highperfocused/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

avutil/hwcontext_cuda: align allocated frames

Browse Source
This commit is contained in:
Timo Rothenpieler 2016-10-01 18:57:44 +02:00
parent 68822da8ff
commit a0d7ce1406
1 changed files with 20 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
libavutil/hwcontext_cuda.c
View File

@ -25,6 +25,8 @@
#include "pixdesc.h" #include "pixdesc.h"
#include "pixfmt.h" #include "pixfmt.h"
#define CUDA_FRAME_ALIGNMENT 256
typedef struct CUDAFramesContext { typedef struct CUDAFramesContext {
int shift_width, shift_height; int shift_width, shift_height;
} CUDAFramesContext; } CUDAFramesContext;
@ -83,6 +85,7 @@ fail:
static int cuda_frames_init(AVHWFramesContext *ctx) static int cuda_frames_init(AVHWFramesContext *ctx)
{ {
CUDAFramesContext *priv = ctx->internal->priv; CUDAFramesContext *priv = ctx->internal->priv;
int aligned_width = FFALIGN(ctx->width, CUDA_FRAME_ALIGNMENT);
int i; int i;
for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) { for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) {
@ -103,10 +106,10 @@ static int cuda_frames_init(AVHWFramesContext *ctx)
switch (ctx->sw_format) { switch (ctx->sw_format) {
case AV_PIX_FMT_NV12: case AV_PIX_FMT_NV12:
case AV_PIX_FMT_YUV420P: case AV_PIX_FMT_YUV420P:
size = ctx->width * ctx->height * 3 / 2; size = aligned_width * ctx->height * 3 / 2;
break; break;
case AV_PIX_FMT_YUV444P: case AV_PIX_FMT_YUV444P:
size = ctx->width * ctx->height * 3; size = aligned_width * ctx->height * 3;
break; break;
} }
@ -120,6 +123,8 @@ static int cuda_frames_init(AVHWFramesContext *ctx)
static int cuda_get_buffer(AVHWFramesContext *ctx, AVFrame *frame) static int cuda_get_buffer(AVHWFramesContext *ctx, AVFrame *frame)
{ {
int aligned_width = FFALIGN(ctx->width, CUDA_FRAME_ALIGNMENT);
frame->buf[0] = av_buffer_pool_get(ctx->pool); frame->buf[0] = av_buffer_pool_get(ctx->pool);
if (!frame->buf[0]) if (!frame->buf[0])
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
@ -127,25 +132,25 @@ static int cuda_get_buffer(AVHWFramesContext *ctx, AVFrame *frame)
switch (ctx->sw_format) { switch (ctx->sw_format) {
case AV_PIX_FMT_NV12: case AV_PIX_FMT_NV12:
frame->data[0] = frame->buf[0]->data; frame->data[0] = frame->buf[0]->data;
frame->data[1] = frame->data[0] + ctx->width * ctx->height; frame->data[1] = frame->data[0] + aligned_width * ctx->height;
frame->linesize[0] = ctx->width; frame->linesize[0] = aligned_width;
frame->linesize[1] = ctx->width; frame->linesize[1] = aligned_width;
break; break;
case AV_PIX_FMT_YUV420P: case AV_PIX_FMT_YUV420P:
frame->data[0] = frame->buf[0]->data; frame->data[0] = frame->buf[0]->data;
frame->data[2] = frame->data[0] + ctx->width * ctx->height; frame->data[2] = frame->data[0] + aligned_width * ctx->height;
frame->data[1] = frame->data[2] + ctx->width * ctx->height / 4; frame->data[1] = frame->data[2] + aligned_width * ctx->height / 4;
frame->linesize[0] = ctx->width; frame->linesize[0] = aligned_width;
frame->linesize[1] = ctx->width / 2; frame->linesize[1] = aligned_width / 2;
frame->linesize[2] = ctx->width / 2; frame->linesize[2] = aligned_width / 2;
break; break;
case AV_PIX_FMT_YUV444P: case AV_PIX_FMT_YUV444P:
frame->data[0] = frame->buf[0]->data; frame->data[0] = frame->buf[0]->data;
frame->data[1] = frame->data[0] + ctx->width * ctx->height; frame->data[1] = frame->data[0] + aligned_width * ctx->height;
frame->data[2] = frame->data[1] + ctx->width * ctx->height; frame->data[2] = frame->data[1] + aligned_width * ctx->height;
frame->linesize[0] = ctx->width; frame->linesize[0] = aligned_width;
frame->linesize[1] = ctx->width; frame->linesize[1] = aligned_width;
frame->linesize[2] = ctx->width; frame->linesize[2] = aligned_width;
break; break;
default: default:
av_frame_unref(frame); av_frame_unref(frame);