lavc: Add new VDA hwaccel

It leverages the new hwaccel 1.2 features:

- get_buffer2 is never called
- the internal context is automatically initialized/deinitialized

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

7 files changed, 412 insertions, 7 deletions

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 2bce954995..e8f12539b9 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -68,6 +68,7 @@ OBJS-$(CONFIG_RDFT)                    += rdft.o $(RDFT-OBJS-yes)
 OBJS-$(CONFIG_SINEWIN)                 += sinewin.o
 OBJS-$(CONFIG_TPELDSP)                 += tpeldsp.o
 OBJS-$(CONFIG_VAAPI)                   += vaapi.o
+OBJS-$(CONFIG_VDA)                     += vda.o
 OBJS-$(CONFIG_VDPAU)                   += vdpau.o
 OBJS-$(CONFIG_VIDEODSP)                += videodsp.o
 OBJS-$(CONFIG_VP3DSP)                  += vp3dsp.o
diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
index dcd6c4e607..9f4fd2531e 100644
--- a/libavcodec/allcodecs.c
+++ b/libavcodec/allcodecs.c
@@ -79,6 +79,7 @@ void avcodec_register_all(void)
     REGISTER_HWACCEL(H264_DXVA2,        h264_dxva2);
     REGISTER_HWACCEL(H264_VAAPI,        h264_vaapi);
     REGISTER_HWACCEL(H264_VDA,          h264_vda);
+    REGISTER_HWACCEL(H264_VDA_OLD,      h264_vda_old);
     REGISTER_HWACCEL(H264_VDPAU,        h264_vdpau);
     REGISTER_HWACCEL(MPEG1_VDPAU,       mpeg1_vdpau);
     REGISTER_HWACCEL(MPEG2_DXVA2,       mpeg2_dxva2);
diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c
index e727233c58..25f0f4d4af 100644
--- a/libavcodec/h264_slice.c
+++ b/libavcodec/h264_slice.c
@@ -153,6 +153,7 @@ static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
 #endif
 #if CONFIG_H264_VDA_HWACCEL
     AV_PIX_FMT_VDA_VLD,
+    AV_PIX_FMT_VDA,
 #endif
 #if CONFIG_H264_VDPAU_HWACCEL
     AV_PIX_FMT_VDPAU,
@@ -170,6 +171,7 @@ static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
 #endif
 #if CONFIG_H264_VDA_HWACCEL
     AV_PIX_FMT_VDA_VLD,
+    AV_PIX_FMT_VDA,
 #endif
 #if CONFIG_H264_VDPAU_HWACCEL
     AV_PIX_FMT_VDPAU,
diff --git a/libavcodec/vda.c b/libavcodec/vda.c
new file mode 100644
index 0000000000..f71fb16488
--- /dev/null
+++ b/libavcodec/vda.c
@@ -0,0 +1,72 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+
+#include "libavutil/mem.h"
+
+#include "vda.h"
+#include "vda_internal.h"
+
+#if CONFIG_H264_VDA_HWACCEL
+AVVDAContext *av_vda_alloc_context(void)
+{
+    AVVDAContext *ret = av_mallocz(sizeof(*ret));
+
+    if (ret)
+        ret->output_callback = ff_vda_output_callback;
+
+    return ret;
+}
+
+int av_vda_default_init(AVCodecContext *avctx)
+{
+    avctx->hwaccel_context = av_vda_alloc_context();
+    if (!avctx->hwaccel_context)
+        return AVERROR(ENOMEM);
+    return ff_vda_default_init(avctx);
+}
+
+void av_vda_default_free(AVCodecContext *avctx)
+{
+    ff_vda_default_free(avctx);
+    av_freep(&avctx->hwaccel_context);
+}
+
+void ff_vda_default_free(AVCodecContext *avctx)
+{
+    AVVDAContext *vda = avctx->hwaccel_context;
+    if (vda && vda->decoder)
+        VDADecoderDestroy(vda->decoder);
+}
+
+#else
+AVVDAContext *av_vda_alloc_context(void)
+{
+    return NULL;
+}
+
+int av_vda_default_init(AVCodecContext *avctx)
+{
+    return AVERROR(ENOSYS);
+}
+
+void av_vda_default_free(AVCodecContext *ctx)
+{
+}
+#endif
diff --git a/libavcodec/vda.h b/libavcodec/vda.h
index 1189e41fb5..9aa5d297fb 100644
--- a/libavcodec/vda.h
+++ b/libavcodec/vda.h
@@ -29,6 +29,7 @@
  * Public libavcodec VDA header.
  */
 
+#include "libavcodec/avcodec.h"
 #include "libavcodec/version.h"
 
 #include <stdint.h>
@@ -136,6 +137,58 @@ int ff_vda_create_decoder(struct vda_context *vda_ctx,
 int ff_vda_destroy_decoder(struct vda_context *vda_ctx);
 
 /**
+ * This struct holds all the information that needs to be passed
+ * between the caller and libavcodec for initializing VDA decoding.
+ * Its size is not a part of the public ABI, it must be allocated with
+ * av_vda_alloc_context() and freed with av_free().
+ */
+typedef struct AVVDAContext {
+    /**
+     * VDA decoder object. Created and freed by the caller.
+     */
+    VDADecoder decoder;
+
+    /**
+     * The output callback that must be passed to VDADecoderCreate.
+     * Set by av_vda_alloc_context().
+     */
+    VDADecoderOutputCallback output_callback;
+} AVVDAContext;
+
+/**
+ * Allocate and initialize a VDA context.
+ *
+ * This function should be called from the get_format() callback when the caller
+ * selects the AV_PIX_FMT_VDA format. The caller must then create the decoder
+ * object (using the output callback provided by libavcodec) that will be used
+ * for VDA-accelerated decoding.
+ *
+ * When decoding with VDA is finished, the caller must destroy the decoder
+ * object and free the VDA context using av_free().
+ *
+ * @return the newly allocated context or NULL on failure
+ */
+AVVDAContext *av_vda_alloc_context(void);
+
+/**
+ * This is a convenience function that creates and sets up the VDA context using
+ * an internal implementation.
+ *
+ * @param avctx the corresponding codec context
+ *
+ * @return >= 0 on success, a negative AVERROR code on failure
+ */
+int av_vda_default_init(AVCodecContext *avctx);
+
+/**
+ * This function must be called to free the VDA context initialized with
+ * av_vda_default_init().
+ *
+ * @param avctx the corresponding codec context
+ */
+void av_vda_default_free(AVCodecContext *avctx);
+
+/**
  * @}
  */
 
diff --git a/libavcodec/vda_h264.c b/libavcodec/vda_h264.c
index 72b0c78fc9..c7f6a7432e 100644
--- a/libavcodec/vda_h264.c
+++ b/libavcodec/vda_h264.c
@@ -26,7 +26,9 @@
 
 #include "libavutil/avutil.h"
 #include "h264.h"
+#include "internal.h"
 #include "vda.h"
+#include "vda_internal.h"
 
 typedef struct VDAContext {
     // The current bitstream buffer.
@@ -37,6 +39,8 @@ typedef struct VDAContext {
 
     // The reference size used for fast reallocation.
     int                  allocated_size;
+
+    CVImageBufferRef frame;
 } VDAContext;
 
 /* Decoder callback that adds the VDA frame to the queue in display order. */
@@ -78,7 +82,7 @@ static int vda_sync_decode(VDAContext *ctx, struct vda_context *vda_ctx)
 }
 
 
-static int vda_h264_start_frame(AVCodecContext *avctx,
+static int vda_old_h264_start_frame(AVCodecContext *avctx,
                                 av_unused const uint8_t *buffer,
                                 av_unused uint32_t size)
 {
@@ -93,7 +97,7 @@ static int vda_h264_start_frame(AVCodecContext *avctx,
     return 0;
 }
 
-static int vda_h264_decode_slice(AVCodecContext *avctx,
+static int vda_old_h264_decode_slice(AVCodecContext *avctx,
                                  const uint8_t *buffer,
                                  uint32_t size)
 {
@@ -120,7 +124,7 @@ static int vda_h264_decode_slice(AVCodecContext *avctx,
     return 0;
 }
 
-static int vda_h264_end_frame(AVCodecContext *avctx)
+static int vda_old_h264_end_frame(AVCodecContext *avctx)
 {
     H264Context *h                      = avctx->priv_data;
     VDAContext *vda                     = avctx->internal->hwaccel_priv_data;
@@ -208,7 +212,7 @@ int ff_vda_create_decoder(struct vda_context *vda_ctx,
 
     status = VDADecoderCreate(config_info,
                               buffer_attributes,
-                              vda_decoder_callback,
+                              (VDADecoderOutputCallback *)vda_decoder_callback,
                               vda_ctx,
                               &vda_ctx->decoder);
 
@@ -234,17 +238,256 @@ int ff_vda_destroy_decoder(struct vda_context *vda_ctx)
     return status;
 }
 
-static void vda_h264_uninit(AVCodecContext *avctx)
+static int vda_h264_uninit(AVCodecContext *avctx)
 {
-    VDAContext *vda = avctx->internal->priv_data;
+    VDAContext *vda = avctx->internal->hwaccel_priv_data;
     av_freep(&vda->bitstream);
+    if (vda->frame)
+        CVPixelBufferRelease(vda->frame);
+    return 0;
 }
 
-AVHWAccel ff_h264_vda_hwaccel = {
+AVHWAccel ff_h264_vda_old_hwaccel = {
     .name           = "h264_vda",
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_H264,
     .pix_fmt        = AV_PIX_FMT_VDA_VLD,
+    .start_frame    = vda_old_h264_start_frame,
+    .decode_slice   = vda_old_h264_decode_slice,
+    .end_frame      = vda_old_h264_end_frame,
+    .uninit         = vda_h264_uninit,
+    .priv_data_size = sizeof(VDAContext),
+};
+
+void ff_vda_output_callback(void *opaque,
+                            CFDictionaryRef user_info,
+                            OSStatus status,
+                            uint32_t infoFlags,
+                            CVImageBufferRef image_buffer)
+{
+    AVCodecContext *ctx = opaque;
+    VDAContext *vda = ctx->internal->hwaccel_priv_data;
+
+
+    if (vda->frame) {
+        CVPixelBufferRelease(vda->frame);
+        vda->frame = NULL;
+    }
+
+    if (!image_buffer)
+        return;
+
+    vda->frame = CVPixelBufferRetain(image_buffer);
+}
+
+static int vda_h264_start_frame(AVCodecContext *avctx,
+                                const uint8_t *buffer,
+                                uint32_t size)
+{
+    VDAContext *vda = avctx->internal->hwaccel_priv_data;
+
+    vda->bitstream_size = 0;
+
+    return 0;
+}
+
+static int vda_h264_decode_slice(AVCodecContext *avctx,
+                                 const uint8_t *buffer,
+                                 uint32_t size)
+{
+    VDAContext *vda       = avctx->internal->hwaccel_priv_data;
+    void *tmp;
+
+    tmp = av_fast_realloc(vda->bitstream,
+                          &vda->allocated_size,
+                          vda->bitstream_size + size + 4);
+    if (!tmp)
+        return AVERROR(ENOMEM);
+
+    vda->bitstream = tmp;
+
+    AV_WB32(vda->bitstream + vda->bitstream_size, size);
+    memcpy(vda->bitstream + vda->bitstream_size + 4, buffer, size);
+
+    vda->bitstream_size += size + 4;
+
+    return 0;
+}
+
+static void release_buffer(void *opaque, uint8_t *data)
+{
+    CVImageBufferRef frame = (CVImageBufferRef)data;
+    CVPixelBufferRelease(frame);
+}
+
+static int vda_h264_end_frame(AVCodecContext *avctx)
+{
+    H264Context *h        = avctx->priv_data;
+    VDAContext *vda       = avctx->internal->hwaccel_priv_data;
+    AVVDAContext *vda_ctx = avctx->hwaccel_context;
+    AVFrame *frame        = &h->cur_pic_ptr->f;
+    uint32_t flush_flags  = 1 << 0; ///< kVDADecoderFlush_emitFrames
+    CFDataRef coded_frame;
+    OSStatus status;
+
+    if (!vda->bitstream_size)
+        return AVERROR_INVALIDDATA;
+
+
+    coded_frame = CFDataCreate(kCFAllocatorDefault,
+                               vda->bitstream,
+                               vda->bitstream_size);
+
+    status = VDADecoderDecode(vda_ctx->decoder, 0, coded_frame, NULL);
+
+    if (status == kVDADecoderNoErr)
+        status = VDADecoderFlush(vda_ctx->decoder, flush_flags);
+
+    CFRelease(coded_frame);
+
+    if (status != kVDADecoderNoErr) {
+        av_log(avctx, AV_LOG_ERROR, "Failed to decode frame (%d)\n", status);
+        return AVERROR_UNKNOWN;
+    }
+
+    if (vda->frame) {
+        av_buffer_unref(&frame->buf[0]);
+
+        frame->buf[0] = av_buffer_create((uint8_t*)vda->frame,
+                                         sizeof(vda->frame),
+                                         release_buffer, NULL,
+                                         AV_BUFFER_FLAG_READONLY);
+        if (!frame->buf)
+            return AVERROR(ENOMEM);
+
+        frame->data[3] = (uint8_t*)vda->frame;
+        vda->frame = NULL;
+    }
+
+    return 0;
+}
+
+int ff_vda_default_init(AVCodecContext *avctx)
+{
+    AVVDAContext *vda_ctx = avctx->hwaccel_context;
+    OSStatus status = kVDADecoderNoErr;
+    CFNumberRef height;
+    CFNumberRef width;
+    CFNumberRef format;
+    CFDataRef avc_data;
+    CFMutableDictionaryRef config_info;
+    CFMutableDictionaryRef buffer_attributes;
+    CFMutableDictionaryRef io_surface_properties;
+    CFNumberRef cv_pix_fmt;
+    int32_t fmt = 'avc1', pix_fmt = kCVPixelFormatType_422YpCbCr8;
+
+    // kCVPixelFormatType_420YpCbCr8Planar;
+
+    /* Each VCL NAL in the bistream sent to the decoder
+     * is preceded by a 4 bytes length header.
+     * Change the avcC atom header if needed, to signal headers of 4 bytes. */
+    if (avctx->extradata_size >= 4 && (avctx->extradata[4] & 0x03) != 0x03) {
+        uint8_t *rw_extradata;
+
+        if (!(rw_extradata = av_malloc(avctx->extradata_size)))
+            return AVERROR(ENOMEM);
+
+        memcpy(rw_extradata, avctx->extradata, avctx->extradata_size);
+
+        rw_extradata[4] |= 0x03;
+
+        avc_data = CFDataCreate(kCFAllocatorDefault, rw_extradata, avctx->extradata_size);
+
+        av_freep(&rw_extradata);
+    } else {
+        avc_data = CFDataCreate(kCFAllocatorDefault,
+                                avctx->extradata, avctx->extradata_size);
+    }
+
+    config_info = CFDictionaryCreateMutable(kCFAllocatorDefault,
+                                            4,
+                                            &kCFTypeDictionaryKeyCallBacks,
+                                            &kCFTypeDictionaryValueCallBacks);
+
+    height = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &avctx->height);
+    width  = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &avctx->width);
+    format = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &fmt);
+    CFDictionarySetValue(config_info, kVDADecoderConfiguration_Height, height);
+    CFDictionarySetValue(config_info, kVDADecoderConfiguration_Width, width);
+    CFDictionarySetValue(config_info, kVDADecoderConfiguration_avcCData, avc_data);
+    CFDictionarySetValue(config_info, kVDADecoderConfiguration_SourceFormat, format);
+
+    buffer_attributes = CFDictionaryCreateMutable(kCFAllocatorDefault,
+                                                  2,
+                                                  &kCFTypeDictionaryKeyCallBacks,
+                                                  &kCFTypeDictionaryValueCallBacks);
+    io_surface_properties = CFDictionaryCreateMutable(kCFAllocatorDefault,
+                                                      0,
+                                                      &kCFTypeDictionaryKeyCallBacks,
+                                                      &kCFTypeDictionaryValueCallBacks);
+    cv_pix_fmt      = CFNumberCreate(kCFAllocatorDefault,
+                                     kCFNumberSInt32Type,
+                                     &pix_fmt);
+
+    CFDictionarySetValue(buffer_attributes,
+                         kCVPixelBufferPixelFormatTypeKey,
+                         cv_pix_fmt);
+    CFDictionarySetValue(buffer_attributes,
+                         kCVPixelBufferIOSurfacePropertiesKey,
+                         io_surface_properties);
+
+    status = VDADecoderCreate(config_info,
+                              buffer_attributes,
+                              (VDADecoderOutputCallback *)ff_vda_output_callback,
+                              avctx,
+                              &vda_ctx->decoder);
+
+    CFRelease(format);
+    CFRelease(height);
+    CFRelease(width);
+    CFRelease(avc_data);
+    CFRelease(config_info);
+    CFRelease(cv_pix_fmt);
+    CFRelease(io_surface_properties);
+    CFRelease(buffer_attributes);
+
+    if (status != kVDADecoderNoErr) {
+        av_log(avctx, AV_LOG_ERROR, "Cannot initialize VDA %d\n", status);
+    }
+
+    switch (status) {
+    case kVDADecoderHardwareNotSupportedErr:
+    case kVDADecoderFormatNotSupportedErr:
+        return AVERROR(ENOSYS);
+    case kVDADecoderConfigurationError:
+        return AVERROR(EINVAL);
+    case kVDADecoderDecoderFailedErr:
+        return AVERROR_INVALIDDATA;
+    case kVDADecoderNoErr:
+        return 0;
+    default:
+        return AVERROR_UNKNOWN;
+    }
+}
+
+static int vda_h264_alloc_frame(AVCodecContext *avctx, AVFrame *frame)
+{
+    frame->width  = avctx->width;
+    frame->height = avctx->height;
+    frame->format = avctx->pix_fmt;
+    frame->buf[0] = av_buffer_alloc(1);
+
+    if (!frame->buf[0])
+        return AVERROR(ENOMEM);
+    return 0;
+}
+
+AVHWAccel ff_h264_vda_hwaccel = {
+    .name           = "h264_vda",
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = AV_CODEC_ID_H264,
+    .pix_fmt        = AV_PIX_FMT_VDA,
+    .alloc_frame    = vda_h264_alloc_frame,
     .start_frame    = vda_h264_start_frame,
     .decode_slice   = vda_h264_decode_slice,
     .end_frame      = vda_h264_end_frame,
diff --git a/libavcodec/vda_internal.h b/libavcodec/vda_internal.h
new file mode 100644
index 0000000000..9d0ed807bf
--- /dev/null
+++ b/libavcodec/vda_internal.h
@@ -0,0 +1,33 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_VDA_INTERNAL_H
+#define AVCODEC_VDA_INTERNAL_H
+
+#include "vda.h"
+
+void ff_vda_output_callback(void *vda_hw_ctx,
+                            CFDictionaryRef user_info,
+                            OSStatus status,
+                            uint32_t infoFlags,
+                            CVImageBufferRef image_buffer);
+
+int ff_vda_default_init(AVCodecContext *avctx);
+void ff_vda_default_free(AVCodecContext *avctx);
+
+#endif /* AVCODEC_VDA_INTERNAL_H */