]> git.ipfire.org Git - thirdparty/zlib-ng.git/log
thirdparty/zlib-ng.git
7 years ago[Issue #126] Fix implicit cast from unsigned char to int. 127/head
Mika Lindqvist [Thu, 21 Sep 2017 17:16:20 +0000 (20:16 +0300)] 
[Issue #126] Fix implicit cast from unsigned char to int.

7 years agoFix that s->prev is not used uninitialized in insert_string_* 115/head
Mika Lindqvist [Thu, 17 Aug 2017 18:35:15 +0000 (21:35 +0300)] 
Fix that s->prev is not used uninitialized in insert_string_*

7 years agoRevert "x86: use TZCNT (#113)"
Hans Kristian Rosbach [Thu, 24 Aug 2017 10:32:18 +0000 (12:32 +0200)] 
Revert "x86: use TZCNT (#113)"

Reverted after objections to its inclusion.

This reverts commit a7271104bf9a2d82dc6a69090c12442eacd2fd71.

7 years agoMake code an int in compress_block()
Milan Ševčík [Mon, 18 Jul 2016 20:27:59 +0000 (22:27 +0200)] 
Make code an int in compress_block()

send_code() expects int instead of unsigned. Other procedures do pass
int.

7 years agoconfigure: For Windows builds, add the CROSS_PREFIX to $RC and $STRIP.
nmlgc [Tue, 20 Jun 2017 20:12:42 +0000 (22:12 +0200)] 
configure: For Windows builds, add the CROSS_PREFIX to $RC and $STRIP.

zlib's original win32/Makefile.gcc did the same, but this was removed in
7d17132436431d5f62cf5089623073d72d07deb0. It is kind of essential for
cross-compiling a Win32 build on Linux, since `windres` most certainly
doesn't exist, and the regular `strip` may not be able to handle DLLs.

It should probably actually be something like

RC="${RC-${CROSS_PREFIX}windres}"

and

STRIP="${STRIP-${CROSS_PREFIX}strip}"

to be consistent with the assignments of $AR, $RANLIB and $NM, but this
didn't work for some reason.

7 years agoZLIB_COMPAT: add an extra 32 bits of padding in z_stream
R.J.V. Bertin [Tue, 23 May 2017 17:46:55 +0000 (19:46 +0200)] 
ZLIB_COMPAT: add an extra 32 bits of padding in z_stream

zlib "stock" uses an "uLong" for zstream::adler, meaning 4 bytes in 64
bit bits. The padding makes zlib-ng a drop-in replacement for libz; without,
the deflateInit2_() function returns a version error when called from
dependents that were built against "stock" zlib.

Committed from host : Portia.local

7 years agovarious CMake fixes:
R.J.V. Bertin [Tue, 23 May 2017 17:32:53 +0000 (19:32 +0200)] 
various  CMake fixes:

- on Mac, builds can target 1 or more architectures that are not the host
  architecture. Pick the first from the list and ignore the others.
  A more complete implementation would warn if i386 and x86_64 builds are
  mixed via the compiler options.
- use CMake's compiler IDs to detect GCC and Clang (should be applied to
  icc too but I can't test)
- disable PCLMUL optimisation in 32bit Mac builds. It crashes and provides
  very little gain (to builds that are probably increasingly rare)

Committed from host : Portia.local

7 years agoApply trivial CMake fixes based on feedback from RJVB in issue #110
Hans Kristian Rosbach [Thu, 17 Aug 2017 09:38:55 +0000 (11:38 +0200)] 
Apply trivial CMake fixes based on feedback from RJVB in issue #110

7 years agoMake sure we don't export internal functions
Hans Kristian Rosbach [Thu, 17 Aug 2017 09:24:46 +0000 (11:24 +0200)] 
Make sure we don't export internal functions

8 years agox86: use TZCNT (#113)
Mat [Wed, 31 May 2017 16:55:32 +0000 (18:55 +0200)] 
x86: use TZCNT (#113)

x86: use TZCNT instruction
On processors that do not support TZCNT, the instruction byte encoding is executed as BSF.
TZCNT is faster on AMD than BSF.

8 years agoFix: wrong register for BMI1 bit (#112)
Mat [Mon, 29 May 2017 09:06:26 +0000 (11:06 +0200)] 
Fix: wrong register for BMI1 bit (#112)

The BMI1 bit is in the ebx register and not in ecx.
See reference: https://software.intel.com/sites/default/files/article/405250/how-to-detect-new-instruction-support-in-the-4th-generation-intel-core-processor-family.pdf

8 years agoLazily initialize functable members. (#108)
Mika Lindqvist [Wed, 3 May 2017 17:14:57 +0000 (20:14 +0300)] 
Lazily initialize functable members. (#108)

- Split functableInit() function as separate functions for each functable member, so we don't need to initialize full functable in multiple places in the zlib-ng code, or to check for NULL on every invocation.
- Optimized function for each functable member is detected on first invocation and the functable item is updated for subsequent invocations.
- Remove NULL check in adler32() and adler32_z() as it is no longer needed.

8 years ago- Add adler32 to functable
Hans Kristian Rosbach [Mon, 24 Apr 2017 10:47:24 +0000 (12:47 +0200)] 
- Add adler32 to functable
- Add missing call to functableinit from inflateinit
- Fix external direct calls to adler32 functions without calling functableinit

8 years agoStyle cleanup
Hans Kristian Rosbach [Mon, 24 Apr 2017 09:43:17 +0000 (11:43 +0200)] 
Style cleanup

8 years agoAdd missing functable.h dependencies in arch makefiles
Hans Kristian Rosbach [Mon, 24 Apr 2017 09:32:05 +0000 (11:32 +0200)] 
Add missing functable.h dependencies in arch makefiles

8 years agoMerge branch 'hacknslash7' of github.com:Dead2/zlib-ng into hacknslash7
Hans Kristian Rosbach [Mon, 24 Apr 2017 09:24:50 +0000 (11:24 +0200)] 
Merge branch 'hacknslash7' of github.com:Dead2/zlib-ng into hacknslash7

8 years agoARM optimizations part 2 (#107)
Mika Lindqvist [Mon, 24 Apr 2017 09:22:11 +0000 (12:22 +0300)] 
ARM optimizations part 2 (#107)

*  add adler32_neon to main dependency checking and ARM/Windows Makefile
*  split non-optimized adler32 to adler32_c so we can test/compare both without recompiling.
*  add detection of default floating point ABI in gcc
    NOTE: This should avoid build error when gcc supports both ABIs but header for just one ABI is installed.

8 years agoMake -warn use the pedantic parameter that makes warnings, not errors.
Hans Kristian Rosbach [Mon, 24 Apr 2017 09:17:50 +0000 (11:17 +0200)] 
Make -warn use the pedantic parameter that makes warnings, not errors.

8 years agoStyle cleanup
Hans Kristian Rosbach [Mon, 24 Apr 2017 09:17:29 +0000 (11:17 +0200)] 
Style cleanup

8 years agoAdd a struct func_table and function functableInit.
Hans Kristian Rosbach [Mon, 24 Apr 2017 08:53:39 +0000 (10:53 +0200)] 
Add a struct func_table and function functableInit.
The struct contains pointers to select functions to be used by the
rest of zlib, and the init function selects what functions will be
used depending on what optimizations has been compiled in and what
instruction-sets are available at runtime.

Tests done on a haswell cpu running minigzip -6 compression of a
40M file shows a 2.5% decrease in branches, and a 25-30% reduction
in iTLB-loads. The reduction i iTLB-loads is likely mostly due to
the inability to inline functions. This also causes a slight
performance regression of around 1%, this might still be worth it
to make it much easier to implement new optimized functions for
various architectures and instruction sets.

The performance penalty will get smaller for functions that get more
alternative implementations to choose from, since there is no need
to add more branches to every call of the function.
Today insert_string has 1 branch to choose insert_string_sse
or insert_string_c, but if we also add for example insert_string_sse4
then that would have needed another branch, and it would probably
at some point hinder effective inlining too.

8 years agoImplementing NEON-ized Adler32 checksum (#102)
Adenilson Cavalcanti [Sat, 22 Apr 2017 09:41:47 +0000 (02:41 -0700)] 
Implementing NEON-ized Adler32 checksum (#102)

The checksum is calculated in the uncompressed PNG data and can be
made much faster by using SIMD. Tests in ARMv8 yielded an improvement
of about 3x (e.g. walltime was 350ms x 125ms for a 4096x4096 bytes
executed 30 times).

This yields an improvement in image decoding in Chromium around 18%
(see https://bugs.chromium.org/p/chromium/issues/detail?id=688601).

8 years agoNeon-Optimized hash chain rebase. (#106)
Jun [Thu, 20 Apr 2017 07:54:21 +0000 (15:54 +0800)] 
Neon-Optimized hash chain rebase. (#106)

* Neon-Optimized hash chain rebase.

Signed-off-by: Jun He <jun.he@arm.com>
8 years agoCMakeLists.txt: Fix cross-compiling. (#104)
Mika Lindqvist [Wed, 12 Apr 2017 19:30:58 +0000 (22:30 +0300)] 
CMakeLists.txt: Fix cross-compiling. (#104)

8 years agoMerge pull request #93 from sebpop/develop
Hans Kristian Rosbach [Fri, 31 Mar 2017 12:24:00 +0000 (14:24 +0200)] 
Merge pull request #93 from sebpop/develop

inflate: improve performance of memory copy operations

8 years agoMerge pull request #88 from mtl1979/arm
Hans Kristian Rosbach [Fri, 31 Mar 2017 12:17:43 +0000 (14:17 +0200)] 
Merge pull request #88 from mtl1979/arm

Implement ACLE/NEON optimizations for ARM/AARCH64

8 years agoCMakeLists.txt: We can't use check_c_source_runs() when cross-compiling. 88/head
Mika Lindqvist [Thu, 28 Apr 2016 18:46:17 +0000 (21:46 +0300)] 
CMakeLists.txt: We can't use check_c_source_runs() when cross-compiling.

8 years agoAdd initial support for ARM NEON vector instructions.
Mika Lindqvist [Sun, 10 Apr 2016 16:59:03 +0000 (19:59 +0300)] 
Add initial support for ARM NEON vector instructions.

8 years agoOptimize fill_window_c.
Mika Lindqvist [Sun, 10 Apr 2016 10:34:53 +0000 (13:34 +0300)] 
Optimize fill_window_c.

8 years agoinflate: improve performance of memory copy operations 93/head
Sebastian Pop [Thu, 16 Mar 2017 15:43:36 +0000 (10:43 -0500)] 
inflate: improve performance of memory copy operations

When memory copy operations happen byte by byte, the processors are unable to
fuse the loads and stores together because of aliasing issues.  This patch
clusters some of the memory copy operations in chunks of 16 and 8 bytes.

For byte memset, the compiler knows how to prepare the chunk to be stored.
When the memset pattern is larger than a byte, this patch builds the pattern for
chunk memset using the same technique as in Simon Hosie's patch
https://codereview.chromium.org/2722063002

This patch improves by 50% the performance of zlib decompression of a 50K PNG on
aarch64-linux and x86_64-linux when compiled with gcc-7 or llvm-5.

The number of executed instructions reported by valgrind --tool=cachegrind
on the decompression of a 50K PNG file on aarch64-linux:
- before the patch:
I   refs:      3,783,757,451
D   refs:      1,574,572,882  (869,116,630 rd   + 705,456,252 wr)

- with the patch:
I   refs:      2,391,899,214
D   refs:        899,359,836  (516,666,051 rd   + 382,693,785 wr)

The compression of a 260MB directory containing the code of llvm into a tar.gz
of 35MB and decompressing that with minigzip -d
on i7-4790K x86_64-linux, it takes 0.533s before the patch and 0.493s with the patch,
on Juno-r0 aarch64-linux A57, it takes 2.796s before the patch and 2.467s with the patch,
on Juno-r0 aarch64-linux A53, it takes 4.055s before the patch and 3.604s with the patch.

8 years agoAdd initial support for AARCH64.
Mika Lindqvist [Tue, 5 Apr 2016 12:28:35 +0000 (15:28 +0300)] 
Add initial support for AARCH64.

8 years agoAdd support for ARM ACLE instructions.
Mika Lindqvist [Sat, 2 Apr 2016 00:11:43 +0000 (03:11 +0300)] 
Add support for ARM ACLE instructions.

8 years agoAdd ARM implementation of CTZL for Visual C++.
Mika Lindqvist [Sun, 30 Oct 2016 12:36:09 +0000 (14:36 +0200)] 
Add ARM implementation of CTZL for Visual C++.

8 years agoInflate using wider loads and stores and a minimum of branches. (#95)
Simon Hosie [Wed, 22 Mar 2017 17:48:39 +0000 (10:48 -0700)] 
Inflate using wider loads and stores and a minimum of branches. (#95)

* Inflate using wider loads and stores.

In inflate_fast() the output pointer always has plenty of room to write.  This
means that so long as the target is capable, wide un-aligned loads and stores
can be used to transfer several bytes at once.

When the reference distance is too short simply unroll the data a little to
increase the distance.

Change-Id: I59854eb25d2b1e43561c8a2afaf9175bf10cf674

8 years agoLet fill_window_c() and fill_window_sse() use insert_string().
Hans Kristian Rosbach [Fri, 17 Mar 2017 11:17:20 +0000 (12:17 +0100)] 
Let fill_window_c() and fill_window_sse() use insert_string().
And let the code assume MIN_MATCH is 3, so a bulk insert can be used.

8 years agodeflate_medium: Make sure we have enough lookahead before trying to scan for matches.
Mika Lindqvist [Fri, 17 Mar 2017 19:36:38 +0000 (21:36 +0200)] 
deflate_medium: Make sure we have enough lookahead before trying to scan for matches.
* longest_match: Abort if match is in future

8 years agodeflate_medium: Remove broken overlap test.
Mika Lindqvist [Sun, 12 Mar 2017 00:52:54 +0000 (02:52 +0200)] 
deflate_medium: Remove broken overlap test.

8 years agoDon't pass unnecessary stream to fold_[1-4] and partial_fold.
Hans Kristian Rosbach [Fri, 24 Mar 2017 11:03:55 +0000 (12:03 +0100)] 
Don't pass unnecessary stream to fold_[1-4] and partial_fold.
Also fix some whitespace to make the code easier to read, and
better match the rest of the zlib-ng codebase.

8 years agoPrevent potential division-by-zero in gzfwrite and gzfread.
Hans Kristian Rosbach [Fri, 24 Mar 2017 13:27:59 +0000 (14:27 +0100)] 
Prevent potential division-by-zero in gzfwrite and gzfread.

8 years agoChange so travis compiles with warnings enabled, when using configure.
Hans Kristian Rosbach [Fri, 24 Mar 2017 13:10:33 +0000 (14:10 +0100)] 
Change so travis compiles with warnings enabled, when using configure.

8 years agoFix typos.
Mika Lindqvist [Mon, 13 Mar 2017 22:10:43 +0000 (00:10 +0200)] 
Fix typos.

8 years agocall memset for read after write dependences at distance 1
Sebastian Pop [Mon, 27 Feb 2017 17:21:59 +0000 (11:21 -0600)] 
call memset for read after write dependences at distance 1

On a benchmark using zlib to decompress a PNG image this change shows a 20%
speedup.  It makes sense to special case distance = 1 of read after write
dependences because it is possible to replace the loop kernel with a memset
which is usually implemented in assembly in the libc, and because of the
frequency at which distance = 1 appears during the PNG decompression:

Distance Frequency
1 1009001
6 64500
9 29000
3 25500
144 14500
12 10000
15 3500
7 2000
24 1000
21 1000
18 1000
87 500
22 500
192 500

8 years agoUpdate readme file, and add install instructions
Hans Kristian Rosbach [Tue, 28 Feb 2017 09:53:01 +0000 (10:53 +0100)] 
Update readme file, and add install instructions

8 years agoCMakeLists.txt: Fix tests on MinGW.
Mika Lindqvist [Fri, 24 Feb 2017 15:18:06 +0000 (17:18 +0200)] 
CMakeLists.txt: Fix tests on MinGW.

8 years agoType cleanup...
Mika Lindqvist [Fri, 17 Feb 2017 07:55:52 +0000 (09:55 +0200)] 
Type cleanup...
* gz_statep -> gz_state *

8 years agoCMakeLists.txt: Fix configuring under OS/X and MinGW
Mika Lindqvist [Wed, 15 Feb 2017 09:32:28 +0000 (11:32 +0200)] 
CMakeLists.txt: Fix configuring under OS/X and MinGW
* Enable CMAKE_MACOSX_RPATH
* Fix detection of PCLMULQDQ
* Fix detection of off64_t

NOTE: If test for __off64_t succeeds, it should mean bug in our tests

8 years agominigzip: add support for deflate_stored.
Mika Lindqvist [Fri, 17 Feb 2017 13:08:47 +0000 (15:08 +0200)] 
minigzip: add support for deflate_stored.

8 years agotest/example.c: Don't mix unsigned long and size_t.
Mika Lindqvist [Fri, 17 Feb 2017 13:01:24 +0000 (15:01 +0200)] 
test/example.c: Don't mix unsigned long and size_t.

8 years agoSome tests were missing EXE suffix.
Mika Lindqvist [Fri, 17 Feb 2017 12:11:22 +0000 (14:11 +0200)] 
Some tests were missing EXE suffix.

8 years agoLet all platforms defining UNALIGNED_OK use the optimized put_short
Hans Kristian Rosbach [Thu, 23 Feb 2017 08:04:48 +0000 (09:04 +0100)] 
Let all platforms defining UNALIGNED_OK use the optimized put_short
implementation. Also change from pre-increment to post-increment to
prevent a double-store on non-x86 platforms.

8 years agoLet all x86 and x86_64 archs use the new UPDATE_HASH implementation,
Hans Kristian Rosbach [Sat, 18 Feb 2017 18:26:52 +0000 (19:26 +0100)] 
Let all x86 and x86_64 archs use the new UPDATE_HASH implementation,
this improves compression performance and can often provide slightly
better compression.

8 years agoFix bug when level 0 used with Z_HUFFMAN or Z_RLE.
Mark Adler [Fri, 28 Oct 2016 05:50:43 +0000 (22:50 -0700)] 
Fix bug when level 0 used with Z_HUFFMAN or Z_RLE.

Compression level 0 requests no compression, using only stored
blocks. When Z_HUFFMAN or Z_RLE was used with level 0 (granted,
an odd choice, but permitted), the resulting blocks were mostly
fixed or dynamic. The reason is that deflate_stored() was not
being called in that case. The compressed data was valid, but it
was not what the application requested. This commit assures that
only stored blocks are emitted for compression level 0, regardless
of the strategy selected.

8 years agoClean up flushing override for level 1 if deflate_quick is enabled but sse4.2 not...
Hans Kristian Rosbach [Thu, 16 Feb 2017 10:43:40 +0000 (11:43 +0100)] 
Clean up flushing override for level 1 if deflate_quick is enabled but sse4.2 not supported.

8 years agoFix 32bit compilation on x86_64 platform
Hans Kristian Rosbach [Tue, 14 Feb 2017 15:05:53 +0000 (16:05 +0100)] 
Fix 32bit compilation on x86_64 platform

8 years agoType cleanup...
Mika Lindqvist [Tue, 14 Feb 2017 20:10:02 +0000 (22:10 +0200)] 
Type cleanup...
* uInt -> unsigned int
* ulg -> unsigned long

8 years agoUse insert_string instead of UPDATE_HASH to avoid double hashing.
Mika Lindqvist [Tue, 14 Feb 2017 07:51:32 +0000 (09:51 +0200)] 
Use insert_string instead of UPDATE_HASH to avoid double hashing.

8 years agoAvoid hashing same memory location twice by truncating overlapping byte ranges,
Mika Lindqvist [Tue, 14 Feb 2017 09:40:52 +0000 (11:40 +0200)] 
Avoid hashing same memory location twice by truncating overlapping byte ranges,
it's speed optimization as the inner code also checks that previous hash value
is not same as new hash value. Essentially those two checks together makes the
compression a little more efficient as it can remember matches further apart.
As far as I remember from my tests, the secondary path was triggered only twice
in very long uncompressed file, but the gain in compression rate was still noticeable.

8 years agoFix leftover charf
Hans Kristian Rosbach [Tue, 14 Feb 2017 11:19:09 +0000 (12:19 +0100)] 
Fix leftover charf

8 years agoMerge pull request #85 from mtl1979/cleanup
Hans Kristian Rosbach [Tue, 14 Feb 2017 06:55:30 +0000 (07:55 +0100)] 
Merge pull request #85 from mtl1979/cleanup

Cleanup configure and Makefile.in

8 years agoquick_send_bits: variable 'code' is not used anymore. 85/head
Mika Lindqvist [Mon, 13 Feb 2017 21:41:30 +0000 (23:41 +0200)] 
quick_send_bits: variable 'code' is not used anymore.

8 years agoFix build with nmake.
Mika Lindqvist [Mon, 13 Feb 2017 19:50:12 +0000 (21:50 +0200)] 
Fix build with nmake.

8 years agoAdd --native flag to configure
Mika Lindqvist [Sat, 30 Apr 2016 23:12:36 +0000 (02:12 +0300)] 
Add --native flag to configure
* When using gcc, it will detect instruction sets available on host CPU

8 years agoMerge branch 'hacknslash6' of github.com:Dead2/zlib-ng into hacknslash6
Hans Kristian Rosbach [Mon, 13 Feb 2017 20:07:20 +0000 (21:07 +0100)] 
Merge branch 'hacknslash6' of github.com:Dead2/zlib-ng into hacknslash6

8 years agoUse default osx version for travis build, seems travis has some capacity issues.
Hans Kristian Rosbach [Mon, 13 Feb 2017 20:02:42 +0000 (21:02 +0100)] 
Use default osx version for travis build, seems travis has some capacity issues.

8 years agoMakefile.in: Update rule to reset zconf.h.cmakein
Mika Lindqvist [Mon, 13 Feb 2017 16:07:39 +0000 (18:07 +0200)] 
Makefile.in: Update rule to reset zconf.h.cmakein

8 years agoAdd dependency tracking for ARCH files.
Mika Lindqvist [Fri, 29 Apr 2016 13:20:11 +0000 (16:20 +0300)] 
Add dependency tracking for ARCH files.

8 years agoCleanup configure and makefiles
Mika Lindqvist [Wed, 3 Jun 2015 00:20:20 +0000 (03:20 +0300)] 
Cleanup configure and makefiles

8 years agoMerge pull request #86 from mtl1979/archclean
Hans Kristian Rosbach [Mon, 13 Feb 2017 19:46:50 +0000 (20:46 +0100)] 
Merge pull request #86 from mtl1979/archclean

Replace remaining Z_NULL instances with NULL.

8 years agoFix only one half of a macro is executed in the correct side of the conditional,
Hans Kristian Rosbach [Mon, 13 Feb 2017 19:38:35 +0000 (20:38 +0100)] 
Fix only one half of a macro is executed in the correct side of the conditional,
causing the potential for hash corruption on calls to deflateParam() to change
level from 0 to something else.

8 years agoRestructure travis config and add tests for osx.
Hans Kristian Rosbach [Mon, 13 Feb 2017 19:20:06 +0000 (20:20 +0100)] 
Restructure travis config and add tests for osx.
Also start doing builds on ubuntu trusty (14.04) instead of precise (12.04).

8 years agoMerge pull request #87 from mtl1979/cvefix
Hans Kristian Rosbach [Mon, 13 Feb 2017 19:14:00 +0000 (20:14 +0100)] 
Merge pull request #87 from mtl1979/cvefix

Fix so Travis doesn't unnecessarily fail on CVE tests without --zlib-compat.

8 years agoFix so Travis doesn't fail on CVE tests without --zlib-compat. 87/head
Mika Lindqvist [Mon, 13 Feb 2017 17:26:09 +0000 (19:26 +0200)] 
Fix so Travis doesn't fail on CVE tests without --zlib-compat.

8 years agoReplace remaining Z_NULL instances with NULL. 86/head
Mika Lindqvist [Mon, 13 Feb 2017 16:17:37 +0000 (18:17 +0200)] 
Replace remaining Z_NULL instances with NULL.

8 years agoMerge pull request #84 from mtl1979/archclean
Hans Kristian Rosbach [Mon, 13 Feb 2017 14:17:40 +0000 (15:17 +0100)] 
Merge pull request #84 from mtl1979/archclean

Cleanup architure-dependent code

8 years agoMove Visual C++ workaround for CTZL to arch/x86. 84/head
Mika Lindqvist [Sun, 30 Oct 2016 12:12:42 +0000 (14:12 +0200)] 
Move Visual C++ workaround for CTZL to arch/x86.

8 years agoMSVC: Use _tzcnt_u32() if available.
Mika Lindqvist [Tue, 3 May 2016 18:03:54 +0000 (21:03 +0300)] 
MSVC: Use _tzcnt_u32() if available.

8 years agoMove instruction set specific crc32 code to arch directories.
Mika Lindqvist [Mon, 4 Apr 2016 14:01:39 +0000 (17:01 +0300)] 
Move instruction set specific crc32 code to arch directories.

8 years agoCMakeLists.txt: Cleanup.
Mika Lindqvist [Mon, 2 May 2016 11:14:01 +0000 (14:14 +0300)] 
CMakeLists.txt: Cleanup.

8 years agoAdd support for internal attribute
Mika Lindqvist [Thu, 28 Apr 2016 19:48:15 +0000 (22:48 +0300)] 
Add support for internal attribute
The advantage of this over hidden is for example that the compiler can
safely assume that pointers to functions declared internal can never be
passed externally. This allows the compiler to consider optimizations
otherwise impossible.

8 years agoRequire cmake 2.8.4.
Mika Lindqvist [Wed, 27 Apr 2016 13:43:55 +0000 (16:43 +0300)] 
Require cmake 2.8.4.
* Add duplicate check to add_intrinsics_option().

8 years agoCMakeLists.txt: use check_c_source_runs instead of check_c_source_compiles
René J.V. Bertin [Thu, 11 Jun 2015 20:08:19 +0000 (22:08 +0200)] 
CMakeLists.txt: use check_c_source_runs instead of check_c_source_compiles
to try to avoid using intrinsics and an instruction set the compiler
knows but the host CPU doesn't support.

8 years agoCMakeLists.txt : preliminary support for MSVC and ICC
René J.V. Bertin [Fri, 12 Jun 2015 13:13:49 +0000 (15:13 +0200)] 
CMakeLists.txt : preliminary support for MSVC and ICC

- select the CMAKE_BUILD_TYPE "Release" by default if none has been set,
  to ensure maximum generic optimisation possible on the host platform
- add WITH_NATIVE_INSTRUCTIONS to build with -march=native or its equivalent
  option with other compilers (when we identify those alternatives)
- NATIVEFLAG (-march=native) will be used instead of -msseN/-mpclmul when
  defined/requested

TODO: discuss whether -msseN/-mpclmul should be used only for the files that
need them instead of globally, while NATIVEFLAG can (is supposed to) be used
globally.

8 years agoCMakeLists.txt: better checking for Intel intrinsics.
René J.V. Bertin [Thu, 11 Jun 2015 17:20:41 +0000 (19:20 +0200)] 
CMakeLists.txt: better checking for Intel intrinsics.

The checks currently assume that instructions that build also execute.
This is not necessarily true: building with -msse4 on an AMD CPU (a C60)
that only has SSE4a leads to a crash in deflateInit2 when the compiler
apparently uses an unsupported instruction to set
s->hash_bits = memLevel + 7;

8 years agoarchitecture checks translated from configure to cmake; initial draft
René J.V. Bertin [Tue, 9 Jun 2015 21:59:37 +0000 (23:59 +0200)] 
architecture checks translated from configure to cmake; initial draft

8 years agoInitialize block_open state
Phil Vachon [Mon, 30 Jan 2017 14:32:29 +0000 (15:32 +0100)] 
Initialize block_open state

On deflation context creation, initialize the block_open state to 0 to
ensure that no uninitialized values are used.

8 years agoAdd block_open state for deflate_quick
Phil Vachon [Mon, 30 Jan 2017 14:28:25 +0000 (15:28 +0100)] 
Add block_open state for deflate_quick

By storing whether or not a block has been opened (or terminated), the
static trees used for the block and the end block markers can be emitted
appropriately.

8 years agoFix Partial Symbol Generation for QUICK deflate
Phil Vachon [Mon, 30 Jan 2017 14:20:20 +0000 (15:20 +0100)] 
Fix Partial Symbol Generation for QUICK deflate

When using deflate_quick() in a streaming fashion and the output buffer
runs out of space while the input buffer still has data, deflate_quick()
would emit partial symbols. Force the deflate_quick() loop to terminate
for a flush before any further processing is done, returning to the main
deflate() routine to do its thing.

8 years agozlib 1.2.11
Mark Adler [Sun, 15 Jan 2017 16:22:16 +0000 (08:22 -0800)] 
zlib 1.2.11

8 years agoPermit immediate deflateParams changes before any deflate input.
Mark Adler [Sun, 15 Jan 2017 16:22:16 +0000 (08:22 -0800)] 
Permit immediate deflateParams changes before any deflate input.

This permits deflateParams to change the strategy and level right
after deflateInit, without having to wait until a header has been
written. The parameters can be changed immediately up until the
first deflate call that consumes any input data.

8 years agoUpdate high water mark in deflate_stored.
Mark Adler [Sun, 15 Jan 2017 16:15:55 +0000 (08:15 -0800)] 
Update high water mark in deflate_stored.

This avoids unnecessary filling of bytes in the sliding window
buffer when switching from level zero to a non-zero level. This
also provides a consistent indication of deflate having taken
input for a later commit ...

8 years agoUpdate vestigial comment from very old Info-ZIP deflate.
Mark Adler [Sat, 14 Jan 2017 07:10:03 +0000 (23:10 -0800)] 
Update vestigial comment from very old Info-ZIP deflate.

8 years agoFix deflate stored bug when pulling last block from window.
Mark Adler [Fri, 13 Jan 2017 05:51:20 +0000 (21:51 -0800)] 
Fix deflate stored bug when pulling last block from window.

And some cosmetic cleanups.

8 years agoRemove files to be installed before copying them in Makefile.in.
Hans Kristian Rosbach [Mon, 13 Feb 2017 09:07:29 +0000 (10:07 +0100)] 
Remove files to be installed before copying them in Makefile.in.
Based on upstream commit c7d77756d01210790b6d87442eba83dcc586a305

8 years agoFix bug in gzwrite.c that produced corrupt gzip files.
Mark Adler [Tue, 3 Jan 2017 00:17:43 +0000 (16:17 -0800)] 
Fix bug in gzwrite.c that produced corrupt gzip files.

8 years agoFix bug in deflate_stored() for zero-length input.
Mark Adler [Mon, 2 Jan 2017 23:17:59 +0000 (15:17 -0800)] 
Fix bug in deflate_stored() for zero-length input.

8 years agoMinor edits and clarifications of comments.
Mark Adler [Mon, 2 Jan 2017 04:02:24 +0000 (20:02 -0800)] 
Minor edits and clarifications of comments.

8 years agoAvoid warnings on snprintf() return value.
Mark Adler [Sun, 1 Jan 2017 20:23:04 +0000 (12:23 -0800)] 
Avoid warnings on snprintf() return value.

8 years agozlib 1.2.9
Mark Adler [Sat, 31 Dec 2016 19:41:31 +0000 (11:41 -0800)] 
zlib 1.2.9

8 years agoAdd crc32_z() and adler32_z() functions with size_t lengths.
Mark Adler [Sun, 1 Jan 2017 00:57:26 +0000 (16:57 -0800)] 
Add crc32_z() and adler32_z() functions with size_t lengths.

8 years agoUse a uniform approach for the largest value of an unsigned type.
Mark Adler [Sat, 31 Dec 2016 16:49:17 +0000 (08:49 -0800)] 
Use a uniform approach for the largest value of an unsigned type.

8 years agoAvoid some random compiler warnings on various platforms.
Mark Adler [Sat, 31 Dec 2016 06:05:05 +0000 (22:05 -0800)] 
Avoid some random compiler warnings on various platforms.