DRC [Fri, 5 Feb 2016 00:34:38 +0000 (18:34 -0600)]
Prevent overread when decoding malformed JPEG
The accelerated Huffman decoder was previously invoked if there were
> 128 bytes in the input buffer. However, it is possible to construct a
JPEG image with Huffman blocks > 430 bytes in length
(http://stackoverflow.com/questions/2734678/jpeg-calculating-max-size).
While such images are pathological and could never be created by a
JPEG compressor, it is conceivable that an attacker could use such an
artifially-constructed image to trigger an input buffer overrun in the
libjpeg-turbo decompressor and thus gain access to some of the data on
the calling program's heap.
This patch simply increases the minimum buffer size for the accelerated
Huffman decoder to 512 bytes, which should (hopefully) accommodate any
possible input.
This addresses a major issue (LJT-01-005) identified in a security audit
by Cure53.
DRC [Thu, 4 Feb 2016 16:59:21 +0000 (10:59 -0600)]
Guard against wrap-around in alloc functions
Because of the exposed nature of the libjpeg API, alloc_small() and
alloc_large() can potentially be called by external code. If an
application were to call either of those functions with
sizeofobject > SIZE_MAX - ALIGN_SIZE - 1, then the math in
round_up_pow2() would wrap around to zero, causing that function to
return a small value. That value would likely not exceed
MAX_ALLOC_CHUNK, so the subsequent size checks in alloc_small() and
alloc_large() would not catch the error.
A similar problem could occur in 32-bit builds if alloc_sarray() were
called with
samplesperrow > SIZE_MAX - (2 * ALIGN_SIZE / sizeof(JSAMPLE)) - 1
This patch simply ensures that the size argument to the alloc_*()
functions will never exceed MAX_ALLOC_CHUNK (1 billion). If it did,
then subsequent size checks would eventually catch that error, so we
are instead catching the error before round_up_pow2() is called.
This addresses a minor concern (LJT-01-001) expressed in a security
audit by Cure53.
DRC [Thu, 4 Feb 2016 15:20:41 +0000 (09:20 -0600)]
TurboJPEG: Avoid dangling pointers
This addresses a minor concern (LJT-01-002) expressed in a security
audit by Cure53. _tjInitCompress() and _tjInitDecompress() call
(respectively) jpeg_mem_dest_tj() and jpeg_mem_src_tj() with a pointer
to a dummy buffer, in order to set up the destination/source manager.
The dummy buffer should never be used, but it's still better to make it
static so that the pointer in the destination/source manager always
points to a valid region of memory.
DRC [Wed, 3 Feb 2016 20:02:13 +0000 (14:02 -0600)]
Adjust performance claims
Document the latest benchmarks on the Nexus 5X and change the "2-4x"
overall claim to "2-6x". The peak performance on x86 platforms was
already closer to 5x, and the addition of SIMD-accelerated Huffman
encoding gave it that extra push over the cliff.
DRC [Wed, 3 Feb 2016 05:17:06 +0000 (23:17 -0600)]
Use consistent formatting in ARM NEON SIMD code
There aren't really any best practices to follow here. I tried as best
as I could to adopt a standard that would ease any future maintenance
burdens. The basic tenets of that standard are:
* Assembly instructions always start on Column 5, and operands always
start on Column 21, except:
- The instruction and operand can be indented (usually by 2 spaces)
to indicate a separate instruction stream.
- If the instruction is within an enclosing .if block in a macro,
it should always be indented relative to the .if block.
* Comments are placed with an eye toward readability. There are always
at least 2 spaces between the end of a line of code and the associated
in-line comment. Where it made sense, I tried to line up the comments
in blocks, and some were shifted right to avoid overlap with
neighboring instruction lines. Not an exact science.
* Assembler directives and macros use 2-space indenting rules. .if
blocks are indented relative to the macro, and code within the .if
blocks is indented relative to the .if directive.
* No extraneous spaces between operands. Lining up the operands
vertically did not really improve readability-- personally, I think it
made it worse, since my eye would tend to lose its place in the
uniform columns of characters. Also, code with a lot of vertical
alignment is really hard to maintain, since changing one line could
necessitate changing a bunch of other lines to avoid spoiling the
alignment.
* No extraneous spaces in #defines or other directives. In general, the
only extraneous spaces (other than indenting spaces) are between:
- Instructions and operands
- Operands and in-line comments
This standard should be more or less in keeping with other formatting
standards used within the project.
Decompression speedup relative to libjpeg-turbo 1.4.2 (IFAST IDCT):
48-core ThunderX (RunAbove ARM Cloud), Linux, 64-bit: 51-98% (avg. 75%)
Minimal speedup (1-5%) observed on iPhone 5S (Cortex-A7)
NOTE: This commit avoids the st3 instruction for non-Android and
non-Apple builds, which may cause a performance regression against
libjpeg-turbo 1.4.x on ARM64 systems that are running plain Linux.
Since ThunderX is the only platform known to suffer from slow ld3 and
st3 instructions, it is probably better to check for the CPU type
at run time and disable ld3/st3 only if ThunderX is detected.
This commit also enables the use of ld3 on Android platforms, which
should be a safe bet, at least for now. This speeds up compression on
the afore-mentioned Nexus Cortex-A53 by 5.5-19% (avg. 12%) and on the
Nexus Cortex-A57 by 1.2-14% (avg. 6.3%), relative to the previous
commits.
DRC [Mon, 1 Feb 2016 17:28:55 +0000 (11:28 -0600)]
Update Android build instr. for ARMv8, PIE, etc.
* Include information on how to do a 64-bit ARMv8 build with the latest
NDK
* Suggest -fPIE and -pie as default CFLAGS (required for android-16 and
later.
* Remove -fstrict-aliasing flag (-Wall already includes it)
DRC [Mon, 1 Feb 2016 17:12:22 +0000 (11:12 -0600)]
Update Android build instr. for ARMv8, PIE, etc.
* Include information on how to do a 64-bit ARMv8 build with the latest
NDK
* Suggest -fPIE and -pie as default CFLAGS (required for android-16 and
later.
* Remove -fstrict-aliasing flag (-Wall already includes it)
DRC [Mon, 1 Feb 2016 17:03:39 +0000 (11:03 -0600)]
TJBench: Fix segfault on Android
For whatever reason, the "write" global variable in tjbench.c was
overriding the linkage with the write() system function. This may have
affected other platforms as well but was not known to.
DRC [Mon, 18 Jan 2016 22:40:07 +0000 (16:40 -0600)]
Provide pkg-config (.pc) scripts
This allows a project to use PKG_CHECK_MODULES() in its configure.ac
file to easily check for the presence of libjpeg-turbo and modify the
compiler/linker flags accordingly. Note that if a project relies solely
on pkg-config to check for libjpeg-turbo, then it will not be possible
to build that project using libjpeg or an earlier version of
libjpeg-turbo.
DRC [Sat, 16 Jan 2016 07:53:32 +0000 (01:53 -0600)]
Optimize ARM64 SIMD code for Cavium ThunderX
Per @ssvb:
ThunderX is an ARM64 chip that dedicates most of its transistor real
estate to providing 48 cores, so each core is not as fast as a result.
Each core is dual-issue & in-order for scalar instructions and has only
a single-issue half-width NEON unit, so the peak throughput is one
128-bit instruction per 2 cycles. So careful instruction scheduling is
important. Furthermore, ThunderX has an extremely slow implementation
of ld2 and ld3, so this commit implements the equivalent of those
instructions using ld1.
Compression speedup relative to libjpeg-turbo 1.4.2:
48-core ThunderX (RunAbove ARM Cloud), Linux, 64-bit: 58-85% (avg. 74%)
relative to jpeg-6b: 1.75-2.14x (avg. 1.95x)
Refer to #49 and #51 for discussion.
Closes #51.
This commit also wordsmiths the ChangeLog entry (the ARMv8 SIMD
implementation is "complete" only for compression-- it still lacks some
decompression algorithms, as does the ARMv7 implementation.)
DRC [Fri, 15 Jan 2016 15:29:11 +0000 (09:29 -0600)]
Complete the ARM64 NEON SIMD implementation
This adds 64-bit NEON coverage for all of the algorithms that are
covered by the 32-bit NEON implementation, except for h2v1 (4:2:2) fancy
upsampling (used when decompressing 4:2:2 JPEG images.) It also adds
64-bit NEON SIMD coverage for:
which are not covered in the 32-bit implementation.
Compression speedups relative to libjpeg-turbo 1.4.2:
Apple A7 (iPhone 5S), iOS, 64-bit: 113-150% (reported)
48-core ThunderX (RunAbove ARM Cloud), Linux, 64-bit: 2.1-33% (avg. 15%)
Refer to #44 and #49 for discussion
This commit also removes the unnecessary
if (simd_support & JSIMD_ARM_NEON)
statements from the jsimd* algorithm functions. Since the jsimd_can*()
functions check for the existence of NEON, the corresponding algorithm
functions will never be called if NEON isn't available.
DRC [Wed, 13 Jan 2016 09:13:20 +0000 (03:13 -0600)]
ARM32 NEON SIMD implementation of Huffman encoding
Full-color compression speedups relative to libjpeg-turbo 1.4.2:
800 MHz ARM Cortex-A9, iOS, 32-bit: 26-44% (avg. 32%)
Refer to #42 and #47 for discussion.
This commit also removes the unnecessary
if (simd_support & JSIMD_ARM_NEON)
statements from the jsimd* algorithm functions. Since the jsimd_can*()
functions check for the existence of NEON, the corresponding algorithm
functions will never be called if NEON isn't available. Removing those
if statements improved performance across the board by a couple of
percent.
DRC [Wed, 13 Jan 2016 18:25:03 +0000 (12:25 -0600)]
BUILDING.md: Restore autotools processing instr.
Partially reverts 54014d9c2a41905b7b766057af6728834da64b59. When
building from a git sandbox, as opposed to from an official source
tarball, it is still necessary to run autoreconf.
DRC [Sat, 20 Jun 2015 16:27:51 +0000 (16:27 +0000)]
Update build instructions for new autoconf, GitHub
The Linux build machine has been upgraded to autoconf 2.69, automake
1.15, m4 1.4.17, and libtool 2.4.6, so it is no longer necessary to
recommend running autoreconf prior to building the source, if one is
building from an official source tarball (as opposed to from a git
sandbox.) Also, there is no SVN repository anymore (oops.)
Refer to #42 for discussion. Numerous other approaches were attempted,
but this one proved to be the most performant across all platforms.
This commit also fixes #3 (works around, really-- the clang-compiled version
of jchuff.c still performs 20% worse than its GCC-compiled counterpart, but
that code is now bypassed by the new SSE2 Huffman algorithm.)
DRC [Tue, 12 Jan 2016 04:03:16 +0000 (22:03 -0600)]
Allow JSIMD_FORCENONE=1 env to disable x86-64 SIMD
Traditionally, the x86-64 code did not call init_simd() because it had
no need to (only SSE2 was supported.) However, having the ability to
disable SIMD at run time is a useful testing tool, and all of the other
SIMD implementations have this ability.
DRC [Thu, 7 Jan 2016 01:17:54 +0000 (19:17 -0600)]
Regression: Allow co-install of 32-bit/64-bit RPMs
Fix a regression introduced in 1.4.1 that prevented 32-bit and 64-bit
libjpeg-turbo RPMs from being installed simultaneously on recent Red
Hat/Fedora distributions. This was due to the addition of the
SIZEOF_SIZE_T macro in jconfig.h, which allows the Huffman codec to
determine the word size at compile time. Since that macro differs
between 32-bit and 64-bit builds, this caused a conflict between the
i386 and x86_64 RPMs (any differing files, other than executables, are
not allowed when 32-bit and 64-bit RPMs are installed simultaneously.)
Since the macro is used only internally, it has been moved into
jconfigint.h.
DRC [Wed, 6 Jan 2016 20:02:27 +0000 (14:02 -0600)]
Fix iOS ARM64 build broken by removal of .arch
The unnecessary .arch directive was removed from the ARM64 SIMD code
in d70a5c12fcb72443483456a2cc8dd18a4c238618, thus allowing clang's
integrated assembler to assemble the code on Linux systems. However,
this broke the detection mechanism in acinclude.m4 that tells the build
system whether it needs to use gas-preprocessor.pl. Since one of the
primary motivators for using gas-preprocessor.pl with ARM64 builds is
the lack of .req/.unreq directives in Apple's implementation of clang,
acinclude.m4 now checks whether .req/.unreq can be properly assembled
and uses gas-preprocessor.pl if not.
DRC [Thu, 17 Dec 2015 16:41:51 +0000 (10:41 -0600)]
cjpeg: Adjust claims RE: image quality settings
Quality values > 95 are not useless. They just may not provide as good
of a size vs. perceptual quality tradeoff as lower quality values. This
also displays the default quality value in the cjpeg usage.
DRC [Thu, 15 Oct 2015 03:26:25 +0000 (22:26 -0500)]
Fix additional issues reported by UB sanitizers
Most of these involved overrunning the signed 32-bit JLONG type whenever
building libjpeg-turbo with a 32-bit compiler. These issues are not
believed to represent actual security threats, but eliminating them
makes it easier to detect such threats should they arise in the future.
DRC [Wed, 14 Oct 2015 22:32:39 +0000 (17:32 -0500)]
Replace INT32 with a new internal datatype (JLONG)
These days, INT32 is a commonly-defined datatype in system headers. We
cannot eliminate the definition of that datatype from jmorecfg.h, since
the INT32 typedef has technically been part of the libjpeg API since
version 5 (1994.) However, using INT32 internally is risky, because the
inclusion of a particular header (Xmd.h, for instance) could change the
definition of INT32 from long to int on 64-bit platforms and thus change
the internal behavior of libjpeg-turbo in unexpected ways (for instance,
failing to correctly set __INT32_IS_ACTUALLY_LONG to match the INT32
typedef-- perhaps as a result of including the wrong version of
jpeglib.h-- could cause libjpeg-turbo to produce incorrect results.)
The library has always been built in environments in which INT32 is
effectively long (on Windows, long is always 32-bit, so effectively it's
the same as int), so it makes sense to turn INT32 into an explicitly
long datatype. This ensures that libjpeg-turbo will always behave
consistently, regardless of the headers included at compile time.
DRC [Sat, 10 Oct 2015 15:25:46 +0000 (10:25 -0500)]
Rename README, LICENSE, BUILDING text files
The IJG README file has been renamed to README.ijg, in order to avoid
confusion (many people were assuming that that was our project's README
file and weren't reading README-turbo.txt) and to lay the groundwork for
markdown versions of the libjpeg-turbo README and build instructions.
Fix various issues reported by the UB sanitizers
Most of these involved left shifting a negative number, which is
technically undefined (although every modern compiler I'm aware of
will implement this by treating the signed integer as a 2's complement
unsigned integer-- the LEFT_SHIFT() macro just makes this behavior
explicit in order to shut up ubsan.) This also fixes a couple of
non-issues in the entropy codecs, whereby the sanitizer reported an
out-of-bounds index in the 4th argument of jpeg_make_d_derived_tbl().
In those cases, the index was actually out of bounds (caused by a
malformed JPEG image), but jpeg_make_d_derived_tbl() would have caught
the error and aborted prior to actually using the invalid address. Here
again, the fix was to make our intentions explicit so as to shut up
ubsan.
James Cowgill [Thu, 17 Sep 2015 04:05:46 +0000 (23:05 -0500)]
Fix MIPS DSPr2 4:2:0 upsample bug w/ small images
The DSPr2 code was errantly comparing the residual (t9, width & 0xF)
with the end pointer (t4, out + width) instead of the width directly
(a1). This would give the wrong results with any image whose output
width was less than 16. The other small changes (ulw to lw and removal
of the nop) are just some easy optimizations around this code.
This issue caused a buffer overrun and subsequent segfault on images
whose scaled output height was 1 pixel and whose scaled output width was
< 16 pixels. Note that the "plain" (non-fancy and non-merged) upsample
routine, which was affected by this bug, is normally not used except
when decompressing a non-YCbCr JPEG image, but it is also used when
decompressing a single-row image (because the other upsampling
algorithms require at least two rows.)
Fix x86-64 ABI conformance issue in SIMD code
(descriptions cribbed by DRC from discussion in #20)
In the x86-64 ABI, the high (unused) DWORD of a 32-bit argument's
register is undefined, so it was incorrect to use a 64-bit mov
instruction to transfer a JDIMENSION argument in the 64-bit SSE2 SIMD
functions. The code worked thus far only because the existing compiler
optimizers weren't smart enough to do anything else with the register in
question, so the upper 32 bits happened to be all zeroes-- for the past
6 years, on every x86-64 compiler previously known to mankind.
The bleeding-edge Clang/LLVM compiler has a smarter optimizer, and
under certain circumstances, it will attempt to load-combine adjacent
32-bit integers from one of the libjpeg structures into a single 64-bit
integer and pass that 64-bit integer as a 32-bit argument to one of the
SIMD functions (which is allowed by the ABI, since the upper 32 bits of
the 32-bit argument's register are undefined.) This caused the
libjpeg-turbo regression tests to crash.
Also enhance the documentation of JDIMENSION to explain that its size
is significant to the implementation of the SIMD code.
Closes #20. Refer also to http://crbug.com/532214.
Add file that explains the libjpeg-turbo licenses
Previously this information was found in a page on libjpeg-turbo.org,
but there was still some confusion, because README-turbo.txt wasn't
clear as to which license applied to what.
DRC [Sat, 29 Aug 2015 23:05:43 +0000 (18:05 -0500)]
Fix negative shift with IFAST FDCT and qual=100
With certain images, compressing using quality=100 and the fast integer
forward DCT will cause the divisor passed to compute_reciprocal() to be
1. In those cases, the library already disables the SIMD quantization
algorithm to avoid 16-bit overflow. However, compute_reciprocal()
doesn't properly handle the divisor==1 case, so we need to use special
values in that case so that the C quantization algorithm will behave
like an identity function.
James Cowgill [Sat, 15 Aug 2015 12:30:14 +0000 (13:30 +0100)]
Fix build error when compiling MIPS SIMD w/ -mfpxx
When compiled with -mfpxx (which is now the default on Debian), there are
some restrictions on the use of odd-numbered FP registers. More details
about FPXX can be found here:
https://dmz-portal.mips.com/wiki/MIPS_O32_ABI_-_FR0_and_FR1_Interlinking
This commit simply changes all uses of FP registers to an even-numbered
equivalent like this:
f0 -> f0
f1 -> f2
f2 -> f4
...
f8 -> f16
This commit should have no observable effect except that the MIPS assembly
will now compile with -mfpxx.
DRC [Thu, 13 Aug 2015 16:09:05 +0000 (11:09 -0500)]
Fix cjpeg segfault when Windows BMP width/height<0
rdbmp.c used the ambiguous INT32 datatype, which is sometimes typedef'ed
to long. Windows bitmap headers use 32-bit signed integers for the
width and height, because height can sometimes be negative (this
indicates a top-down bitmap.) If biWidth or biHeight was negative and
INT32 was a 64-bit long, then biWidth and biHeight were read as a
positive integer > INT32_MAX, which failed the test in line 385:
if (biWidth <= 0 || biHeight <= 0)
ERREXIT(cinfo, JERR_BMP_EMPTY);
This commit refactors rdbmp.c so that it uses the datatypes specified by
Microsoft for the Windows BMP header.
This closes #9 and also provides a better solution for mozilla/mozjpeg#153.
DRC [Fri, 14 Aug 2015 01:06:03 +0000 (20:06 -0500)]
Declare source buffers in TurboJPEG C API as const
This reassures the caller that the buffers will not be modified and also
allows read-only buffers to be passed to the functions.
DRC [Thu, 13 Aug 2015 21:51:08 +0000 (16:51 -0500)]
Merge pull request #10 from pornel/const
Declare inbuffer arg in jpeg_mem_src() to be const
This reassures the caller that the buffer will not be modified and also
allows read-only buffers to be passed to the function.
DRC [Thu, 13 Aug 2015 16:09:05 +0000 (11:09 -0500)]
Fix cjpeg segfault when Windows BMP width/height<0
rdbmp.c used the ambiguous INT32 datatype, which is sometimes typedef'ed
to long. Windows bitmap headers use 32-bit signed integers for the
width and height, because height can sometimes be negative (this
indicates a top-down bitmap.) If biWidth or biHeight was negative and
INT32 was a 64-bit long, then biWidth and biHeight were read as a
positive integer > INT32_MAX, which failed the test in line 385:
if (biWidth <= 0 || biHeight <= 0)
ERREXIT(cinfo, JERR_BMP_EMPTY);
This commit refactors rdbmp.c so that it uses the datatypes specified by
Microsoft for the Windows BMP header.
This closes #9 and also provides a better solution for mozilla/mozjpeg#153.
DRC [Tue, 4 Aug 2015 04:59:27 +0000 (23:59 -0500)]
Update x86[-64] assembler recommendations
NASM 2.11.08 has a bug that prevents it from properly assembling a
macho64 version of libjpeg-turbo (the resulting binary generates corrupt
images.) 2.11.09 works properly. YASM also works properly and has been
a supported alternative since libjpeg-turbo 1.2.
DRC [Tue, 4 Aug 2015 04:59:12 +0000 (23:59 -0500)]
Update x86[-64] assembler recommendations
NASM 2.11.08 has a bug that prevents it from properly assembling a
macho64 version of libjpeg-turbo (the resulting binary generates corrupt
images.) 2.11.09 works properly. YASM also works properly and has been
a supported alternative since libjpeg-turbo 1.2.
DRC [Tue, 4 Aug 2015 04:58:40 +0000 (23:58 -0500)]
Update x86[-64] assembler recommendations
NASM 2.11.08 has a bug that prevents it from properly assembling a
macho64 version of libjpeg-turbo (the resulting binary generates corrupt
images.) 2.11.09 works properly. YASM also works properly and has been
a supported alternative since libjpeg-turbo 1.2.
DRC [Tue, 4 Aug 2015 04:56:09 +0000 (23:56 -0500)]
Update x86[-64] assembler recommendations
NASM 2.11.08 has a bug that prevents it from properly assembling a
macho64 version of libjpeg-turbo (the resulting binary generates corrupt
images.) 2.11.09 works properly. YASM also works properly and has been
a supported alternative since libjpeg-turbo 1.2.
Fix rare bug: right shift by a negative # of bits
Under very rare circumstances, decompressing specific corrupt JPEG
images would create a situation whereby GET_BITS(1) was invoked
from within HUFF_DECODE_FAST() when bits_left=0. This produced a right
shift by a negative number of bits, which is undefined in C.
Fix rare bug: right shift by a negative # of bits
Under very rare circumstances, decompressing specific corrupt JPEG
images would create a situation whereby GET_BITS(1) was invoked
from within HUFF_DECODE_FAST() when bits_left=0. This produced a right
shift by a negative number of bits, which is undefined in C.
Fix rare bug: right shift by a negative # of bits
Under very rare circumstances, decompressing specific corrupt JPEG
images would create a situation whereby GET_BITS(1) was invoked
from within HUFF_DECODE_FAST() when bits_left=0. This produced a right
shift by a negative number of bits, which is undefined in C.
Further improvements to partial image decoding
When using context-based upsampling, use a dummy color conversion
routine instead of a dummy row buffer. This improves performance
(since the actual color conversion routine no longer has to be called),
and it also fixes valgrind errors when decompressing to RGB565.
Valgrind previously complained, because using the RGB565 color
converter with the dummy row buffer was causing a table lookup with
undefined indices.
Fix rare bug: right shift by a negative # of bits
Under very rare circumstances, decompressing specific corrupt JPEG
images would create a situation whereby GET_BITS(1) was invoked
from within HUFF_DECODE_FAST() when bits_left=0. This produced a right
shift by a negative number of bits, which is undefined in C.
Further exception cleanup
Use a new checked exception type (TJException) when passing through
errors from the underlying C library. This gives the application a
choice of catching all exceptions or just those from TurboJPEG.
Throw IllegalArgumentException at the JNI level when arguments to the
JNI function are incorrect, and when one of the TurboJPEG "utility"
functions returns an error (because, per the C API specification, those
functions will only return an error if one of their arguments is out of
range.)
Remove "throws Exception" from the signature of any methods that no
longer pass through an error from the TurboJPEG C library.
Throw idiomatic unchecked exceptions from the Java classes and JNI wrapper if there is an unrecoverable error caused by incorrect API usage (such as illegal arguments, etc.), and throw Errors if there is an unrecoverable error at the C level (such as a failed malloc() call.)
Change the behavior of the bailif0() macro in the JNI wrapper so that it doesn't throw an exception for an unexpected NULL condition. In fact, in all cases, the underlying JNI API function (such as GetFieldID(), etc.) will throw an Error on its own whenever it returns NULL, so our custom exceptions were never being thrown in that case anyhow. All we need to do is just detect the error and bail out of the C code.
This also corrects a couple of formatting issues (semicolons aren't needed at the end of class definitions, and @Override should be specified for the methods we're overriding from super-classes, so the compiler can sanity-check that we're actually overriding a method and not declaring a new one.)
Add additional protections against defining INT32 if another header has already defined it (code borrowed from libjpeg v8.) This isn't necessary when using the libjpeg-turbo build system on Windows, because the CMake generated jconfig.h defines INT32 and then defines XMD_H to trick jmorecfg.h into not redefining it. However, some projects build libjpeg-turbo using their own build systems.
projects / libjpeg-turbo / log