summaryrefslogtreecommitdiff
path: root/libavcodec/libx264.c
diff options
context:
space:
mode:
Diffstat (limited to 'libavcodec/libx264.c')
-rw-r--r--libavcodec/libx264.c148
1 files changed, 74 insertions, 74 deletions
diff --git a/libavcodec/libx264.c b/libavcodec/libx264.c
index 894a63eadb..7410112172 100644
--- a/libavcodec/libx264.c
+++ b/libavcodec/libx264.c
@@ -127,30 +127,6 @@ static int encode_nals(AVCodecContext *ctx, uint8_t *buf, int size,
return p - buf;
}
-static int avfmt2_csp(int avfmt)
-{
- switch (avfmt) {
- case PIX_FMT_YUV420P:
- case PIX_FMT_YUVJ420P:
- case PIX_FMT_YUV420P9:
- case PIX_FMT_YUV420P10:
- return X264_CSP_I420;
-#ifdef X264_CSP_I444
- case PIX_FMT_YUV444P:
- return X264_CSP_I444;
-#endif
-#ifdef X264_CSP_BGR
- case PIX_FMT_BGR24:
- return X264_CSP_BGR;
-
- case PIX_FMT_RGB24:
- return X264_CSP_RGB;
-#endif
- default:
- return X264_CSP_NONE;
- }
-}
-
static int avfmt2_num_planes(int avfmt)
{
switch (avfmt) {
@@ -181,7 +157,7 @@ static int X264_frame(AVCodecContext *ctx, uint8_t *buf,
int bufsize;
x264_picture_init( &x4->pic );
- x4->pic.img.i_csp = avfmt2_csp(ctx->pix_fmt);
+ x4->pic.img.i_csp = x4->params.i_csp;
if (x264_bit_depth > 8)
x4->pic.img.i_csp |= X264_CSP_HIGH_DEPTH;
x4->pic.img.i_plane = avfmt2_num_planes(ctx->pix_fmt);
@@ -271,6 +247,29 @@ static av_cold int X264_close(AVCodecContext *avctx)
} \
} while (0);
+static int convert_pix_fmt(enum PixelFormat pix_fmt)
+{
+ switch (pix_fmt) {
+ case PIX_FMT_YUV420P:
+ case PIX_FMT_YUVJ420P:
+ case PIX_FMT_YUV420P9:
+ case PIX_FMT_YUV420P10: return X264_CSP_I420;
+ case PIX_FMT_YUV422P:
+ case PIX_FMT_YUV422P10: return X264_CSP_I422;
+ case PIX_FMT_YUV444P:
+ case PIX_FMT_YUV444P9:
+ case PIX_FMT_YUV444P10: return X264_CSP_I444;
+#ifdef X264_CSP_BGR
+ case PIX_FMT_BGR24:
+ return X264_CSP_BGR;
+
+ case PIX_FMT_RGB24:
+ return X264_CSP_RGB;
+#endif
+ };
+ return 0;
+}
+
#define PARSE_X264_OPT(name, var)\
if (x4->var && x264_param_parse(&x4->params, name, x4->var) < 0) {\
av_log(avctx, AV_LOG_ERROR, "Error parsing option '%s' with value '%s'.\n", name, x4->var);\
@@ -300,6 +299,7 @@ static av_cold int X264_init(AVCodecContext *avctx)
x4->params.pf_log = X264_log;
x4->params.p_log_private = avctx;
x4->params.i_log_level = X264_LOG_DEBUG;
+ x4->params.i_csp = convert_pix_fmt(avctx->pix_fmt);
OPT_STR("weightp", x4->wpredp);
@@ -533,8 +533,6 @@ static av_cold int X264_init(AVCodecContext *avctx)
avctx->crf = x4->params.rc.f_rf_constant;
#endif
- x4->params.i_csp = avfmt2_csp(avctx->pix_fmt);
-
x4->enc = x264_encoder_open(&x4->params);
if (!x4->enc)
return -1;
@@ -561,9 +559,8 @@ static av_cold int X264_init(AVCodecContext *avctx)
static const enum PixelFormat pix_fmts_8bit[] = {
PIX_FMT_YUV420P,
PIX_FMT_YUVJ420P,
-#ifdef X264_CSP_I444
+ PIX_FMT_YUV422P,
PIX_FMT_YUV444P,
-#endif
#ifdef X264_CSP_BGR
PIX_FMT_BGR24,
PIX_FMT_RGB24,
@@ -572,10 +569,13 @@ static const enum PixelFormat pix_fmts_8bit[] = {
};
static const enum PixelFormat pix_fmts_9bit[] = {
PIX_FMT_YUV420P9,
+ PIX_FMT_YUV444P9,
PIX_FMT_NONE
};
static const enum PixelFormat pix_fmts_10bit[] = {
PIX_FMT_YUV420P10,
+ PIX_FMT_YUV422P10,
+ PIX_FMT_YUV444P10,
PIX_FMT_NONE
};
@@ -592,52 +592,52 @@ static av_cold void X264_init_static(AVCodec *codec)
#define OFFSET(x) offsetof(X264Context, x)
#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
static const AVOption options[] = {
- { "preset", "Set the encoding preset (cf. x264 --fullhelp)", OFFSET(preset), FF_OPT_TYPE_STRING, { .str = "medium" }, 0, 0, VE},
- { "tune", "Tune the encoding params (cf. x264 --fullhelp)", OFFSET(tune), FF_OPT_TYPE_STRING, { 0 }, 0, 0, VE},
- { "profile", "Set profile restrictions (cf. x264 --fullhelp) ", OFFSET(profile), FF_OPT_TYPE_STRING, { 0 }, 0, 0, VE},
- { "fastfirstpass", "Use fast settings when encoding first pass", OFFSET(fastfirstpass), FF_OPT_TYPE_INT, { 1 }, 0, 1, VE},
- {"level", "Specify level (as defined by Annex A)", OFFSET(level), FF_OPT_TYPE_STRING, {.str=NULL}, 0, 0, VE},
- {"passlogfile", "Filename for 2 pass stats", OFFSET(stats), FF_OPT_TYPE_STRING, {.str=NULL}, 0, 0, VE},
- {"wpredp", "Weighted prediction for P-frames", OFFSET(wpredp), FF_OPT_TYPE_STRING, {.str=NULL}, 0, 0, VE},
- {"x264opts", "x264 options", OFFSET(x264opts), FF_OPT_TYPE_STRING, {.str=NULL}, 0, 0, VE},
- { "crf", "Select the quality for constant quality mode", OFFSET(crf), FF_OPT_TYPE_FLOAT, {-1 }, -1, FLT_MAX, VE },
- { "crf_max", "In CRF mode, prevents VBV from lowering quality beyond this point.",OFFSET(crf_max), FF_OPT_TYPE_FLOAT, {-1 }, -1, FLT_MAX, VE },
- { "qp", "Constant quantization parameter rate control method",OFFSET(cqp), FF_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE },
- { "aq-mode", "AQ method", OFFSET(aq_mode), FF_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE, "aq_mode"},
- { "none", NULL, 0, FF_OPT_TYPE_CONST, {X264_AQ_NONE}, INT_MIN, INT_MAX, VE, "aq_mode" },
- { "variance", "Variance AQ (complexity mask)", 0, FF_OPT_TYPE_CONST, {X264_AQ_VARIANCE}, INT_MIN, INT_MAX, VE, "aq_mode" },
- { "autovariance", "Auto-variance AQ (experimental)", 0, FF_OPT_TYPE_CONST, {X264_AQ_AUTOVARIANCE}, INT_MIN, INT_MAX, VE, "aq_mode" },
- { "aq-strength", "AQ strength. Reduces blocking and blurring in flat and textured areas.", OFFSET(aq_strength), FF_OPT_TYPE_FLOAT, {-1}, -1, FLT_MAX, VE},
- { "psy", "Use psychovisual optimizations.", OFFSET(psy), FF_OPT_TYPE_INT, {-1 }, -1, 1, VE },
- { "psy-rd", "Strength of psychovisual optimization, in <psy-rd>:<psy-trellis> format.", OFFSET(psy_rd), FF_OPT_TYPE_STRING, {0 }, 0, 0, VE},
- { "rc-lookahead", "Number of frames to look ahead for frametype and ratecontrol", OFFSET(rc_lookahead), FF_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE },
- { "weightb", "Weighted prediction for B-frames.", OFFSET(weightb), FF_OPT_TYPE_INT, {-1 }, -1, 1, VE },
- { "weightp", "Weighted prediction analysis method.", OFFSET(weightp), FF_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE, "weightp" },
- { "none", NULL, 0, FF_OPT_TYPE_CONST, {X264_WEIGHTP_NONE}, INT_MIN, INT_MAX, VE, "weightp" },
- { "simple", NULL, 0, FF_OPT_TYPE_CONST, {X264_WEIGHTP_SIMPLE}, INT_MIN, INT_MAX, VE, "weightp" },
- { "smart", NULL, 0, FF_OPT_TYPE_CONST, {X264_WEIGHTP_SMART}, INT_MIN, INT_MAX, VE, "weightp" },
- { "ssim", "Calculate and print SSIM stats.", OFFSET(ssim), FF_OPT_TYPE_INT, {-1 }, -1, 1, VE },
- { "intra-refresh", "Use Periodic Intra Refresh instead of IDR frames.",OFFSET(intra_refresh),FF_OPT_TYPE_INT, {-1 }, -1, 1, VE },
- { "b-bias", "Influences how often B-frames are used", OFFSET(b_bias), FF_OPT_TYPE_INT, {INT_MIN}, INT_MIN, INT_MAX, VE },
- { "b-pyramid", "Keep some B-frames as references.", OFFSET(b_pyramid), FF_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE, "b_pyramid" },
- { "none", NULL, 0, FF_OPT_TYPE_CONST, {X264_B_PYRAMID_NONE}, INT_MIN, INT_MAX, VE, "b_pyramid" },
- { "strict", "Strictly hierarchical pyramid", 0, FF_OPT_TYPE_CONST, {X264_B_PYRAMID_STRICT}, INT_MIN, INT_MAX, VE, "b_pyramid" },
- { "normal", "Non-strict (not Blu-ray compatible)", 0, FF_OPT_TYPE_CONST, {X264_B_PYRAMID_NORMAL}, INT_MIN, INT_MAX, VE, "b_pyramid" },
- { "mixed-refs", "One reference per partition, as opposed to one reference per macroblock", OFFSET(mixed_refs), FF_OPT_TYPE_INT, {-1}, -1, 1, VE },
- { "8x8dct", "High profile 8x8 transform.", OFFSET(dct8x8), FF_OPT_TYPE_INT, {-1 }, -1, 1, VE},
- { "fast-pskip", NULL, OFFSET(fast_pskip), FF_OPT_TYPE_INT, {-1 }, -1, 1, VE},
- { "aud", "Use access unit delimiters.", OFFSET(aud), FF_OPT_TYPE_INT, {-1 }, -1, 1, VE},
- { "mbtree", "Use macroblock tree ratecontrol.", OFFSET(mbtree), FF_OPT_TYPE_INT, {-1 }, -1, 1, VE},
- { "deblock", "Loop filter parameters, in <alpha:beta> form.", OFFSET(deblock), FF_OPT_TYPE_STRING, { 0 }, 0, 0, VE},
- { "cplxblur", "Reduce fluctuations in QP (before curve compression)", OFFSET(cplxblur), FF_OPT_TYPE_FLOAT, {-1 }, -1, FLT_MAX, VE},
+ { "preset", "Set the encoding preset (cf. x264 --fullhelp)", OFFSET(preset), AV_OPT_TYPE_STRING, { .str = "medium" }, 0, 0, VE},
+ { "tune", "Tune the encoding params (cf. x264 --fullhelp)", OFFSET(tune), AV_OPT_TYPE_STRING, { 0 }, 0, 0, VE},
+ { "profile", "Set profile restrictions (cf. x264 --fullhelp) ", OFFSET(profile), AV_OPT_TYPE_STRING, { 0 }, 0, 0, VE},
+ { "fastfirstpass", "Use fast settings when encoding first pass", OFFSET(fastfirstpass), AV_OPT_TYPE_INT, { 1 }, 0, 1, VE},
+ {"level", "Specify level (as defined by Annex A)", OFFSET(level), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, VE},
+ {"passlogfile", "Filename for 2 pass stats", OFFSET(stats), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, VE},
+ {"wpredp", "Weighted prediction for P-frames", OFFSET(wpredp), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, VE},
+ {"x264opts", "x264 options", OFFSET(x264opts), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, VE},
+ { "crf", "Select the quality for constant quality mode", OFFSET(crf), AV_OPT_TYPE_FLOAT, {-1 }, -1, FLT_MAX, VE },
+ { "crf_max", "In CRF mode, prevents VBV from lowering quality beyond this point.",OFFSET(crf_max), AV_OPT_TYPE_FLOAT, {-1 }, -1, FLT_MAX, VE },
+ { "qp", "Constant quantization parameter rate control method",OFFSET(cqp), AV_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE },
+ { "aq-mode", "AQ method", OFFSET(aq_mode), AV_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE, "aq_mode"},
+ { "none", NULL, 0, AV_OPT_TYPE_CONST, {X264_AQ_NONE}, INT_MIN, INT_MAX, VE, "aq_mode" },
+ { "variance", "Variance AQ (complexity mask)", 0, AV_OPT_TYPE_CONST, {X264_AQ_VARIANCE}, INT_MIN, INT_MAX, VE, "aq_mode" },
+ { "autovariance", "Auto-variance AQ (experimental)", 0, AV_OPT_TYPE_CONST, {X264_AQ_AUTOVARIANCE}, INT_MIN, INT_MAX, VE, "aq_mode" },
+ { "aq-strength", "AQ strength. Reduces blocking and blurring in flat and textured areas.", OFFSET(aq_strength), AV_OPT_TYPE_FLOAT, {-1}, -1, FLT_MAX, VE},
+ { "psy", "Use psychovisual optimizations.", OFFSET(psy), AV_OPT_TYPE_INT, {-1 }, -1, 1, VE },
+ { "psy-rd", "Strength of psychovisual optimization, in <psy-rd>:<psy-trellis> format.", OFFSET(psy_rd), AV_OPT_TYPE_STRING, {0 }, 0, 0, VE},
+ { "rc-lookahead", "Number of frames to look ahead for frametype and ratecontrol", OFFSET(rc_lookahead), AV_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE },
+ { "weightb", "Weighted prediction for B-frames.", OFFSET(weightb), AV_OPT_TYPE_INT, {-1 }, -1, 1, VE },
+ { "weightp", "Weighted prediction analysis method.", OFFSET(weightp), AV_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE, "weightp" },
+ { "none", NULL, 0, AV_OPT_TYPE_CONST, {X264_WEIGHTP_NONE}, INT_MIN, INT_MAX, VE, "weightp" },
+ { "simple", NULL, 0, AV_OPT_TYPE_CONST, {X264_WEIGHTP_SIMPLE}, INT_MIN, INT_MAX, VE, "weightp" },
+ { "smart", NULL, 0, AV_OPT_TYPE_CONST, {X264_WEIGHTP_SMART}, INT_MIN, INT_MAX, VE, "weightp" },
+ { "ssim", "Calculate and print SSIM stats.", OFFSET(ssim), AV_OPT_TYPE_INT, {-1 }, -1, 1, VE },
+ { "intra-refresh", "Use Periodic Intra Refresh instead of IDR frames.",OFFSET(intra_refresh),AV_OPT_TYPE_INT, {-1 }, -1, 1, VE },
+ { "b-bias", "Influences how often B-frames are used", OFFSET(b_bias), AV_OPT_TYPE_INT, {INT_MIN}, INT_MIN, INT_MAX, VE },
+ { "b-pyramid", "Keep some B-frames as references.", OFFSET(b_pyramid), AV_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE, "b_pyramid" },
+ { "none", NULL, 0, AV_OPT_TYPE_CONST, {X264_B_PYRAMID_NONE}, INT_MIN, INT_MAX, VE, "b_pyramid" },
+ { "strict", "Strictly hierarchical pyramid", 0, AV_OPT_TYPE_CONST, {X264_B_PYRAMID_STRICT}, INT_MIN, INT_MAX, VE, "b_pyramid" },
+ { "normal", "Non-strict (not Blu-ray compatible)", 0, AV_OPT_TYPE_CONST, {X264_B_PYRAMID_NORMAL}, INT_MIN, INT_MAX, VE, "b_pyramid" },
+ { "mixed-refs", "One reference per partition, as opposed to one reference per macroblock", OFFSET(mixed_refs), AV_OPT_TYPE_INT, {-1}, -1, 1, VE },
+ { "8x8dct", "High profile 8x8 transform.", OFFSET(dct8x8), AV_OPT_TYPE_INT, {-1 }, -1, 1, VE},
+ { "fast-pskip", NULL, OFFSET(fast_pskip), AV_OPT_TYPE_INT, {-1 }, -1, 1, VE},
+ { "aud", "Use access unit delimiters.", OFFSET(aud), AV_OPT_TYPE_INT, {-1 }, -1, 1, VE},
+ { "mbtree", "Use macroblock tree ratecontrol.", OFFSET(mbtree), AV_OPT_TYPE_INT, {-1 }, -1, 1, VE},
+ { "deblock", "Loop filter parameters, in <alpha:beta> form.", OFFSET(deblock), AV_OPT_TYPE_STRING, { 0 }, 0, 0, VE},
+ { "cplxblur", "Reduce fluctuations in QP (before curve compression)", OFFSET(cplxblur), AV_OPT_TYPE_FLOAT, {-1 }, -1, FLT_MAX, VE},
{ "partitions", "A comma-separated list of partitions to consider. "
- "Possible values: p8x8, p4x4, b8x8, i8x8, i4x4, none, all", OFFSET(partitions), FF_OPT_TYPE_STRING, { 0 }, 0, 0, VE},
- { "direct-pred", "Direct MV prediction mode", OFFSET(direct_pred), FF_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE, "direct-pred" },
- { "none", NULL, 0, FF_OPT_TYPE_CONST, { X264_DIRECT_PRED_NONE }, 0, 0, VE, "direct-pred" },
- { "spatial", NULL, 0, FF_OPT_TYPE_CONST, { X264_DIRECT_PRED_SPATIAL }, 0, 0, VE, "direct-pred" },
- { "temporal", NULL, 0, FF_OPT_TYPE_CONST, { X264_DIRECT_PRED_TEMPORAL }, 0, 0, VE, "direct-pred" },
- { "auto", NULL, 0, FF_OPT_TYPE_CONST, { X264_DIRECT_PRED_AUTO }, 0, 0, VE, "direct-pred" },
- { "slice-max-size","Constant quantization parameter rate control method",OFFSET(slice_max_size), FF_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE },
+ "Possible values: p8x8, p4x4, b8x8, i8x8, i4x4, none, all", OFFSET(partitions), AV_OPT_TYPE_STRING, { 0 }, 0, 0, VE},
+ { "direct-pred", "Direct MV prediction mode", OFFSET(direct_pred), AV_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE, "direct-pred" },
+ { "none", NULL, 0, AV_OPT_TYPE_CONST, { X264_DIRECT_PRED_NONE }, 0, 0, VE, "direct-pred" },
+ { "spatial", NULL, 0, AV_OPT_TYPE_CONST, { X264_DIRECT_PRED_SPATIAL }, 0, 0, VE, "direct-pred" },
+ { "temporal", NULL, 0, AV_OPT_TYPE_CONST, { X264_DIRECT_PRED_TEMPORAL }, 0, 0, VE, "direct-pred" },
+ { "auto", NULL, 0, AV_OPT_TYPE_CONST, { X264_DIRECT_PRED_AUTO }, 0, 0, VE, "direct-pred" },
+ { "slice-max-size","Constant quantization parameter rate control method",OFFSET(slice_max_size), AV_OPT_TYPE_INT, {-1 }, -1, INT_MAX, VE },
{ NULL },
};