| Message ID | 20210601073315.9104-1-michael@niedermayer.cc |
|---|---|
| State | Accepted |
| Headers | show |
| Series | [FFmpeg-devel,1/3] avcodec/lpc: check for zero err in normalization in compute_lpc_coefs() | expand |
| Context | Check | Description |
|---|---|---|
| andriy/x86_make | success | Make finished |
| andriy/x86_make_fate | success | Make fate finished |
| andriy/PPC64_make | success | Make finished |
| andriy/PPC64_make_fate | success | Make fate finished |
Quoting Michael Niedermayer (2021-06-01 09:33:13) > Fixes: floating point division by 0 > Fixes: Ticket8213 > > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> > --- > libavcodec/lpc.h | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h > index 52170fd623..c99e568794 100644 > --- a/libavcodec/lpc.h > +++ b/libavcodec/lpc.h > @@ -186,7 +186,8 @@ static inline int AAC_RENAME(compute_lpc_coefs)(const LPC_TYPE *autoc, int max_o > for(j=0; j<i; j++) > r -= lpc_last[j] * autoc[i-j-1]; > > - r /= err; > + if (r || err) Why check for non-zero r?
On Sat, Jun 05, 2021 at 03:44:49PM +0200, Anton Khirnov wrote: > Quoting Michael Niedermayer (2021-06-01 09:33:13) > > Fixes: floating point division by 0 > > Fixes: Ticket8213 > > > > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> > > --- > > libavcodec/lpc.h | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h > > index 52170fd623..c99e568794 100644 > > --- a/libavcodec/lpc.h > > +++ b/libavcodec/lpc.h > > @@ -186,7 +186,8 @@ static inline int AAC_RENAME(compute_lpc_coefs)(const LPC_TYPE *autoc, int max_o > > for(j=0; j<i; j++) > > r -= lpc_last[j] * autoc[i-j-1]; > > > > - r /= err; > > + if (r || err) > > Why check for non-zero r? The idea was to make it clear what the case was that this checked for i can drop the r check if you prefer ? [...]
Quoting Michael Niedermayer (2021-06-05 16:14:24) > On Sat, Jun 05, 2021 at 03:44:49PM +0200, Anton Khirnov wrote: > > Quoting Michael Niedermayer (2021-06-01 09:33:13) > > > Fixes: floating point division by 0 > > > Fixes: Ticket8213 > > > > > > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> > > > --- > > > libavcodec/lpc.h | 3 ++- > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h > > > index 52170fd623..c99e568794 100644 > > > --- a/libavcodec/lpc.h > > > +++ b/libavcodec/lpc.h > > > @@ -186,7 +186,8 @@ static inline int AAC_RENAME(compute_lpc_coefs)(const LPC_TYPE *autoc, int max_o > > > for(j=0; j<i; j++) > > > r -= lpc_last[j] * autoc[i-j-1]; > > > > > > - r /= err; > > > + if (r || err) > > > > Why check for non-zero r? > > The idea was to make it clear what the case was that this checked for > i can drop the r check if you prefer ? So something guarantees that when r is non-zero then err is also non-zero? It's not obvious to me, but I don't understand that code. Maybe a comment would be enough.
On Sat, Jun 05, 2021 at 06:12:47PM +0200, Anton Khirnov wrote: > Quoting Michael Niedermayer (2021-06-05 16:14:24) > > On Sat, Jun 05, 2021 at 03:44:49PM +0200, Anton Khirnov wrote: > > > Quoting Michael Niedermayer (2021-06-01 09:33:13) > > > > Fixes: floating point division by 0 > > > > Fixes: Ticket8213 > > > > > > > > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> > > > > --- > > > > libavcodec/lpc.h | 3 ++- > > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h > > > > index 52170fd623..c99e568794 100644 > > > > --- a/libavcodec/lpc.h > > > > +++ b/libavcodec/lpc.h > > > > @@ -186,7 +186,8 @@ static inline int AAC_RENAME(compute_lpc_coefs)(const LPC_TYPE *autoc, int max_o > > > > for(j=0; j<i; j++) > > > > r -= lpc_last[j] * autoc[i-j-1]; > > > > > > > > - r /= err; > > > > + if (r || err) > > > > > > Why check for non-zero r? > > > > The idea was to make it clear what the case was that this checked for > > i can drop the r check if you prefer ? > > So something guarantees that when r is non-zero then err is also > non-zero? It's not obvious to me, but I don't understand that code. > Maybe a comment would be enough. With rounding any combination can likely occur so i will just check for err. I should have done this in the initial patch, what i did in it was insufficient. thx [...]
diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h index 52170fd623..c99e568794 100644 --- a/libavcodec/lpc.h +++ b/libavcodec/lpc.h @@ -186,7 +186,8 @@ static inline int AAC_RENAME(compute_lpc_coefs)(const LPC_TYPE *autoc, int max_o for(j=0; j<i; j++) r -= lpc_last[j] * autoc[i-j-1]; - r /= err; + if (r || err) + r /= err; err *= FIXR(1.0) - (r * r); }
Fixes: floating point division by 0 Fixes: Ticket8213 Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> --- libavcodec/lpc.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)