gpujpeg_to_dxt.cpp

Go to the documentation of this file.

 
/*
 * Copyright (c) 2012-2023, CESNET z. s. p. o.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, is permitted provided that the following conditions
 * are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 *
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * 3. Neither the name of CESNET nor the names of its contributors may be
 *    used to endorse or promote products derived from this software without
 *    specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING,
 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY
 * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
 * EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
#ifdef HAVE_CONFIG_H
#include "config.h"
#include "config_unix.h"
#include "config_win32.h"
#endif // HAVE_CONFIG_H
 
#include "cuda_dxt/cuda_dxt.h"
 
#include <pthread.h>
#include <stdlib.h>
 
#include "libgpujpeg/gpujpeg_decoder.h"
#include "libgpujpeg/gpujpeg_version.h"
 
#include "debug.h"
#include "lib_common.h"
#include "host.h"
#include "utils/synchronized_queue.h"
#include "video.h"
#include "video_decompress.h"
 
#define MOD_NAME "[GPUJPEG to DXT] "
 
namespace {
 
struct thread_data {
        thread_data() :
                gpujpeg_decoder(0), desc(), out_codec(), ppb(), dxt_out_buff(0),
                cuda_dev_index(-1)
        {}
        synchronized_queue<msg *, 1> m_in;
        // currently only for output frames
        synchronized_queue<msg *, 1> m_out;
 
        struct gpujpeg_decoder  *gpujpeg_decoder;
        struct video_desc        desc;
        codec_t                  out_codec;
        int                      ppb;
        char                    *dxt_out_buff;
 
        int                      cuda_dev_index;
};
 
struct msg_reconfigure : public msg {
        struct video_desc        desc;
        int                      ppb; // pixels per byte
        codec_t                  out_codec;
};
 
struct msg_reconfigure_status : public msg {
        msg_reconfigure_status(bool val) : success(val) {}
 
        bool                     success;
};
 
struct msg_frame : public msg {
        msg_frame(int len) {
                data = new char[len];
                data_len = len;
        }
        ~msg_frame() {
                delete data;
        }
        char *data;
        int data_len;
};
 
struct state_decompresss_gpujpeg_to_dxt {
        struct thread_data       thread_data[MAX_CUDA_DEVICES];
        pthread_t                thread_id[MAX_CUDA_DEVICES];
 
        struct video_desc        desc;
        int                      ppb;
 
        // which card is free to process next image
        unsigned int             free;
        unsigned int             occupied_count; // number of GPUs occupied
 
};
 
} // namespace
 
static int reconfigure_thread(struct thread_data *s, struct video_desc desc, int ppb);
 
static void *worker_thread(void *arg)
{
        struct thread_data *s =
                (struct thread_data *) arg;
 
        while(1) {
                msg *message = s->m_in.pop();
 
                if (dynamic_cast<msg_quit *>(message)) {
                        delete message;
                        break;
                } else if (dynamic_cast<msg_reconfigure *>(message)) {
                        msg_reconfigure *reconf = dynamic_cast<msg_reconfigure *>(message);
                        bool ret = reconfigure_thread(s, reconf->desc, reconf->ppb);
                        s->desc = reconf->desc;
                        s->out_codec = reconf->out_codec;
                        s->ppb = reconf->ppb;
 
                        msg_reconfigure_status *status = new msg_reconfigure_status(ret);
                        s->m_out.push(status);
 
                        delete message;
                } else if (dynamic_cast<msg_frame *>(message)) {
                        msg_frame *frame_msg = dynamic_cast<msg_frame *>(message);
                        struct gpujpeg_decoder_output decoder_output;
 
                        gpujpeg_decoder_output_set_cuda_buffer(&decoder_output);
                        gpujpeg_decoder_decode(s->gpujpeg_decoder, (uint8_t *) frame_msg->data, frame_msg->data_len,
                                        &decoder_output);
 
                        if (s->out_codec == DXT1) {
                                cuda_rgb_to_dxt1(decoder_output.data, s->dxt_out_buff, s->desc.width,
                                                -s->desc.height, 0);
                        } else {
                                cuda_rgb_to_dxt6(decoder_output.data, s->dxt_out_buff, s->desc.width,
                                                -s->desc.height, 0);
                        }
 
                        msg_frame *output_frame = new msg_frame(s->desc.width * s->desc.height / s->ppb);
 
                        if (cuda_wrapper_memcpy((char*) output_frame->data, s->dxt_out_buff,
                                                output_frame->data_len,
                                                CUDA_WRAPPER_MEMCPY_DEVICE_TO_HOST) !=
                                        CUDA_WRAPPER_SUCCESS) {
                                log_msg(LOG_LEVEL_ERROR, MOD_NAME "unable to copy from device.");
                        }
                        s->m_out.push(output_frame);
 
                        delete frame_msg;
                }
        }
 
        if(s->gpujpeg_decoder) {
                gpujpeg_decoder_destroy(s->gpujpeg_decoder);
        }
 
        cuda_wrapper_free(s->dxt_out_buff);
 
        return NULL;
}
 
static void * gpujpeg_to_dxt_decompress_init(void)
{
        struct state_decompresss_gpujpeg_to_dxt *s;
 
        s = new state_decompresss_gpujpeg_to_dxt;
 
        s->free = 0;
        s->occupied_count = 0;
 
        for(unsigned int i = 0; i < cuda_devices_count; ++i) {
                int ret = gpujpeg_init_device(cuda_devices[i], TRUE);
                if(ret != 0) {
                        log_msg(LOG_LEVEL_ERROR, MOD_NAME "initializing CUDA device %d failed.\n", cuda_devices[0]);
                        delete s;
                        return NULL;
                }
        }
 
        for(unsigned int i = 0; i < cuda_devices_count; ++i) {
                s->thread_data[i].cuda_dev_index = i;
                int ret = pthread_create(&s->thread_id[i], NULL, worker_thread,  &s->thread_data[i]);
                assert(ret == 0);
        }
 
        return s;
}
 
static void flush(struct state_decompresss_gpujpeg_to_dxt *s)
{
        if (s->occupied_count > 0) {
                for (unsigned int i = 0; i < cuda_devices_count; ++i) {
                        if (i == s->free)
                                continue;
                        struct msg_frame *completed =
                                dynamic_cast<msg_frame *>(s->thread_data[(s->free+1)%cuda_devices_count].m_out.pop());
                        delete completed;
                }
                s->occupied_count = 0;
        }
        s->free = 0;
}
 
static int gpujpeg_to_dxt_decompress_reconfigure(void *state, struct video_desc desc,
                int rshift, int gshift, int bshift, int pitch, codec_t out_codec)
{
        struct state_decompresss_gpujpeg_to_dxt *s = (struct state_decompresss_gpujpeg_to_dxt *) state;
 
        UNUSED(rshift);
        UNUSED(gshift);
        UNUSED(bshift);
        assert(out_codec == DXT1 || out_codec == DXT5);
        if(out_codec == DXT1) {
                s->ppb = 2;
        } else { // DXT5
                s->ppb = 1;
        }
        s->desc = desc;
        assert(pitch == (int) desc.width / s->ppb); // default for DXT1
 
        flush(s);
 
        for(unsigned int i = 0; i < cuda_devices_count; ++i) {
                msg_reconfigure *reconf = new msg_reconfigure;
                reconf->desc = desc;
                reconf->out_codec = out_codec;
                reconf->ppb = s->ppb;
 
                s->thread_data[i].m_in.push(reconf);
 
                msg_reconfigure_status *status =
                        dynamic_cast<msg_reconfigure_status *>(s->thread_data[i].m_out.pop());
                assert(status != NULL);
                if (!status->success) {
                        delete status;
                        return FALSE;
                }
                delete status;
        }
 
        return TRUE;
}
 
static int reconfigure_thread(struct thread_data *s, struct video_desc desc, int ppb)
{
        if(s->gpujpeg_decoder != NULL) {
                gpujpeg_decoder_destroy(s->gpujpeg_decoder);
                s->gpujpeg_decoder = NULL;
        } else {
                gpujpeg_init_device(cuda_devices[s->cuda_dev_index], 0);
        }
 
        if(s->dxt_out_buff != NULL) {
                cuda_wrapper_free(s->dxt_out_buff);
                s->dxt_out_buff = NULL;
        }
 
        if(cuda_wrapper_malloc_host((void **) &s->dxt_out_buff, desc.width * desc.height / ppb)
                        != CUDA_WRAPPER_SUCCESS) {
                fprintf(stderr, "Could not allocate CUDA output buffer.\n");
                return false;
        }
        //gpujpeg_init_device(cuda_device, GPUJPEG_OPENGL_INTEROPERABILITY);
 
#if LIBGPUJPEG_API_VERSION <= 2
        s->gpujpeg_decoder = gpujpeg_decoder_create();
#else
        s->gpujpeg_decoder = gpujpeg_decoder_create(NULL);
#endif
        if(!s->gpujpeg_decoder) {
                log_msg(LOG_LEVEL_ERROR, "Creating GPUJPEG decoder failed.\n");
                return false;
        }
 
        return true;
}
 
static decompress_status gpujpeg_to_dxt_decompress(void *state, unsigned char *dst, unsigned char *buffer,
                unsigned int src_len, int frame_seq, video_frame_callbacks * /* callbacks */, codec_t * /* internal_codec */)
{
        struct state_decompresss_gpujpeg_to_dxt *s = (struct state_decompresss_gpujpeg_to_dxt *) state;
        UNUSED(frame_seq);
 
        msg_frame *message = new msg_frame(src_len);
        memcpy(message->data, buffer, src_len);
 
        if(s->occupied_count < cuda_devices_count - 1) {
                s->thread_data[s->free].m_in.push(message);
                s->free = s->free + 1; // should not exceed cuda_device_count
                s->occupied_count += 1;
        } else {
                s->thread_data[s->free].m_in.push(message);
 
                s->free = (s->free + 1) % cuda_devices_count;
 
                struct msg_frame *completed =
                        dynamic_cast<msg_frame *>(s->thread_data[s->free].m_out.pop());
                assert(completed != NULL);
                memcpy(dst, completed->data, s->desc.width * s->desc.height / s->ppb);
 
                delete completed;
        }
 
        return DECODER_GOT_FRAME;
}
 
static int gpujpeg_to_dxt_decompress_get_property(void *state, int property, void *val, size_t *len)
{
        UNUSED(state);
        UNUSED(property);
        UNUSED(val);
        UNUSED(len);
 
        return FALSE;
}
 
static void gpujpeg_to_dxt_decompress_done(void *state)
{
        struct state_decompresss_gpujpeg_to_dxt *s = (struct state_decompresss_gpujpeg_to_dxt *) state;
 
        if(!state) {
                return;
        }
 
        flush(s);
 
        for(unsigned int i = 0; i < cuda_devices_count; ++i) {
                msg_quit *quit_msg = new msg_quit;
                s->thread_data[i].m_in.push(quit_msg);
        }
 
        for(unsigned int i = 0; i < cuda_devices_count; ++i) {
                pthread_join(s->thread_id[i], NULL);
        }
 
        delete s;
}
 
static int gpujpeg_to_dxt_decompress_get_priority(codec_t compression, struct pixfmt_desc internal, codec_t ugc) {
        return compression == JPEG && (ugc == DXT1 || ugc == DXT5) ? 900 : -1;
}
 
static const struct video_decompress_info gpujpeg_to_dxt_info = {
        gpujpeg_to_dxt_decompress_init,
        gpujpeg_to_dxt_decompress_reconfigure,
        gpujpeg_to_dxt_decompress,
        gpujpeg_to_dxt_decompress_get_property,
        gpujpeg_to_dxt_decompress_done,
        gpujpeg_to_dxt_decompress_get_priority,
};
 
REGISTER_MODULE(gpujpeg_to_dxt, &gpujpeg_to_dxt_info, LIBRARY_CLASS_VIDEO_DECOMPRESS, VIDEO_DECOMPRESS_ABI_VERSION);