X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fnvenc.c;h=9a96bf2bbafe5d0100a5c468742a46f76425ad8b;hb=56c8856966ecc7a016aa0edb5472a902a7fbd49a;hp=304a684e0c2816513a675ce4517a91a29621a7c8;hpb=5ca7eb36b7353f9e6af05a5a952eead5f6d326dd;p=ffmpeg

diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c
index 304a684e0c2..9a96bf2bbaf 100644
--- a/libavcodec/nvenc.c
+++ b/libavcodec/nvenc.c
@@ -110,19 +110,49 @@ static int nvenc_map_error(NVENCSTATUS err, const char **desc)
     return AVERROR_UNKNOWN;
 }
 
-static int nvenc_print_error(void *log_ctx, NVENCSTATUS err,
+static int nvenc_print_error(AVCodecContext *avctx, NVENCSTATUS err,
                              const char *error_string)
 {
     const char *desc;
-    int ret;
-    ret = nvenc_map_error(err, &desc);
-    av_log(log_ctx, AV_LOG_ERROR, "%s: %s (%d)\n", error_string, desc, err);
+    const char *details = "(no details)";
+    int ret = nvenc_map_error(err, &desc);
+
+#ifdef NVENC_HAVE_GETLASTERRORSTRING
+    NvencContext *ctx = avctx->priv_data;
+    NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &ctx->nvenc_dload_funcs.nvenc_funcs;
+
+    if (p_nvenc && ctx->nvencoder)
+        details = p_nvenc->nvEncGetLastErrorString(ctx->nvencoder);
+#endif
+
+    av_log(avctx, AV_LOG_ERROR, "%s: %s (%d): %s\n", error_string, desc, err, details);
+
     return ret;
 }
 
 static void nvenc_print_driver_requirement(AVCodecContext *avctx, int level)
 {
-#if NVENCAPI_CHECK_VERSION(8, 1)
+#if NVENCAPI_CHECK_VERSION(9, 2)
+    const char *minver = "(unknown)";
+#elif NVENCAPI_CHECK_VERSION(9, 1)
+# if defined(_WIN32) || defined(__CYGWIN__)
+    const char *minver = "436.15";
+# else
+    const char *minver = "435.21";
+# endif
+#elif NVENCAPI_CHECK_VERSION(9, 0)
+# if defined(_WIN32) || defined(__CYGWIN__)
+    const char *minver = "418.81";
+# else
+    const char *minver = "418.30";
+# endif
+#elif NVENCAPI_CHECK_VERSION(8, 2)
+# if defined(_WIN32) || defined(__CYGWIN__)
+    const char *minver = "397.93";
+# else
+    const char *minver = "396.24";
+#endif
+#elif NVENCAPI_CHECK_VERSION(8, 1)
 # if defined(_WIN32) || defined(__CYGWIN__)
     const char *minver = "390.77";
 # else
@@ -291,39 +321,39 @@ static int nvenc_check_capabilities(AVCodecContext *avctx)
 
     ret = nvenc_check_codec_support(avctx);
     if (ret < 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "Codec not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "Codec not supported\n");
         return ret;
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_YUV444_ENCODE);
     if (IS_YUV444(ctx->data_pix_fmt) && ret <= 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "YUV444P not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "YUV444P not supported\n");
         return AVERROR(ENOSYS);
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_LOSSLESS_ENCODE);
     if (ctx->preset >= PRESET_LOSSLESS_DEFAULT && ret <= 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "Lossless encoding not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "Lossless encoding not supported\n");
         return AVERROR(ENOSYS);
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_WIDTH_MAX);
     if (ret < avctx->width) {
-        av_log(avctx, AV_LOG_VERBOSE, "Width %d exceeds %d\n",
+        av_log(avctx, AV_LOG_WARNING, "Width %d exceeds %d\n",
                avctx->width, ret);
         return AVERROR(ENOSYS);
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_HEIGHT_MAX);
     if (ret < avctx->height) {
-        av_log(avctx, AV_LOG_VERBOSE, "Height %d exceeds %d\n",
+        av_log(avctx, AV_LOG_WARNING, "Height %d exceeds %d\n",
                avctx->height, ret);
         return AVERROR(ENOSYS);
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_NUM_MAX_BFRAMES);
     if (ret < avctx->max_b_frames) {
-        av_log(avctx, AV_LOG_VERBOSE, "Max B-frames %d exceed %d\n",
+        av_log(avctx, AV_LOG_WARNING, "Max B-frames %d exceed %d\n",
                avctx->max_b_frames, ret);
 
         return AVERROR(ENOSYS);
@@ -331,7 +361,7 @@ static int nvenc_check_capabilities(AVCodecContext *avctx)
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_FIELD_ENCODING);
     if (ret < 1 && avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
-        av_log(avctx, AV_LOG_VERBOSE,
+        av_log(avctx, AV_LOG_WARNING,
                "Interlaced encoding is not supported. Supported level: %d\n",
                ret);
         return AVERROR(ENOSYS);
@@ -339,46 +369,59 @@ static int nvenc_check_capabilities(AVCodecContext *avctx)
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_10BIT_ENCODE);
     if (IS_10BIT(ctx->data_pix_fmt) && ret <= 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "10 bit encode not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "10 bit encode not supported\n");
         return AVERROR(ENOSYS);
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_LOOKAHEAD);
     if (ctx->rc_lookahead > 0 && ret <= 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "RC lookahead not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "RC lookahead not supported\n");
         return AVERROR(ENOSYS);
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_TEMPORAL_AQ);
     if (ctx->temporal_aq > 0 && ret <= 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "Temporal AQ not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "Temporal AQ not supported\n");
         return AVERROR(ENOSYS);
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_WEIGHTED_PREDICTION);
     if (ctx->weighted_pred > 0 && ret <= 0) {
-        av_log (avctx, AV_LOG_VERBOSE, "Weighted Prediction not supported\n");
+        av_log (avctx, AV_LOG_WARNING, "Weighted Prediction not supported\n");
         return AVERROR(ENOSYS);
     }
 
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_CABAC);
     if (ctx->coder == NV_ENC_H264_ENTROPY_CODING_MODE_CABAC && ret <= 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "CABAC entropy coding not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "CABAC entropy coding not supported\n");
         return AVERROR(ENOSYS);
     }
 
 #ifdef NVENC_HAVE_BFRAME_REF_MODE
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_BFRAME_REF_MODE);
     if (ctx->b_ref_mode == NV_ENC_BFRAME_REF_MODE_EACH && ret != 1) {
-        av_log(avctx, AV_LOG_VERBOSE, "Each B frame as reference is not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "Each B frame as reference is not supported\n");
         return AVERROR(ENOSYS);
     } else if (ctx->b_ref_mode != NV_ENC_BFRAME_REF_MODE_DISABLED && ret == 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "B frames as references are not supported\n");
+        av_log(avctx, AV_LOG_WARNING, "B frames as references are not supported\n");
         return AVERROR(ENOSYS);
     }
 #else
     if (ctx->b_ref_mode != 0) {
-        av_log(avctx, AV_LOG_VERBOSE, "B frames as references need SDK 8.1 at build time\n");
+        av_log(avctx, AV_LOG_WARNING, "B frames as references need SDK 8.1 at build time\n");
+        return AVERROR(ENOSYS);
+    }
+#endif
+
+#ifdef NVENC_HAVE_MULTIPLE_REF_FRAMES
+    ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_MULTIPLE_REF_FRAMES);
+    if(avctx->refs != NV_ENC_NUM_REF_FRAMES_AUTOSELECT && ret <= 0) {
+        av_log(avctx, AV_LOG_WARNING, "Multiple reference frames are not supported by the device\n");
+        return AVERROR(ENOSYS);
+    }
+#else
+    if(avctx->refs != 0) {
+        av_log(avctx, AV_LOG_WARNING, "Multiple reference frames need SDK 9.1 at build time\n");
         return AVERROR(ENOSYS);
     }
 #endif
@@ -427,6 +470,7 @@ static av_cold int nvenc_check_device(AVCodecContext *avctx, int idx)
         goto fail;
 
     ctx->cu_context = ctx->cu_context_internal;
+    ctx->cu_stream = NULL;
 
     if ((ret = nvenc_pop_context(avctx)) < 0)
         goto fail2;
@@ -513,6 +557,7 @@ static av_cold int nvenc_setup_device(AVCodecContext *avctx)
 
         if (cuda_device_hwctx) {
             ctx->cu_context = cuda_device_hwctx->cuda_ctx;
+            ctx->cu_stream = cuda_device_hwctx->stream;
         }
 #if CONFIG_D3D11VA
         else if (d3d11_device_hwctx) {
@@ -556,7 +601,7 @@ static av_cold int nvenc_setup_device(AVCodecContext *avctx)
             return AVERROR_EXIT;
 
         if (!dl_fn->nvenc_device_count) {
-            av_log(avctx, AV_LOG_FATAL, "No NVENC capable devices found\n");
+            av_log(avctx, AV_LOG_FATAL, "No capable devices found\n");
             return AVERROR_EXTERNAL;
         }
 
@@ -929,9 +974,9 @@ static av_cold int nvenc_setup_h264_config(AVCodecContext *avctx)
     h264->repeatSPSPPS  = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
     h264->outputAUD     = ctx->aud;
 
-    if (avctx->refs >= 0) {
+    if (ctx->dpb_size >= 0) {
         /* 0 means "let the hardware decide" */
-        h264->maxNumRefFrames = avctx->refs;
+        h264->maxNumRefFrames = ctx->dpb_size;
     }
     if (avctx->gop_size >= 0) {
         h264->idrPeriod = cc->gopLength;
@@ -990,6 +1035,11 @@ static av_cold int nvenc_setup_h264_config(AVCodecContext *avctx)
     h264->useBFramesAsRef = ctx->b_ref_mode;
 #endif
 
+#ifdef NVENC_HAVE_MULTIPLE_REF_FRAMES
+    h264->numRefL0 = avctx->refs;
+    h264->numRefL1 = avctx->refs;
+#endif
+
     return 0;
 }
 
@@ -1021,9 +1071,9 @@ static av_cold int nvenc_setup_hevc_config(AVCodecContext *avctx)
     hevc->repeatSPSPPS  = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
     hevc->outputAUD     = ctx->aud;
 
-    if (avctx->refs >= 0) {
+    if (ctx->dpb_size >= 0) {
         /* 0 means "let the hardware decide" */
-        hevc->maxNumRefFramesInDPB = avctx->refs;
+        hevc->maxNumRefFramesInDPB = ctx->dpb_size;
     }
     if (avctx->gop_size >= 0) {
         hevc->idrPeriod = cc->gopLength;
@@ -1074,6 +1124,11 @@ static av_cold int nvenc_setup_hevc_config(AVCodecContext *avctx)
     hevc->useBFramesAsRef = ctx->b_ref_mode;
 #endif
 
+#ifdef NVENC_HAVE_MULTIPLE_REF_FRAMES
+    hevc->numRefL0 = avctx->refs;
+    hevc->numRefL1 = avctx->refs;
+#endif
+
     return 0;
 }
 
@@ -1155,7 +1210,7 @@ static av_cold int nvenc_setup_encoder(AVCodecContext *avctx)
 
     if (ctx->bluray_compat) {
         ctx->aud = 1;
-        avctx->refs = FFMIN(FFMAX(avctx->refs, 0), 6);
+        ctx->dpb_size = FFMIN(FFMAX(avctx->refs, 0), 6);
         avctx->max_b_frames = FFMIN(avctx->max_b_frames, 3);
         switch (avctx->codec->id) {
         case AV_CODEC_ID_H264:
@@ -1202,15 +1257,25 @@ static av_cold int nvenc_setup_encoder(AVCodecContext *avctx)
         return res;
 
     nv_status = p_nvenc->nvEncInitializeEncoder(ctx->nvencoder, &ctx->init_encode_params);
+    if (nv_status != NV_ENC_SUCCESS) {
+        nvenc_pop_context(avctx);
+        return nvenc_print_error(avctx, nv_status, "InitializeEncoder failed");
+    }
+
+#ifdef NVENC_HAVE_CUSTREAM_PTR
+    if (ctx->cu_context) {
+        nv_status = p_nvenc->nvEncSetIOCudaStreams(ctx->nvencoder, &ctx->cu_stream, &ctx->cu_stream);
+        if (nv_status != NV_ENC_SUCCESS) {
+            nvenc_pop_context(avctx);
+            return nvenc_print_error(avctx, nv_status, "SetIOCudaStreams failed");
+        }
+    }
+#endif
 
     res = nvenc_pop_context(avctx);
     if (res < 0)
         return res;
 
-    if (nv_status != NV_ENC_SUCCESS) {
-        return nvenc_print_error(avctx, nv_status, "InitializeEncoder failed");
-    }
-
     if (ctx->encode_config.frameIntervalP > 1)
         avctx->has_b_frames = 2;
 
@@ -1548,19 +1613,23 @@ static int nvenc_find_free_reg_resource(AVCodecContext *avctx)
     NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
     NVENCSTATUS nv_status;
 
-    int i;
+    int i, first_round;
 
     if (ctx->nb_registered_frames == FF_ARRAY_ELEMS(ctx->registered_frames)) {
-        for (i = 0; i < ctx->nb_registered_frames; i++) {
-            if (!ctx->registered_frames[i].mapped) {
-                if (ctx->registered_frames[i].regptr) {
-                    nv_status = p_nvenc->nvEncUnregisterResource(ctx->nvencoder, ctx->registered_frames[i].regptr);
-                    if (nv_status != NV_ENC_SUCCESS)
-                        return nvenc_print_error(avctx, nv_status, "Failed unregistering unused input resource");
-                    ctx->registered_frames[i].ptr = NULL;
-                    ctx->registered_frames[i].regptr = NULL;
+        for (first_round = 1; first_round >= 0; first_round--) {
+            for (i = 0; i < ctx->nb_registered_frames; i++) {
+                if (!ctx->registered_frames[i].mapped) {
+                    if (ctx->registered_frames[i].regptr) {
+                        if (first_round)
+                            continue;
+                        nv_status = p_nvenc->nvEncUnregisterResource(ctx->nvencoder, ctx->registered_frames[i].regptr);
+                        if (nv_status != NV_ENC_SUCCESS)
+                            return nvenc_print_error(avctx, nv_status, "Failed unregistering unused input resource");
+                        ctx->registered_frames[i].ptr = NULL;
+                        ctx->registered_frames[i].regptr = NULL;
+                    }
+                    return i;
                 }
-                return i;
             }
         }
     } else {
@@ -1812,7 +1881,11 @@ static int process_output_surface(AVCodecContext *avctx, AVPacket *pkt, NvencSur
         goto error;
     }
 
-    if (res = ff_alloc_packet2(avctx, pkt, lock_params.bitstreamSizeInBytes,0)) {
+    res = pkt->data ?
+        ff_alloc_packet2(avctx, pkt, lock_params.bitstreamSizeInBytes, lock_params.bitstreamSizeInBytes) :
+        av_new_packet(pkt, lock_params.bitstreamSizeInBytes);
+
+    if (res < 0) {
         p_nvenc->nvEncUnlockBitstream(ctx->nvencoder, tmpoutsurf->output_surface);
         goto error;
     }
@@ -1834,13 +1907,6 @@ static int process_output_surface(AVCodecContext *avctx, AVPacket *pkt, NvencSur
                 res = nvenc_print_error(avctx, nv_status, "Failed unmapping input resource");
                 goto error;
             }
-            nv_status = p_nvenc->nvEncUnregisterResource(ctx->nvencoder, ctx->registered_frames[tmpoutsurf->reg_idx].regptr);
-            if (nv_status != NV_ENC_SUCCESS) {
-                res = nvenc_print_error(avctx, nv_status, "Failed unregistering input resource");
-                goto error;
-            }
-            ctx->registered_frames[tmpoutsurf->reg_idx].ptr = NULL;
-            ctx->registered_frames[tmpoutsurf->reg_idx].regptr = NULL;
         } else if (ctx->registered_frames[tmpoutsurf->reg_idx].mapped < 0) {
             res = AVERROR_BUG;
             goto error;
@@ -2196,3 +2262,8 @@ int ff_nvenc_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
 
     return 0;
 }
+
+av_cold void ff_nvenc_encode_flush(AVCodecContext *avctx)
+{
+    ff_nvenc_send_frame(avctx, NULL);
+}