replaygain_synthesis.c

这是著名的TCPMP播放器在WINDWOWS,和WINCE下编译通过的源程序.笔者对其中的LIBMAD库做了针对ARM MPU的优化. 并增加了词幕功能.

 * the following is based on parts of wavegain.c
 */

static FLAC__INLINE FLAC__int64 dither_output_(DitherContext *d, FLAC__bool do_dithering, int shapingtype, int i, double Sum, int k)
{
	union {
		double d;
		FLAC__int64 i;
	} doubletmp;
	double Sum2;
	FLAC__int64 val;

#define ROUND64(x)   ( doubletmp.d = (x) + d->Add + (FLAC__int64)FLAC__I64L(0x001FFFFD80000000), doubletmp.i - (FLAC__int64)FLAC__I64L(0x433FFFFD80000000) )

	if(do_dithering) {
		if(shapingtype == 0) {
			double  tmp = random_equi_(d->Dither);
			Sum2 = tmp - d->LastRandomNumber [k];
			d->LastRandomNumber [k] = (int)tmp;
			Sum2 = Sum += Sum2;
			val = ROUND64(Sum2) & d->Mask;
		}
		else {
			Sum2 = random_triangular_(d->Dither) - scalar16_(d->DitherHistory[k], d->FilterCoeff + i);
			Sum += d->DitherHistory [k] [(-1-i)&15] = (float)Sum2;
			Sum2 = Sum + scalar16_(d->ErrorHistory [k], d->FilterCoeff + i);
			val = ROUND64(Sum2) & d->Mask;
			d->ErrorHistory [k] [(-1-i)&15] = (float)(Sum - val);
		}
		return val;
	}
	else
		return ROUND64(Sum);

#undef ROUND64
}

#if 0
	float        peak = 0.f,
	             new_peak,
	             factor_clip
	double       scale,
	             dB;

	...

	peak is in the range -32768.0 .. 32767.0

	/* calculate factors for ReplayGain and ClippingPrevention */
	*track_gain = GetTitleGain() + settings->man_gain;
	scale = (float) pow(10., *track_gain * 0.05);
	if(settings->clip_prev) {
		factor_clip  = (float) (32767./( peak + 1));
		if(scale < factor_clip)
			factor_clip = 1.f;
		else
			factor_clip /= scale;
		scale *= factor_clip;
	}
	new_peak = (float) peak * scale;

	dB = 20. * log10(scale);
	*track_gain = (float) dB;

 	const double scale = pow(10., (double)gain * 0.05);
#endif


size_t FLAC__replaygain_synthesis__apply_gain(FLAC__byte *data_out, FLAC__bool little_endian_data_out, FLAC__bool unsigned_data_out, const FLAC__int32 * const input[], unsigned wide_samples, unsigned channels, const unsigned source_bps, const unsigned target_bps, const double scale, const FLAC__bool hard_limit, FLAC__bool do_dithering, DitherContext *dither_context)
{
	static const FLAC__int32 conv_factors_[33] = {
		-1, /* 0 bits-per-sample (not supported) */
		-1, /* 1 bits-per-sample (not supported) */
		-1, /* 2 bits-per-sample (not supported) */
		-1, /* 3 bits-per-sample (not supported) */
		268435456, /* 4 bits-per-sample */
		134217728, /* 5 bits-per-sample */
		67108864, /* 6 bits-per-sample */
		33554432, /* 7 bits-per-sample */
		16777216, /* 8 bits-per-sample */
		8388608, /* 9 bits-per-sample */
		4194304, /* 10 bits-per-sample */
		2097152, /* 11 bits-per-sample */
		1048576, /* 12 bits-per-sample */
		524288, /* 13 bits-per-sample */
		262144, /* 14 bits-per-sample */
		131072, /* 15 bits-per-sample */
		65536, /* 16 bits-per-sample */
		32768, /* 17 bits-per-sample */
		16384, /* 18 bits-per-sample */
		8192, /* 19 bits-per-sample */
		4096, /* 20 bits-per-sample */
		2048, /* 21 bits-per-sample */
		1024, /* 22 bits-per-sample */
		512, /* 23 bits-per-sample */
		256, /* 24 bits-per-sample */
		128, /* 25 bits-per-sample */
		64, /* 26 bits-per-sample */
		32, /* 27 bits-per-sample */
		16, /* 28 bits-per-sample */
		8, /* 29 bits-per-sample */
		4, /* 30 bits-per-sample */
		2, /* 31 bits-per-sample */
		1 /* 32 bits-per-sample */
	};
	static const FLAC__int64 hard_clip_factors_[33] = {
		0, /* 0 bits-per-sample (not supported) */
		0, /* 1 bits-per-sample (not supported) */
		0, /* 2 bits-per-sample (not supported) */
		0, /* 3 bits-per-sample (not supported) */
		-8, /* 4 bits-per-sample */
		-16, /* 5 bits-per-sample */
		-32, /* 6 bits-per-sample */
		-64, /* 7 bits-per-sample */
		-128, /* 8 bits-per-sample */
		-256, /* 9 bits-per-sample */
		-512, /* 10 bits-per-sample */
		-1024, /* 11 bits-per-sample */
		-2048, /* 12 bits-per-sample */
		-4096, /* 13 bits-per-sample */
		-8192, /* 14 bits-per-sample */
		-16384, /* 15 bits-per-sample */
		-32768, /* 16 bits-per-sample */
		-65536, /* 17 bits-per-sample */
		-131072, /* 18 bits-per-sample */
		-262144, /* 19 bits-per-sample */
		-524288, /* 20 bits-per-sample */
		-1048576, /* 21 bits-per-sample */
		-2097152, /* 22 bits-per-sample */
		-4194304, /* 23 bits-per-sample */
		-8388608, /* 24 bits-per-sample */
		-16777216, /* 25 bits-per-sample */
		-33554432, /* 26 bits-per-sample */
		-67108864, /* 27 bits-per-sample */
		-134217728, /* 28 bits-per-sample */
		-268435456, /* 29 bits-per-sample */
		-536870912, /* 30 bits-per-sample */
		-1073741824, /* 31 bits-per-sample */
		(FLAC__int64)(-1073741824) * 2 /* 32 bits-per-sample */
	};
	const FLAC__int32 conv_factor = conv_factors_[target_bps];
	const FLAC__int64 hard_clip_factor = hard_clip_factors_[target_bps];
	/*
	 * The integer input coming in has a varying range based on the
	 * source_bps.  We want to normalize it to [-1.0, 1.0) so instead
	 * of doing two multiplies on each sample, we just multiple
	 * 'scale' by 1/(2^(source_bps-1))
	 */
	const double multi_scale = scale / (double)(1u << (source_bps-1));

	FLAC__byte * const start = data_out;
	unsigned i, channel;
	const FLAC__int32 *input_;
	double sample;
	const unsigned bytes_per_sample = target_bps / 8;
	const unsigned last_history_index = dither_context->LastHistoryIndex;
	NoiseShaping noise_shaping = dither_context->ShapingType;
	FLAC__int64 val64;
	FLAC__int32 val32;
	FLAC__int32 uval32;
	const FLAC__uint32 twiggle = 1u << (target_bps - 1);

	FLAC__ASSERT(channels > 0 && channels <= FLAC_SHARE__MAX_SUPPORTED_CHANNELS);
	FLAC__ASSERT(source_bps >= 4);
	FLAC__ASSERT(target_bps >= 4);
	FLAC__ASSERT(source_bps <= 32);
	FLAC__ASSERT(target_bps < 32);
	FLAC__ASSERT((target_bps & 7) == 0);

	for(channel = 0; channel < channels; channel++) {
		const unsigned incr = bytes_per_sample * channels;
		data_out = start + bytes_per_sample * channel;
		input_ = input[channel];
		for(i = 0; i < wide_samples; i++, data_out += incr) {
			sample = (double)input_[i] * multi_scale;

			if(hard_limit) {
				/* hard 6dB limiting */
				if(sample < -0.5)
					sample = tanh((sample + 0.5) / (1-0.5)) * (1-0.5) - 0.5;
				else if(sample > 0.5)
					sample = tanh((sample - 0.5) / (1-0.5)) * (1-0.5) + 0.5;
			}
			sample *= 2147483647.f;

			val64 = dither_output_(dither_context, do_dithering, noise_shaping, (i + last_history_index) % 32, sample, channel) / conv_factor;

			val32 = (FLAC__int32)val64;
			if(val64 >= -hard_clip_factor)
				val32 = (FLAC__int32)(-(hard_clip_factor+1));
			else if(val64 < hard_clip_factor)
				val32 = (FLAC__int32)hard_clip_factor;

			uval32 = (FLAC__uint32)val32;
			if (unsigned_data_out)
				uval32 ^= twiggle;

			if (little_endian_data_out) {
				switch(target_bps) {
					case 24:
						data_out[2] = (FLAC__byte)(uval32 >> 16);
						/* fall through */
					case 16:
						data_out[1] = (FLAC__byte)(uval32 >> 8);
						/* fall through */
					case 8:
						data_out[0] = (FLAC__byte)uval32;
						break;
				}
			}
			else {
				switch(target_bps) {
					case 24:
						data_out[0] = (FLAC__byte)(uval32 >> 16);
						data_out[1] = (FLAC__byte)(uval32 >> 8);
						data_out[2] = (FLAC__byte)uval32;
						break;
					case 16:
						data_out[0] = (FLAC__byte)(uval32 >> 8);
						data_out[1] = (FLAC__byte)uval32;
						break;
					case 8:
						data_out[0] = (FLAC__byte)uval32;
						break;
				}
			}
		}
	}
	dither_context->LastHistoryIndex = (last_history_index + wide_samples) % 32;

	return wide_samples * channels * (target_bps/8);
}