/* Copyright (C) 1996-1997 Id Software, Inc. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ // snd_mem.c: sound caching #include "quakedef.h" #include "winquake.h" #include "fs.h" typedef struct { int format; int rate; int width; int numchannels; int loopstart; int samples; int dataofs; // chunk starts this many bytes from file start } wavinfo_t; static wavinfo_t GetWavinfo (char *name, qbyte *wav, int wavlength); int cache_full_cycle; qbyte *S_Alloc (int size); #define LINEARUPSCALE(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = inrate / (double)outrate; \ infrac = floor(scale * 65536); \ outsamps = insamps / scale; \ inaccum = 0; \ outnlsamps = floor(1.0 / scale); \ outsamps -= outnlsamps; \ \ while (outsamps) \ { \ *out = ((0xFFFF - inaccum)*in[0] + inaccum*in[1]) >> (16 - outlshift + outrshift); \ inaccum += infrac; \ in += (inaccum >> 16); \ inaccum &= 0xFFFF; \ out++; \ outsamps--; \ } \ while (outnlsamps) \ { \ *out = (*in >> outrshift) << outlshift; \ out++; \ outnlsamps--; \ } \ } #define LINEARUPSCALESTEREO(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = inrate / (double)outrate; \ infrac = floor(scale * 65536); \ outsamps = insamps / scale; \ inaccum = 0; \ outnlsamps = floor(1.0 / scale); \ outsamps -= outnlsamps; \ \ while (outsamps) \ { \ out[0] = ((0xFFFF - inaccum)*in[0] + inaccum*in[2]) >> (16 - outlshift + outrshift); \ out[1] = ((0xFFFF - inaccum)*in[1] + inaccum*in[3]) >> (16 - outlshift + outrshift); \ inaccum += infrac; \ in += (inaccum >> 16) * 2; \ inaccum &= 0xFFFF; \ out += 2; \ outsamps--; \ } \ while (outnlsamps) \ { \ out[0] = (in[0] >> outrshift) << outlshift; \ out[1] = (in[1] >> outrshift) << outlshift; \ out += 2; \ outnlsamps--; \ } \ } #define LINEARUPSCALESTEREOTOMONO(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = inrate / (double)outrate; \ infrac = floor(scale * 65536); \ outsamps = insamps / scale; \ inaccum = 0; \ outnlsamps = floor(1.0 / scale); \ outsamps -= outnlsamps; \ \ while (outsamps) \ { \ *out = ((((0xFFFF - inaccum)*in[0] + inaccum*in[2]) >> (16 - outlshift + outrshift)) + \ (((0xFFFF - inaccum)*in[1] + inaccum*in[3]) >> (16 - outlshift + outrshift))) >> 1; \ inaccum += infrac; \ in += (inaccum >> 16) * 2; \ inaccum &= 0xFFFF; \ out++; \ outsamps--; \ } \ while (outnlsamps) \ { \ out[0] = (((in[0] >> outrshift) << outlshift) + ((in[1] >> outrshift) << outlshift)) >> 1; \ out++; \ outnlsamps--; \ } \ } #define LINEARDOWNSCALE(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = outrate / (double)inrate; \ infrac = floor(scale * 65536); \ inaccum = 0; \ insamps--; \ outsampleft = 0; \ \ while (insamps) \ { \ inaccum += infrac; \ if (inaccum >> 16) \ { \ inaccum &= 0xFFFF; \ outsampleft += (infrac - inaccum) * (*in); \ *out = outsampleft >> (16 - outlshift + outrshift); \ out++; \ outsampleft = inaccum * (*in); \ } \ else \ outsampleft += infrac * (*in); \ in++; \ insamps--; \ } \ outsampleft += (0xFFFF - inaccum) * (*in);\ *out = outsampleft >> (16 - outlshift + outrshift); \ } #define LINEARDOWNSCALESTEREO(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = outrate / (double)inrate; \ infrac = floor(scale * 65536); \ inaccum = 0; \ insamps--; \ outsampleft = 0; \ outsampright = 0; \ \ while (insamps) \ { \ inaccum += infrac; \ if (inaccum >> 16) \ { \ inaccum &= 0xFFFF; \ outsampleft += (infrac - inaccum) * in[0]; \ outsampright += (infrac - inaccum) * in[1]; \ out[0] = outsampleft >> (16 - outlshift + outrshift); \ out[1] = outsampright >> (16 - outlshift + outrshift); \ out += 2; \ outsampleft = inaccum * in[0]; \ outsampright = inaccum * in[1]; \ } \ else \ { \ outsampleft += infrac * in[0]; \ outsampright += infrac * in[1]; \ } \ in += 2; \ insamps--; \ } \ outsampleft += (0xFFFF - inaccum) * in[0];\ outsampright += (0xFFFF - inaccum) * in[1];\ out[0] = outsampleft >> (16 - outlshift + outrshift); \ out[1] = outsampright >> (16 - outlshift + outrshift); \ } #define LINEARDOWNSCALESTEREOTOMONO(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = outrate / (double)inrate; \ infrac = floor(scale * 65536); \ inaccum = 0; \ insamps--; \ outsampleft = 0; \ \ while (insamps) \ { \ inaccum += infrac; \ if (inaccum >> 16) \ { \ inaccum &= 0xFFFF; \ outsampleft += (infrac - inaccum) * ((in[0] + in[1]) >> 1); \ *out = outsampleft >> (16 - outlshift + outrshift); \ out++; \ outsampleft = inaccum * ((in[0] + in[1]) >> 1); \ } \ else \ outsampleft += infrac * ((in[0] + in[1]) >> 1); \ in += 2; \ insamps--; \ } \ outsampleft += (0xFFFF - inaccum) * ((in[0] + in[1]) >> 1);\ *out = outsampleft >> (16 - outlshift + outrshift); \ } #define STANDARDRESCALE(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = inrate / (double)outrate; \ infrac = floor(scale * 65536); \ outsamps = insamps / scale; \ inaccum = 0; \ \ while (outsamps) \ { \ *out = (*in >> outrshift) << outlshift; \ inaccum += infrac; \ in += (inaccum >> 16); \ inaccum &= 0xFFFF; \ out++; \ outsamps--; \ } \ } #define STANDARDRESCALESTEREO(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = inrate / (double)outrate; \ infrac = floor(scale * 65536); \ outsamps = insamps / scale; \ inaccum = 0; \ \ while (outsamps) \ { \ out[0] = (in[0] >> outrshift) << outlshift; \ out[1] = (in[1] >> outrshift) << outlshift; \ inaccum += infrac; \ in += (inaccum >> 16) * 2; \ inaccum &= 0xFFFF; \ out += 2; \ outsamps--; \ } \ } #define STANDARDRESCALESTEREOTOMONO(in, inrate, insamps, out, outrate, outlshift, outrshift) \ { \ scale = inrate / (double)outrate; \ infrac = floor(scale * 65536); \ outsamps = insamps / scale; \ inaccum = 0; \ \ while (outsamps) \ { \ out[0] = (((in[0] >> outrshift) << outlshift) + ((in[1] >> outrshift) << outlshift)) >> 1; \ inaccum += infrac; \ in += (inaccum >> 16) * 2; \ inaccum &= 0xFFFF; \ out++; \ outsamps--; \ } \ } #define QUICKCONVERT(in, insamps, out, outlshift, outrshift) \ { \ while (insamps) \ { \ *out = (*in >> outrshift) << outlshift; \ out++; \ in++; \ insamps--; \ } \ } #define QUICKCONVERTSTEREOTOMONO(in, insamps, out, outlshift, outrshift) \ { \ while (insamps) \ { \ *out = (((in[0] >> outrshift) << outlshift) + ((in[1] >> outrshift) << outlshift)) >> 1; \ out++; \ in += 2; \ insamps--; \ } \ } // SND_ResampleStream: takes a sound stream and converts with given parameters. Limited to // 8-16-bit signed conversions and mono-to-mono/stereo-to-stereo conversions. // Not an in-place algorithm. void SND_ResampleStream (void *in, int inrate, int inwidth, int inchannels, int insamps, void *out, int outrate, int outwidth, int outchannels, int resampstyle) { double scale; signed char *in8 = (signed char *)in; short *in16 = (short *)in; signed char *out8 = (signed char *)out; short *out16 = (short *)out; int outsamps, outnlsamps, outsampleft, outsampright; int infrac, inaccum; if (insamps <= 0) return; if (inchannels == outchannels && inwidth == outwidth && inrate == outrate) { memcpy(out, in, inwidth*insamps*inchannels); return; } if (inchannels == 1 && outchannels == 1) { if (inwidth == 1) { if (outwidth == 1) { if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALE(in8, inrate, insamps, out8, outrate, 0, 0) else STANDARDRESCALE(in8, inrate, insamps, out8, outrate, 0, 0) } else // downsample { if (resampstyle > 1) LINEARDOWNSCALE(in8, inrate, insamps, out8, outrate, 0, 0) else STANDARDRESCALE(in8, inrate, insamps, out8, outrate, 0, 0) } return; } else { if (inrate == outrate) // quick convert QUICKCONVERT(in8, insamps, out16, 8, 0) else if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALE(in8, inrate, insamps, out16, outrate, 8, 0) else STANDARDRESCALE(in8, inrate, insamps, out16, outrate, 8, 0) } else // downsample { if (resampstyle > 1) LINEARDOWNSCALE(in8, inrate, insamps, out16, outrate, 8, 0) else STANDARDRESCALE(in8, inrate, insamps, out16, outrate, 8, 0) } return; } } else // 16-bit { if (outwidth == 2) { if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALE(in16, inrate, insamps, out16, outrate, 0, 0) else STANDARDRESCALE(in16, inrate, insamps, out16, outrate, 0, 0) } else // downsample { if (resampstyle > 1) LINEARDOWNSCALE(in16, inrate, insamps, out16, outrate, 0, 0) else STANDARDRESCALE(in16, inrate, insamps, out16, outrate, 0, 0) } return; } else { if (inrate == outrate) // quick convert QUICKCONVERT(in16, insamps, out8, 0, 8) else if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALE(in16, inrate, insamps, out8, outrate, 0, 8) else STANDARDRESCALE(in16, inrate, insamps, out8, outrate, 0, 8) } else // downsample { if (resampstyle > 1) LINEARDOWNSCALE(in16, inrate, insamps, out8, outrate, 0, 8) else STANDARDRESCALE(in16, inrate, insamps, out8, outrate, 0, 8) } return; } } } else if (outchannels == 2 && inchannels == 2) { if (inwidth == 1) { if (outwidth == 1) { if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALESTEREO(in8, inrate, insamps, out8, outrate, 0, 0) else STANDARDRESCALESTEREO(in8, inrate, insamps, out8, outrate, 0, 0) } else // downsample { if (resampstyle > 1) LINEARDOWNSCALESTEREO(in8, inrate, insamps, out8, outrate, 0, 0) else STANDARDRESCALESTEREO(in8, inrate, insamps, out8, outrate, 0, 0) } } else { if (inrate == outrate) // quick convert { insamps *= 2; QUICKCONVERT(in8, insamps, out16, 8, 0) } else if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALESTEREO(in8, inrate, insamps, out16, outrate, 8, 0) else STANDARDRESCALESTEREO(in8, inrate, insamps, out16, outrate, 8, 0) } else // downsample { if (resampstyle > 1) LINEARDOWNSCALESTEREO(in8, inrate, insamps, out16, outrate, 8, 0) else STANDARDRESCALESTEREO(in8, inrate, insamps, out16, outrate, 8, 0) } } } else // 16-bit { if (outwidth == 2) { if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALESTEREO(in16, inrate, insamps, out16, outrate, 0, 0) else STANDARDRESCALESTEREO(in16, inrate, insamps, out16, outrate, 0, 0) } else // downsample { if (resampstyle > 1) LINEARDOWNSCALESTEREO(in16, inrate, insamps, out16, outrate, 0, 0) else STANDARDRESCALESTEREO(in16, inrate, insamps, out16, outrate, 0, 0) } } else { if (inrate == outrate) // quick convert { insamps *= 2; QUICKCONVERT(in16, insamps, out8, 0, 8) } else if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALESTEREO(in16, inrate, insamps, out8, outrate, 0, 8) else STANDARDRESCALESTEREO(in16, inrate, insamps, out8, outrate, 0, 8) } else // downsample { if (resampstyle > 1) LINEARDOWNSCALESTEREO(in16, inrate, insamps, out8, outrate, 0, 8) else STANDARDRESCALESTEREO(in16, inrate, insamps, out8, outrate, 0, 8) } } } } #if 0 else if (outchannels == 1 && inchannels == 2) { if (inwidth == 1) { if (outwidth == 1) { if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALESTEREOTOMONO(in8, inrate, insamps, out8, outrate, 0, 0) else STANDARDRESCALESTEREOTOMONO(in8, inrate, insamps, out8, outrate, 0, 0) } else // downsample STANDARDRESCALESTEREOTOMONO(in8, inrate, insamps, out8, outrate, 0, 0) } else { if (inrate == outrate) // quick convert QUICKCONVERTSTEREOTOMONO(in8, insamps, out16, 8, 0) else if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALESTEREOTOMONO(in8, inrate, insamps, out16, outrate, 8, 0) else STANDARDRESCALESTEREOTOMONO(in8, inrate, insamps, out16, outrate, 8, 0) } else // downsample STANDARDRESCALESTEREOTOMONO(in8, inrate, insamps, out16, outrate, 8, 0) } } else // 16-bit { if (outwidth == 2) { if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALESTEREOTOMONO(in16, inrate, insamps, out16, outrate, 0, 0) else STANDARDRESCALESTEREOTOMONO(in16, inrate, insamps, out16, outrate, 0, 0) } else // downsample STANDARDRESCALESTEREOTOMONO(in16, inrate, insamps, out16, outrate, 0, 0) } else { if (inrate == outrate) // quick convert QUICKCONVERTSTEREOTOMONO(in16, insamps, out8, 0, 8) else if (inrate < outrate) // upsample { if (resampstyle) LINEARUPSCALESTEREOTOMONO(in16, inrate, insamps, out8, outrate, 0, 8) else STANDARDRESCALESTEREOTOMONO(in16, inrate, insamps, out8, outrate, 0, 8) } else // downsample STANDARDRESCALESTEREOTOMONO(in16, inrate, insamps, out8, outrate, 0, 8) } } } #endif } /* ================ ResampleSfx ================ */ static qboolean ResampleSfx (sfx_t *sfx, int inrate, int inchannels, int inwidth, int insamps, int inloopstart, qbyte *data) { extern cvar_t snd_linearresample; double scale; sfxcache_t *sc; int outsamps; int len; int outwidth; scale = snd_speed / (double)inrate; outsamps = insamps * scale; if (loadas8bit.ival < 0) outwidth = 2; else if (loadas8bit.ival) outwidth = 1; else outwidth = inwidth; len = outsamps * outwidth * inchannels; sfx->decoder.buf = sc = BZ_Malloc(len + sizeof(sfxcache_t)); if (!sc) { return false; } sc->numchannels = inchannels; sc->width = outwidth; sc->speed = snd_speed; sc->length = outsamps; sc->soundoffset = 0; sc->data = (qbyte*)(sc+1); if (inloopstart == -1) sfx->loopstart = inloopstart; else sfx->loopstart = inloopstart * scale; SND_ResampleStream (data, inrate, inwidth, inchannels, insamps, sc->data, sc->speed, sc->width, sc->numchannels, snd_linearresample.ival); return true; } //============================================================================= #ifdef PACKAGE_DOOMWAD #define DSPK_RATE 140 #define DSPK_BASE 170.0 #define DSPK_EXP 0.0433 /* qboolean QDECL S_LoadDoomSpeakerSound (sfx_t *s, qbyte *data, size_t datalen, int sndspeed) { sfxcache_t *sc; // format data from Unofficial Doom Specs v1.6 unsigned short *dataus; int samples, len, inrate, inaccum; qbyte *outdata; qbyte towrite; double timeraccum, timerfreq; if (datalen < 4) return NULL; dataus = (unsigned short*)data; if (LittleShort(dataus[0]) != 0) return NULL; samples = LittleShort(dataus[1]); data += 4; datalen -= 4; if (datalen != samples) return NULL; len = (int)((double)samples * (double)snd_speed / DSPK_RATE); sc = Cache_Alloc (&s->cache, len + sizeof(sfxcache_t), s->name); if (!sc) { return NULL; } sc->length = len; s->loopstart = -1; sc->numchannels = 1; sc->width = 1; sc->speed = snd_speed; timeraccum = 0; outdata = sc->data; towrite = 0x40; inrate = (int)((double)snd_speed / DSPK_RATE); inaccum = inrate; if (*data) timerfreq = DSPK_BASE * pow((double)2.0, DSPK_EXP * (*data)); else timerfreq = 0; while (len > 0) { timeraccum += timerfreq; if (timeraccum > (float)snd_speed) { towrite ^= 0xFF; // swap speaker component timeraccum -= (float)snd_speed; } inaccum--; if (!inaccum) { data++; if (*data) timerfreq = DSPK_BASE * pow((double)2.0, DSPK_EXP * (*data)); inaccum = inrate; } *outdata = towrite; outdata++; len--; } return sc; } */ static qboolean QDECL S_LoadDoomSound (sfx_t *s, qbyte *data, size_t datalen, int sndspeed) { // format data from Unofficial Doom Specs v1.6 unsigned short *dataus; int samples, rate; if (datalen < 8) return false; dataus = (unsigned short*)data; if (LittleShort(dataus[0]) != 3) return false; rate = LittleShort(dataus[1]); samples = LittleShort(dataus[2]); data += 8; datalen -= 8; if (datalen != samples) return false; COM_CharBias(data, datalen); return ResampleSfx (s, rate, 1, 1, samples, -1, data); } #endif void S_ShortedLittleFloats(void *p, size_t samples) { short *out = p; float *in = p; int t; while(samples --> 0) { t = LittleFloat(*in++) * 32767; t = bound(-32768, t, 32767); *out++ = t; } } static qboolean QDECL S_LoadWavSound (sfx_t *s, qbyte *data, size_t datalen, int sndspeed) { wavinfo_t info; if (datalen < 4 || strncmp(data, "RIFF", 4)) return false; info = GetWavinfo (s->name, data, datalen); if (info.numchannels < 1 || info.numchannels > 2) { s->loadstate = SLS_FAILED; Con_Printf ("%s has an unsupported quantity of channels.\n",s->name); return false; } if (info.format == 1 && info.width == 1) COM_CharBias(data + info.dataofs, info.samples*info.numchannels); else if (info.format == 1 && info.width == 2) COM_SwapLittleShortBlock((short *)(data + info.dataofs), info.samples*info.numchannels); else if (info.format == 3 && info.width == 4) { S_ShortedLittleFloats(data + info.dataofs, info.samples*info.numchannels); info.width = 2; } else { s->loadstate = SLS_FAILED; switch(info.format) { case 1: case 3: Con_Printf ("%s has an unsupported width (%i bits).\n", s->name, info.width*8); break; case 6: Con_Printf ("%s uses unsupported a-law format.\n", s->name); break; case 7: Con_Printf ("%s uses unsupported mu-law format.\n", s->name); break; default: Con_Printf ("%s has an unsupported format.\n", s->name); break; } return false; } return ResampleSfx (s, info.rate, info.numchannels, info.width, info.samples, info.loopstart, data + info.dataofs); } qboolean QDECL S_LoadOVSound (sfx_t *s, qbyte *data, size_t datalen, int sndspeed); #ifdef FTE_TARGET_WEB //web browsers contain their own decoding libraries that our openal stuff can use. static qboolean QDECL S_LoadBrowserFile (sfx_t *s, qbyte *data, size_t datalen, int sndspeed) { sfxcache_t *sc; s->decoder.buf = sc = BZ_Malloc(sizeof(sfxcache_t) + datalen); s->loopstart = -1; sc->data = (qbyte*)(sc+1); sc->length = datalen; sc->width = 0; //ie: not pcm sc->speed = sndspeed; sc->numchannels = 2; sc->soundoffset = 0; memcpy(sc->data, data, datalen); return true; } #endif //highest priority is last. static S_LoadSound_t AudioInputPlugins[10] = { #ifdef FTE_TARGET_WEB S_LoadBrowserFile, #endif #ifdef AVAIL_OGGVORBIS S_LoadOVSound, #endif S_LoadWavSound, #ifdef PACKAGE_DOOMWAD S_LoadDoomSound, // S_LoadDoomSpeakerSound, #endif }; qboolean S_RegisterSoundInputPlugin(S_LoadSound_t loadfnc) { int i; for (i = 0; i < sizeof(AudioInputPlugins)/sizeof(AudioInputPlugins[0]); i++) { if (!AudioInputPlugins[i]) { AudioInputPlugins[i] = loadfnc; return true; } } return false; } static void S_LoadedOrFailed (void *ctx, void *ctxdata, size_t a, size_t b) { sfx_t *s = ctx; s->loadstate = a; } /* ============== S_LoadSound ============== */ static void S_LoadSoundWorker (void *ctx, void *ctxdata, size_t a, size_t b) { sfx_t *s = ctx; char namebuffer[256]; qbyte *data; int i; size_t result; char *name = s->name; size_t filesize; s->loopstart = -1; if (s->syspath) { vfsfile_t *f; if ((f = VFSOS_Open(name, "rb"))) { filesize = VFS_GETLEN(f); data = BZ_Malloc (filesize); result = VFS_READ(f, data, filesize); if (result != filesize) Con_SafePrintf("S_LoadSound() fread: Filename: %s, expected %"PRIuSIZE", result was %"PRIuSIZE"\n", name, filesize, result); VFS_CLOSE(f); } else { Con_SafePrintf ("Couldn't load %s\n", namebuffer); COM_AddWork(WG_MAIN, S_LoadedOrFailed, s, NULL, SLS_FAILED, 0); return; } } else { //Con_Printf ("S_LoadSound: %x\n", (int)stackbuf); // load it in const char *prefixes[] = {"sound/", ""}; const char *extensions[] = { ".wav", #ifdef AVAIL_OGGOPUS ".opus", #endif #ifdef AVAIL_OGGVORBIS ".ogg", #endif }; char altname[sizeof(namebuffer)]; char orig[16]; size_t pre, ex; data = NULL; filesize = 0; if (*name == '*') //q2 sexed sounds { //clq2_parsestartsound detects this also, and should not try playing these sounds. s->loadstate = SLS_FAILED; return; } for (pre = 0; !data && pre < countof(prefixes); pre++) { if (name[0] == '.' && name[1] == '.' && name[2] == '/') { //someone's being specific. disable prefixes entirely. if (pre) break; //not relative to sound/ Q_snprintfz(namebuffer, sizeof(namebuffer), "%s", name+3); } else Q_snprintfz(namebuffer, sizeof(namebuffer), "%s%s", prefixes[pre], name); data = COM_LoadFile(namebuffer, 5, &filesize); if (data) break; COM_FileExtension(namebuffer, orig, sizeof(orig)); COM_StripExtension(namebuffer, altname, sizeof(altname)); for (ex = 0; ex < countof(extensions); ex++) { if (!strcmp(orig, extensions[ex]+1)) continue; Q_snprintfz(namebuffer, sizeof(namebuffer), "%s%s", altname, extensions[ex]); data = COM_LoadFile(namebuffer, 5, &filesize); if (data) { Con_DPrintf("found a mangled name: %s\n", namebuffer); break; } } } } if (!data) { //FIXME: check to see if queued for download. if (name[0] == '.' && name[1] == '.' && name[2] == '/') Con_DPrintf ("Couldn't load %s\n", name+3); else Con_DPrintf ("Couldn't load sound/%s\n", name); COM_AddWork(WG_MAIN, S_LoadedOrFailed, s, NULL, SLS_FAILED, 0); return; } for (i = sizeof(AudioInputPlugins)/sizeof(AudioInputPlugins[0])-1; i >= 0; i--) { if (AudioInputPlugins[i]) { if (AudioInputPlugins[i](s, data, filesize, snd_speed)) { //wake up the main thread in case it decided to wait for us. COM_AddWork(WG_MAIN, S_LoadedOrFailed, s, NULL, SLS_LOADED, 0); BZ_Free(data); return; } } } if (s->loadstate != SLS_FAILED) Con_Printf ("Format not recognised: %s\n", namebuffer); COM_AddWork(WG_MAIN, S_LoadedOrFailed, s, NULL, SLS_FAILED, 0); BZ_Free(data); return; } qboolean S_LoadSound (sfx_t *s) { if (s->loadstate == SLS_NOTLOADED && sndcardinfo) { s->loadstate = SLS_LOADING; COM_AddWork(WG_LOADER, S_LoadSoundWorker, s, NULL, 0, 0); } if (s->loadstate == SLS_FAILED) return false; //it failed to load once before, don't bother trying again. return true; //loaded okay, or still loading } /* =============================================================================== WAV loading =============================================================================== */ typedef struct { char *wavname; qbyte *data_p; qbyte *iff_end; qbyte *last_chunk; qbyte *iff_data; int iff_chunk_len; } wavctx_t; static short GetLittleShort(wavctx_t *ctx) { short val = 0; val = *ctx->data_p; val = val + (*(ctx->data_p+1)<<8); ctx->data_p += 2; return val; } static int GetLittleLong(wavctx_t *ctx) { int val = 0; val = *ctx->data_p; val = val + (*(ctx->data_p+1)<<8); val = val + (*(ctx->data_p+2)<<16); val = val + (*(ctx->data_p+3)<<24); ctx->data_p += 4; return val; } static unsigned int FindNextChunk(wavctx_t *ctx, char *name) { unsigned int dataleft; while (1) { dataleft = ctx->iff_end - ctx->last_chunk; if (dataleft < 8) { // didn't find the chunk ctx->data_p = NULL; return 0; } ctx->data_p=ctx->last_chunk; ctx->data_p += 4; dataleft-= 8; ctx->iff_chunk_len = GetLittleLong(ctx); if (ctx->iff_chunk_len < 0) { ctx->data_p = NULL; return 0; } if (ctx->iff_chunk_len > dataleft) { Con_DPrintf ("\"%s\" seems truncated by %i bytes\n", ctx->wavname, ctx->iff_chunk_len-dataleft); #if 1 ctx->iff_chunk_len = dataleft; #else ctx->data_p = NULL; return 0; #endif } dataleft-= ctx->iff_chunk_len; // if (iff_chunk_len > 1024*1024) // Sys_Error ("FindNextChunk: %i length is past the 1 meg sanity limit", iff_chunk_len); ctx->data_p -= 8; ctx->last_chunk = ctx->data_p + 8 + ctx->iff_chunk_len; if ((ctx->iff_chunk_len&1) && dataleft) ctx->last_chunk++; if (!Q_strncmp(ctx->data_p, name, 4)) return ctx->iff_chunk_len; } } static unsigned int FindChunk(wavctx_t *ctx, char *name) { ctx->last_chunk = ctx->iff_data; return FindNextChunk (ctx, name); } #if 0 static void DumpChunks(void) { char str[5]; str[4] = 0; data_p=iff_data; do { memcpy (str, data_p, 4); data_p += 4; iff_chunk_len = GetLittleLong(); Con_Printf ("0x%x : %s (%d)\n", (int)(data_p - 4), str, iff_chunk_len); data_p += (iff_chunk_len + 1) & ~1; } while (data_p < iff_end); } #endif /* ============ GetWavinfo ============ */ static wavinfo_t GetWavinfo (char *name, qbyte *wav, int wavlength) { wavinfo_t info; int i; int samples; int chunklen; wavctx_t ctx; memset (&info, 0, sizeof(info)); if (!wav) return info; ctx.data_p = NULL; ctx.last_chunk = NULL; ctx.iff_chunk_len = 0; ctx.iff_data = wav; ctx.iff_end = wav + wavlength; ctx.wavname = name; // find "RIFF" chunk chunklen = FindChunk(&ctx, "RIFF"); if (chunklen < 4 || Q_strncmp(ctx.data_p+8, "WAVE", 4)) { Con_Printf("Missing RIFF/WAVE chunks in %s\n", name); return info; } // get "fmt " chunk ctx.iff_data = ctx.data_p + 12; // DumpChunks (); chunklen = FindChunk(&ctx, "fmt "); if (chunklen < 24-8) { Con_Printf("Missing/truncated fmt chunk\n"); return info; } ctx.data_p += 8; info.format = GetLittleShort(&ctx); info.numchannels = GetLittleShort(&ctx); info.rate = GetLittleLong(&ctx); ctx.data_p += 4+2; info.width = GetLittleShort(&ctx) / 8; // get cue chunk chunklen = FindChunk(&ctx, "cue "); if (chunklen >= 36-8) { ctx.data_p += 32; info.loopstart = GetLittleLong(&ctx); // Con_Printf("loopstart=%d\n", sfx->loopstart); // if the next chunk is a LIST chunk, look for a cue length marker chunklen = FindNextChunk (&ctx, "LIST"); if (chunklen >= 32-8) { if (!strncmp (ctx.data_p + 28, "mark", 4)) { // this is not a proper parse, but it works with cooledit... ctx.data_p += 24; i = GetLittleLong (&ctx); // samples in loop info.samples = info.loopstart + i; // Con_Printf("looped length: %i\n", i); } } } else info.loopstart = -1; // find data chunk chunklen = FindChunk(&ctx, "data"); if (!ctx.data_p) { Con_Printf("Missing data chunk in %s\n", name); return info; } ctx.data_p += 8; samples = chunklen / info.width /info.numchannels; if (info.samples) { if (samples < info.samples) { info.samples = samples; Con_Printf ("Sound %s has a bad loop length\n", name); } } else info.samples = samples; if (info.loopstart > info.samples) { Con_Printf ("Sound %s has a bad loop start\n", name); info.loopstart = info.samples; } info.dataofs = ctx.data_p - wav; return info; }