diff options
author | Jean-Marc Valin <jmvalin@jmvalin.ca> | 2016-07-21 19:40:23 -0400 |
---|---|---|
committer | Jean-Marc Valin <jmvalin@jmvalin.ca> | 2016-07-22 15:30:19 -0400 |
commit | 273906404c1647ef9f43c27fc1efced3c7af96f4 (patch) | |
tree | 1f19bfdacd528081b1f2c9debf82113cc29ff0c6 /celt/kiss_fft.c | |
parent | ae3321849596094d204f9c3dbe485b5fde6e5782 (diff) | |
download | opus-273906404c1647ef9f43c27fc1efced3c7af96f4.tar.gz |
silencing overflows in MDCT and FFT
Diffstat (limited to 'celt/kiss_fft.c')
-rw-r--r-- | celt/kiss_fft.c | 56 |
1 files changed, 28 insertions, 28 deletions
diff --git a/celt/kiss_fft.c b/celt/kiss_fft.c index 1f8fd053..83775165 100644 --- a/celt/kiss_fft.c +++ b/celt/kiss_fft.c @@ -82,8 +82,8 @@ static void kf_bfly2( C_SUB( Fout2[0] , Fout[0] , t ); C_ADDTO( Fout[0] , t ); - t.r = S_MUL(Fout2[1].r+Fout2[1].i, tw); - t.i = S_MUL(Fout2[1].i-Fout2[1].r, tw); + t.r = S_MUL(ADD32_ovflw(Fout2[1].r, Fout2[1].i), tw); + t.i = S_MUL(SUB32_ovflw(Fout2[1].i, Fout2[1].r), tw); C_SUB( Fout2[1] , Fout[1] , t ); C_ADDTO( Fout[1] , t ); @@ -92,8 +92,8 @@ static void kf_bfly2( C_SUB( Fout2[2] , Fout[2] , t ); C_ADDTO( Fout[2] , t ); - t.r = S_MUL(Fout2[3].i-Fout2[3].r, tw); - t.i = S_MUL(-Fout2[3].i-Fout2[3].r, tw); + t.r = S_MUL(SUB32_ovflw(Fout2[3].i, Fout2[3].r), tw); + t.i = S_MUL(NEG32_ovflw(ADD32_ovflw(Fout2[3].i, Fout2[3].r)), tw); C_SUB( Fout2[3] , Fout[3] , t ); C_ADDTO( Fout[3] , t ); Fout += 8; @@ -126,10 +126,10 @@ static void kf_bfly4( C_ADDTO( *Fout , scratch1 ); C_SUB( scratch1 , Fout[1] , Fout[3] ); - Fout[1].r = scratch0.r + scratch1.i; - Fout[1].i = scratch0.i - scratch1.r; - Fout[3].r = scratch0.r - scratch1.i; - Fout[3].i = scratch0.i + scratch1.r; + Fout[1].r = ADD32_ovflw(scratch0.r, scratch1.i); + Fout[1].i = SUB32_ovflw(scratch0.i, scratch1.r); + Fout[3].r = SUB32_ovflw(scratch0.r, scratch1.i); + Fout[3].i = ADD32_ovflw(scratch0.i, scratch1.r); Fout+=4; } } else { @@ -160,10 +160,10 @@ static void kf_bfly4( tw3 += fstride*3; C_ADDTO( *Fout , scratch[3] ); - Fout[m].r = scratch[5].r + scratch[4].i; - Fout[m].i = scratch[5].i - scratch[4].r; - Fout[m3].r = scratch[5].r - scratch[4].i; - Fout[m3].i = scratch[5].i + scratch[4].r; + Fout[m].r = ADD32_ovflw(scratch[5].r, scratch[4].i); + Fout[m].i = SUB32_ovflw(scratch[5].i, scratch[4].r); + Fout[m3].r = SUB32_ovflw(scratch[5].r, scratch[4].i); + Fout[m3].i = ADD32_ovflw(scratch[5].i, scratch[4].r); ++Fout; } } @@ -212,18 +212,18 @@ static void kf_bfly3( tw1 += fstride; tw2 += fstride*2; - Fout[m].r = Fout->r - HALF_OF(scratch[3].r); - Fout[m].i = Fout->i - HALF_OF(scratch[3].i); + Fout[m].r = SUB32_ovflw(Fout->r, HALF_OF(scratch[3].r)); + Fout[m].i = SUB32_ovflw(Fout->i, HALF_OF(scratch[3].i)); C_MULBYSCALAR( scratch[0] , epi3.i ); C_ADDTO(*Fout,scratch[3]); - Fout[m2].r = Fout[m].r + scratch[0].i; - Fout[m2].i = Fout[m].i - scratch[0].r; + Fout[m2].r = ADD32_ovflw(Fout[m].r, scratch[0].i); + Fout[m2].i = SUB32_ovflw(Fout[m].i, scratch[0].r); - Fout[m].r -= scratch[0].i; - Fout[m].i += scratch[0].r; + Fout[m].r = SUB32_ovflw(Fout[m].r, scratch[0].i); + Fout[m].i = ADD32_ovflw(Fout[m].i, scratch[0].r); ++Fout; } while(--k); @@ -282,22 +282,22 @@ static void kf_bfly5( C_ADD( scratch[8],scratch[2],scratch[3]); C_SUB( scratch[9],scratch[2],scratch[3]); - Fout0->r += scratch[7].r + scratch[8].r; - Fout0->i += scratch[7].i + scratch[8].i; + Fout0->r = ADD32_ovflw(Fout0->r, ADD32_ovflw(scratch[7].r, scratch[8].r)); + Fout0->i = ADD32_ovflw(Fout0->i, ADD32_ovflw(scratch[7].i, scratch[8].i)); - scratch[5].r = scratch[0].r + S_MUL(scratch[7].r,ya.r) + S_MUL(scratch[8].r,yb.r); - scratch[5].i = scratch[0].i + S_MUL(scratch[7].i,ya.r) + S_MUL(scratch[8].i,yb.r); + scratch[5].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,ya.r), S_MUL(scratch[8].r,yb.r))); + scratch[5].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,ya.r), S_MUL(scratch[8].i,yb.r))); - scratch[6].r = S_MUL(scratch[10].i,ya.i) + S_MUL(scratch[9].i,yb.i); - scratch[6].i = -S_MUL(scratch[10].r,ya.i) - S_MUL(scratch[9].r,yb.i); + scratch[6].r = ADD32_ovflw(S_MUL(scratch[10].i,ya.i), S_MUL(scratch[9].i,yb.i)); + scratch[6].i = NEG32_ovflw(ADD32_ovflw(S_MUL(scratch[10].r,ya.i), S_MUL(scratch[9].r,yb.i))); C_SUB(*Fout1,scratch[5],scratch[6]); C_ADD(*Fout4,scratch[5],scratch[6]); - scratch[11].r = scratch[0].r + S_MUL(scratch[7].r,yb.r) + S_MUL(scratch[8].r,ya.r); - scratch[11].i = scratch[0].i + S_MUL(scratch[7].i,yb.r) + S_MUL(scratch[8].i,ya.r); - scratch[12].r = - S_MUL(scratch[10].i,yb.i) + S_MUL(scratch[9].i,ya.i); - scratch[12].i = S_MUL(scratch[10].r,yb.i) - S_MUL(scratch[9].r,ya.i); + scratch[11].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,yb.r), S_MUL(scratch[8].r,ya.r))); + scratch[11].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,yb.r), S_MUL(scratch[8].i,ya.r))); + scratch[12].r = SUB32_ovflw(S_MUL(scratch[9].i,ya.i), S_MUL(scratch[10].i,yb.i)); + scratch[12].i = SUB32_ovflw(S_MUL(scratch[10].r,yb.i), S_MUL(scratch[9].r,ya.i)); C_ADD(*Fout2,scratch[11],scratch[12]); C_SUB(*Fout3,scratch[11],scratch[12]); |