File libavcodec_nvenc.patch of Package avidemux3

File libavcodec_nvenc.patch of Package avidemux3 (Revision 19)

Currently displaying revision 19 , Show latest

--- libavcodec/nvenc.c.orig	2015-09-10 16:03:16.151393593 +0200
+++ libavcodec/nvenc.c	2015-09-10 16:03:16.198393966 +0200
@@ -960,19 +960,19 @@
 
         switch (avctx->pix_fmt) {
         case AV_PIX_FMT_YUV420P:
-            allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_YV12_PL;
-            break;
+            //allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_YV12_PL;
+            //break;
 
         case AV_PIX_FMT_NV12:
             allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_NV12_PL;
             break;
 
-        case AV_PIX_FMT_YUV444P:
-            allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_YUV444_PL;
-            break;
+        //case AV_PIX_FMT_YUV444P:
+            //allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_YUV444_PL;
+            //break;
 
         default:
-            av_log(avctx, AV_LOG_FATAL, "Invalid input pixel format\n");
+            av_log(avctx, AV_LOG_FATAL, "NVENC: Invalid input pixel format\n");
             res = AVERROR(EINVAL);
             goto error;
         }
@@ -1198,6 +1198,28 @@
 
     return res;
 }
+/**
+        MEANX
+*/
+static void interleave_nv12(unsigned char *dest, int dest_stride,unsigned char *u,int u_stride,unsigned char *v,int v_stride,int width, int height)
+{
+        int x,y;
+        unsigned char *d;
+        unsigned char *pu;
+        unsigned char *pv;
+        for(y=0;y<height;y++)
+        {
+                d=dest;dest+=dest_stride;
+                pu=u;u+=u_stride;
+                pv=v;v+=v_stride;
+                for(x=0;x<width;x++)
+                {
+                        *(d++)=*(pu++);
+                        *(d++)=*(pv++);
+                }
+
+        }
+} 
 
 static int nvenc_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
     const AVFrame *frame, int *got_packet)
@@ -1235,62 +1257,33 @@
         if (nv_status != NV_ENC_SUCCESS) {
             av_log(avctx, AV_LOG_ERROR, "Failed locking nvenc input buffer\n");
             return 0;
-        }
-
-        if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
-            uint8_t *buf = lockBufferParams.bufferDataPtr;
-
-            av_image_copy_plane(buf, lockBufferParams.pitch,
-                frame->data[0], frame->linesize[0],
-                avctx->width, avctx->height);
-
-            buf += inSurf->height * lockBufferParams.pitch;
-
-            av_image_copy_plane(buf, lockBufferParams.pitch >> 1,
-                frame->data[2], frame->linesize[2],
-                avctx->width >> 1, avctx->height >> 1);
-
-            buf += (inSurf->height * lockBufferParams.pitch) >> 2;
+        }  
+        { 
+        uint8_t *buf = lockBufferParams.bufferDataPtr;
 
-            av_image_copy_plane(buf, lockBufferParams.pitch >> 1,
-                frame->data[1], frame->linesize[1],
-                avctx->width >> 1, avctx->height >> 1);
-        } else if (avctx->pix_fmt == AV_PIX_FMT_NV12) {
-            uint8_t *buf = lockBufferParams.bufferDataPtr;
-
-            av_image_copy_plane(buf, lockBufferParams.pitch,
+        av_image_copy_plane(buf, lockBufferParams.pitch,
                 frame->data[0], frame->linesize[0],
                 avctx->width, avctx->height);
+// MEANX : We accept YV12 and interleave while copying, it saves us one intermediate image buffer
+        buf += inSurf->height * lockBufferParams.pitch;
+        switch(avctx->pix_fmt)
+        {
+                case AV_PIX_FMT_YUV420P:
+                        // Interleave U & V to NV12
+                        interleave_nv12(buf, lockBufferParams.pitch,frame->data[1],frame->linesize[1],frame->data[2],frame->linesize[2],avctx->width>>1, avctx->height>>1); 
+                        break;
+
+                case AV_PIX_FMT_NV12:
+                        av_image_copy_plane(buf, lockBufferParams.pitch,
+                                frame->data[1], frame->linesize[1],
+                                avctx->width, avctx->height >> 1);
+                        break;
+                default:
+                        av_log(avctx, AV_LOG_FATAL, "NVENC: Invalid pixel format!\n");
+                        return AVERROR(EINVAL);
 
-            buf += inSurf->height * lockBufferParams.pitch;
-
-            av_image_copy_plane(buf, lockBufferParams.pitch,
-                frame->data[1], frame->linesize[1],
-                avctx->width, avctx->height >> 1);
-        } else if (avctx->pix_fmt == AV_PIX_FMT_YUV444P) {
-            uint8_t *buf = lockBufferParams.bufferDataPtr;
-
-            av_image_copy_plane(buf, lockBufferParams.pitch,
-                frame->data[0], frame->linesize[0],
-                avctx->width, avctx->height);
-
-            buf += inSurf->height * lockBufferParams.pitch;
-
-            av_image_copy_plane(buf, lockBufferParams.pitch,
-                frame->data[1], frame->linesize[1],
-                avctx->width, avctx->height);
-
-            buf += inSurf->height * lockBufferParams.pitch;
-
-            av_image_copy_plane(buf, lockBufferParams.pitch,
-                frame->data[2], frame->linesize[2],
-                avctx->width, avctx->height);
-        } else {
-            av_log(avctx, AV_LOG_FATAL, "Invalid pixel format!\n");
-            return AVERROR(EINVAL);
-        }
-
-        nv_status = p_nvenc->nvEncUnlockInputBuffer(ctx->nvencoder, inSurf->input_surface);
+       }
+       nv_status = p_nvenc->nvEncUnlockInputBuffer(ctx->nvencoder, inSurf->input_surface);
         if (nv_status != NV_ENC_SUCCESS) {
             av_log(avctx, AV_LOG_FATAL, "Failed unlocking input buffer!\n");
             return AVERROR_EXTERNAL;