Name Description Size
backends.h this is exposed up here because we need it for static modes. Lookups for each backend aren't exposed because there's no reason to do so 5151
bitrate.h encode side bitrate tracking 1963
codebook.h This structure encapsulates huffman and VQ style encoding books; it doesn't do anything specific to either. valuelist/quantlist are nonNULL (and q_* significant) only if there's entry->value mapping to be done. If encode-side mapping must be done (and thus the entry needs to be hunted), the auxiliary encode pointer will point to a decision tree. This is true of both VQ and huffman, but is mostly useful with VQ. 4958
codec_internal.h this is a pointer into local storage 5392
envelope.h a bit less than short block 2116
highlevel.h bitrate management below all settable 1840
lookup.h 1249
lookup_data.h 10036
lpc.h simple linear scale LPC code 1174
lsp.h 1182
masking.h more detailed ATH; the bass if flat to save stressing the floor overly for only a bin or two of savings. 39455
mdct.h #define MDCT_INTEGERIZED <- be warned there could be some hurt left here 1955
misc.h 1988
os.h _V_IFDEFJAIL_H_ 5176
psy.h psychoacoustic setup ******************************************* 4628
registry.h 1247
scales.h MS Visual Studio doesn't have C99 inline keyword. 2761
smallft.h 1218
vorbis_analysis.c decides between modes, dispatches to the appropriate mapping. 3088
vorbis_bitrate.c compute bitrate tracking setup 8478
vorbis_block.c pcm accumulator examples (not exhaustive): <-------------- lW ----------------> <--------------- W ----------------> : .....|..... _______________ | : .''' | '''_--- | |\ | :.....''' |_____--- '''......| | \_______| :.................|__________________|_______|__|______| |<------ Sl ------>| > Sr < |endW |beginSl |endSl | |endSr |beginW |endlW |beginSr |< lW >| <--------------- W ----------------> | | .. ______________ | | | ' `/ | ---_ | |___.'___/`. | ---_____| |_______|__|_______|_________________| | >|Sl|< |<------ Sr ----->|endW | | |endSl |beginSr |endSr |beginW | |endlW mult[0] |beginSl mult[n] <-------------- lW -----------------> |<--W-->| : .............. ___ | | : .''' |`/ \ | | :.....''' |/`....\|...| :.........................|___|___|___| |Sl |Sr |endW | | |endSr | |beginSr | |endSl |beginSl |beginW 30592
vorbis_codebook.c packs the given codebook into the bitstream ************************* 13408
vorbis_envelope.c not random 10672
vorbis_floor0.c / static void floor0_free_info(vorbis_info_floor *i){ vorbis_info_floor0 *info=(vorbis_info_floor0 *)i; if(info){ memset(info,0,sizeof(*info)); _ogg_free(info); } } static void floor0_free_look(vorbis_look_floor *i){ vorbis_look_floor0 *look=(vorbis_look_floor0 *)i; if(look){ if(look->linearmap){ if(look->linearmap[0])_ogg_free(look->linearmap[0]); if(look->linearmap[1])_ogg_free(look->linearmap[1]); _ogg_free(look->linearmap); } memset(look,0,sizeof(*look)); _ogg_free(look); } } static vorbis_info_floor *floor0_unpack (vorbis_info *vi,oggpack_buffer *opb){ codec_setup_info *ci=vi->codec_setup; int j; vorbis_info_floor0 *info=_ogg_malloc(sizeof(*info)); info->order=oggpack_read(opb,8); info->rate=oggpack_read(opb,16); info->barkmap=oggpack_read(opb,16); info->ampbits=oggpack_read(opb,6); info->ampdB=oggpack_read(opb,8); info->numbooks=oggpack_read(opb,4)+1; if(info->order<1)goto err_out; if(info->rate<1)goto err_out; if(info->barkmap<1)goto err_out; if(info->numbooks<1)goto err_out; for(j=0;j<info->numbooks;j++){ info->books[j]=oggpack_read(opb,8); if(info->books[j]<0 || info->books[j]>=ci->books)goto err_out; if(ci->book_param[info->books[j]]->maptype==0)goto err_out; if(ci->book_param[info->books[j]]->dim<1)goto err_out; } return(info); err_out: floor0_free_info(info); return(NULL); } /* initialize Bark scale and normalization lookups. We could do this with static tables, but Vorbis allows a number of possible combinations, so it's best to do it computationally. The below is authoritative in terms of defining scale mapping. Note that the scale depends on the sampling rate as well as the linear block and mapping sizes 6889
vorbis_floor1.c floor 1 fixed at -140dB to 0dB range 30922
vorbis_info.c general handling of the header and the vorbis_info structure (and substructures) 19866
vorbis_lookup.c interpolated lookup based cos function, domain 0 to PI only 3246
vorbis_lpc.c Some of these routines (autocorrelator, LPC coefficient estimator) are derived from code written by Jutta Degener and Carsten Bormann; thus we include their copyright below. The entirety of this file is freely redistributable on the condition that both of these copyright notices are preserved without modification. 4486
vorbis_lsp.c Note that the lpc-lsp conversion finds the roots of polynomial with an iterative root polisher (CACM algorithm 283). It *is* possible to confuse this algorithm into not converging; that should only happen with absurdly closely spaced roots (very sharp peaks in the LPC f response) which in turn should be impossible in our use of the code. If this *does* happen anyway, it's a bug in the floor finder; find the cause of the confusion (probably a single bin spike or accidental near-float-limit resolution problems) and correct it. 12599
vorbis_mapping0.c simplistic, wasteful way of doing this (unique lookup for each mode/submapping); there should be a central repository for identical lookups. That will require minor work, so I'm putting it off as low priority. Why a lookup for each backend in a given mode? Because the blocksize is set by the mode, and low backend lookups may require parameters from other areas of the mode/mapping 28765
vorbis_mdct.c this can also be run as an integer transform by uncommenting a define in mdct.h; the integerization is a first pass and although it's likely stable for Vorbis, the dynamic range is constrained and roundoff isn't done (so it's noisy). Consider it functional, but only a starting point. There's no point on a machine with an FPU 15493
vorbis_psy.c we add back in the ATH to avoid low level curves falling off to -infinity and unnecessarily cutting off high level curves in the curve limiting (last step). 35900
vorbis_registry.c seems like major overkill now; the backend numbers will grow into the infrastructure soon enough 1772
vorbis_res0.c Slow, slow, slow, simpleminded and did I mention it was slow? The encode/decode loops are coded for clarity and performance is not yet even a nagging little idea lurking in the shadows. Oh and BTW, it's slow. 24915
vorbis_sharedbook.c pack/unpack helpers ***************************************** 17806
vorbis_smallft.c FFT implementation from OggSquish, minus cosine transforms, minus all but radix 2/4 case. In Vorbis we only need this cut-down version. To do more than just power-of-two sized vectors, see the full version I wrote for NetLib. Note that the packing is a little strange; rather than the FFT r/i packing following R_0, I_n, R_1, I_1, R_2, I_2 ... R_n-1, I_n-1, it follows R_0, R_1, I_1, R_2, I_2 ... R_n-1, I_n-1, I_n like the FORTRAN version 22185
vorbis_synthesis.c first things first. Make sure decode is ready 5180
vorbis_window.c 128646
vorbisenc.c careful with this; it's using static array sizing to make managing all the modes a little less annoying. If we use a residue backend with > 12 partition types, or a different division of iteration, this needs to be updated. 39310
window.h 1086