[FORMATTING][NTOS][HAL][DRIVERS][SDK] Fix trailing spaces

This commit is contained in:
Victor Perevertkin 2021-06-11 15:29:21 +03:00
parent 4b4ffa92f5
commit 34593d933b
No known key found for this signature in database
GPG key ID: C750B7222E9C7830
952 changed files with 12942 additions and 12942 deletions

View file

@ -258,9 +258,9 @@
int synth_1to1_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
{
short *samples = (short *) (fr->buffer.data+fr->buffer.fill);
real *b0, **buf;
int clip;
int clip;
int bo1;
#ifndef NO_EQUALIZER
if(fr->have_eq_settings) do_equalizer(bandPtr,channel,fr->equalizer);
@ -276,7 +276,7 @@ int synth_1to1_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
samples++;
buf = fr->real_buffs[1];
}
if(fr->bo & 0x1)
{
b0 = buf[0];
@ -289,12 +289,12 @@ int synth_1to1_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
bo1 = fr->bo+1;
dct64_altivec(buf[0]+fr->bo,buf[1]+fr->bo+1,bandPtr);
}
{
register int j;
real *window = fr->decwin + 16 - bo1;
ALIGNED(16) int clip_tmp[4];
vector float v1,v2,v3,v4,v5,v6,v7,v8,v9;
vector unsigned char vperm1,vperm2,vperm3,vperm4;
@ -314,18 +314,18 @@ int synth_1to1_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
vmin = (vector float){-32768.0f,-32768.0f,-32768.0f,-32768.0f};
vperm4 = (vector unsigned char){0,1,18,19,2,3,22,23,4,5,26,27,6,7,30,31};
#endif
vperm1 = vec_lvsl(0,window);
vperm2 = vec_lvsl(0,samples);
vperm3 = vec_lvsr(0,samples);
for (j=4;j;j--)
{
SYNTH_ALTIVEC(16);
vsum = vec_sub(v5,v6);
v9 = vec_sub(v7,v8);
vsum = vec_add(vsum,v9);
v3 = vec_round(vsum);
v3 = (vector float)vec_cts(v3,0);
v1 = (vector float)vec_cmpgt(vsum,vmax);
@ -341,21 +341,21 @@ int synth_1to1_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
vec_st((vector signed short)v7,15,samples);
vec_st((vector signed short)v8,0,samples);
samples += 8;
v1 = (vector float)vec_sr((vector unsigned int)v1, vshift);
v2 = (vector float)vec_sr((vector unsigned int)v2, vshift);
v1 = (vector float)vec_add((vector unsigned int)v1,(vector unsigned int)v2);
vclip = vec_sums((vector signed int)v1,vclip);
}
for (j=4;j;j--)
{
SYNTH_ALTIVEC(-16);
vsum = vec_add(v5,v6);
v9 = vec_add(v7,v8);
vsum = vec_add(vsum,v9);
v3 = vec_round(vsum);
v3 = (vector float)vec_cts(v3,0);
v1 = (vector float)vec_cmpgt(vsum,vmax);
@ -371,7 +371,7 @@ int synth_1to1_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
vec_st((vector signed short)v7,15,samples);
vec_st((vector signed short)v8,0,samples);
samples += 8;
v1 = (vector float)vec_sr((vector unsigned int)v1, vshift);
v2 = (vector float)vec_sr((vector unsigned int)v2, vshift);
v1 = (vector float)vec_add((vector unsigned int)v1,(vector unsigned int)v2);
@ -382,16 +382,16 @@ int synth_1to1_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
clip = clip_tmp[3];
}
if(final) fr->buffer.fill += 128;
return clip;
}
int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr)
{
short *samples = (short *) (fr->buffer.data+fr->buffer.fill);
real *b0l, *b0r, **bufl, **bufr;
int clip;
int clip;
int bo1;
#ifndef NO_EQUALIZER
if(fr->have_eq_settings)
@ -404,7 +404,7 @@ int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *f
fr->bo &= 0xf;
bufl = fr->real_buffs[0];
bufr = fr->real_buffs[1];
if(fr->bo & 0x1)
{
b0l = bufl[0];
@ -421,12 +421,12 @@ int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *f
dct64_altivec(bufl[0]+fr->bo,bufl[1]+fr->bo+1,bandPtr_l);
dct64_altivec(bufr[0]+fr->bo,bufr[1]+fr->bo+1,bandPtr_r);
}
{
register int j;
real *window = fr->decwin + 16 - bo1;
ALIGNED(16) int clip_tmp[4];
vector float v1,v2,v3,v4,v5,v6,v7,v8,v9,v10,v11,v12,v13;
vector unsigned char vperm1,vperm2;
@ -444,21 +444,21 @@ int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *f
vmax = (vector float){32767.0f,32767.0f,32767.0f,32767.0f};
vmin = (vector float){-32768.0f,-32768.0f,-32768.0f,-32768.0f};
#endif
vperm1 = vec_lvsl(0,window);
vperm2 = vec_lvsr(0,samples);
vprev = vec_perm(vec_ld(0,samples),vec_ld(0,samples),vec_lvsl(0,samples));
for (j=4;j;j--)
{
SYNTH_STEREO_ALTIVEC(16);
vsum = vec_sub(vsum,vsum2);
vsum2 = vec_sub(vsum5,vsum6);
vsum3 = vec_sub(vsum3,vsum4);
vsum4 = vec_sub(vsum7,vsum8);
vsum = vec_add(vsum,vsum3);
vsum2 = vec_add(vsum2,vsum4);
v1 = vec_round(vsum);
v2 = vec_round(vsum2);
v1 = (vector float)vec_cts(v1,0);
@ -474,7 +474,7 @@ int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *f
v4 = (vector float)vec_cmplt(vsum2,vmin);
vec_st((vector signed short)v6,0,samples);
samples += 8;
v1 = (vector float)vec_sr((vector unsigned int)v1, vshift);
v2 = (vector float)vec_sr((vector unsigned int)v2, vshift);
v3 = (vector float)vec_sr((vector unsigned int)v3, vshift);
@ -484,18 +484,18 @@ int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *f
vclip = vec_sums((vector signed int)v1,vclip);
vclip = vec_sums((vector signed int)v2,vclip);
}
for (j=4;j;j--)
{
SYNTH_STEREO_ALTIVEC(-16);
vsum = vec_add(vsum,vsum2);
vsum2 = vec_add(vsum5,vsum6);
vsum3 = vec_add(vsum3,vsum4);
vsum4 = vec_add(vsum7,vsum8);
vsum = vec_add(vsum,vsum3);
vsum2 = vec_add(vsum2,vsum4);
v1 = vec_round(vsum);
v2 = vec_round(vsum2);
v1 = (vector float)vec_cts(v1,0);
@ -511,7 +511,7 @@ int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *f
v4 = (vector float)vec_cmplt(vsum2,vmin);
vec_st((vector signed short)v6,0,samples);
samples += 8;
v1 = (vector float)vec_sr((vector unsigned int)v1, vshift);
v2 = (vector float)vec_sr((vector unsigned int)v2, vshift);
v3 = (vector float)vec_sr((vector unsigned int)v3, vshift);
@ -521,7 +521,7 @@ int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *f
vclip = vec_sums((vector signed int)v1,vclip);
vclip = vec_sums((vector signed int)v2,vclip);
}
if((size_t)samples & 0xf)
{
v1 = (vector float)vec_perm(vec_ld(0,samples),vec_ld(0,samples),vec_lvsl(0,samples));
@ -533,14 +533,14 @@ int synth_1to1_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *f
clip = clip_tmp[3];
}
fr->buffer.fill += 128;
return clip;
}
int synth_1to1_real_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
{
real *samples = (real *) (fr->buffer.data+fr->buffer.fill);
real *b0, **buf;
int bo1;
#ifndef NO_EQUALIZER
@ -557,7 +557,7 @@ int synth_1to1_real_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fin
samples++;
buf = fr->real_buffs[1];
}
if(fr->bo & 0x1)
{
b0 = buf[0];
@ -570,12 +570,12 @@ int synth_1to1_real_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fin
bo1 = fr->bo+1;
dct64_altivec(buf[0]+fr->bo,buf[1]+fr->bo+1,bandPtr);
}
{
register int j;
real *window = fr->decwin + 16 - bo1;
vector float v1,v2,v3,v4,v5,v6,v7,v8,v9;
vector unsigned char vperm1,vperm2,vperm3,vperm4, vperm5;
vector float vsum,vsum2,vsum3,vsum4,vscale,vzero;
@ -590,19 +590,19 @@ int synth_1to1_real_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fin
vperm4 = (vector unsigned char){0,1,2,3,20,21,22,23,4,5,6,7,28,29,30,31};
vperm5 = (vector unsigned char){8,9,10,11,20,21,22,23,12,13,14,15,28,29,30,31};
#endif
vperm1 = vec_lvsl(0,window);
vperm2 = vec_lvsl(0,samples);
vperm3 = vec_lvsr(0,samples);
for (j=4;j;j--)
{
SYNTH_ALTIVEC(16);
vsum = vec_sub(v5,v6);
v9 = vec_sub(v7,v8);
vsum = vec_add(vsum,v9);
vsum = vec_madd(vsum, vscale, vzero);
vsample1 = vec_ld(0,samples);
vsample2 = vec_ld(16,samples);
vsample3 = vec_ld(31,samples);
@ -620,16 +620,16 @@ int synth_1to1_real_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fin
vec_st(v7,0,samples);
samples += 8;
}
for (j=4;j;j--)
{
SYNTH_ALTIVEC(-16);
vsum = vec_add(v5,v6);
v9 = vec_add(v7,v8);
vsum = vec_add(vsum,v9);
vsum = vec_madd(vsum, vscale, vzero);
vsample1 = vec_ld(0,samples);
vsample2 = vec_ld(16,samples);
vsample3 = vec_ld(31,samples);
@ -649,14 +649,14 @@ int synth_1to1_real_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fin
}
}
if(final) fr->buffer.fill += 256;
return 0;
}
int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr)
{
real *samples = (real *) (fr->buffer.data+fr->buffer.fill);
real *b0l, *b0r, **bufl, **bufr;
int bo1;
#ifndef NO_EQUALIZER
@ -670,7 +670,7 @@ int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr
fr->bo &= 0xf;
bufl = fr->real_buffs[0];
bufr = fr->real_buffs[1];
if(fr->bo & 0x1)
{
b0l = bufl[0];
@ -687,12 +687,12 @@ int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr
dct64_altivec(bufl[0]+fr->bo,bufl[1]+fr->bo+1,bandPtr_l);
dct64_altivec(bufr[0]+fr->bo,bufr[1]+fr->bo+1,bandPtr_r);
}
{
register int j;
real *window = fr->decwin + 16 - bo1;
vector float v1,v2,v3,v4,v5,v6,v7,v8,v9,v10,v11,v12,v13;
vector unsigned char vperm1,vperm2;
vector float vsum,vsum2,vsum3,vsum4,vsum5,vsum6,vsum7,vsum8,vscale,vzero;
@ -703,14 +703,14 @@ int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr
#else
vscale = (vector float){1.0f/32768.0f,1.0f/32768.0f,1.0f/32768.0f,1.0f/32768.0f};
#endif
vperm1 = vec_lvsl(0,window);
vperm2 = vec_lvsr(0,samples);
vprev = vec_perm(vec_ld(0,samples),vec_ld(0,samples),vec_lvsl(0,samples));
for (j=4;j;j--)
{
SYNTH_STEREO_ALTIVEC(16);
vsum = vec_sub(vsum,vsum2);
vsum2 = vec_sub(vsum5,vsum6);
vsum3 = vec_sub(vsum3,vsum4);
@ -719,7 +719,7 @@ int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr
vsum2 = vec_add(vsum2,vsum4);
vsum = vec_madd(vsum, vscale, vzero);
vsum2 = vec_madd(vsum2, vscale, vzero);
v1 = vec_mergeh(vsum, vsum2);
v2 = vec_mergel(vsum, vsum2);
v3 = vec_perm(vprev,v1,vperm2);
@ -729,11 +729,11 @@ int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr
vec_st(v4,16,samples);
samples += 8;
}
for (j=4;j;j--)
{
SYNTH_STEREO_ALTIVEC(-16);
vsum = vec_add(vsum,vsum2);
vsum2 = vec_add(vsum5,vsum6);
vsum3 = vec_add(vsum3,vsum4);
@ -742,7 +742,7 @@ int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr
vsum2 = vec_add(vsum2,vsum4);
vsum = vec_madd(vsum, vscale, vzero);
vsum2 = vec_madd(vsum2, vscale, vzero);
v1 = vec_mergeh(vsum, vsum2);
v2 = vec_mergel(vsum, vsum2);
v3 = vec_perm(vprev,v1,vperm2);
@ -752,7 +752,7 @@ int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr
vec_st(v4,16,samples);
samples += 8;
}
if((size_t)samples & 0xf)
{
v1 = (vector float)vec_perm(vec_ld(0,samples),vec_ld(0,samples),vec_lvsl(0,samples));
@ -761,14 +761,14 @@ int synth_1to1_fltst_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr
}
}
fr->buffer.fill += 256;
return 0;
}
int synth_1to1_s32_altivec(real *bandPtr,int channel,mpg123_handle *fr, int final)
{
int32_t *samples = (int32_t *) (fr->buffer.data+fr->buffer.fill);
real *b0, **buf;
int clip;
int bo1;
@ -786,7 +786,7 @@ int synth_1to1_s32_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fina
samples++;
buf = fr->real_buffs[1];
}
if(fr->bo & 0x1)
{
b0 = buf[0];
@ -799,12 +799,12 @@ int synth_1to1_s32_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fina
bo1 = fr->bo+1;
dct64_altivec(buf[0]+fr->bo,buf[1]+fr->bo+1,bandPtr);
}
{
register int j;
real *window = fr->decwin + 16 - bo1;
ALIGNED(16) int clip_tmp[4];
vector float v1,v2,v3,v4,v5,v6,v7,v8,v9;
vector unsigned char vperm1,vperm2,vperm3,vperm4,vperm5;
@ -826,21 +826,21 @@ int synth_1to1_s32_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fina
vperm4 = (vector unsigned char){0,1,2,3,20,21,22,23,4,5,6,7,28,29,30,31};
vperm5 = (vector unsigned char){8,9,10,11,20,21,22,23,12,13,14,15,28,29,30,31};
#endif
vperm1 = vec_lvsl(0,window);
vperm2 = vec_lvsl(0,samples);
vperm3 = vec_lvsr(0,samples);
for (j=4;j;j--)
{
SYNTH_ALTIVEC(16);
vsum = vec_sub(v5,v6);
v9 = vec_sub(v7,v8);
v1 = vec_add(vsum,v9);
vsum = (vector float)vec_cts(v1,16);
v8 = (vector float)vec_cmpgt(v1,vmax);
v9 = (vector float)vec_cmplt(v1,vmin);
vsample1 = vec_ld(0,samples);
vsample2 = vec_ld(16,samples);
vsample3 = vec_ld(31,samples);
@ -857,24 +857,24 @@ int synth_1to1_s32_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fina
vec_st((vector signed int)v6,16,samples);
vec_st((vector signed int)v7,0,samples);
samples += 8;
v1 = (vector float)vec_sr((vector unsigned int)v8, vshift);
v2 = (vector float)vec_sr((vector unsigned int)v9, vshift);
v1 = (vector float)vec_add((vector unsigned int)v1,(vector unsigned int)v2);
vclip = vec_sums((vector signed int)v1,vclip);
}
for (j=4;j;j--)
{
SYNTH_ALTIVEC(-16);
vsum = vec_add(v5,v6);
v9 = vec_add(v7,v8);
v1 = vec_add(vsum,v9);
vsum = (vector float)vec_cts(v1,16);
v8 = (vector float)vec_cmpgt(v1,vmax);
v9 = (vector float)vec_cmplt(v1,vmin);
vsample1 = vec_ld(0,samples);
vsample2 = vec_ld(16,samples);
vsample3 = vec_ld(31,samples);
@ -891,18 +891,18 @@ int synth_1to1_s32_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fina
vec_st((vector signed int)v6,16,samples);
vec_st((vector signed int)v7,0,samples);
samples += 8;
v1 = (vector float)vec_sr((vector unsigned int)v8, vshift);
v2 = (vector float)vec_sr((vector unsigned int)v9, vshift);
v1 = (vector float)vec_add((vector unsigned int)v1,(vector unsigned int)v2);
vclip = vec_sums((vector signed int)v1,vclip);
}
vec_st(vclip,0,clip_tmp);
clip = clip_tmp[3];
}
if(final) fr->buffer.fill += 256;
return clip;
}
@ -910,7 +910,7 @@ int synth_1to1_s32_altivec(real *bandPtr,int channel,mpg123_handle *fr, int fina
int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handle *fr)
{
int32_t *samples = (int32_t *) (fr->buffer.data+fr->buffer.fill);
real *b0l, *b0r, **bufl, **bufr;
int clip;
int bo1;
@ -925,7 +925,7 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
fr->bo &= 0xf;
bufl = fr->real_buffs[0];
bufr = fr->real_buffs[1];
if(fr->bo & 0x1)
{
b0l = bufl[0];
@ -942,12 +942,12 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
dct64_altivec(bufl[0]+fr->bo,bufl[1]+fr->bo+1,bandPtr_l);
dct64_altivec(bufr[0]+fr->bo,bufr[1]+fr->bo+1,bandPtr_r);
}
{
register int j;
real *window = fr->decwin + 16 - bo1;
ALIGNED(16) int clip_tmp[4];
vector float v1,v2,v3,v4,v5,v6,v7,v8,v9,v10,v11,v12,v13;
vector unsigned char vperm1,vperm2;
@ -965,14 +965,14 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
vmax = (vector float){32767.999f,32767.999f,32767.999f,32767.999f};
vmin = (vector float){-32768.0f,-32768.0f,-32768.0f,-32768.0f};
#endif
vperm1 = vec_lvsl(0,window);
vperm2 = vec_lvsr(0,samples);
vprev = (vector float)vec_perm(vec_ld(0,samples),vec_ld(0,samples),vec_lvsl(0,samples));
for (j=4;j;j--)
{
SYNTH_STEREO_ALTIVEC(16);
vsum = vec_sub(vsum,vsum2);
vsum2 = vec_sub(vsum5,vsum6);
vsum3 = vec_sub(vsum3,vsum4);
@ -985,7 +985,7 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
v6 = (vector float)vec_cmplt(v1,vmin);
v7 = (vector float)vec_cmpgt(v2,vmax);
v8 = (vector float)vec_cmplt(v2,vmin);
v1 = vec_mergeh(vsum, vsum2);
v2 = vec_mergel(vsum, vsum2);
v3 = vec_perm(vprev,v1,vperm2);
@ -994,7 +994,7 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
vec_st((vector signed int)v3,0,samples);
vec_st((vector signed int)v4,16,samples);
samples += 8;
v1 = (vector float)vec_sr((vector unsigned int)v5, vshift);
v2 = (vector float)vec_sr((vector unsigned int)v6, vshift);
v3 = (vector float)vec_sr((vector unsigned int)v7, vshift);
@ -1004,11 +1004,11 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
vclip = vec_sums((vector signed int)v1,vclip);
vclip = vec_sums((vector signed int)v2,vclip);
}
for (j=4;j;j--)
{
SYNTH_STEREO_ALTIVEC(-16);
vsum = vec_add(vsum,vsum2);
vsum2 = vec_add(vsum5,vsum6);
vsum3 = vec_add(vsum3,vsum4);
@ -1021,7 +1021,7 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
v6 = (vector float)vec_cmplt(v1,vmin);
v7 = (vector float)vec_cmpgt(v2,vmax);
v8 = (vector float)vec_cmplt(v2,vmin);
v1 = vec_mergeh(vsum, vsum2);
v2 = vec_mergel(vsum, vsum2);
v3 = vec_perm(vprev,v1,vperm2);
@ -1030,7 +1030,7 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
vec_st((vector signed int)v3,0,samples);
vec_st((vector signed int)v4,16,samples);
samples += 8;
v1 = (vector float)vec_sr((vector unsigned int)v5, vshift);
v2 = (vector float)vec_sr((vector unsigned int)v6, vshift);
v3 = (vector float)vec_sr((vector unsigned int)v7, vshift);
@ -1040,18 +1040,18 @@ int synth_1to1_s32_stereo_altivec(real *bandPtr_l, real *bandPtr_r, mpg123_handl
vclip = vec_sums((vector signed int)v1,vclip);
vclip = vec_sums((vector signed int)v2,vclip);
}
if((size_t)samples & 0xf)
{
v1 = (vector float)vec_perm(vec_ld(0,samples),vec_ld(0,samples),vec_lvsl(0,samples));
v2 = (vector float)vec_perm(vprev,v1,vperm2);
vec_st((vector signed int)v2,0,samples);
}
vec_st(vclip,0,clip_tmp);
clip = clip_tmp[3];
}
fr->buffer.fill += 256;
return clip;
}