[FFmpeg-devel] [PATCH 1/2] avutil/hwcontext_cuda: align allocated frames
Philip Langdale
philipl at overt.org
Sun Oct 2 20:07:15 EEST 2016
On Sun, 2 Oct 2016 18:58:12 +0200
Timo Rothenpieler <timo at rothenpieler.org> wrote:
> ---
> libavutil/hwcontext_cuda.c | 43
> ++++++++++++++++++++++++++++--------------- 1 file changed, 28
> insertions(+), 15 deletions(-)
>
> diff --git a/libavutil/hwcontext_cuda.c b/libavutil/hwcontext_cuda.c
> index 40d2971..706d195 100644
> --- a/libavutil/hwcontext_cuda.c
> +++ b/libavutil/hwcontext_cuda.c
> @@ -16,6 +16,8 @@
> * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
> 02110-1301 USA */
>
> +#include "libavutil/intmath.h"
> +
> #include "buffer.h"
> #include "common.h"
> #include "hwcontext.h"
> @@ -35,6 +37,14 @@ static const enum AVPixelFormat
> supported_formats[] = { AV_PIX_FMT_YUV444P,
> };
>
> +static unsigned int next_pow2(unsigned int a)
> +{
> + if (a <= 1)
> + return 1;
> + a = 1 << (sizeof(a) * 8 - ff_clz(a - 1));
> + return FFALIGN(a, 256);
> +}
> +
> static void cuda_buffer_free(void *opaque, uint8_t *data)
> {
> AVHWFramesContext *ctx = opaque;
> @@ -83,6 +93,7 @@ fail:
> static int cuda_frames_init(AVHWFramesContext *ctx)
> {
> CUDAFramesContext *priv = ctx->internal->priv;
> + int aligned_width = next_pow2(ctx->width);
> int i;
>
> for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) {
> @@ -103,10 +114,10 @@ static int cuda_frames_init(AVHWFramesContext
> *ctx) switch (ctx->sw_format) {
> case AV_PIX_FMT_NV12:
> case AV_PIX_FMT_YUV420P:
> - size = ctx->width * ctx->height * 3 / 2;
> + size = aligned_width * ctx->height * 3 / 2;
> break;
> case AV_PIX_FMT_YUV444P:
> - size = ctx->width * ctx->height * 3;
> + size = aligned_width * ctx->height * 3;
> break;
> }
>
> @@ -120,6 +131,8 @@ static int cuda_frames_init(AVHWFramesContext
> *ctx)
> static int cuda_get_buffer(AVHWFramesContext *ctx, AVFrame *frame)
> {
> + int aligned_width = next_pow2(ctx->width);
> +
> frame->buf[0] = av_buffer_pool_get(ctx->pool);
> if (!frame->buf[0])
> return AVERROR(ENOMEM);
> @@ -127,25 +140,25 @@ static int cuda_get_buffer(AVHWFramesContext
> *ctx, AVFrame *frame) switch (ctx->sw_format) {
> case AV_PIX_FMT_NV12:
> frame->data[0] = frame->buf[0]->data;
> - frame->data[1] = frame->data[0] + ctx->width *
> ctx->height;
> - frame->linesize[0] = ctx->width;
> - frame->linesize[1] = ctx->width;
> + frame->data[1] = frame->data[0] + aligned_width *
> ctx->height;
> + frame->linesize[0] = aligned_width;
> + frame->linesize[1] = aligned_width;
> break;
> case AV_PIX_FMT_YUV420P:
> frame->data[0] = frame->buf[0]->data;
> - frame->data[2] = frame->data[0] + ctx->width *
> ctx->height;
> - frame->data[1] = frame->data[2] + ctx->width *
> ctx->height / 4;
> - frame->linesize[0] = ctx->width;
> - frame->linesize[1] = ctx->width / 2;
> - frame->linesize[2] = ctx->width / 2;
> + frame->data[2] = frame->data[0] + aligned_width *
> ctx->height;
> + frame->data[1] = frame->data[2] + aligned_width *
> ctx->height / 4;
> + frame->linesize[0] = aligned_width;
> + frame->linesize[1] = aligned_width / 2;
> + frame->linesize[2] = aligned_width / 2;
> break;
> case AV_PIX_FMT_YUV444P:
> frame->data[0] = frame->buf[0]->data;
> - frame->data[1] = frame->data[0] + ctx->width *
> ctx->height;
> - frame->data[2] = frame->data[1] + ctx->width *
> ctx->height;
> - frame->linesize[0] = ctx->width;
> - frame->linesize[1] = ctx->width;
> - frame->linesize[2] = ctx->width;
> + frame->data[1] = frame->data[0] + aligned_width *
> ctx->height;
> + frame->data[2] = frame->data[1] + aligned_width *
> ctx->height;
> + frame->linesize[0] = aligned_width;
> + frame->linesize[1] = aligned_width;
> + frame->linesize[2] = aligned_width;
> break;
> default:
> av_frame_unref(frame);
Looks good to me (and I tried it out).
--phil
More information about the ffmpeg-devel
mailing list