From 84949792cbc547ac62ece52747b25ab3a7f9aca3 Mon Sep 17 00:00:00 2001 From: Oliver Collyer Date: Wed, 24 Aug 2016 10:46:55 +0100 Subject: [PATCH] NVENC - added support for 10 bit HEVC encoding --- configure | 4 +- libavcodec/nvenc.c | 133 ++++++++++++++++++++++++++++++++++++++++++++++-- libavcodec/nvenc.h | 5 ++ libavcodec/nvenc_hevc.c | 5 +- libavcodec/version.h | 2 +- 5 files changed, 140 insertions(+), 9 deletions(-) diff --git a/configure b/configure index 5b017fd..8342ba8 100755 --- a/configure +++ b/configure @@ -5777,8 +5777,8 @@ enabled mmal && check_func_headers interface/mmal/mmal.h "MMAL_PARAMETER_VIDEO_M enabled netcdf && require_pkg_config netcdf netcdf.h nc_inq_libvers enabled nvenc && { check_header nvEncodeAPI.h || die "ERROR: nvEncodeAPI.h not found."; } && - { check_cpp_condition nvEncodeAPI.h "NVENCAPI_MAJOR_VERSION >= 6" || - die "ERROR: NVENC API version 5 or older is not supported"; } && + { check_cpp_condition nvEncodeAPI.h "NVENCAPI_MAJOR_VERSION >= 7" || + die "ERROR: NVENC API version 6 or older is not supported"; } && { [ $target_os != cygwin ] || die "ERROR: NVENC is not supported on Cygwin currently."; } enabled openal && { { for al_libs in "${OPENAL_LIBS}" "-lopenal" "-lOpenAL32"; do check_lib 'AL/al.h' alGetError "${al_libs}" && break; done } || diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c index 984dd3b..d12d91b 100644 --- a/libavcodec/nvenc.c +++ b/libavcodec/nvenc.c @@ -75,14 +75,24 @@ const enum AVPixelFormat ff_nvenc_pix_fmts[] = { AV_PIX_FMT_YUV420P, + AV_PIX_FMT_YUV420P10, AV_PIX_FMT_NV12, + AV_PIX_FMT_P010, AV_PIX_FMT_YUV444P, + AV_PIX_FMT_YUV444P16, #if CONFIG_CUDA AV_PIX_FMT_CUDA, #endif AV_PIX_FMT_NONE }; +#define IS_10BIT(pix_fmt) (pix_fmt == AV_PIX_FMT_YUV420P10 || \ + pix_fmt == AV_PIX_FMT_P010 || \ + pix_fmt == AV_PIX_FMT_YUV444P16) + +#define IS_YUV444(pix_fmt) (pix_fmt == AV_PIX_FMT_YUV444P || \ + pix_fmt == AV_PIX_FMT_YUV444P16) + static const struct { NVENCSTATUS nverr; int averr; @@ -273,7 +283,7 @@ static int nvenc_check_capabilities(AVCodecContext *avctx) } ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_YUV444_ENCODE); - if (ctx->data_pix_fmt == AV_PIX_FMT_YUV444P && ret <= 0) { + if (IS_YUV444(ctx->data_pix_fmt) && ret <= 0) { av_log(avctx, AV_LOG_VERBOSE, "YUV444P not supported\n"); return AVERROR(ENOSYS); } @@ -314,6 +324,12 @@ static int nvenc_check_capabilities(AVCodecContext *avctx) return AVERROR(ENOSYS); } + 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"); + return AVERROR(ENOSYS); + } + return 0; } @@ -800,9 +816,26 @@ static av_cold int nvenc_setup_hevc_config(AVCodecContext *avctx) hevc->outputPictureTimingSEI = 1; } - /* No other profile is supported in the current SDK version 5 */ - cc->profileGUID = NV_ENC_HEVC_PROFILE_MAIN_GUID; - avctx->profile = FF_PROFILE_HEVC_MAIN; + switch(ctx->profile) { + case NV_ENC_HEVC_PROFILE_MAIN: + cc->profileGUID = NV_ENC_HEVC_PROFILE_MAIN_GUID; + avctx->profile = FF_PROFILE_HEVC_MAIN; + break; + case NV_ENC_HEVC_PROFILE_MAIN_10: + cc->profileGUID = NV_ENC_HEVC_PROFILE_MAIN10_GUID; + avctx->profile = FF_PROFILE_HEVC_MAIN_10; + break; + } + + // force setting profile as main10 if input is 10 bit + if (IS_10BIT(ctx->data_pix_fmt)) { + cc->profileGUID = NV_ENC_HEVC_PROFILE_MAIN10_GUID; + avctx->profile = FF_PROFILE_HEVC_MAIN_10; + } + + hevc->chromaFormatIDC = IS_YUV444(ctx->data_pix_fmt) ? 3 : 1; + + hevc->pixelBitDepthMinus8 = IS_10BIT(ctx->data_pix_fmt) ? 2 : 0; hevc->level = ctx->level; @@ -954,14 +987,26 @@ static av_cold int nvenc_alloc_surface(AVCodecContext *avctx, int idx) ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_YV12_PL; break; + case AV_PIX_FMT_YUV420P10: + ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_YUV420_10BIT; + break; + case AV_PIX_FMT_NV12: ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_NV12_PL; break; + case AV_PIX_FMT_P010: + ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_YUV420_10BIT; + break; + case AV_PIX_FMT_YUV444P: ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_YUV444_PL; break; + case AV_PIX_FMT_YUV444P16: + ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_YUV444_10BIT; + break; + default: av_log(avctx, AV_LOG_FATAL, "Invalid input pixel format\n"); return AVERROR(EINVAL); @@ -1206,6 +1251,49 @@ static NvencSurface *get_free_frame(NvencContext *ctx) return NULL; } +static void copy_single_10bit_plane(uint8_t *dst, int dst_linesize, + const uint8_t *src, int src_linesize, + int width, int height) +{ + if (!dst || !src) + return; + av_assert0(abs(src_linesize) >= width << 1); + av_assert0(abs(dst_linesize) >= width << 1); + for (;height > 0; height--) { + uint16_t* tdst = (uint16_t*)dst; + uint16_t* tsrc = (uint16_t*)src; + for (int w = width; w > 0; w--) { + *tdst++ = *tsrc++ << 6; + } + dst += dst_linesize; + src += src_linesize; + } +} + +static void interleave_10bit_planes(uint8_t *dst, int dst_linesize, + const uint8_t *src1, int src1_linesize, + const uint8_t *src2, int src2_linesize, + int width, int height) +{ + if (!dst || !src1 || !src2) + return; + av_assert0(abs(src1_linesize) >= width); + av_assert0(abs(src2_linesize) >= width); + av_assert0(abs(dst_linesize) >= width << 1); + for (;height > 0; height--) { + uint16_t* tdst = (uint16_t*)dst; + uint16_t* tsrc1 = (uint16_t*)src1; + uint16_t* tsrc2 = (uint16_t*)src2; + for (int w = width; w > 0; w-=2) { + *tdst++ = *tsrc1++ << 6; + *tdst++ = *tsrc2++ << 6; + } + dst += dst_linesize; + src1 += src1_linesize; + src2 += src2_linesize; + } +} + static int nvenc_copy_frame(AVCodecContext *avctx, NvencSurface *inSurf, NV_ENC_LOCK_INPUT_BUFFER *lockBufferParams, const AVFrame *frame) { @@ -1228,6 +1316,17 @@ static int nvenc_copy_frame(AVCodecContext *avctx, NvencSurface *inSurf, av_image_copy_plane(buf, lockBufferParams->pitch >> 1, frame->data[1], frame->linesize[1], avctx->width >> 1, avctx->height >> 1); + } else if (frame->format == AV_PIX_FMT_YUV420P10) { + copy_single_10bit_plane(buf, lockBufferParams->pitch, + frame->data[0], frame->linesize[0], + avctx->width, avctx->height); + + buf += off; + + interleave_10bit_planes(buf, lockBufferParams->pitch, + frame->data[1], frame->linesize[1], + frame->data[2], frame->linesize[2], + avctx->width, avctx->height >> 1); } else if (frame->format == AV_PIX_FMT_NV12) { av_image_copy_plane(buf, lockBufferParams->pitch, frame->data[0], frame->linesize[0], @@ -1238,6 +1337,16 @@ static int nvenc_copy_frame(AVCodecContext *avctx, NvencSurface *inSurf, av_image_copy_plane(buf, lockBufferParams->pitch, frame->data[1], frame->linesize[1], avctx->width, avctx->height >> 1); + } else if (frame->format == AV_PIX_FMT_P010) { + av_image_copy_plane(buf, lockBufferParams->pitch, + frame->data[0], frame->linesize[0], + avctx->width << 1, avctx->height); + + buf += off; + + av_image_copy_plane(buf, lockBufferParams->pitch, + frame->data[1], frame->linesize[1], + avctx->width << 1, avctx->height >> 1); } else if (frame->format == AV_PIX_FMT_YUV444P) { av_image_copy_plane(buf, lockBufferParams->pitch, frame->data[0], frame->linesize[0], @@ -1254,6 +1363,22 @@ static int nvenc_copy_frame(AVCodecContext *avctx, NvencSurface *inSurf, av_image_copy_plane(buf, lockBufferParams->pitch, frame->data[2], frame->linesize[2], avctx->width, avctx->height); + } else if (frame->format == AV_PIX_FMT_YUV444P16) { + av_image_copy_plane(buf, lockBufferParams->pitch, + frame->data[0], frame->linesize[0], + avctx->width << 1, avctx->height); + + buf += off; + + av_image_copy_plane(buf, lockBufferParams->pitch, + frame->data[1], frame->linesize[1], + avctx->width << 1, avctx->height); + + buf += off; + + av_image_copy_plane(buf, lockBufferParams->pitch, + frame->data[2], frame->linesize[2], + avctx->width << 1, avctx->height); } else { av_log(avctx, AV_LOG_FATAL, "Invalid pixel format!\n"); return AVERROR(EINVAL); diff --git a/libavcodec/nvenc.h b/libavcodec/nvenc.h index 961cbc7..3cf0309 100644 --- a/libavcodec/nvenc.h +++ b/libavcodec/nvenc.h @@ -117,6 +117,11 @@ enum { }; enum { + NV_ENC_HEVC_PROFILE_MAIN, + NV_ENC_HEVC_PROFILE_MAIN_10, +}; + +enum { NVENC_LOWLATENCY = 1, NVENC_LOSSLESS = 2, NVENC_ONE_PASS = 4, diff --git a/libavcodec/nvenc_hevc.c b/libavcodec/nvenc_hevc.c index 1ce7c89..d9a4bf6 100644 --- a/libavcodec/nvenc_hevc.c +++ b/libavcodec/nvenc_hevc.c @@ -39,8 +39,9 @@ static const AVOption options[] = { { "llhp", "low latency hp", 0, AV_OPT_TYPE_CONST, { .i64 = PRESET_LOW_LATENCY_HP }, 0, 0, VE, "preset" }, { "lossless", "lossless", 0, AV_OPT_TYPE_CONST, { .i64 = PRESET_LOSSLESS_DEFAULT }, 0, 0, VE, "preset" }, { "losslesshp", "lossless hp", 0, AV_OPT_TYPE_CONST, { .i64 = PRESET_LOSSLESS_HP }, 0, 0, VE, "preset" }, - { "profile", "Set the encoding profile", OFFSET(profile), AV_OPT_TYPE_INT, { .i64 = FF_PROFILE_HEVC_MAIN }, FF_PROFILE_HEVC_MAIN, FF_PROFILE_HEVC_MAIN, VE, "profile" }, - { "main", "", 0, AV_OPT_TYPE_CONST, { .i64 = FF_PROFILE_HEVC_MAIN }, 0, 0, VE, "profile" }, + { "profile", "Set the encoding profile", OFFSET(profile), AV_OPT_TYPE_INT, { .i64 = NV_ENC_HEVC_PROFILE_MAIN }, NV_ENC_HEVC_PROFILE_MAIN, FF_PROFILE_HEVC_MAIN_10, VE, "profile" }, + { "main", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_HEVC_PROFILE_MAIN }, 0, 0, VE, "profile" }, + { "main10", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_HEVC_PROFILE_MAIN_10 }, 0, 0, VE, "profile" }, { "level", "Set the encoding level restriction", OFFSET(level), AV_OPT_TYPE_INT, { .i64 = NV_ENC_LEVEL_AUTOSELECT }, NV_ENC_LEVEL_AUTOSELECT, NV_ENC_LEVEL_HEVC_62, VE, "level" }, { "auto", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_LEVEL_AUTOSELECT }, 0, 0, VE, "level" }, { "1", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_LEVEL_HEVC_1 }, 0, 0, VE, "level" }, diff --git a/libavcodec/version.h b/libavcodec/version.h index 7ee5b5a..f7a7344 100644 --- a/libavcodec/version.h +++ b/libavcodec/version.h @@ -29,7 +29,7 @@ #define LIBAVCODEC_VERSION_MAJOR 57 #define LIBAVCODEC_VERSION_MINOR 54 -#define LIBAVCODEC_VERSION_MICRO 100 +#define LIBAVCODEC_VERSION_MICRO 101 #define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \ LIBAVCODEC_VERSION_MINOR, \ -- 2.7.4