In some architectures GLib macros to change endianness use the
argument multiple times causing possible side effects.
This happens for instance using Debian SID and MIPS.
This fixes https://gitlab.freedesktop.org/spice/spice-common/issues/1.
Reported-by: Laurent Bigonville <bigon@debian.org>
Tested-by: Laurent Bigonville <bigon@debian.org>
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>
Now that the 2 template files are the same (except for whitespace differences),
we can use a single file.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Most users of spice_common.h don't need it, or only need log.h. It only
has a few users outside of spice-common. It's not very well defined
which headers it should contain. This commit removes spice_common.h in
favour of direct inclusion of the needed headers.
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
golomb_coding is always followed by a encode call.
Simplify code calling directly it, no reason to pass back output
using pointers.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
Different processors has specific instructions to count leading
zero bits. This includes: x86. x64, arm, ppc.
For portability reason the behaviour of __builtin_clz is not
defined if the value is zero so test for it.
Currently the function is not called with the value or 0.
This increase performance decoding of about 4-5% on a x64 machine
(code size decreases a little too, but about 0.1%).
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>
Nothing uses it since the GL backend was removed in 384698a
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
The image is going to network and network protocol is little endian
so the numbers has to be little endian. Note that this is already done
during decoding.
Tested on a ppc64 machine.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>
Some assert in the code are doing some paranoid test and in code
paths quite hot.
The encoding time is reduced by 30-50% while the decoding time
is reduced by a 20-30%.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>
The quic code goes through a function pointer in two places in order to
try to prevent the compiler from inlining code.
Doing performance measurements this trick does not work anymore
and just make code less readable.
This patch and message was based on a previous work of
Christophe Fergeau.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>
We don't need 2 different implementations when the only difference is
the CommonState which is being used.
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
It's always set, no need to have conditional compilation based on it.
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
It's always set, no need to have conditional compilation based on it.
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
It's always set, no need to have conditional compilation based on it.
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
It's hardcoded at compile-time, and I don't think it was changed in
years...
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
It's hardcoded at compile-time, and I don't think it was changed in
years...
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
encodes_ones is called to encode a long sequence of 1 bits.
In some conditions (I manage to reproduce with a 85000x4 pixel
image fill with a single color) encodes_ones is called with a
"n" value >= 32.
This cause encode to be called with a "len" value of 32 which
trigger this assert:
spice_assert(len > 0 && len < 32);
causing a crash. Instead of calling encode with a constant
"len" as 32 call encode_32 which is supposed to encode
exactly 32 bit.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
In most occurrences bppmask is converted to 32 bit anyway.
In the left one a possible more bigger precision is not needed.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
This field is easily accessible from Encoder structure.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
We never change its value, so we basically have 2 constants for the
same thing, DEFwmimax and wmimax. This commit removes the latter.
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
We never change its value, so we basically have 2 constants for the
same thing, DEFwminext and wminext. This commit removes the latter
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
We never change 'evol' value, and this is currently causing issues with
gcc 7.1.1: quic.c is checking at compile-time that 'evol' is 1, 3 or 5.
This is a constant, so a static check should be good, but the compiler (gcc
7.1.1) is unable to know 'evol' value at compile-time. Since the removal
of spice_static_assert in favour of SPICE_VERIFY, this causes a
compile-time failure.
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
The macro was misused and not doing static check.
Spice have other working static check macros to use.
The macro is used only by spice-common so removing it
does not cause issues to other depending projects.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Marc-André Lureau <marcandre.lureau@redhat.com>
On Windows long is always 32 bit so under x64 the cast from pointer to
"unsigned long" cause a warning.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
Avoid having to call function at runtime to inialize static.
Old functions are defined as deprecated for compatibility.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
Code is in the slow path, this reduce space needed for data+code.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
Runtime conversion from a string to uint32 is storing the magic with the same
endianness on both LE and BE machines. This requires aditional byte swap
when sending magic between LE/BE machines.
Changing quic magic to a constant will ensure, that it will be always stored in
native endianness, and the second byte swap won't be needed.
Check for MELCSTATES - 1 to get inside the branch, otherwise
(...)->rgb_state.melcstate may be up to MELCSTATES after the
pre-incrementing, which would result in an access to a position
that is out bounds of the array size MELCSTATES.
find_model_params() is first doing *nbuckets = 0; and it then checks
nbuckets for NULL. This is redundant as the dereferencing would cause a
segfault if nbuckets was NULL, so the if (nbuckets) test can't be false.
As Uri pointed out, the "/* bucket start */" comment on the same line
probably implies that the test was meant to be 'if (*nbuckets)'
I've ran a few test and I did not observe issues because of it...
Unhandled values call an error callback, and then fall through the default:
case, which will call again the error callback. This commit adds some
break; after these cases to avoid this.
We can avoid repetitive computation by using two precomputed array, of
8k each.
before:
1.79% lt-spicy-stats libspice-client-glib-2.0.so.8.4.0 [.]
golomb_code_len_8bpc
after:
0.79% lt-spicy-stats libspice-client-glib-2.0.so.8.4.0 [.]
golomb_code_len_8bpc
The quic code has been changed recently this way:
- ASSERT(encoder->usr, line);
+ if (line == NULL) {
+ spice_warn_if_reached();
+ return QUIC_ERROR;
+ }
It appears that the only caller of quic_encode() gives a NULL line and
rely on the more_lines() callback to return new lines instead.
Adjust the code accordingly, adding a few more checks to verify the
caller gives/returns correct values.
Be more library friendly, by not aborting in library errors.
spice_common now includes a proper log handler that will abort by
default when reaching a warning.
SPICE_ABORT_LEVEL can be changed to modify run-time abort level.
SPICE_DEBUG_LEVEL can be changed to be more verbose. By default, only
log level more importants than WARNING.
Only memory-related functions are allowed to abort(), since they are
not recoverable errors in the library.
