| Message ID | ab046eb4-81ec-ce1d-5c58-bec1b3092710@gmail.com |
|---|---|
| State | New |
| Headers | show |
On 2017/5/14 12:26, Jun Zhao wrote: > V3: - Fix build error with old VAAPI version. > V2: - Refine the name/value type to mb_rate_control/bool. > - Only supported GEN9+ (SKL/APL/KBL/...) > - i965 driver default use frame-level rate control algorithm (generate the QP for each frame), > when enable mb_rate_control, it's will enable the MB-level RC algorithm (generate the QP for each MB). > - enables MB-level bitrate control that generally improves subjective visual quality, > but have negative impact on performance and objective visual quality metric. > Ping ?
On 19/05/17 00:43, Jun Zhao wrote: > > On 2017/5/14 12:26, Jun Zhao wrote: >> V3: - Fix build error with old VAAPI version. >> V2: - Refine the name/value type to mb_rate_control/bool. >> - Only supported GEN9+ (SKL/APL/KBL/...) >> - i965 driver default use frame-level rate control algorithm (generate the QP for each frame), >> when enable mb_rate_control, it's will enable the MB-level RC algorithm (generate the QP for each MB). >> - enables MB-level bitrate control that generally improves subjective visual quality, >> but have negative impact on performance and objective visual quality metric. >> > > Ping ? I fixed the default/enable/disable inconsistency in <https://github.com/01org/intel-vaapi-driver/commit/0dd6ba4e26d3abaa7e66265284ef81c746c13f20>. Using 0/1 as default/on is probably fine, but it doesn't match the statement that it's definitely disabled if the option is zero. (I don't mind which way around that gets fixed.) I tested with this patch and it works fine on a Skylake GT3 (6260U), but it killed the whole machine when enabled on a Skylake GT2 (6300). I haven't yet had chance to investigate that further (rebooting repeatedly is kindof inconvenient), but I will try to do so soon. What platforms have you tested on? - Mark
On 19/05/17 14:10, Mark Thompson wrote: > On 19/05/17 00:43, Jun Zhao wrote: >> >> On 2017/5/14 12:26, Jun Zhao wrote: >>> V3: - Fix build error with old VAAPI version. >>> V2: - Refine the name/value type to mb_rate_control/bool. >>> - Only supported GEN9+ (SKL/APL/KBL/...) >>> - i965 driver default use frame-level rate control algorithm (generate the QP for each frame), >>> when enable mb_rate_control, it's will enable the MB-level RC algorithm (generate the QP for each MB). >>> - enables MB-level bitrate control that generally improves subjective visual quality, >>> but have negative impact on performance and objective visual quality metric. >>> >> >> Ping ? > > I fixed the default/enable/disable inconsistency in <https://github.com/01org/intel-vaapi-driver/commit/0dd6ba4e26d3abaa7e66265284ef81c746c13f20>. Using 0/1 as default/on is probably fine, but it doesn't match the statement that it's definitely disabled if the option is zero. (I don't mind which way around that gets fixed.) > > I tested with this patch and it works fine on a Skylake GT3 (6260U), but it killed the whole machine when enabled on a Skylake GT2 (6300). I haven't yet had chance to investigate that further (rebooting repeatedly is kindof inconvenient), but I will try to do so soon. This failure is completely consistent. Reported here: <https://github.com/01org/intel-vaapi-driver/issues/172>. You will understand that I am reluctant to commit an option which nukes my machine, so lets hold off on this patch while the Intel driver people look at it. Thanks, - Mark
On 2017/5/20 20:32, Mark Thompson wrote: > On 19/05/17 14:10, Mark Thompson wrote: >> On 19/05/17 00:43, Jun Zhao wrote: >>> >>> On 2017/5/14 12:26, Jun Zhao wrote: >>>> V3: - Fix build error with old VAAPI version. >>>> V2: - Refine the name/value type to mb_rate_control/bool. >>>> - Only supported GEN9+ (SKL/APL/KBL/...) >>>> - i965 driver default use frame-level rate control algorithm (generate the QP for each frame), >>>> when enable mb_rate_control, it's will enable the MB-level RC algorithm (generate the QP for each MB). >>>> - enables MB-level bitrate control that generally improves subjective visual quality, >>>> but have negative impact on performance and objective visual quality metric. >>>> >>> >>> Ping ? >> >> I fixed the default/enable/disable inconsistency in <https://github.com/01org/intel-vaapi-driver/commit/0dd6ba4e26d3abaa7e66265284ef81c746c13f20>. Using 0/1 as default/on is probably fine, but it doesn't match the statement that it's definitely disabled if the option is zero. (I don't mind which way around that gets fixed.) >> >> I tested with this patch and it works fine on a Skylake GT3 (6260U), but it killed the whole machine when enabled on a Skylake GT2 (6300). I haven't yet had chance to investigate that further (rebooting repeatedly is kindof inconvenient), but I will try to do so soon. > > This failure is completely consistent. Reported here: <https://github.com/01org/intel-vaapi-driver/issues/172>. > > You will understand that I am reluctant to commit an option which nukes my machine, so lets hold off on this patch while the Intel driver people look at it. > > Thanks, > Got it，Thanks , Mark. BTW, I have enabled the multi slice encoding for VAAPI AVC/HEVC, maybe I can send a early version for code review. > - Mark > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel >
diff --git a/libavcodec/vaapi_encode_h264.c b/libavcodec/vaapi_encode_h264.c index 92e29554ed..44223ada55 100644 --- a/libavcodec/vaapi_encode_h264.c +++ b/libavcodec/vaapi_encode_h264.c @@ -168,6 +168,7 @@ typedef struct VAAPIEncodeH264Options { int qp; int quality; int low_power; + int mb_rate_control; } VAAPIEncodeH264Options; @@ -1133,8 +1134,22 @@ static av_cold int vaapi_encode_h264_configure(AVCodecContext *avctx) priv->fixed_qp_p = 26; priv->fixed_qp_b = 26; - av_log(avctx, AV_LOG_DEBUG, "Using %s-bitrate = %"PRId64" bps.\n", + if (opt->mb_rate_control) { +#if VA_CHECK_VERSION(0, 39, 2) + ctx->rc_params.rc.rc_flags.bits.mb_rate_control = opt->mb_rate_control; +#else + av_log(avctx, AV_LOG_WARNING, "The MB rate control option is not " + "supported with this VAAPI version.\n"); +#endif + } + + av_log(avctx, AV_LOG_DEBUG, "Using %s-bitrate %s MB rate control = %"PRId64" bps.\n", ctx->va_rc_mode == VA_RC_CBR ? "constant" : "variable", +#if VA_CHECK_VERSION(0, 39, 2) + opt->mb_rate_control ? "with" : "without", +#else + "without", +#endif avctx->bit_rate); } else { @@ -1283,6 +1298,8 @@ static const AVOption vaapi_encode_h264_options[] = { { "low_power", "Use low-power encoding mode (experimental: only supported " "on some platforms, does not support all features)", OFFSET(low_power), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS }, + { "mb_rate_control", "MB level bitrate control (only supported on GEN9+)", + OFFSET(mb_rate_control), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS, "mb_rate_control" }, { NULL }, };
V3: - Fix build error with old VAAPI version. V2: - Refine the name/value type to mb_rate_control/bool. - Only supported GEN9+ (SKL/APL/KBL/...) - i965 driver default use frame-level rate control algorithm (generate the QP for each frame), when enable mb_rate_control, it's will enable the MB-level RC algorithm (generate the QP for each MB). - enables MB-level bitrate control that generally improves subjective visual quality, but have negative impact on performance and objective visual quality metric. From 219b06cec04a6c3f719ce80084a98b19456498e0 Mon Sep 17 00:00:00 2001 From: Jun Zhao <jun.zhao@intel.com> Date: Tue, 9 May 2017 08:19:16 +0800 Subject: [PATCH V3] lavc/vaapi_encode_h264: Enable MB rate control. Enables macroblock-level bitrate control that generally improves subjective visual quality. It may have a negative impact on performance and objective visual quality metrics. Default is off and can't compatible with Constant QP. Signed-off-by: Jun Zhao <jun.zhao@intel.com> --- libavcodec/vaapi_encode_h264.c | 19 ++++++++++++++++++- 1 file changed, 18 insertions(+), 1 deletion(-)