swresample. Incorrect work of swr_set_matrix() if the SwrContext was never initialized

[v0lt]

Summary of the bug:

Function swr_set_matrix() muted front channels when mixing from AV_CH_LAYOUT_6POINT1 in AV_CH_LAYOUT_STEREO if the SwrContext was never initialized.

How to reproduce:

SwrContext* m_pSWRCxt = swr_alloc();

for (int i = 1; i <= 2; i++) {
	int ret = 0;

	swr_close(m_pSWRCxt);

	av_opt_set_int(m_pSWRCxt, "in_sample_fmt", AV_SAMPLE_FMT_S16, 0);
	av_opt_set_int(m_pSWRCxt, "out_sample_fmt", AV_SAMPLE_FMT_FLT, 0);
	av_opt_set_int(m_pSWRCxt, "in_channel_layout", AV_CH_LAYOUT_6POINT1, 0);
	av_opt_set_int(m_pSWRCxt, "out_channel_layout", AV_CH_LAYOUT_STEREO, 0);
	av_opt_set_int(m_pSWRCxt, "in_sample_rate", 48000, 0);
	av_opt_set_int(m_pSWRCxt, "out_sample_rate", 48000, 0);

	const int in_ch  = av_popcount(AV_CH_LAYOUT_6POINT1);
	const int out_ch = av_popcount(AV_CH_LAYOUT_STEREO);
	double* m_matrix_dbl = (double*)av_mallocz(in_ch * out_ch * sizeof(*m_matrix_dbl));

	const double center_mix_level   = M_SQRT1_2;
	const double surround_mix_level = 1.0;
	const double lfe_mix_level      = 1.0;
	const double rematrix_maxval    = INT_MAX;
	const double rematrix_volume    = 0.0;
	ret = swr_build_matrix(
		m_in_layout, m_out_layout,
		center_mix_level, surround_mix_level, lfe_mix_level,
		rematrix_maxval, rematrix_volume,
		m_matrix_dbl, in_ch,
		AV_MATRIX_ENCODING_NONE, NULL);

	ret = swr_set_matrix(m_pSWRCxt, m_matrix_dbl, in_ch);

	ret = swr_init(m_pSWRCxt);

	// working...
	// if i = 1 get the silence.
	// if i = 2 get the sound.
}

swr_free(&m_pSWRCxt);
av_free(m_matrix_dbl);

[Carl Eugen Hoyos]

Is this a regression?

[Carl Eugen Hoyos]

Please send your patch to the development mailing list, patches are ignored on this bug tracker.

[v0lt]

I do not understand this.

[Carl Eugen Hoyos]

I believe you attached a patch here on the bug tracker that fixes the issue that you reported. (Is this correct?)
Unfortunately, patches are ignored on this bug tracker, so please send the patch to the development mailing list to have it reviewed: ​http://ffmpeg.org/mailman/listinfo/ffmpeg-devel

[v0lt]

Yes. Aleksoid1978 says that the patch solves the problem. I confirm the sound appears.

[Carl Eugen Hoyos]

Please ask Aleksoid1978 to send his patch to the development mailing list to get this ticket fixed.

[llogan]

Fixed in 9445e7e6d562e14a6fbc3f73cd282136a239a611.