SURFACE_DEF yuvSurfaceDef = {0} ; struct copybit_context_t *ctx; char fbName[64]; ctx = (struct copybit_context_t *)malloc(sizeof(struct copybit_context_t)); if(!ctx) { ALOGE("%s: malloc failed", __FUNCTION__); return COPYBIT_FAILURE; } /* initialize drawstate */ memset(ctx, 0, sizeof(*ctx)); ctx->libc2d2 = ::dlopen("libC2D2.so", RTLD_NOW); if (!ctx->libc2d2) { ALOGE("FATAL ERROR: could not dlopen libc2d2.so: %s", dlerror()); clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } *(void **)&LINK_c2dCreateSurface = ::dlsym(ctx->libc2d2, "c2dCreateSurface"); *(void **)&LINK_c2dUpdateSurface = ::dlsym(ctx->libc2d2, "c2dUpdateSurface"); *(void **)&LINK_c2dReadSurface = ::dlsym(ctx->libc2d2, "c2dReadSurface"); *(void **)&LINK_c2dDraw = ::dlsym(ctx->libc2d2, "c2dDraw"); *(void **)&LINK_c2dFlush = ::dlsym(ctx->libc2d2, "c2dFlush"); *(void **)&LINK_c2dFinish = ::dlsym(ctx->libc2d2, "c2dFinish"); *(void **)&LINK_c2dWaitTimestamp = ::dlsym(ctx->libc2d2, "c2dWaitTimestamp"); *(void **)&LINK_c2dDestroySurface = ::dlsym(ctx->libc2d2, "c2dDestroySurface"); *(void **)&LINK_c2dMapAddr = ::dlsym(ctx->libc2d2, "c2dMapAddr"); *(void **)&LINK_c2dUnMapAddr = ::dlsym(ctx->libc2d2, "c2dUnMapAddr"); *(void **)&LINK_c2dGetDriverCapabilities = ::dlsym(ctx->libc2d2, "c2dGetDriverCapabilities"); *(void **)&LINK_c2dCreateFenceFD = ::dlsym(ctx->libc2d2, "c2dCreateFenceFD"); *(void **)&LINK_c2dFillSurface = ::dlsym(ctx->libc2d2, "c2dFillSurface"); if (!LINK_c2dCreateSurface || !LINK_c2dUpdateSurface || !LINK_c2dReadSurface || !LINK_c2dDraw || !LINK_c2dFlush || !LINK_c2dWaitTimestamp || !LINK_c2dFinish || !LINK_c2dDestroySurface || !LINK_c2dGetDriverCapabilities || !LINK_c2dCreateFenceFD || !LINK_c2dFillSurface) { ALOGE("%s: dlsym ERROR", __FUNCTION__); clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } ctx->device.common.tag = HARDWARE_DEVICE_TAG; ctx->device.common.version = 1; ctx->device.common.module = (hw_module_t*)(module); ctx->device.common.close = close_copybit; ctx->device.set_parameter = set_parameter_copybit; ctx->device.get = get; ctx->device.blit = blit_copybit; ctx->device.stretch = stretch_copybit; ctx->device.finish = finish_copybit; ctx->device.flush_get_fence = flush_get_fence_copybit; ctx->device.clear = clear_copybit; /* Create RGB Surface */ surfDefinition.buffer = (void*)0xdddddddd; surfDefinition.phys = (void*)0xdddddddd; surfDefinition.stride = 1 * 4; surfDefinition.width = 1; surfDefinition.height = 1; surfDefinition.format = C2D_COLOR_FORMAT_8888_ARGB; if (LINK_c2dCreateSurface(&(ctx->dst[RGB_SURFACE]), C2D_TARGET | C2D_SOURCE, (C2D_SURFACE_TYPE)(C2D_SURFACE_RGB_HOST | C2D_SURFACE_WITH_PHYS | C2D_SURFACE_WITH_PHYS_DUMMY ), &surfDefinition)) { ALOGE("%s: create ctx->dst_surface[RGB_SURFACE] failed", __FUNCTION__); ctx->dst[RGB_SURFACE] = 0; clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } unsigned int surface_id = 0; for (int i = 0; i < MAX_RGB_SURFACES; i++) { if (LINK_c2dCreateSurface(&surface_id, C2D_TARGET | C2D_SOURCE, (C2D_SURFACE_TYPE)(C2D_SURFACE_RGB_HOST | C2D_SURFACE_WITH_PHYS | C2D_SURFACE_WITH_PHYS_DUMMY ), &surfDefinition)) { ALOGE("%s: create RGB source surface %d failed", __FUNCTION__, i); ctx->blit_rgb_object[i].surface_id = 0; status = COPYBIT_FAILURE; break; } else { ctx->blit_rgb_object[i].surface_id = surface_id; ALOGW("%s i = %d surface_id=%d", __FUNCTION__, i, ctx->blit_rgb_object[i].surface_id); } } if (status == COPYBIT_FAILURE) { clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } // Create 2 plane YUV surfaces yuvSurfaceDef.format = C2D_COLOR_FORMAT_420_NV12; yuvSurfaceDef.width = 4; yuvSurfaceDef.height = 4; yuvSurfaceDef.plane0 = (void*)0xaaaaaaaa; yuvSurfaceDef.phys0 = (void*) 0xaaaaaaaa; yuvSurfaceDef.stride0 = 4; yuvSurfaceDef.plane1 = (void*)0xaaaaaaaa; yuvSurfaceDef.phys1 = (void*) 0xaaaaaaaa; yuvSurfaceDef.stride1 = 4; if (LINK_c2dCreateSurface(&(ctx->dst[YUV_SURFACE_2_PLANES]), C2D_TARGET | C2D_SOURCE, (C2D_SURFACE_TYPE)(C2D_SURFACE_YUV_HOST | C2D_SURFACE_WITH_PHYS | C2D_SURFACE_WITH_PHYS_DUMMY), &yuvSurfaceDef)) { ALOGE("%s: create ctx->dst[YUV_SURFACE_2_PLANES] failed", __FUNCTION__); ctx->dst[YUV_SURFACE_2_PLANES] = 0; clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } for (int i=0; i < MAX_YUV_2_PLANE_SURFACES; i++) { if (LINK_c2dCreateSurface(&surface_id, C2D_TARGET | C2D_SOURCE, (C2D_SURFACE_TYPE)(C2D_SURFACE_YUV_HOST | C2D_SURFACE_WITH_PHYS | C2D_SURFACE_WITH_PHYS_DUMMY ), &yuvSurfaceDef)) { ALOGE("%s: create YUV source %d failed", __FUNCTION__, i); ctx->blit_yuv_2_plane_object[i].surface_id = 0; status = COPYBIT_FAILURE; break; } else { ctx->blit_yuv_2_plane_object[i].surface_id = surface_id; ALOGW("%s: 2 Plane YUV i=%d surface_id=%d", __FUNCTION__, i, ctx->blit_yuv_2_plane_object[i].surface_id); } } if (status == COPYBIT_FAILURE) { clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } // Create YUV 3 plane surfaces yuvSurfaceDef.format = C2D_COLOR_FORMAT_420_YV12; yuvSurfaceDef.plane2 = (void*)0xaaaaaaaa; yuvSurfaceDef.phys2 = (void*) 0xaaaaaaaa; yuvSurfaceDef.stride2 = 4; if (LINK_c2dCreateSurface(&(ctx->dst[YUV_SURFACE_3_PLANES]), C2D_TARGET | C2D_SOURCE, (C2D_SURFACE_TYPE)(C2D_SURFACE_YUV_HOST | C2D_SURFACE_WITH_PHYS | C2D_SURFACE_WITH_PHYS_DUMMY), &yuvSurfaceDef)) { ALOGE("%s: create ctx->dst[YUV_SURFACE_3_PLANES] failed", __FUNCTION__); ctx->dst[YUV_SURFACE_3_PLANES] = 0; clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } for (int i=0; i < MAX_YUV_3_PLANE_SURFACES; i++) { if (LINK_c2dCreateSurface(&(surface_id), C2D_TARGET | C2D_SOURCE, (C2D_SURFACE_TYPE)(C2D_SURFACE_YUV_HOST | C2D_SURFACE_WITH_PHYS | C2D_SURFACE_WITH_PHYS_DUMMY), &yuvSurfaceDef)) { ALOGE("%s: create 3 plane YUV surface %d failed", __FUNCTION__, i); ctx->blit_yuv_3_plane_object[i].surface_id = 0; status = COPYBIT_FAILURE; break; } else { ctx->blit_yuv_3_plane_object[i].surface_id = surface_id; ALOGW("%s: 3 Plane YUV i=%d surface_id=%d", __FUNCTION__, i, ctx->blit_yuv_3_plane_object[i].surface_id); } } if (status == COPYBIT_FAILURE) { clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } if (LINK_c2dGetDriverCapabilities(&(ctx->c2d_driver_info))) { ALOGE("%s: LINK_c2dGetDriverCapabilities failed", __FUNCTION__); clean_up(ctx); status = COPYBIT_FAILURE; *device = NULL; return status; } // Initialize context variables. ctx->trg_transform = C2D_TARGET_ROTATE_0; ctx->temp_src_buffer.fd = -1; ctx->temp_src_buffer.base = 0; ctx->temp_src_buffer.size = 0; ctx->temp_dst_buffer.fd = -1; ctx->temp_dst_buffer.base = 0; ctx->temp_dst_buffer.size = 0; ctx->fb_width = 0; ctx->fb_height = 0; ctx->blit_rgb_count = 0; ctx->blit_yuv_2_plane_count = 0; ctx->blit_yuv_3_plane_count = 0; ctx->blit_count = 0; ctx->wait_timestamp = false; ctx->stop_thread = false; pthread_mutex_init(&(ctx->wait_cleanup_lock), NULL); pthread_cond_init(&(ctx->wait_cleanup_cond), NULL); /* Start the wait thread */ pthread_attr_t attr; pthread_attr_init(&attr); pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); pthread_create(&ctx->wait_thread_id, &attr, &c2d_wait_loop, (void *)ctx); pthread_attr_destroy(&attr); *device = &ctx->device.common; return status; }