Merge branch 'libjpeg-turbo'

* libjpeg-turbo: (39 commits)
  Oops.  Delete the duplicate copy of [lib]turbojpeg.dll in the binary directory when uninstalling the package.
  AltiVec SIMD implementation of sample conversion and integer quantization
  Document the fact that the AltiVec implementation uses the same modified algorithms as the SSE2 implementation
  Use intrinsics for loading/storing data in the DCT/IDCT functions.  This has no effect on the performance of the aligned loads/stores, but it makes it more obvious what that code is doing.  Using intrinsics for the unaligned stores in the inverse DCT functions increases overall decompression performance by 1-2%.
  AltiVec SIMD implementation of RGB-to-Grayscale color conversion
  Remove unneeded code;  Make sure jccolor-altivec.o will be rebuilt if jccolext-altivec.c changes.
  AltiVec SIMD implementation of RGB-to-YCC color conversion
  Make test a phony target so things don't go haywire if there is a file named test.c in the current directory.
  Maintain the traditional order of the regression tests while allowing the TurboJPEG and libjpeg portions to be executed separately
  Make comments more consistent
  Add a "quicktest" pseudo-target, for those times when you just don't want to sit through 11 iterations of TJUnitTest.
  Cosmetic tweaks to the PowerPC SIMD stubs
  Split AltiVec algorithms into separate files for ease of maintenance;  Rename constants using lowercase so they are not confused with macros
  Optimizations to the AltiVec DCT algorithms (pre-compute constants and combine multiply/add operations)
  AltiVec SIMD implementation of slow integer inverse DCT
  Use macros to allocate constants statically, rather than reading them from a table using vec_splat*().  This improves code readability and probably improves performance a bit as well.
  Swap the order of the IFAST and ISLOW FDCT functions so that it matches the order of the prototypes in jsimd.h and the stubs in jsimd_powerpc.c.
  Include ARMv8 binaries when generating a combined OS X/iOS package using 'make iosdmg'
  In the output of the configure script, indicate whether gas-preprocessor.pl is being used along with the assembler.
  Modify the ARM64 assembly file so that it uses only syntax that the clang assembler in XCode 5.x can understand.  These changes should all be cosmetic in nature-- they do not change the meaning or readability of the code nor the ability to build it for Linux.  Actually, the code is now more in compliance with the ARM64 programming manual.  In addition to these changes, there were a couple of instructions that clang simply doesn't support, so gas-preprocessor.pl was modified so that it now converts those into equivalent instructions that clang can handle.
  ...

Conflicts:
	BUILDING.txt
	ChangeLog.txt
	cjpeg.c
	jpegtran.c

BUILDING.txt

@@ -322,6 +322,9 @@ Additional build requirements:
   (https://sourceforge.net/p/libjpeg-turbo/code/HEAD/tree/gas-preprocessor)
   should be installed in your PATH.
 
+
+ARM 32-bit Build (Xcode 4.6.x and earlier, LLVM-GCC):
+
 Set the following shell variables for simplicity:
 
   Xcode 4.2 and earlier:
@@ -330,47 +333,80 @@ Set the following shell variables for simplicity:
   IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
 
   IOS_SYSROOT=$IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk
-
-  Xcode 4.6.x and earlier:
   IOS_GCC=$IOS_PLATFORMDIR/Developer/usr/bin/arm-apple-darwin10-llvm-gcc-4.2
-  Xcode 5.0.x and later:
-  IOS_GCC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
 
-  ARM v6 (code will run on all iOS devices, not SIMD-accelerated):
+  ARMv6 (code will run on all iOS devices, not SIMD-accelerated):
   [NOTE: Requires Xcode 4.4.x or earlier]
   IOS_CFLAGS="-march=armv6 -mcpu=arm1176jzf-s -mfpu=vfp"
 
-  ARM v7 (code will run on iPhone 3GS-4S/iPad 1st-3rd Generation and newer):
+  ARMv7 (code will run on iPhone 3GS-4S/iPad 1st-3rd Generation and newer):
-  Xcode 4.6.x and earlier:
   IOS_CFLAGS="-march=armv7 -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon"
-  Xcode 5.0.x and later:
-  IOS_CFLAGS="-arch armv7"
 
-  ARM v7s (code will run on iPhone 5/iPad 4th Generation and newer):
+  ARMv7s (code will run on iPhone 5/iPad 4th Generation and newer):
   [NOTE: Requires Xcode 4.5 or later]
-  Xcode 4.6.x and earlier:
   IOS_CFLAGS="-march=armv7s -mcpu=swift -mtune=swift -mfpu=neon"
-  Xcode 5.0.x and later:
-  IOS_CFLAGS="-arch armv7s"
 
 Follow the procedure under "Building mozjpeg" above, adding
 
-  --host arm-apple-darwin10 --enable-static --disable-shared \
+  --host arm-apple-darwin10 \
     CC="$IOS_GCC" LD="$IOS_GCC" \
     CFLAGS="-mfloat-abi=softfp -isysroot $IOS_SYSROOT -O3 $IOS_CFLAGS" \
     LDFLAGS="-mfloat-abi=softfp -isysroot $IOS_SYSROOT $IOS_CFLAGS"
 
-to the configure command line.  If using Xcode 5.0.x or later, also add
+to the configure command line.
 
+
+ARM 32-bit Build (Xcode 5.0.x and later, Clang):
+
+Set the following shell variables for simplicity:
+
+  IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
+  IOS_SYSROOT=$IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk
+  IOS_GCC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+
+  ARMv7 (code will run on iPhone 3GS-4S/iPad 1st-3rd Generation and newer):
+  IOS_CFLAGS="-arch armv7"
+
+  ARMv7s (code will run on iPhone 5/iPad 4th Generation and newer):
+  IOS_CFLAGS="-arch armv7s"
+
+Follow the procedure under "Building libjpeg-turbo" above, adding
+
+  --host arm-apple-darwin10 \
+    CC="$IOS_GCC" LD="$IOS_GCC" \
+    CFLAGS="-mfloat-abi=softfp -isysroot $IOS_SYSROOT -O3 $IOS_CFLAGS" \
+    LDFLAGS="-mfloat-abi=softfp -isysroot $IOS_SYSROOT $IOS_CFLAGS" \
     CCASFLAGS="-no-integrated-as $IOS_CFLAGS"
 
 to the configure command line.
 
+
+ARMv8 64-bit Build (Xcode 5.0.x and later, Clang):
+
+Code will run on iPhone 5S/iPad Mini 2 and newer.
+
+Set the following shell variables for simplicity:
+
+  IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
+  IOS_SYSROOT=$IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk
+  IOS_GCC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+  IOS_CFLAGS="-arch arm64"
+
+Follow the procedure under "Building libjpeg-turbo" above, adding
+
+  --host aarch64-apple-darwin \
+    CC="$IOS_GCC" LD="$IOS_GCC" \
+    CFLAGS="-isysroot $IOS_SYSROOT -O3 $IOS_CFLAGS" \
+    LDFLAGS="-isysroot $IOS_SYSROOT $IOS_CFLAGS"
+
+to the configure command line.
+
+
 NOTE:  You can also add -miphoneos-version-min={version} to $IOS_CFLAGS above
 in order to support older versions of iOS than the default version supported by
 the SDK.
 
-Once built, lipo can be used to combine the ARM v6, v7, and/or v7s variants
+Once built, lipo can be used to combine the ARMv6, v7, v7s, and/or v8 variants
 into a universal library.
 
 
@@ -745,26 +781,32 @@ make udmg [BUILDDIR32={32-bit build directory}]
   make command line as shown above.
 
 make iosdmg [BUILDDIR32={32-bit build directory}] \
-  [BUILDDIRARMV6={ARM v6 build directory}] \
+  [BUILDDIRARMV6={ARMv6 build directory}] \
-  [BUILDDIRARMV7={ARM v7 build directory}] \
+  [BUILDDIRARMV7={ARMv7 build directory}] \
-  [BUILDDIRARMV7S={ARM v7s build directory}]
+  [BUILDDIRARMV7S={ARMv7s build directory}] \
+  [BUILDDIRARMV8={ARMv8 build directory}]
 
   On OS X systems, this creates a Macintosh package and disk image in which the
   mozjpeg static libraries contain ARM architectures necessary to build
   iOS applications.  If building on an x86-64 system, the binaries will also
   contain the i386 architecture, as with 'make udmg' above.  You should first
-  configure ARM v6, ARM v7, and/or ARM v7s out-of-tree builds of mozjpeg
+  configure ARMv6, ARMv7, ARMv7s, and/or ARMv8 out-of-tree builds of
-  (see "Building mozjpeg for iOS" above.)  If you are building an x86-64
+  mozjpeg (see "Building mozjpeg for iOS" above.)  If you are
-  version of mozjpeg, you should configure a 32-bit out-of-tree build as
+  building an x86-64 version of mozjpeg, you should configure a 32-bit
-  well.  Next, build mozjpeg as you would normally, using an out-of-tree
+  out-of-tree build as well.  Next, build mozjpeg as you would normally,
-  build.  When it is built, run 'make iosdmg' from the build directory.  The
+  using an out-of-tree build.  When it is built, run 'make iosdmg' from the
-  build system will look for the ARM v6 build under {source_directory}/iosarmv6
+  build directory.  The build system will look for the ARMv6 build under
-  by default, the ARM v7 build under {source_directory}/iosarmv7 by default,
+  {source_directory}/iosarmv6 by default, the ARMv7 build under
-  the ARM v7s build under {source_directory}/iosarmv7s by default, and (if
+  {source_directory}/iosarmv7 by default, the ARMv7s build under
-  applicable) the 32-bit build under {source_directory}/osxx86 by default, but
+  {source_directory}/iosarmv7s by default, the ARMv8 build under
-  you can override this by setting the BUILDDIR32, BUILDDIRARMV6,
+  {source_directory}/iosarmv8 by default, and (if applicable) the 32-bit build
-  BUILDDIRARMV7, and/or BUILDDIRARMV7S variables on the make command line as
+  under {source_directory}/osxx86 by default, but you can override this by
-  shown above.
+  setting the BUILDDIR32, BUILDDIRARMV6, BUILDDIRARMV7, BUILDDIRARMV7S, and/or
+  BUILDDIRARMV8 variables on the make command line as shown above.
+
+  NOTE: If including an ARMv8 build in the package, then you may need to use
+  Xcode's version of lipo instead of the operating system's.  To do this, pass
+  an argument of LIPO="xcrun lipo" on the make command line.
 
 make cygwinpkg
 

CMakeLists.txt

@@ -557,7 +557,7 @@ foreach(libtype shared static)
   add_test(djpeg${suffix}-gray-islow-rgb
     ${dir}djpeg${suffix} -dct int -rgb -outfile testout_gray_islow_rgb.ppm
       testout_gray_islow.jpg)
-  add_test(cjpeg${suffix}-gray-islow-rgb-cmp
+  add_test(djpeg${suffix}-gray-islow-rgb-cmp
     ${CMAKE_COMMAND} -DMD5=${MD5_PPM_GRAY_ISLOW_RGB}
       -DFILE=testout_gray_islow_rgb.ppm
       -P ${CMAKE_SOURCE_DIR}/cmakescripts/md5cmp.cmake)
@@ -637,10 +637,10 @@ foreach(libtype shared static)
   endif()
   if(WITH_ARITH_DEC)
     # CC: RGB->YCC  SAMP: h2v2 merged  IDCT: ifast  ENT: arith
-    add_test(cjpeg${suffix}-420m-ifast-ari
+    add_test(djpeg${suffix}-420m-ifast-ari
       ${dir}djpeg${suffix} -fast -ppm -outfile testout_420m_ifast_ari.ppm
         ${CMAKE_SOURCE_DIR}/testimages/testimgari.jpg)
-    add_test(cjpeg${suffix}-420m-ifast-ari-cmp
+    add_test(djpeg${suffix}-420m-ifast-ari-cmp
       ${CMAKE_COMMAND} -DMD5=${MD5_PPM_420M_IFAST_ARI}
         -DFILE=testout_420m_ifast_ari.ppm
         -P ${CMAKE_SOURCE_DIR}/cmakescripts/md5cmp.cmake)

ChangeLog.txt

@@ -13,6 +13,25 @@ instead of -1 if componentID was > 0 and subsamp was TJSAMP_GRAY.
 [3] Fixed an issue in tjBufSizeYUV2() wherby it would erroneously return 0
 instead of -1 if width was < 1.
 
+[5] The Huffman encoder now uses clz and bsr instructions for bit counting on
+ARM64 platforms (see 1.4 beta1 [5].)
+
+[6] The close() method in the TJCompressor and TJDecompressor Java classes is
+now idempotent.  Previously, that method would call the native tjDestroy()
+function even if the TurboJPEG instance had already been destroyed.  This
+caused an exception to be thrown during finalization, if the close() method had
+already been called.  The exception was caught, but it was still an expensive
+operation.
+
+[7] The TurboJPEG API previously generated an error ("Could not determine
+subsampling type for JPEG image") when attempting to decompress grayscale JPEG
+images that were compressed with a sampling factor other than 1 (for instance,
+with 'cjpeg -grayscale -sample 2x2').  Subsampling technically has no meaning
+with grayscale JPEGs, and thus the horizontal and vertical sampling factors
+for such images are ignored by the decompressor.  However, the TurboJPEG API
+was being too rigid and was expecting the sampling factors to be equal to 1
+before it treated the image as a grayscale JPEG.
+
 [8] cjpeg, djpeg, and jpegtran now accept an argument of -version, which will
 print the library version and exit.
 
@@ -28,6 +47,26 @@ order), the Huffman encoder can produce encoded blocks that approach double the
 size of the unencoded blocks.  Thus, the Huffman local buffer was increased to
 256 bytes, which should prevent any such issue from re-occurring in the future.
 
+[10] The new tjPlaneSizeYUV(), tjPlaneWidth(), and tjPlaneHeight() functions
+were not actually usable on any platform except OS X and Windows, because
+those functions were not included in the libturbojpeg mapfile.  This has been
+fixed.
+
+[11] Restored the JPP(), JMETHOD(), and FAR macros in the libjpeg-turbo header
+files.  The JPP() and JMETHOD() macros were originally implemented in libjpeg
+as a way of supporting non-ANSI compilers that lacked support for prototype
+parameters.  libjpeg-turbo has never supported such compilers, but some
+software packages still use the macros to define their own prototypes.
+Similarly, libjpeg-turbo has never supported MS-DOS and other platforms that
+have far symbols, but some software packages still use the FAR macro.  A pretty
+good argument can be made that this is a bad practice on the part of the
+software in question, but since this affects more than one package, it's just
+easier to fix it here.
+
+[12] Fixed issues that were preventing the ARM 64-bit SIMD code from compiling
+for iOS, and included an ARMv8 architecture in all of the binaries installed by
+the "official" libjpeg-turbo SDK for OS X.
+
 
 1.3.90 (1.4 beta1)
 ==================
@@ -280,7 +319,7 @@ configure/CMake switch in order to retain strict API/ABI compatibility with the
 libjpeg v6b or v7 API/ABI (or with previous versions of libjpeg-turbo.)  See
 README-turbo.txt for more details.
 
-[13] Added ARM v7s architecture to libjpeg.a and libturbojpeg.a in the official
+[13] Added ARMv7s architecture to libjpeg.a and libturbojpeg.a in the official
 libjpeg-turbo binary package for OS X, so that those libraries can be used to
 build applications that leverage the faster CPUs in the iPhone 5 and iPad 4.
 
@@ -363,7 +402,7 @@ K component is assigned a component ID of 1 instead of 4.  Although these files
 are in violation of the spec, other JPEG implementations handle them
 correctly.
 
-[7] Added ARM v6 and ARM v7 architectures to libjpeg.a and libturbojpeg.a in
+[7] Added ARMv6 and ARMv7 architectures to libjpeg.a and libturbojpeg.a in
 the official libjpeg-turbo binary package for OS X, so that those libraries can
 be used to build both OS X and iOS applications.
 

Makefile.am

@@ -276,7 +276,10 @@ MD5_JPEG_CROP = b4197f377e621c4e9b1d20471432610d
 
 endif
 
-test: testclean all
+.PHONY: test
+test: tjquicktest bittest
+
+tjquicktest: testclean all
 
 if WITH_TURBOJPEG
 if WITH_JAVA
@@ -294,6 +297,8 @@ endif
 	./tjunittest -yuv -noyuvpad
 endif
 
+bittest: testclean all
+
 # These tests are carefully crafted to provide full coverage of as many of the
 # underlying algorithms as possible (including all of the SIMD-accelerated
 # ones.)
@@ -598,12 +603,12 @@ udmg: all pkgscripts/makemacpkg pkgscripts/uninstall
 	sh pkgscripts/makemacpkg -build32 ${BUILDDIR32}
 
 iosdmg: all pkgscripts/makemacpkg pkgscripts/uninstall
-	sh pkgscripts/makemacpkg -build32 ${BUILDDIR32} -buildarmv6 ${BUILDDIRARMV6} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S}
+	sh pkgscripts/makemacpkg -build32 ${BUILDDIR32} -buildarmv6 ${BUILDDIRARMV6} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S} -buildarmv8 ${BUILDDIRARMV8} -lipo "${LIPO}"
 
 else
 
 iosdmg: all pkgscripts/makemacpkg pkgscripts/uninstall
-	sh pkgscripts/makemacpkg -buildarmv6 ${BUILDDIRARMV6} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S}
+	sh pkgscripts/makemacpkg -buildarmv6 ${BUILDDIRARMV6} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S} -buildarmv8 ${BUILDDIRARMV8} -lipo "${LIPO}"
 
 endif
 

configure.ac

@@ -443,7 +443,11 @@ if test "x${with_simd}" != "xno"; then
       AC_MSG_RESULT([yes (arm)])
       AC_MSG_CHECKING([if the assembler is GNU-compatible and can be used])
       AC_CHECK_COMPATIBLE_ARM_ASSEMBLER_IFELSE(
-        [AC_MSG_RESULT([yes])
+        [if test "x$ac_use_gas_preprocessor" = "xyes"; then
+           AC_MSG_RESULT([yes (with gas-preprocessor)])
+         else
+           AC_MSG_RESULT([yes])
+         fi
          simd_arch=arm],
         [AC_MSG_RESULT([no])
          with_simd=no])
@@ -459,7 +463,11 @@ if test "x${with_simd}" != "xno"; then
       AC_MSG_RESULT([yes (arm64)])
       AC_MSG_CHECKING([if the assembler is GNU-compatible and can be used])
       AC_CHECK_COMPATIBLE_ARM64_ASSEMBLER_IFELSE(
-        [AC_MSG_RESULT([yes])
+        [if test "x$ac_use_gas_preprocessor" = "xyes"; then
+           AC_MSG_RESULT([yes (with gas-preprocessor)])
+         else
+           AC_MSG_RESULT([yes])
+         fi
          simd_arch=aarch64],
         [AC_MSG_RESULT([no])
          with_simd=no])

java/org/libjpegturbo/turbojpeg/TJCompressor.java

@@ -567,6 +567,7 @@ public class TJCompressor {
    * Free the native structures associated with this compressor instance.
    */
   public void close() throws Exception {
+    if (handle != 0)
       destroy();
   }
 

java/org/libjpegturbo/turbojpeg/TJDecompressor.java

@@ -834,6 +834,7 @@ public class TJDecompressor {
    * Free the native structures associated with this decompressor instance.
    */
   public void close() throws Exception {
+    if (handle != 0)
       destroy();
   }
 

jchuff.c

@@ -37,7 +37,7 @@
  */
 
 /* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined __GNUC__ && defined __arm__
+#if defined __GNUC__ && (defined __arm__ || defined __aarch64__)
 #if !defined __thumb__ || defined __thumb2__
 #define USE_CLZ_INTRINSIC
 #endif

jcmainct.c

@@ -17,14 +17,6 @@
 #include "jpeglib.h"
 
 
-/* Note: currently, there is no operating mode in which a full-image buffer
- * is needed at this step.  If there were, that mode could not be used with
- * "raw data" input, since this module is bypassed in that case.  However,
- * we've left the code here for possible use in special applications.
- */
-#undef FULL_MAIN_BUFFER_SUPPORTED
-
-
 /* Private buffer controller object */
 
 typedef struct {
@@ -40,13 +32,6 @@ typedef struct {
    * points to the currently accessible strips of the virtual arrays.
    */
   JSAMPARRAY buffer[MAX_COMPONENTS];
-
-#ifdef FULL_MAIN_BUFFER_SUPPORTED
-  /* If using full-image storage, this array holds pointers to virtual-array
-   * control blocks for each component.  Unused if not full-image storage.
-   */
-  jvirt_sarray_ptr whole_image[MAX_COMPONENTS];
-#endif
 } my_main_controller;
 
 typedef my_main_controller * my_main_ptr;
@@ -56,11 +41,6 @@ typedef my_main_controller * my_main_ptr;
 METHODDEF(void) process_data_simple_main
         (j_compress_ptr cinfo, JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
          JDIMENSION in_rows_avail);
-#ifdef FULL_MAIN_BUFFER_SUPPORTED
-METHODDEF(void) process_data_buffer_main
-        (j_compress_ptr cinfo, JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
-         JDIMENSION in_rows_avail);
-#endif
 
 
 /*
@@ -76,32 +56,14 @@ start_pass_main (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
   if (cinfo->raw_data_in)
     return;
 
+  if (pass_mode != JBUF_PASS_THRU)
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+
   main_ptr->cur_iMCU_row = 0;   /* initialize counters */
   main_ptr->rowgroup_ctr = 0;
   main_ptr->suspended = FALSE;
   main_ptr->pass_mode = pass_mode;      /* save mode for use by process_data */
-
-  switch (pass_mode) {
-  case JBUF_PASS_THRU:
-#ifdef FULL_MAIN_BUFFER_SUPPORTED
-    if (main_ptr->whole_image[0] != NULL)
-      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
-#endif
   main_ptr->pub.process_data = process_data_simple_main;
-    break;
-#ifdef FULL_MAIN_BUFFER_SUPPORTED
-  case JBUF_SAVE_SOURCE:
-  case JBUF_CRANK_DEST:
-  case JBUF_SAVE_AND_PASS:
-    if (main_ptr->whole_image[0] == NULL)
-      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
-    main_ptr->pub.process_data = process_data_buffer_main;
-    break;
-#endif
-  default:
-    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
-    break;
-  }
 }
 
 
@@ -160,85 +122,6 @@ process_data_simple_main (j_compress_ptr cinfo,
 }
 
 
-#ifdef FULL_MAIN_BUFFER_SUPPORTED
-
-/*
- * Process some data.
- * This routine handles all of the modes that use a full-size buffer.
- */
-
-METHODDEF(void)
-process_data_buffer_main (j_compress_ptr cinfo,
-                          JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
-                          JDIMENSION in_rows_avail)
-{
-  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
-  int ci;
-  jpeg_component_info *compptr;
-  boolean writing = (main_ptr->pass_mode != JBUF_CRANK_DEST);
-
-  while (main_ptr->cur_iMCU_row < cinfo->total_iMCU_rows) {
-    /* Realign the virtual buffers if at the start of an iMCU row. */
-    if (main_ptr->rowgroup_ctr == 0) {
-      for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
-           ci++, compptr++) {
-        main_ptr->buffer[ci] = (*cinfo->mem->access_virt_sarray)
-          ((j_common_ptr) cinfo, main_ptr->whole_image[ci],
-           main_ptr->cur_iMCU_row * (compptr->v_samp_factor * DCTSIZE),
-           (JDIMENSION) (compptr->v_samp_factor * DCTSIZE), writing);
-      }
-      /* In a read pass, pretend we just read some source data. */
-      if (! writing) {
-        *in_row_ctr += cinfo->max_v_samp_factor * DCTSIZE;
-        main_ptr->rowgroup_ctr = DCTSIZE;
-      }
-    }
-
-    /* If a write pass, read input data until the current iMCU row is full. */
-    /* Note: preprocessor will pad if necessary to fill the last iMCU row. */
-    if (writing) {
-      (*cinfo->prep->pre_process_data) (cinfo,
-                                        input_buf, in_row_ctr, in_rows_avail,
-                                        main_ptr->buffer, &main_ptr->rowgroup_ctr,
-                                        (JDIMENSION) DCTSIZE);
-      /* Return to application if we need more data to fill the iMCU row. */
-      if (main_ptr->rowgroup_ctr < DCTSIZE)
-        return;
-    }
-
-    /* Emit data, unless this is a sink-only pass. */
-    if (main_ptr->pass_mode != JBUF_SAVE_SOURCE) {
-      if (! (*cinfo->coef->compress_data) (cinfo, main_ptr->buffer)) {
-        /* If compressor did not consume the whole row, then we must need to
-         * suspend processing and return to the application.  In this situation
-         * we pretend we didn't yet consume the last input row; otherwise, if
-         * it happened to be the last row of the image, the application would
-         * think we were done.
-         */
-        if (! main_ptr->suspended) {
-          (*in_row_ctr)--;
-          main_ptr->suspended = TRUE;
-        }
-        return;
-      }
-      /* We did finish the row.  Undo our little suspension hack if a previous
-       * call suspended; then mark the main buffer empty.
-       */
-      if (main_ptr->suspended) {
-        (*in_row_ctr)++;
-        main_ptr->suspended = FALSE;
-      }
-    }
-
-    /* If get here, we are done with this iMCU row.  Mark buffer empty. */
-    main_ptr->rowgroup_ctr = 0;
-    main_ptr->cur_iMCU_row++;
-  }
-}
-
-#endif /* FULL_MAIN_BUFFER_SUPPORTED */
-
-
 /*
  * Initialize main buffer controller.
  */
@@ -264,25 +147,8 @@ jinit_c_main_controller (j_compress_ptr cinfo, boolean need_full_buffer)
    * may be of a different size.
    */
   if (need_full_buffer) {
-#ifdef FULL_MAIN_BUFFER_SUPPORTED
-    /* Allocate a full-image virtual array for each component */
-    /* Note we pad the bottom to a multiple of the iMCU height */
-    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
-         ci++, compptr++) {
-      main_ptr->whole_image[ci] = (*cinfo->mem->request_virt_sarray)
-        ((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
-         compptr->width_in_blocks * DCTSIZE,
-         (JDIMENSION) jround_up((long) compptr->height_in_blocks,
-                                (long) compptr->v_samp_factor) * DCTSIZE,
-         (JDIMENSION) (compptr->v_samp_factor * DCTSIZE));
-    }
-#else
     ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
-#endif
   } else {
-#ifdef FULL_MAIN_BUFFER_SUPPORTED
-    main_ptr->whole_image[0] = NULL; /* flag for no virtual arrays */
-#endif
     /* Allocate a strip buffer for each component */
     for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
          ci++, compptr++) {

jcsample.c

@@ -514,8 +514,9 @@ jinit_downsampler (j_compress_ptr cinfo)
 #endif
           downsample->methods[ci] = h2v2_smooth_downsample;
         downsample->pub.need_context_rows = TRUE;
-      } else {
+      } else
 #endif
+      {
         if (jsimd_can_h2v2_downsample())
           downsample->methods[ci] = jsimd_h2v2_downsample;
         else

jmorecfg.h

@@ -180,6 +180,24 @@ typedef unsigned int JDIMENSION;
 #define EXTERN(type)            extern type
 
 
+/* Originally, this macro was used as a way of defining function prototypes
+ * for both modern compilers as well as older compilers that did not support
+ * prototype parameters.  libjpeg-turbo has never supported these older,
+ * non-ANSI compilers, but the macro is still included because there is some
+ * software out there that uses it.
+ */
+
+#define JMETHOD(type,methodname,arglist)  type (*methodname) arglist
+
+
+/* libjpeg-turbo no longer supports platforms that have far symbols (MS-DOS),
+ * but again, some software relies on this macro.
+ */
+
+#undef FAR
+#define FAR
+
+
 /*
  * On a few systems, type boolean and/or its values FALSE, TRUE may appear
  * in standard header files.  Or you may have conflicts with application-

jpeglib.h

@@ -923,6 +923,16 @@ struct jpeg_memory_mgr {
 typedef boolean (*jpeg_marker_parser_method) (j_decompress_ptr cinfo);
 
 
+/* Originally, this macro was used as a way of defining function prototypes
+ * for both modern compilers as well as older compilers that did not support
+ * prototype parameters.  libjpeg-turbo has never supported these older,
+ * non-ANSI compilers, but the macro is still included because there is some
+ * software out there that uses it.
+ */
+
+#define JPP(arglist)    arglist
+
+
 /* Default error-management setup */
 EXTERN(struct jpeg_error_mgr *) jpeg_std_error (struct jpeg_error_mgr * err);
 

release/makemacpkg.in

@@ -17,7 +17,7 @@ onexit()
 
 usage()
 {
-	echo "$0 [-build32 [32-bit build dir]] [-buildarmv6 [ARM v6 build dir]] [-buildarmv7 [ARM v7 build dir]] [-buildarmv7s [ARM v7s build dir]]"
+	echo "$0 [-build32 [32-bit build dir]] [-buildarmv6 [ARMv6 build dir]] [-buildarmv7 [ARMv7 build dir]] [-buildarmv7s [ARMv7s build dir] [-buildarmv8 [ARMv8 build dir]] [-lipo [path to lipo]]"
 	exit 1
 }
 
@@ -33,7 +33,10 @@ BUILDDIRARMV7=@abs_top_srcdir@/iosarmv7
 BUILDARMV7=0
 BUILDDIRARMV7S=@abs_top_srcdir@/iosarmv7s
 BUILDARMV7S=0
+BUILDDIRARMV8=@abs_top_srcdir@/iosarmv8
+BUILDARMV8=0
 WITH_JAVA=@WITH_JAVA@
+LIPO=lipo
 
 PREFIX=%{__prefix}
 BINDIR=%{__bindir}
@@ -75,6 +78,21 @@ while [ $# -gt 0 ]; do
 			fi
 		fi
 		;;
+	-buildarmv8)
+		BUILDARMV8=1
+		if [ $# -gt 1 ]; then
+			if [[ ! "$2" =~ -.* ]]; then
+				BUILDDIRARMV8=$2;  shift
+			fi
+		fi
+		;;
+	-lipo)
+		if [ $# -gt 1 ]; then
+			if [[ ! "$2" =~ -.* ]]; then
+				LIPO=$2;  shift
+			fi
+		fi
+		;;
 	esac
 	shift
 done
@@ -110,50 +128,50 @@ if [ $BUILD32 = 1 ]; then
 	popd
 	if [ ! -h $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
-		lipo -create \
+		$LIPO -create \
 			-arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-arch x86_64 $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
 	elif [ ! -h $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
-		lipo -create \
+		$LIPO -create \
 			-arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-arch x86_64 $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
 	fi
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.a \
 		-arch x86_64 $PKGROOT/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$LIBDIR/libturbojpeg.0.dylib \
 		-arch x86_64 $PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$LIBDIR/libturbojpeg.a \
 		-arch x86_64 $PKGROOT/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/cjpeg \
 		-arch x86_64 $PKGROOT/$BINDIR/cjpeg \
 		-output $PKGROOT/$BINDIR/cjpeg
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/djpeg \
 		-arch x86_64 $PKGROOT/$BINDIR/djpeg \
 		-output $PKGROOT/$BINDIR/djpeg
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/jpegtran \
 		-arch x86_64 $PKGROOT/$BINDIR/jpegtran \
 		-output $PKGROOT/$BINDIR/jpegtran
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/tjbench \
 		-arch x86_64 $PKGROOT/$BINDIR/tjbench \
 		-output $PKGROOT/$BINDIR/tjbench
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/rdjpgcom \
 		-arch x86_64 $PKGROOT/$BINDIR/rdjpgcom \
 		-output $PKGROOT/$BINDIR/rdjpgcom
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/wrjpgcom \
 		-arch x86_64 $PKGROOT/$BINDIR/wrjpgcom \
 		-output $PKGROOT/$BINDIR/wrjpgcom
@@ -162,71 +180,258 @@ fi
 
 if [ $BUILDARMV6 = 1 ]; then
 	if [ ! -d $BUILDDIRARMV6 ]; then
-		echo ERROR: ARM v6 build directory $BUILDDIRARMV6 does not exist
+		echo ERROR: ARMv6 build directory $BUILDDIRARMV6 does not exist
 		exit 1
 	fi
 	if [ ! -f $BUILDDIRARMV6/Makefile ]; then
-		echo ERROR: ARM v6 build directory $BUILDDIRARMV6 is not configured
+		echo ERROR: ARMv6 build directory $BUILDDIRARMV6 is not configured
 		exit 1
 	fi
 	mkdir -p $TMPDIR/dist.armv6
 	pushd $BUILDDIRARMV6
 	make install DESTDIR=$TMPDIR/dist.armv6
 	popd
-	lipo -create \
+	if [ ! -h $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
+		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
+		$LIPO -create \
+			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
+			-arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
+			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
+	elif [ ! -h $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
+		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
+		$LIPO -create \
+			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
+			-arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
+			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
+	fi
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libjpeg.a \
 		-arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
-	lipo -create \
+	$LIPO -create \
+		$PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
+		-arch arm $TMPDIR/dist.armv6/$LIBDIR/libturbojpeg.0.dylib \
+		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.a \
 		-arch arm $TMPDIR/dist.armv6/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
+	$LIPO -create \
+		$PKGROOT/$BINDIR/cjpeg \
+		-arch arm $TMPDIR/dist.armv6/$BINDIR/cjpeg \
+		-output $PKGROOT/$BINDIR/cjpeg
+	$LIPO -create \
+		$PKGROOT/$BINDIR/djpeg \
+		-arch arm $TMPDIR/dist.armv6/$BINDIR/djpeg \
+		-output $PKGROOT/$BINDIR/djpeg
+	$LIPO -create \
+		$PKGROOT/$BINDIR/jpegtran \
+		-arch arm $TMPDIR/dist.armv6/$BINDIR/jpegtran \
+		-output $PKGROOT/$BINDIR/jpegtran
+	$LIPO -create \
+		$PKGROOT/$BINDIR/tjbench \
+		-arch arm $TMPDIR/dist.armv6/$BINDIR/tjbench \
+		-output $PKGROOT/$BINDIR/tjbench
+	$LIPO -create \
+		$PKGROOT/$BINDIR/rdjpgcom \
+		-arch arm $TMPDIR/dist.armv6/$BINDIR/rdjpgcom \
+		-output $PKGROOT/$BINDIR/rdjpgcom
+	$LIPO -create \
+		$PKGROOT/$BINDIR/wrjpgcom \
+		-arch arm $TMPDIR/dist.armv6/$BINDIR/wrjpgcom \
+		-output $PKGROOT/$BINDIR/wrjpgcom
 fi
 
 if [ $BUILDARMV7 = 1 ]; then
 	if [ ! -d $BUILDDIRARMV7 ]; then
-		echo ERROR: ARM v7 build directory $BUILDDIRARMV7 does not exist
+		echo ERROR: ARMv7 build directory $BUILDDIRARMV7 does not exist
 		exit 1
 	fi
 	if [ ! -f $BUILDDIRARMV7/Makefile ]; then
-		echo ERROR: ARM v7 build directory $BUILDDIRARMV7 is not configured
+		echo ERROR: ARMv7 build directory $BUILDDIRARMV7 is not configured
 		exit 1
 	fi
 	mkdir -p $TMPDIR/dist.armv7
 	pushd $BUILDDIRARMV7
 	make install DESTDIR=$TMPDIR/dist.armv7
 	popd
-	lipo -create \
+	if [ ! -h $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
+		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
+		$LIPO -create \
+			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
+			-arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
+			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
+	elif [ ! -h $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
+		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
+		$LIPO -create \
+			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
+			-arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
+			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
+	fi
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libjpeg.a \
 		-arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
-	lipo -create \
+	$LIPO -create \
+		$PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
+		-arch arm $TMPDIR/dist.armv7/$LIBDIR/libturbojpeg.0.dylib \
+		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.a \
 		-arch arm $TMPDIR/dist.armv7/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
+	$LIPO -create \
+		$PKGROOT/$BINDIR/cjpeg \
+		-arch arm $TMPDIR/dist.armv7/$BINDIR/cjpeg \
+		-output $PKGROOT/$BINDIR/cjpeg
+	$LIPO -create \
+		$PKGROOT/$BINDIR/djpeg \
+		-arch arm $TMPDIR/dist.armv7/$BINDIR/djpeg \
+		-output $PKGROOT/$BINDIR/djpeg
+	$LIPO -create \
+		$PKGROOT/$BINDIR/jpegtran \
+		-arch arm $TMPDIR/dist.armv7/$BINDIR/jpegtran \
+		-output $PKGROOT/$BINDIR/jpegtran
+	$LIPO -create \
+		$PKGROOT/$BINDIR/tjbench \
+		-arch arm $TMPDIR/dist.armv7/$BINDIR/tjbench \
+		-output $PKGROOT/$BINDIR/tjbench
+	$LIPO -create \
+		$PKGROOT/$BINDIR/rdjpgcom \
+		-arch arm $TMPDIR/dist.armv7/$BINDIR/rdjpgcom \
+		-output $PKGROOT/$BINDIR/rdjpgcom
+	$LIPO -create \
+		$PKGROOT/$BINDIR/wrjpgcom \
+		-arch arm $TMPDIR/dist.armv7/$BINDIR/wrjpgcom \
+		-output $PKGROOT/$BINDIR/wrjpgcom
 fi
 
 if [ $BUILDARMV7S = 1 ]; then
 	if [ ! -d $BUILDDIRARMV7S ]; then
-		echo ERROR: ARM v7s build directory $BUILDDIRARMV7S does not exist
+		echo ERROR: ARMv7s build directory $BUILDDIRARMV7S does not exist
 		exit 1
 	fi
 	if [ ! -f $BUILDDIRARMV7S/Makefile ]; then
-		echo ERROR: ARM v7s build directory $BUILDDIRARMV7S is not configured
+		echo ERROR: ARMv7s build directory $BUILDDIRARMV7S is not configured
 		exit 1
 	fi
 	mkdir -p $TMPDIR/dist.armv7s
 	pushd $BUILDDIRARMV7S
 	make install DESTDIR=$TMPDIR/dist.armv7s
 	popd
-	lipo -create \
+	if [ ! -h $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
+		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
+		$LIPO -create \
+			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
+			-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
+			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
+	elif [ ! -h $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
+		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
+		$LIPO -create \
+			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
+			-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
+			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
+	fi
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libjpeg.a \
 		-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
-	lipo -create \
+	$LIPO -create \
+		$PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
+		-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libturbojpeg.0.dylib \
+		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.a \
 		-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
+	$LIPO -create \
+		$PKGROOT/$BINDIR/cjpeg \
+		-arch arm $TMPDIR/dist.armv7s/$BINDIR/cjpeg \
+		-output $PKGROOT/$BINDIR/cjpeg
+	$LIPO -create \
+		$PKGROOT/$BINDIR/djpeg \
+		-arch arm $TMPDIR/dist.armv7s/$BINDIR/djpeg \
+		-output $PKGROOT/$BINDIR/djpeg
+	$LIPO -create \
+		$PKGROOT/$BINDIR/jpegtran \
+		-arch arm $TMPDIR/dist.armv7s/$BINDIR/jpegtran \
+		-output $PKGROOT/$BINDIR/jpegtran
+	$LIPO -create \
+		$PKGROOT/$BINDIR/tjbench \
+		-arch arm $TMPDIR/dist.armv7s/$BINDIR/tjbench \
+		-output $PKGROOT/$BINDIR/tjbench
+	$LIPO -create \
+		$PKGROOT/$BINDIR/rdjpgcom \
+		-arch arm $TMPDIR/dist.armv7s/$BINDIR/rdjpgcom \
+		-output $PKGROOT/$BINDIR/rdjpgcom
+	$LIPO -create \
+		$PKGROOT/$BINDIR/wrjpgcom \
+		-arch arm $TMPDIR/dist.armv7s/$BINDIR/wrjpgcom \
+		-output $PKGROOT/$BINDIR/wrjpgcom
+fi
+
+if [ $BUILDARMV8 = 1 ]; then
+	if [ ! -d $BUILDDIRARMV8 ]; then
+		echo ERROR: ARMv8 build directory $BUILDDIRARMV8 does not exist
+		exit 1
+	fi
+	if [ ! -f $BUILDDIRARMV8/Makefile ]; then
+		echo ERROR: ARMv8 build directory $BUILDDIRARMV8 is not configured
+		exit 1
+	fi
+	mkdir -p $TMPDIR/dist.armv8
+	pushd $BUILDDIRARMV8
+	make install DESTDIR=$TMPDIR/dist.armv8
+	popd
+	if [ ! -h $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
+		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
+		$LIPO -create \
+			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
+			-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
+			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
+	elif [ ! -h $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
+		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
+		$LIPO -create \
+			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
+			-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
+			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
+	fi
+	$LIPO -create \
+		$PKGROOT/$LIBDIR/libjpeg.a \
+		-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.a \
+		-output $PKGROOT/$LIBDIR/libjpeg.a
+	$LIPO -create \
+		$PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
+		-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libturbojpeg.0.dylib \
+		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
+	$LIPO -create \
+		$PKGROOT/$LIBDIR/libturbojpeg.a \
+		-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libturbojpeg.a \
+		-output $PKGROOT/$LIBDIR/libturbojpeg.a
+	$LIPO -create \
+		$PKGROOT/$BINDIR/cjpeg \
+		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/cjpeg \
+		-output $PKGROOT/$BINDIR/cjpeg
+	$LIPO -create \
+		$PKGROOT/$BINDIR/djpeg \
+		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/djpeg \
+		-output $PKGROOT/$BINDIR/djpeg
+	$LIPO -create \
+		$PKGROOT/$BINDIR/jpegtran \
+		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/jpegtran \
+		-output $PKGROOT/$BINDIR/jpegtran
+	$LIPO -create \
+		$PKGROOT/$BINDIR/tjbench \
+		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/tjbench \
+		-output $PKGROOT/$BINDIR/tjbench
+	$LIPO -create \
+		$PKGROOT/$BINDIR/rdjpgcom \
+		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/rdjpgcom \
+		-output $PKGROOT/$BINDIR/rdjpgcom
+	$LIPO -create \
+		$PKGROOT/$BINDIR/wrjpgcom \
+		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/wrjpgcom \
+		-output $PKGROOT/$BINDIR/wrjpgcom
 fi
 
 install_name_tool -id $LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib

release/mozjpeg.nsi.in

@@ -111,6 +111,7 @@ Section "Uninstall"
 
 !ifdef GCC
 	Delete $INSTDIR\bin\libjpeg-@DLL_VERSION@.dll
+	Delete $INSTDIR\bin\libturbojpeg.dll
 	Delete $SYSDIR\libturbojpeg.dll
 	Delete $INSTDIR\lib\libturbojpeg.dll.a"
 	Delete $INSTDIR\lib\libturbojpeg.a"
@@ -118,6 +119,7 @@ Section "Uninstall"
 	Delete $INSTDIR\lib\libjpeg.a"
 !else
 	Delete $INSTDIR\bin\jpeg@DLL_VERSION@.dll
+	Delete $INSTDIR\bin\turbojpeg.dll
 	Delete $SYSDIR\turbojpeg.dll
 	Delete $INSTDIR\lib\jpeg.lib
 	Delete $INSTDIR\lib\jpeg-static.lib

simd/Makefile.am

@@ -72,9 +72,16 @@ endif
 
 if SIMD_POWERPC
 
-libsimd_la_SOURCES = jsimd_powerpc.c jsimd_powerpc_altivec.c
+libsimd_la_SOURCES = jsimd_powerpc.c \
+	jccolor-altivec.c     jcgray-altivec.c \
+	jfdctfst-altivec.c    jfdctint-altivec.c \
+	jidctfst-altivec.c    jidctint-altivec.c \
+  jquanti-altivec.c
 libsimd_la_CFLAGS = -maltivec
 
+jccolor-altivec.lo:  jccolext-altivec.c
+jcgray-altivec.lo:   jcgryext-altivec.c
+
 endif
 
 AM_CPPFLAGS = -I$(top_srcdir)

simd/jccolext-altivec.c

@@ -0,0 +1,250 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * Copyright (C) 2014, Jay Foad.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* This file is included by jccolor-altivec.c */
+
+
+void jsimd_rgb_ycc_convert_altivec (JDIMENSION img_width, JSAMPARRAY input_buf,
+                                    JSAMPIMAGE output_buf,
+                                    JDIMENSION output_row, int num_rows)
+{
+  JSAMPROW inptr;
+  JSAMPROW outptr0, outptr1, outptr2;
+  int pitch;
+  __vector unsigned char rgb0, rgb1 = {0}, rgb2 = {0}, rgb3 = {0}, rgbg0,
+    rgbg1, rgbg2, rgbg3, y, cb, cr;
+#if RGB_PIXELSIZE == 4
+  __vector unsigned char rgb4;
+#endif
+  __vector short rg0, rg1, rg2, rg3, bg0, bg1, bg2, bg3;
+  __vector unsigned short y01, y23, cr01, cr23, cb01, cb23;
+  __vector int y0, y1, y2, y3, cr0, cr1, cr2, cr3, cb0, cb1, cb2, cb3;
+
+  /* Constants */
+  __vector short pw_f0299_f0337 = { __4X2(F_0_299, F_0_337) },
+    pw_f0114_f0250 = { __4X2(F_0_114, F_0_250) },
+    pw_mf016_mf033 = { __4X2(-F_0_168, -F_0_331) },
+    pw_mf008_mf041 = { __4X2(-F_0_081, -F_0_418) };
+  __vector unsigned short pw_f050_f000 = { __4X2(F_0_500, 0) };
+  __vector int pd_onehalf = { __4X(ONE_HALF) },
+    pd_onehalfm1_cj = { __4X(ONE_HALF - 1 + (CENTERJSAMPLE << SCALEBITS)) };
+  __vector unsigned char zero = { __16X(0) },
+    shift_pack_index =
+      { 0, 1, 4, 5, 8, 9, 12, 13, 16, 17, 20, 21, 24, 25, 28, 29};
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    output_row++;
+
+    for (pitch = img_width * RGB_PIXELSIZE; pitch > 0;
+         pitch -= RGB_PIXELSIZE * 16, inptr += RGB_PIXELSIZE * 16,
+         outptr0 += 16, outptr1 += 16, outptr2 += 16) {
+
+#if RGB_PIXELSIZE == 3
+      /* Load 16 pixels == 48 bytes */
+      if ((size_t)inptr & 15) {
+        __vector unsigned char unaligned_shift_index;
+        rgb0 = vec_ld(0, inptr);
+        if (pitch > 16)
+          rgb1 = vec_ld(16, inptr);
+        else
+          rgb1 = vec_ld(-1, inptr + pitch);
+        if (pitch > 32)
+          rgb2 = vec_ld(32, inptr);
+        else
+          rgb2 = vec_ld(-1, inptr + pitch);
+        if (pitch > 48)
+          rgb3 = vec_ld(48, inptr);
+        else
+          rgb3 = vec_ld(-1, inptr + pitch);
+        unaligned_shift_index = vec_lvsl(0, inptr);
+        rgb0 = vec_perm(rgb0, rgb1, unaligned_shift_index);
+        rgb1 = vec_perm(rgb1, rgb2, unaligned_shift_index);
+        rgb2 = vec_perm(rgb2, rgb3, unaligned_shift_index);
+      } else {
+        rgb0 = vec_ld(0, inptr);
+        if (pitch > 16)
+          rgb1 = vec_ld(16, inptr);
+        if (pitch > 32)
+          rgb2 = vec_ld(32, inptr);
+      }
+
+      /* rgb0 = R0 G0 B0 R1 G1 B1 R2 G2 B2 R3 G3 B3 R4 G4 B4 R5
+       * rgb1 = G5 B5 R6 G6 B6 R7 G7 B7 R8 G8 B8 R9 G9 B9 Ra Ga
+       * rgb2 = Ba Rb Gb Bb Rc Gc Bc Rd Gd Bd Re Ge Be Rf Gf Bf
+       *
+       * rgbg0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 G0 B1 G1 B2 G2 B3 G3
+       * rgbg1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 G4 B5 G5 B6 G6 B7 G7
+       * rgbg2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 G8 B9 G9 Ba Ga Bb Gb
+       * rgbg3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Gc Bd Gd Be Ge Bf Gf
+       */
+      rgbg0 = vec_perm(rgb0, rgb0, (__vector unsigned char)RGBG_INDEX0);
+      rgbg1 = vec_perm(rgb0, rgb1, (__vector unsigned char)RGBG_INDEX1);
+      rgbg2 = vec_perm(rgb1, rgb2, (__vector unsigned char)RGBG_INDEX2);
+      rgbg3 = vec_perm(rgb2, rgb2, (__vector unsigned char)RGBG_INDEX3);
+#else
+      /* Load 16 pixels == 64 bytes */
+      if ((size_t)inptr & 15) {
+        __vector unsigned char unaligned_shift_index;
+        rgb0 = vec_ld(0, inptr);
+        if (pitch > 16)
+          rgb1 = vec_ld(16, inptr);
+        else
+          rgb1 = vec_ld(-1, inptr + pitch);
+        if (pitch > 32)
+          rgb2 = vec_ld(32, inptr);
+        else
+          rgb2 = vec_ld(-1, inptr + pitch);
+        if (pitch > 48)
+          rgb3 = vec_ld(48, inptr);
+        else
+          rgb3 = vec_ld(-1, inptr + pitch);
+        if (pitch > 64)
+          rgb4 = vec_ld(64, inptr);
+        else
+          rgb4 = vec_ld(-1, inptr + pitch);
+        unaligned_shift_index = vec_lvsl(0, inptr);
+        rgb0 = vec_perm(rgb0, rgb1, unaligned_shift_index);
+        rgb1 = vec_perm(rgb1, rgb2, unaligned_shift_index);
+        rgb2 = vec_perm(rgb2, rgb3, unaligned_shift_index);
+        rgb3 = vec_perm(rgb3, rgb4, unaligned_shift_index);
+      } else {
+        rgb0 = vec_ld(0, inptr);
+        if (pitch > 16)
+          rgb1 = vec_ld(16, inptr);
+        if (pitch > 32)
+          rgb2 = vec_ld(32, inptr);
+        if (pitch > 48)
+          rgb3 = vec_ld(48, inptr);
+      }
+
+      /* rgb0 = R0 G0 B0 X0 R1 G1 B1 X1 R2 G2 B2 X2 R3 G3 B3 X3
+       * rgb0 = R4 G4 B4 X4 R5 G5 B5 X5 R6 G6 B6 X6 R7 G7 B7 X7
+       * rgb0 = R8 G8 B8 X8 R9 G9 B9 X9 Ra Ga Ba Xa Rb Gb Bb Xb
+       * rgb0 = Rc Gc Bc Xc Rd Gd Bd Xd Re Ge Be Xe Rf Gf Bf Xf
+       *
+       * rgbg0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 G0 B1 G1 B2 G2 B3 G3
+       * rgbg1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 G4 B5 G5 B6 G6 B7 G7
+       * rgbg2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 G8 B9 G9 Ba Ga Bb Gb
+       * rgbg3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Gc Bd Gd Be Ge Bf Gf
+       */
+      rgbg0 = vec_perm(rgb0, rgb0, (__vector unsigned char)RGBG_INDEX);
+      rgbg1 = vec_perm(rgb1, rgb1, (__vector unsigned char)RGBG_INDEX);
+      rgbg2 = vec_perm(rgb2, rgb2, (__vector unsigned char)RGBG_INDEX);
+      rgbg3 = vec_perm(rgb3, rgb3, (__vector unsigned char)RGBG_INDEX);
+#endif
+
+      /* rg0 = R0 G0 R1 G1 R2 G2 R3 G3
+       * bg0 = B0 G0 B1 G1 B2 G2 B3 G3
+       * ...
+       *
+       * NOTE: We have to use vec_merge*() here because vec_unpack*() doesn't
+       * support unsigned vectors.
+       */
+      rg0 = (__vector signed short)vec_mergeh(zero, rgbg0);
+      bg0 = (__vector signed short)vec_mergel(zero, rgbg0);
+      rg1 = (__vector signed short)vec_mergeh(zero, rgbg1);
+      bg1 = (__vector signed short)vec_mergel(zero, rgbg1);
+      rg2 = (__vector signed short)vec_mergeh(zero, rgbg2);
+      bg2 = (__vector signed short)vec_mergel(zero, rgbg2);
+      rg3 = (__vector signed short)vec_mergeh(zero, rgbg3);
+      bg3 = (__vector signed short)vec_mergel(zero, rgbg3);
+
+      /* (Original)
+       * Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
+       * Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+       * Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+       *
+       * (This implementation)
+       * Y  =  0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+       * Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+       * Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+       */
+
+      /* Calculate Y values */
+
+      y0 = vec_msums(rg0, pw_f0299_f0337, pd_onehalf);
+      y1 = vec_msums(rg1, pw_f0299_f0337, pd_onehalf);
+      y2 = vec_msums(rg2, pw_f0299_f0337, pd_onehalf);
+      y3 = vec_msums(rg3, pw_f0299_f0337, pd_onehalf);
+      y0 = vec_msums(bg0, pw_f0114_f0250, y0);
+      y1 = vec_msums(bg1, pw_f0114_f0250, y1);
+      y2 = vec_msums(bg2, pw_f0114_f0250, y2);
+      y3 = vec_msums(bg3, pw_f0114_f0250, y3);
+      /* Clever way to avoid 4 shifts + 2 packs.  This packs the high word from
+       * each dword into a new 16-bit vector, which is the equivalent of
+       * descaling the 32-bit results (right-shifting by 16 bits) and then
+       * packing them.
+       */
+      y01 = vec_perm((__vector unsigned short)y0, (__vector unsigned short)y1,
+                     shift_pack_index);
+      y23 = vec_perm((__vector unsigned short)y2, (__vector unsigned short)y3,
+                     shift_pack_index);
+      y = vec_pack(y01, y23);
+      vec_st(y, 0, outptr0);
+
+      /* Calculate Cb values */
+      cb0 = vec_msums(rg0, pw_mf016_mf033, pd_onehalfm1_cj);
+      cb1 = vec_msums(rg1, pw_mf016_mf033, pd_onehalfm1_cj);
+      cb2 = vec_msums(rg2, pw_mf016_mf033, pd_onehalfm1_cj);
+      cb3 = vec_msums(rg3, pw_mf016_mf033, pd_onehalfm1_cj);
+      cb0 = (__vector int)vec_msum((__vector unsigned short)bg0, pw_f050_f000,
+                                   (__vector unsigned int)cb0);
+      cb1 = (__vector int)vec_msum((__vector unsigned short)bg1, pw_f050_f000,
+                                   (__vector unsigned int)cb1);
+      cb2 = (__vector int)vec_msum((__vector unsigned short)bg2, pw_f050_f000,
+                                   (__vector unsigned int)cb2);
+      cb3 = (__vector int)vec_msum((__vector unsigned short)bg3, pw_f050_f000,
+                                   (__vector unsigned int)cb3);
+      cb01 = vec_perm((__vector unsigned short)cb0,
+                      (__vector unsigned short)cb1, shift_pack_index);
+      cb23 = vec_perm((__vector unsigned short)cb2,
+                      (__vector unsigned short)cb3, shift_pack_index);
+      cb = vec_pack(cb01, cb23);
+      vec_st(cb, 0, outptr1);
+
+      /* Calculate Cr values */
+      cr0 = vec_msums(bg0, pw_mf008_mf041, pd_onehalfm1_cj);
+      cr1 = vec_msums(bg1, pw_mf008_mf041, pd_onehalfm1_cj);
+      cr2 = vec_msums(bg2, pw_mf008_mf041, pd_onehalfm1_cj);
+      cr3 = vec_msums(bg3, pw_mf008_mf041, pd_onehalfm1_cj);
+      cr0 = (__vector int)vec_msum((__vector unsigned short)rg0, pw_f050_f000,
+                                   (__vector unsigned int)cr0);
+      cr1 = (__vector int)vec_msum((__vector unsigned short)rg1, pw_f050_f000,
+                                   (__vector unsigned int)cr1);
+      cr2 = (__vector int)vec_msum((__vector unsigned short)rg2, pw_f050_f000,
+                                   (__vector unsigned int)cr2);
+      cr3 = (__vector int)vec_msum((__vector unsigned short)rg3, pw_f050_f000,
+                                   (__vector unsigned int)cr3);
+      cr01 = vec_perm((__vector unsigned short)cr0,
+                      (__vector unsigned short)cr1, shift_pack_index);
+      cr23 = vec_perm((__vector unsigned short)cr2,
+                      (__vector unsigned short)cr3, shift_pack_index);
+      cr = vec_pack(cr01, cr23);
+      vec_st(cr, 0, outptr2);
+    }
+  }
+}

simd/jccolor-altivec.c

@@ -0,0 +1,104 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* RGB --> YCC CONVERSION */
+
+#include "jsimd_altivec.h"
+
+
+#define F_0_081 5329                 /* FIX(0.08131) */
+#define F_0_114 7471                 /* FIX(0.11400) */
+#define F_0_168 11059                /* FIX(0.16874) */
+#define F_0_250 16384                /* FIX(0.25000) */
+#define F_0_299 19595                /* FIX(0.29900) */
+#define F_0_331 21709                /* FIX(0.33126) */
+#define F_0_418 27439                /* FIX(0.41869) */
+#define F_0_500 32768                /* FIX(0.50000) */
+#define F_0_587 38470                /* FIX(0.58700) */
+#define F_0_337 (F_0_587 - F_0_250)  /* FIX(0.58700) - FIX(0.25000) */
+
+#define SCALEBITS 16
+#define ONE_HALF (1 << (SCALEBITS - 1))
+
+
+#define RGBG_INDEX0 {0,1,3,4,6,7,9,10,2,1,5,4,8,7,11,10}
+#define RGBG_INDEX1 {12,13,15,16,18,19,21,22,14,13,17,16,20,19,23,22}
+#define RGBG_INDEX2 {8,9,11,12,14,15,17,18,10,9,13,12,16,15,19,18}
+#define RGBG_INDEX3 {4,5,7,8,10,11,13,14,6,5,9,8,12,11,15,14}
+#include "jccolext-altivec.c"
+#undef RGB_PIXELSIZE
+
+#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+#define jsimd_rgb_ycc_convert_altivec jsimd_extrgb_ycc_convert_altivec
+#include "jccolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX0
+#undef RGBG_INDEX1
+#undef RGBG_INDEX2
+#undef RGBG_INDEX3
+#undef jsimd_rgb_ycc_convert_altivec
+
+#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+#define RGBG_INDEX {0,1,4,5,8,9,12,13,2,1,6,5,10,9,14,13}
+#define jsimd_rgb_ycc_convert_altivec jsimd_extrgbx_ycc_convert_altivec
+#include "jccolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX
+#undef jsimd_rgb_ycc_convert_altivec
+
+#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+#define RGBG_INDEX0 {2,1,5,4,8,7,11,10,0,1,3,4,6,7,9,10}
+#define RGBG_INDEX1 {14,13,17,16,20,19,23,22,12,13,15,16,18,19,21,22}
+#define RGBG_INDEX2 {10,9,13,12,16,15,19,18,8,9,11,12,14,15,17,18}
+#define RGBG_INDEX3 {6,5,9,8,12,11,15,14,4,5,7,8,10,11,13,14}
+#define jsimd_rgb_ycc_convert_altivec jsimd_extbgr_ycc_convert_altivec
+#include "jccolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX0
+#undef RGBG_INDEX1
+#undef RGBG_INDEX2
+#undef RGBG_INDEX3
+#undef jsimd_rgb_ycc_convert_altivec
+
+#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+#define RGBG_INDEX {2,1,6,5,10,9,14,13,0,1,4,5,8,9,12,13}
+#define jsimd_rgb_ycc_convert_altivec jsimd_extbgrx_ycc_convert_altivec
+#include "jccolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX
+#undef jsimd_rgb_ycc_convert_altivec
+
+#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+#define RGBG_INDEX {3,2,7,6,11,10,15,14,1,2,5,6,9,10,13,14}
+#define jsimd_rgb_ycc_convert_altivec jsimd_extxbgr_ycc_convert_altivec
+#include "jccolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX
+#undef jsimd_rgb_ycc_convert_altivec
+
+#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+#define RGBG_INDEX {1,2,5,6,9,10,13,14,3,2,7,6,11,10,15,14}
+#define jsimd_rgb_ycc_convert_altivec jsimd_extxrgb_ycc_convert_altivec
+#include "jccolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX
+#undef jsimd_rgb_ycc_convert_altivec

simd/jcgray-altivec.c

@@ -0,0 +1,99 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* RGB --> GRAYSCALE CONVERSION */
+
+#include "jsimd_altivec.h"
+
+
+#define F_0_114 7471                 /* FIX(0.11400) */
+#define F_0_250 16384                /* FIX(0.25000) */
+#define F_0_299 19595                /* FIX(0.29900) */
+#define F_0_587 38470                /* FIX(0.58700) */
+#define F_0_337 (F_0_587 - F_0_250)  /* FIX(0.58700) - FIX(0.25000) */
+
+#define SCALEBITS 16
+#define ONE_HALF (1 << (SCALEBITS - 1))
+
+
+#define RGBG_INDEX0 {0,1,3,4,6,7,9,10,2,1,5,4,8,7,11,10}
+#define RGBG_INDEX1 {12,13,15,16,18,19,21,22,14,13,17,16,20,19,23,22}
+#define RGBG_INDEX2 {8,9,11,12,14,15,17,18,10,9,13,12,16,15,19,18}
+#define RGBG_INDEX3 {4,5,7,8,10,11,13,14,6,5,9,8,12,11,15,14}
+#include "jcgryext-altivec.c"
+#undef RGB_PIXELSIZE
+
+#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+#define jsimd_rgb_gray_convert_altivec jsimd_extrgb_gray_convert_altivec
+#include "jcgryext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX0
+#undef RGBG_INDEX1
+#undef RGBG_INDEX2
+#undef RGBG_INDEX3
+#undef jsimd_rgb_gray_convert_altivec
+
+#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+#define RGBG_INDEX {0,1,4,5,8,9,12,13,2,1,6,5,10,9,14,13}
+#define jsimd_rgb_gray_convert_altivec jsimd_extrgbx_gray_convert_altivec
+#include "jcgryext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX
+#undef jsimd_rgb_gray_convert_altivec
+
+#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+#define RGBG_INDEX0 {2,1,5,4,8,7,11,10,0,1,3,4,6,7,9,10}
+#define RGBG_INDEX1 {14,13,17,16,20,19,23,22,12,13,15,16,18,19,21,22}
+#define RGBG_INDEX2 {10,9,13,12,16,15,19,18,8,9,11,12,14,15,17,18}
+#define RGBG_INDEX3 {6,5,9,8,12,11,15,14,4,5,7,8,10,11,13,14}
+#define jsimd_rgb_gray_convert_altivec jsimd_extbgr_gray_convert_altivec
+#include "jcgryext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX0
+#undef RGBG_INDEX1
+#undef RGBG_INDEX2
+#undef RGBG_INDEX3
+#undef jsimd_rgb_gray_convert_altivec
+
+#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+#define RGBG_INDEX {2,1,6,5,10,9,14,13,0,1,4,5,8,9,12,13}
+#define jsimd_rgb_gray_convert_altivec jsimd_extbgrx_gray_convert_altivec
+#include "jcgryext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX
+#undef jsimd_rgb_gray_convert_altivec
+
+#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+#define RGBG_INDEX {3,2,7,6,11,10,15,14,1,2,5,6,9,10,13,14}
+#define jsimd_rgb_gray_convert_altivec jsimd_extxbgr_gray_convert_altivec
+#include "jcgryext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX
+#undef jsimd_rgb_gray_convert_altivec
+
+#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+#define RGBG_INDEX {1,2,5,6,9,10,13,14,3,2,7,6,11,10,15,14}
+#define jsimd_rgb_gray_convert_altivec jsimd_extxrgb_gray_convert_altivec
+#include "jcgryext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGBG_INDEX
+#undef jsimd_rgb_gray_convert_altivec

simd/jcgryext-altivec.c

@@ -0,0 +1,200 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * Copyright (C) 2014, Jay Foad.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* This file is included by jcgray-altivec.c */
+
+
+void jsimd_rgb_gray_convert_altivec (JDIMENSION img_width,
+                                     JSAMPARRAY input_buf,
+                                     JSAMPIMAGE output_buf,
+                                     JDIMENSION output_row, int num_rows)
+{
+  JSAMPROW inptr, outptr;
+  int pitch;
+  __vector unsigned char rgb0, rgb1 = {0}, rgb2 = {0}, rgb3 = {0}, rgbg0,
+    rgbg1, rgbg2, rgbg3, y;
+#if RGB_PIXELSIZE == 4
+  __vector unsigned char rgb4;
+#endif
+  __vector short rg0, rg1, rg2, rg3, bg0, bg1, bg2, bg3;
+  __vector unsigned short y01, y23;
+  __vector int y0, y1, y2, y3;
+
+  /* Constants */
+  __vector short pw_f0299_f0337 = { __4X2(F_0_299, F_0_337) },
+    pw_f0114_f0250 = { __4X2(F_0_114, F_0_250) };
+  __vector int pd_onehalf = { __4X(ONE_HALF) };
+  __vector unsigned char zero = { __16X(0) },
+    shift_pack_index =
+      { 0, 1, 4, 5, 8, 9, 12, 13, 16, 17, 20, 21, 24, 25, 28, 29};
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr = output_buf[0][output_row];
+    output_row++;
+
+    for (pitch = img_width * RGB_PIXELSIZE; pitch > 0;
+         pitch -= RGB_PIXELSIZE * 16, inptr += RGB_PIXELSIZE * 16,
+         outptr += 16) {
+
+#if RGB_PIXELSIZE == 3
+      /* Load 16 pixels == 48 bytes */
+      if ((size_t)inptr & 15) {
+        __vector unsigned char unaligned_shift_index;
+        rgb0 = vec_ld(0, inptr);
+        if (pitch > 16)
+          rgb1 = vec_ld(16, inptr);
+        else
+          rgb1 = vec_ld(-1, inptr + pitch);
+        if (pitch > 32)
+          rgb2 = vec_ld(32, inptr);
+        else
+          rgb2 = vec_ld(-1, inptr + pitch);
+        if (pitch > 48)
+          rgb3 = vec_ld(48, inptr);
+        else
+          rgb3 = vec_ld(-1, inptr + pitch);
+        unaligned_shift_index = vec_lvsl(0, inptr);
+        rgb0 = vec_perm(rgb0, rgb1, unaligned_shift_index);
+        rgb1 = vec_perm(rgb1, rgb2, unaligned_shift_index);
+        rgb2 = vec_perm(rgb2, rgb3, unaligned_shift_index);
+      } else {
+        rgb0 = vec_ld(0, inptr);
+        if (pitch > 16)
+          rgb1 = vec_ld(16, inptr);
+        if (pitch > 32)
+          rgb2 = vec_ld(32, inptr);
+      }
+
+      /* rgb0 = R0 G0 B0 R1 G1 B1 R2 G2 B2 R3 G3 B3 R4 G4 B4 R5
+       * rgb1 = G5 B5 R6 G6 B6 R7 G7 B7 R8 G8 B8 R9 G9 B9 Ra Ga
+       * rgb2 = Ba Rb Gb Bb Rc Gc Bc Rd Gd Bd Re Ge Be Rf Gf Bf
+       *
+       * rgbg0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 G0 B1 G1 B2 G2 B3 G3
+       * rgbg1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 G4 B5 G5 B6 G6 B7 G7
+       * rgbg2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 G8 B9 G9 Ba Ga Bb Gb
+       * rgbg3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Gc Bd Gd Be Ge Bf Gf
+       */
+      rgbg0 = vec_perm(rgb0, rgb0, (__vector unsigned char)RGBG_INDEX0);
+      rgbg1 = vec_perm(rgb0, rgb1, (__vector unsigned char)RGBG_INDEX1);
+      rgbg2 = vec_perm(rgb1, rgb2, (__vector unsigned char)RGBG_INDEX2);
+      rgbg3 = vec_perm(rgb2, rgb2, (__vector unsigned char)RGBG_INDEX3);
+#else
+      /* Load 16 pixels == 64 bytes */
+      if ((size_t)inptr & 15) {
+        __vector unsigned char unaligned_shift_index;
+        rgb0 = vec_ld(0, inptr);
+        if (pitch > 16)
+          rgb1 = vec_ld(16, inptr);
+        else
+          rgb1 = vec_ld(-1, inptr + pitch);
+        if (pitch > 32)
+          rgb2 = vec_ld(32, inptr);
+        else
+          rgb2 = vec_ld(-1, inptr + pitch);
+        if (pitch > 48)
+          rgb3 = vec_ld(48, inptr);
+        else
+          rgb3 = vec_ld(-1, inptr + pitch);
+        if (pitch > 64)
+          rgb4 = vec_ld(64, inptr);
+        else
+          rgb4 = vec_ld(-1, inptr + pitch);
+        unaligned_shift_index = vec_lvsl(0, inptr);
+        rgb0 = vec_perm(rgb0, rgb1, unaligned_shift_index);
+        rgb1 = vec_perm(rgb1, rgb2, unaligned_shift_index);
+        rgb2 = vec_perm(rgb2, rgb3, unaligned_shift_index);
+        rgb3 = vec_perm(rgb3, rgb4, unaligned_shift_index);
+      } else {
+        rgb0 = vec_ld(0, inptr);
+        if (pitch > 16)
+          rgb1 = vec_ld(16, inptr);
+        if (pitch > 32)
+          rgb2 = vec_ld(32, inptr);
+        if (pitch > 48)
+          rgb3 = vec_ld(48, inptr);
+      }
+
+      /* rgb0 = R0 G0 B0 X0 R1 G1 B1 X1 R2 G2 B2 X2 R3 G3 B3 X3
+       * rgb0 = R4 G4 B4 X4 R5 G5 B5 X5 R6 G6 B6 X6 R7 G7 B7 X7
+       * rgb0 = R8 G8 B8 X8 R9 G9 B9 X9 Ra Ga Ba Xa Rb Gb Bb Xb
+       * rgb0 = Rc Gc Bc Xc Rd Gd Bd Xd Re Ge Be Xe Rf Gf Bf Xf
+       *
+       * rgbg0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 G0 B1 G1 B2 G2 B3 G3
+       * rgbg1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 G4 B5 G5 B6 G6 B7 G7
+       * rgbg2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 G8 B9 G9 Ba Ga Bb Gb
+       * rgbg3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Gc Bd Gd Be Ge Bf Gf
+       */
+      rgbg0 = vec_perm(rgb0, rgb0, (__vector unsigned char)RGBG_INDEX);
+      rgbg1 = vec_perm(rgb1, rgb1, (__vector unsigned char)RGBG_INDEX);
+      rgbg2 = vec_perm(rgb2, rgb2, (__vector unsigned char)RGBG_INDEX);
+      rgbg3 = vec_perm(rgb3, rgb3, (__vector unsigned char)RGBG_INDEX);
+#endif
+
+      /* rg0 = R0 G0 R1 G1 R2 G2 R3 G3
+       * bg0 = B0 G0 B1 G1 B2 G2 B3 G3
+       * ...
+       *
+       * NOTE: We have to use vec_merge*() here because vec_unpack*() doesn't
+       * support unsigned vectors.
+       */
+      rg0 = (__vector signed short)vec_mergeh(zero, rgbg0);
+      bg0 = (__vector signed short)vec_mergel(zero, rgbg0);
+      rg1 = (__vector signed short)vec_mergeh(zero, rgbg1);
+      bg1 = (__vector signed short)vec_mergel(zero, rgbg1);
+      rg2 = (__vector signed short)vec_mergeh(zero, rgbg2);
+      bg2 = (__vector signed short)vec_mergel(zero, rgbg2);
+      rg3 = (__vector signed short)vec_mergeh(zero, rgbg3);
+      bg3 = (__vector signed short)vec_mergel(zero, rgbg3);
+
+      /* (Original)
+       * Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
+       *
+       * (This implementation)
+       * Y  =  0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+       */
+
+      /* Calculate Y values */
+
+      y0 = vec_msums(rg0, pw_f0299_f0337, pd_onehalf);
+      y1 = vec_msums(rg1, pw_f0299_f0337, pd_onehalf);
+      y2 = vec_msums(rg2, pw_f0299_f0337, pd_onehalf);
+      y3 = vec_msums(rg3, pw_f0299_f0337, pd_onehalf);
+      y0 = vec_msums(bg0, pw_f0114_f0250, y0);
+      y1 = vec_msums(bg1, pw_f0114_f0250, y1);
+      y2 = vec_msums(bg2, pw_f0114_f0250, y2);
+      y3 = vec_msums(bg3, pw_f0114_f0250, y3);
+      /* Clever way to avoid 4 shifts + 2 packs.  This packs the high word from
+       * each dword into a new 16-bit vector, which is the equivalent of
+       * descaling the 32-bit results (right-shifting by 16 bits) and then
+       * packing them.
+       */
+      y01 = vec_perm((__vector unsigned short)y0, (__vector unsigned short)y1,
+                     shift_pack_index);
+      y23 = vec_perm((__vector unsigned short)y2, (__vector unsigned short)y3,
+                     shift_pack_index);
+      y = vec_pack(y01, y23);
+      vec_st(y, 0, outptr);
+    }
+  }
+}

simd/jfdctfst-altivec.c

@@ -0,0 +1,156 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* FAST INTEGER FORWARD DCT
+ *
+ * This is similar to the SSE2 implementation, except that we left-shift the
+ * constants by 1 less bit (the -1 in CONST_SHIFT.)  This is because
+ * vec_madds(arg1, arg2, arg3) generates the 16-bit saturated sum of:
+ *   the elements in arg3 + the most significant 17 bits of
+ *     (the elements in arg1 * the elements in arg2).
+ */
+
+#include "jsimd_altivec.h"
+
+
+#define F_0_382 98   /* FIX(0.382683433) */
+#define F_0_541 139  /* FIX(0.541196100) */
+#define F_0_707 181  /* FIX(0.707106781) */
+#define F_1_306 334  /* FIX(1.306562965) */
+
+#define CONST_BITS 8
+#define PRE_MULTIPLY_SCALE_BITS 2
+#define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS - 1)
+
+
+#define DO_FDCT()  \
+{  \
+  /* Even part */  \
+  \
+  tmp10 = vec_add(tmp0, tmp3);  \
+  tmp13 = vec_sub(tmp0, tmp3);  \
+  tmp11 = vec_add(tmp1, tmp2);  \
+  tmp12 = vec_sub(tmp1, tmp2);  \
+  \
+  out0  = vec_add(tmp10, tmp11);  \
+  out4  = vec_sub(tmp10, tmp11);  \
+  \
+  z1 = vec_add(tmp12, tmp13);  \
+  z1 = vec_sl(z1, pre_multiply_scale_bits);  \
+  z1 = vec_madds(z1, pw_0707, zero);  \
+  \
+  out2 = vec_add(tmp13, z1);  \
+  out6 = vec_sub(tmp13, z1);  \
+  \
+  /* Odd part */  \
+  \
+  tmp10 = vec_add(tmp4, tmp5);  \
+  tmp11 = vec_add(tmp5, tmp6);  \
+  tmp12 = vec_add(tmp6, tmp7);  \
+  \
+  tmp10 = vec_sl(tmp10, pre_multiply_scale_bits);  \
+  tmp12 = vec_sl(tmp12, pre_multiply_scale_bits);  \
+  z5 = vec_sub(tmp10, tmp12);  \
+  z5 = vec_madds(z5, pw_0382, zero);  \
+  \
+  z2 = vec_madds(tmp10, pw_0541, z5);  \
+  z4 = vec_madds(tmp12, pw_1306, z5);  \
+  \
+  tmp11 = vec_sl(tmp11, pre_multiply_scale_bits);  \
+  z3 = vec_madds(tmp11, pw_0707, zero);  \
+  \
+  z11 = vec_add(tmp7, z3);  \
+  z13 = vec_sub(tmp7, z3);  \
+  \
+  out5 = vec_add(z13, z2);  \
+  out3 = vec_sub(z13, z2);  \
+  out1 = vec_add(z11, z4);  \
+  out7 = vec_sub(z11, z4);  \
+}
+
+
+void
+jsimd_fdct_ifast_altivec (DCTELEM *data)
+{
+  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
+    col0, col1, col2, col3, col4, col5, col6, col7,
+    tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp10, tmp11, tmp12, tmp13,
+    z1, z2, z3, z4, z5, z11, z13,
+    out0, out1, out2, out3, out4, out5, out6, out7;
+
+  /* Constants */
+  __vector short zero = vec_splat_s16(0),
+    pw_0382 = { __8X(F_0_382 << CONST_SHIFT) },
+    pw_0541 = { __8X(F_0_541 << CONST_SHIFT) },
+    pw_0707 = { __8X(F_0_707 << CONST_SHIFT) },
+    pw_1306 = { __8X(F_1_306 << CONST_SHIFT) };
+  __vector unsigned short
+    pre_multiply_scale_bits = { __8X(PRE_MULTIPLY_SCALE_BITS) };
+
+  /* Pass 1: process rows */
+
+  row0 = vec_ld(0, data);
+  row1 = vec_ld(16, data);
+  row2 = vec_ld(32, data);
+  row3 = vec_ld(48, data);
+  row4 = vec_ld(64, data);
+  row5 = vec_ld(80, data);
+  row6 = vec_ld(96, data);
+  row7 = vec_ld(112, data);
+
+  TRANSPOSE(row, col);
+
+  tmp0 = vec_add(col0, col7);
+  tmp7 = vec_sub(col0, col7);
+  tmp1 = vec_add(col1, col6);
+  tmp6 = vec_sub(col1, col6);
+  tmp2 = vec_add(col2, col5);
+  tmp5 = vec_sub(col2, col5);
+  tmp3 = vec_add(col3, col4);
+  tmp4 = vec_sub(col3, col4);
+
+  DO_FDCT();
+
+  /* Pass 2: process columns */
+
+  TRANSPOSE(out, row);
+
+  tmp0 = vec_add(row0, row7);
+  tmp7 = vec_sub(row0, row7);
+  tmp1 = vec_add(row1, row6);
+  tmp6 = vec_sub(row1, row6);
+  tmp2 = vec_add(row2, row5);
+  tmp5 = vec_sub(row2, row5);
+  tmp3 = vec_add(row3, row4);
+  tmp4 = vec_sub(row3, row4);
+
+  DO_FDCT();
+
+  vec_st(out0, 0, data);
+  vec_st(out1, 16, data);
+  vec_st(out2, 32, data);
+  vec_st(out3, 48, data);
+  vec_st(out4, 64, data);
+  vec_st(out5, 80, data);
+  vec_st(out6, 96, data);
+  vec_st(out7, 112, data);
+}

simd/jfdctint-altivec.c

@@ -0,0 +1,262 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* SLOW INTEGER FORWARD DCT */
+
+#include "jsimd_altivec.h"
+
+
+#define F_0_298 2446   /* FIX(0.298631336) */
+#define F_0_390 3196   /* FIX(0.390180644) */
+#define F_0_541 4433   /* FIX(0.541196100) */
+#define F_0_765 6270   /* FIX(0.765366865) */
+#define F_0_899 7373   /* FIX(0.899976223) */
+#define F_1_175 9633   /* FIX(1.175875602) */
+#define F_1_501 12299  /* FIX(1.501321110) */
+#define F_1_847 15137  /* FIX(1.847759065) */
+#define F_1_961 16069  /* FIX(1.961570560) */
+#define F_2_053 16819  /* FIX(2.053119869) */
+#define F_2_562 20995  /* FIX(2.562915447) */
+#define F_3_072 25172  /* FIX(3.072711026) */
+
+#define CONST_BITS 13
+#define PASS1_BITS 2
+#define DESCALE_P1 (CONST_BITS - PASS1_BITS)
+#define DESCALE_P2 (CONST_BITS + PASS1_BITS)
+
+
+#define DO_FDCT_COMMON(PASS)  \
+{  \
+  /* (Original)  \
+   * z1 = (tmp12 + tmp13) * 0.541196100;  \
+   * data2 = z1 + tmp13 * 0.765366865;  \
+   * data6 = z1 + tmp12 * -1.847759065;  \
+   *  \
+   * (This implementation)  \
+   * data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;  \
+   * data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);  \
+   */  \
+  \
+  tmp1312l = vec_mergeh(tmp13, tmp12);  \
+  tmp1312h = vec_mergel(tmp13, tmp12);  \
+  \
+  out2l = vec_msums(tmp1312l, pw_f130_f054, pd_descale_p##PASS);  \
+  out2h = vec_msums(tmp1312h, pw_f130_f054, pd_descale_p##PASS);  \
+  out6l = vec_msums(tmp1312l, pw_f054_mf130, pd_descale_p##PASS);  \
+  out6h = vec_msums(tmp1312h, pw_f054_mf130, pd_descale_p##PASS);  \
+  \
+  out2l = vec_sra(out2l, descale_p##PASS);  \
+  out2h = vec_sra(out2h, descale_p##PASS);  \
+  out6l = vec_sra(out6l, descale_p##PASS);  \
+  out6h = vec_sra(out6h, descale_p##PASS);  \
+  \
+  out2 = vec_pack(out2l, out2h);  \
+  out6 = vec_pack(out6l, out6h);  \
+  \
+  /* Odd part */  \
+  \
+  z3 = vec_add(tmp4, tmp6);  \
+  z4 = vec_add(tmp5, tmp7);  \
+  \
+  /* (Original)  \
+   * z5 = (z3 + z4) * 1.175875602;  \
+   * z3 = z3 * -1.961570560;  z4 = z4 * -0.390180644;  \
+   * z3 += z5;  z4 += z5;  \
+   *  \
+   * (This implementation)  \
+   * z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;  \
+   * z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);  \
+   */  \
+  \
+  z34l = vec_mergeh(z3, z4);  \
+  z34h = vec_mergel(z3, z4);  \
+  \
+  z3l = vec_msums(z34l, pw_mf078_f117, pd_descale_p##PASS);  \
+  z3h = vec_msums(z34h, pw_mf078_f117, pd_descale_p##PASS);  \
+  z4l = vec_msums(z34l, pw_f117_f078, pd_descale_p##PASS);  \
+  z4h = vec_msums(z34h, pw_f117_f078, pd_descale_p##PASS);  \
+  \
+  /* (Original)  \
+   * z1 = tmp4 + tmp7;  z2 = tmp5 + tmp6;  \
+   * tmp4 = tmp4 * 0.298631336;  tmp5 = tmp5 * 2.053119869;  \
+   * tmp6 = tmp6 * 3.072711026;  tmp7 = tmp7 * 1.501321110;  \
+   * z1 = z1 * -0.899976223;  z2 = z2 * -2.562915447;  \
+   * data7 = tmp4 + z1 + z3;  data5 = tmp5 + z2 + z4;  \
+   * data3 = tmp6 + z2 + z3;  data1 = tmp7 + z1 + z4;  \
+   *  \
+   * (This implementation)  \
+   * tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;  \
+   * tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;  \
+   * tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);  \
+   * tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);  \
+   * data7 = tmp4 + z3;  data5 = tmp5 + z4;  \
+   * data3 = tmp6 + z3;  data1 = tmp7 + z4;  \
+   */  \
+  \
+  tmp47l = vec_mergeh(tmp4, tmp7);  \
+  tmp47h = vec_mergel(tmp4, tmp7);  \
+  \
+  out7l = vec_msums(tmp47l, pw_mf060_mf089, z3l);  \
+  out7h = vec_msums(tmp47h, pw_mf060_mf089, z3h);  \
+  out1l = vec_msums(tmp47l, pw_mf089_f060, z4l);  \
+  out1h = vec_msums(tmp47h, pw_mf089_f060, z4h);  \
+  \
+  out7l = vec_sra(out7l, descale_p##PASS);  \
+  out7h = vec_sra(out7h, descale_p##PASS);  \
+  out1l = vec_sra(out1l, descale_p##PASS);  \
+  out1h = vec_sra(out1h, descale_p##PASS);  \
+  \
+  out7 = vec_pack(out7l, out7h);  \
+  out1 = vec_pack(out1l, out1h);  \
+  \
+  tmp56l = vec_mergeh(tmp5, tmp6);  \
+  tmp56h = vec_mergel(tmp5, tmp6);  \
+  \
+  out5l = vec_msums(tmp56l, pw_mf050_mf256, z4l);  \
+  out5h = vec_msums(tmp56h, pw_mf050_mf256, z4h);  \
+  out3l = vec_msums(tmp56l, pw_mf256_f050, z3l);  \
+  out3h = vec_msums(tmp56h, pw_mf256_f050, z3h);  \
+  \
+  out5l = vec_sra(out5l, descale_p##PASS);  \
+  out5h = vec_sra(out5h, descale_p##PASS);  \
+  out3l = vec_sra(out3l, descale_p##PASS);  \
+  out3h = vec_sra(out3h, descale_p##PASS);  \
+  \
+  out5 = vec_pack(out5l, out5h);  \
+  out3 = vec_pack(out3l, out3h);  \
+}
+
+#define DO_FDCT_ROWS()  \
+{  \
+  /* Even part */  \
+  \
+  tmp10 = vec_add(tmp0, tmp3);  \
+  tmp13 = vec_sub(tmp0, tmp3);  \
+  tmp11 = vec_add(tmp1, tmp2);  \
+  tmp12 = vec_sub(tmp1, tmp2);  \
+  \
+  out0  = vec_add(tmp10, tmp11);  \
+  out0  = vec_sl(out0, pass1_bits);  \
+  out4  = vec_sub(tmp10, tmp11);  \
+  out4  = vec_sl(out4, pass1_bits);  \
+  \
+  DO_FDCT_COMMON(1);  \
+}
+
+#define DO_FDCT_COLS()  \
+{  \
+  /* Even part */  \
+  \
+  tmp10 = vec_add(tmp0, tmp3);  \
+  tmp13 = vec_sub(tmp0, tmp3);  \
+  tmp11 = vec_add(tmp1, tmp2);  \
+  tmp12 = vec_sub(tmp1, tmp2);  \
+  \
+  out0  = vec_add(tmp10, tmp11);  \
+  out0  = vec_add(out0, pw_descale_p2x);  \
+  out0  = vec_sra(out0, pass1_bits);  \
+  out4  = vec_sub(tmp10, tmp11);  \
+  out4  = vec_add(out4, pw_descale_p2x);  \
+  out4  = vec_sra(out4, pass1_bits);  \
+  \
+  DO_FDCT_COMMON(2);  \
+}
+
+
+void
+jsimd_fdct_islow_altivec (DCTELEM *data)
+{
+  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
+    col0, col1, col2, col3, col4, col5, col6, col7,
+    tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp10, tmp11, tmp12, tmp13,
+    tmp47l, tmp47h, tmp56l, tmp56h, tmp1312l, tmp1312h,
+    z3, z4, z34l, z34h,
+    out0, out1, out2, out3, out4, out5, out6, out7;
+  __vector int z3l, z3h, z4l, z4h,
+    out1l, out1h, out2l, out2h, out3l, out3h, out5l, out5h, out6l, out6h,
+    out7l, out7h;
+
+  /* Constants */
+  __vector short
+    pw_f130_f054 = { __4X2(F_0_541 + F_0_765, F_0_541) },
+    pw_f054_mf130 = { __4X2(F_0_541, F_0_541 - F_1_847) },
+    pw_mf078_f117 = { __4X2(F_1_175 - F_1_961, F_1_175) },
+    pw_f117_f078 = { __4X2(F_1_175, F_1_175 - F_0_390) },
+    pw_mf060_mf089 = { __4X2(F_0_298 - F_0_899, -F_0_899) },
+    pw_mf089_f060 = { __4X2(-F_0_899, F_1_501 - F_0_899) },
+    pw_mf050_mf256 = { __4X2(F_2_053 - F_2_562, -F_2_562) },
+    pw_mf256_f050 = { __4X2(-F_2_562, F_3_072 - F_2_562) },
+    pw_descale_p2x = { __8X(1 << (PASS1_BITS - 1)) };
+  __vector unsigned short pass1_bits = { __8X(PASS1_BITS) };
+  __vector int pd_descale_p1 = { __4X(1 << (DESCALE_P1 - 1)) },
+    pd_descale_p2 = { __4X(1 << (DESCALE_P2 - 1)) };
+  __vector unsigned int descale_p1 = { __4X(DESCALE_P1) },
+    descale_p2 = { __4X(DESCALE_P2) };
+
+  /* Pass 1: process rows */
+
+  row0 = vec_ld(0, data);
+  row1 = vec_ld(16, data);
+  row2 = vec_ld(32, data);
+  row3 = vec_ld(48, data);
+  row4 = vec_ld(64, data);
+  row5 = vec_ld(80, data);
+  row6 = vec_ld(96, data);
+  row7 = vec_ld(112, data);
+
+  TRANSPOSE(row, col);
+
+  tmp0 = vec_add(col0, col7);
+  tmp7 = vec_sub(col0, col7);
+  tmp1 = vec_add(col1, col6);
+  tmp6 = vec_sub(col1, col6);
+  tmp2 = vec_add(col2, col5);
+  tmp5 = vec_sub(col2, col5);
+  tmp3 = vec_add(col3, col4);
+  tmp4 = vec_sub(col3, col4);
+
+  DO_FDCT_ROWS();
+
+  /* Pass 2: process columns */
+
+  TRANSPOSE(out, row);
+
+  tmp0 = vec_add(row0, row7);
+  tmp7 = vec_sub(row0, row7);
+  tmp1 = vec_add(row1, row6);
+  tmp6 = vec_sub(row1, row6);
+  tmp2 = vec_add(row2, row5);
+  tmp5 = vec_sub(row2, row5);
+  tmp3 = vec_add(row3, row4);
+  tmp4 = vec_sub(row3, row4);
+
+  DO_FDCT_COLS();
+
+  vec_st(out0, 0, data);
+  vec_st(out1, 16, data);
+  vec_st(out2, 32, data);
+  vec_st(out3, 48, data);
+  vec_st(out4, 64, data);
+  vec_st(out5, 80, data);
+  vec_st(out6, 96, data);
+  vec_st(out7, 112, data);
+}

simd/jidctfst-altivec.c

@@ -0,0 +1,256 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* FAST INTEGER INVERSE DCT
+ *
+ * This is similar to the SSE2 implementation, except that we left-shift the
+ * constants by 1 less bit (the -1 in CONST_SHIFT.)  This is because
+ * vec_madds(arg1, arg2, arg3) generates the 16-bit saturated sum of:
+ *   the elements in arg3 + the most significant 17 bits of
+ *     (the elements in arg1 * the elements in arg2).
+ */
+
+#include "jsimd_altivec.h"
+
+
+#define F_1_082 277              /* FIX(1.082392200) */
+#define F_1_414 362              /* FIX(1.414213562) */
+#define F_1_847 473              /* FIX(1.847759065) */
+#define F_2_613 669              /* FIX(2.613125930) */
+#define F_1_613 (F_2_613 - 256)  /* FIX(2.613125930) - FIX(1) */
+
+#define CONST_BITS 8
+#define PASS1_BITS 2
+#define PRE_MULTIPLY_SCALE_BITS 2
+#define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS - 1)
+
+
+#define DO_IDCT(in)  \
+{  \
+  /* Even part */  \
+  \
+  tmp10 = vec_add(in##0, in##4);  \
+  tmp11 = vec_sub(in##0, in##4);  \
+  tmp13 = vec_add(in##2, in##6);  \
+  \
+  tmp12 = vec_sub(in##2, in##6);  \
+  tmp12 = vec_sl(tmp12, pre_multiply_scale_bits);  \
+  tmp12 = vec_madds(tmp12, pw_F1414, zero);  \
+  tmp12 = vec_sub(tmp12, tmp13);  \
+  \
+  tmp0 = vec_add(tmp10, tmp13);  \
+  tmp3 = vec_sub(tmp10, tmp13);  \
+  tmp1 = vec_add(tmp11, tmp12);  \
+  tmp2 = vec_sub(tmp11, tmp12);  \
+  \
+  /* Odd part */  \
+  \
+  z13 = vec_add(in##5, in##3);  \
+  z10 = vec_sub(in##5, in##3);  \
+  z10s = vec_sl(z10, pre_multiply_scale_bits);  \
+  z11 = vec_add(in##1, in##7);  \
+  z12s = vec_sub(in##1, in##7);  \
+  z12s = vec_sl(z12s, pre_multiply_scale_bits);  \
+  \
+  tmp11 = vec_sub(z11, z13);  \
+  tmp11 = vec_sl(tmp11, pre_multiply_scale_bits);  \
+  tmp11 = vec_madds(tmp11, pw_F1414, zero);  \
+  \
+  tmp7 = vec_add(z11, z13);  \
+  \
+  /* To avoid overflow...  \
+   *  \
+   * (Original)  \
+   * tmp12 = -2.613125930 * z10 + z5;  \
+   *  \
+   * (This implementation)  \
+   * tmp12 = (-1.613125930 - 1) * z10 + z5;  \
+   *       = -1.613125930 * z10 - z10 + z5;  \
+   */  \
+  \
+  z5 = vec_add(z10s, z12s);  \
+  z5 = vec_madds(z5, pw_F1847, zero);  \
+  \
+  tmp10 = vec_madds(z12s, pw_F1082, zero);  \
+  tmp10 = vec_sub(tmp10, z5);  \
+  tmp12 = vec_madds(z10s, pw_MF1613, z5);  \
+  tmp12 = vec_sub(tmp12, z10);  \
+  \
+  tmp6 = vec_sub(tmp12, tmp7);  \
+  tmp5 = vec_sub(tmp11, tmp6);  \
+  tmp4 = vec_add(tmp10, tmp5);  \
+  \
+  out0 = vec_add(tmp0, tmp7);  \
+  out1 = vec_add(tmp1, tmp6);  \
+  out2 = vec_add(tmp2, tmp5);  \
+  out3 = vec_sub(tmp3, tmp4);  \
+  out4 = vec_add(tmp3, tmp4);  \
+  out5 = vec_sub(tmp2, tmp5);  \
+  out6 = vec_sub(tmp1, tmp6);  \
+  out7 = vec_sub(tmp0, tmp7);  \
+}
+
+
+void
+jsimd_idct_ifast_altivec (void * dct_table_, JCOEFPTR coef_block,
+                          JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  short *dct_table = (short *)dct_table_;
+  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
+    col0, col1, col2, col3, col4, col5, col6, col7,
+    quant0, quant1, quant2, quant3, quant4, quant5, quant6, quant7,
+    tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp10, tmp11, tmp12, tmp13,
+    z5, z10, z10s, z11, z12s, z13,
+    out0, out1, out2, out3, out4, out5, out6, out7;
+  __vector signed char outb;
+  int *outptr;
+
+  /* Constants */
+  __vector short zero = { __8X(0) },
+    pw_F1414 = { __8X(F_1_414 << CONST_SHIFT) },
+    pw_F1847 = { __8X(F_1_847 << CONST_SHIFT) },
+    pw_MF1613 = { __8X(-F_1_613 << CONST_SHIFT) },
+    pw_F1082 = { __8X(F_1_082 << CONST_SHIFT) };
+  __vector unsigned short
+    pre_multiply_scale_bits = { __8X(PRE_MULTIPLY_SCALE_BITS) },
+    pass1_bits3 = { __8X(PASS1_BITS + 3) };
+  __vector signed char pb_centerjsamp = { __16X(CENTERJSAMPLE) };
+
+  /* Pass 1: process columns */
+
+  col0 = vec_ld(0, coef_block);
+  col1 = vec_ld(16, coef_block);
+  col2 = vec_ld(32, coef_block);
+  col3 = vec_ld(48, coef_block);
+  col4 = vec_ld(64, coef_block);
+  col5 = vec_ld(80, coef_block);
+  col6 = vec_ld(96, coef_block);
+  col7 = vec_ld(112, coef_block);
+
+  tmp1 = vec_or(col1, col2);
+  tmp2 = vec_or(col3, col4);
+  tmp1 = vec_or(tmp1, tmp2);
+  tmp3 = vec_or(col5, col6);
+  tmp3 = vec_or(tmp3, col7);
+  tmp1 = vec_or(tmp1, tmp3);
+
+  quant0 = *(__vector short *)&dct_table[0];
+  col0 = vec_mladd(col0, quant0, zero);
+
+  if (vec_all_eq(tmp1, zero)) {
+    /* AC terms all zero */
+
+    row0 = vec_splat(col0, 0);
+    row1 = vec_splat(col0, 1);
+    row2 = vec_splat(col0, 2);
+    row3 = vec_splat(col0, 3);
+    row4 = vec_splat(col0, 4);
+    row5 = vec_splat(col0, 5);
+    row6 = vec_splat(col0, 6);
+    row7 = vec_splat(col0, 7);
+
+  } else {
+
+    quant1 = *(__vector short *)&dct_table[8];
+    quant2 = *(__vector short *)&dct_table[16];
+    quant3 = *(__vector short *)&dct_table[24];
+    quant4 = *(__vector short *)&dct_table[32];
+    quant5 = *(__vector short *)&dct_table[40];
+    quant6 = *(__vector short *)&dct_table[48];
+    quant7 = *(__vector short *)&dct_table[56];
+
+    col1 = vec_mladd(col1, quant1, zero);
+    col2 = vec_mladd(col2, quant2, zero);
+    col3 = vec_mladd(col3, quant3, zero);
+    col4 = vec_mladd(col4, quant4, zero);
+    col5 = vec_mladd(col5, quant5, zero);
+    col6 = vec_mladd(col6, quant6, zero);
+    col7 = vec_mladd(col7, quant7, zero);
+
+    DO_IDCT(col);
+
+    TRANSPOSE(out, row);
+  }
+
+  /* Pass 2: process rows */
+
+  DO_IDCT(row);
+
+  out0 = vec_sra(out0, pass1_bits3);
+  out1 = vec_sra(out1, pass1_bits3);
+  out2 = vec_sra(out2, pass1_bits3);
+  out3 = vec_sra(out3, pass1_bits3);
+  out4 = vec_sra(out4, pass1_bits3);
+  out5 = vec_sra(out5, pass1_bits3);
+  out6 = vec_sra(out6, pass1_bits3);
+  out7 = vec_sra(out7, pass1_bits3);
+
+  TRANSPOSE(out, col);
+
+  outb = vec_packs(col0, col0);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[0] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col1, col1);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[1] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col2, col2);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[2] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col3, col3);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[3] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col4, col4);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[4] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col5, col5);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[5] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col6, col6);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[6] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col7, col7);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[7] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+}

simd/jidctint-altivec.c

@@ -0,0 +1,358 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* SLOW INTEGER INVERSE DCT */
+
+#include "jsimd_altivec.h"
+
+
+#define F_0_298 2446   /* FIX(0.298631336) */
+#define F_0_390 3196   /* FIX(0.390180644) */
+#define F_0_541 4433   /* FIX(0.541196100) */
+#define F_0_765 6270   /* FIX(0.765366865) */
+#define F_0_899 7373   /* FIX(0.899976223) */
+#define F_1_175 9633   /* FIX(1.175875602) */
+#define F_1_501 12299  /* FIX(1.501321110) */
+#define F_1_847 15137  /* FIX(1.847759065) */
+#define F_1_961 16069  /* FIX(1.961570560) */
+#define F_2_053 16819  /* FIX(2.053119869) */
+#define F_2_562 20995  /* FIX(2.562915447) */
+#define F_3_072 25172  /* FIX(3.072711026) */
+
+#define CONST_BITS 13
+#define PASS1_BITS 2
+#define DESCALE_P1 (CONST_BITS - PASS1_BITS)
+#define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3)
+
+
+#define DO_IDCT(in, PASS)  \
+{  \
+  /* Even part  \
+   *  \
+   * (Original)  \
+   * z1 = (z2 + z3) * 0.541196100;  \
+   * tmp2 = z1 + z3 * -1.847759065;  \
+   * tmp3 = z1 + z2 * 0.765366865;  \
+   *  \
+   * (This implementation)  \
+   * tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);  \
+   * tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;  \
+   */  \
+  \
+  in##26l = vec_mergeh(in##2, in##6);  \
+  in##26h = vec_mergel(in##2, in##6);  \
+  \
+  tmp3l = vec_msums(in##26l, pw_f130_f054, zero32);  \
+  tmp3h = vec_msums(in##26h, pw_f130_f054, zero32);  \
+  tmp2l = vec_msums(in##26l, pw_f054_mf130, zero32);  \
+  tmp2h = vec_msums(in##26h, pw_f054_mf130, zero32);  \
+  \
+  tmp0 = vec_add(in##0, in##4);  \
+  tmp1 = vec_sub(in##0, in##4);  \
+  \
+  tmp0l = vec_unpackh(tmp0);  \
+  tmp0h = vec_unpackl(tmp0);  \
+  tmp0l = vec_sl(tmp0l, const_bits);  \
+  tmp0h = vec_sl(tmp0h, const_bits);  \
+  tmp0l = vec_add(tmp0l, pd_descale_p##PASS);  \
+  tmp0h = vec_add(tmp0h, pd_descale_p##PASS);  \
+  \
+  tmp10l = vec_add(tmp0l, tmp3l);  \
+  tmp10h = vec_add(tmp0h, tmp3h);  \
+  tmp13l = vec_sub(tmp0l, tmp3l);  \
+  tmp13h = vec_sub(tmp0h, tmp3h);  \
+  \
+  tmp1l = vec_unpackh(tmp1);  \
+  tmp1h = vec_unpackl(tmp1);  \
+  tmp1l = vec_sl(tmp1l, const_bits);  \
+  tmp1h = vec_sl(tmp1h, const_bits);  \
+  tmp1l = vec_add(tmp1l, pd_descale_p##PASS);  \
+  tmp1h = vec_add(tmp1h, pd_descale_p##PASS);  \
+  \
+  tmp11l = vec_add(tmp1l, tmp2l);  \
+  tmp11h = vec_add(tmp1h, tmp2h);  \
+  tmp12l = vec_sub(tmp1l, tmp2l);  \
+  tmp12h = vec_sub(tmp1h, tmp2h);  \
+  \
+  /* Odd part */  \
+  \
+  z3 = vec_add(in##3, in##7);  \
+  z4 = vec_add(in##1, in##5);  \
+  \
+  /* (Original)  \
+   * z5 = (z3 + z4) * 1.175875602;  \
+   * z3 = z3 * -1.961570560;  z4 = z4 * -0.390180644;  \
+   * z3 += z5;  z4 += z5;  \
+   *  \
+   * (This implementation)  \
+   * z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;  \
+   * z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);  \
+   */  \
+  \
+  z34l = vec_mergeh(z3, z4);  \
+  z34h = vec_mergel(z3, z4);  \
+  \
+  z3l = vec_msums(z34l, pw_mf078_f117, zero32);  \
+  z3h = vec_msums(z34h, pw_mf078_f117, zero32);  \
+  z4l = vec_msums(z34l, pw_f117_f078, zero32);  \
+  z4h = vec_msums(z34h, pw_f117_f078, zero32);  \
+  \
+  /* (Original)  \
+   * z1 = tmp0 + tmp3;  z2 = tmp1 + tmp2;  \
+   * tmp0 = tmp0 * 0.298631336;  tmp1 = tmp1 * 2.053119869;  \
+   * tmp2 = tmp2 * 3.072711026;  tmp3 = tmp3 * 1.501321110;  \
+   * z1 = z1 * -0.899976223;  z2 = z2 * -2.562915447;  \
+   * tmp0 += z1 + z3;  tmp1 += z2 + z4;  \
+   * tmp2 += z2 + z3;  tmp3 += z1 + z4;  \
+   *  \
+   * (This implementation)  \
+   * tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;  \
+   * tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;  \
+   * tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);  \
+   * tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);  \
+   * tmp0 += z3;  tmp1 += z4;  \
+   * tmp2 += z3;  tmp3 += z4;  \
+   */  \
+  \
+  in##71l = vec_mergeh(in##7, in##1);  \
+  in##71h = vec_mergel(in##7, in##1);  \
+  \
+  tmp0l = vec_msums(in##71l, pw_mf060_mf089, z3l);  \
+  tmp0h = vec_msums(in##71h, pw_mf060_mf089, z3h);  \
+  tmp3l = vec_msums(in##71l, pw_mf089_f060, z4l);  \
+  tmp3h = vec_msums(in##71h, pw_mf089_f060, z4h);  \
+  \
+  in##53l = vec_mergeh(in##5, in##3);  \
+  in##53h = vec_mergel(in##5, in##3);  \
+  \
+  tmp1l = vec_msums(in##53l, pw_mf050_mf256, z4l);  \
+  tmp1h = vec_msums(in##53h, pw_mf050_mf256, z4h);  \
+  tmp2l = vec_msums(in##53l, pw_mf256_f050, z3l);  \
+  tmp2h = vec_msums(in##53h, pw_mf256_f050, z3h);  \
+  \
+  /* Final output stage */  \
+  \
+  out0l = vec_add(tmp10l, tmp3l);  \
+  out0h = vec_add(tmp10h, tmp3h);  \
+  out7l = vec_sub(tmp10l, tmp3l);  \
+  out7h = vec_sub(tmp10h, tmp3h);  \
+  \
+  out0l = vec_sra(out0l, descale_p##PASS);  \
+  out0h = vec_sra(out0h, descale_p##PASS);  \
+  out7l = vec_sra(out7l, descale_p##PASS);  \
+  out7h = vec_sra(out7h, descale_p##PASS);  \
+  \
+  out0 = vec_pack(out0l, out0h);  \
+  out7 = vec_pack(out7l, out7h);  \
+  \
+  out1l = vec_add(tmp11l, tmp2l);  \
+  out1h = vec_add(tmp11h, tmp2h);  \
+  out6l = vec_sub(tmp11l, tmp2l);  \
+  out6h = vec_sub(tmp11h, tmp2h);  \
+  \
+  out1l = vec_sra(out1l, descale_p##PASS);  \
+  out1h = vec_sra(out1h, descale_p##PASS);  \
+  out6l = vec_sra(out6l, descale_p##PASS);  \
+  out6h = vec_sra(out6h, descale_p##PASS);  \
+  \
+  out1 = vec_pack(out1l, out1h);  \
+  out6 = vec_pack(out6l, out6h);  \
+  \
+  out2l = vec_add(tmp12l, tmp1l);  \
+  out2h = vec_add(tmp12h, tmp1h);  \
+  out5l = vec_sub(tmp12l, tmp1l);  \
+  out5h = vec_sub(tmp12h, tmp1h);  \
+  \
+  out2l = vec_sra(out2l, descale_p##PASS);  \
+  out2h = vec_sra(out2h, descale_p##PASS);  \
+  out5l = vec_sra(out5l, descale_p##PASS);  \
+  out5h = vec_sra(out5h, descale_p##PASS);  \
+  \
+  out2 = vec_pack(out2l, out2h);  \
+  out5 = vec_pack(out5l, out5h);  \
+  \
+  out3l = vec_add(tmp13l, tmp0l);  \
+  out3h = vec_add(tmp13h, tmp0h);  \
+  out4l = vec_sub(tmp13l, tmp0l);  \
+  out4h = vec_sub(tmp13h, tmp0h);  \
+  \
+  out3l = vec_sra(out3l, descale_p##PASS);  \
+  out3h = vec_sra(out3h, descale_p##PASS);  \
+  out4l = vec_sra(out4l, descale_p##PASS);  \
+  out4h = vec_sra(out4h, descale_p##PASS);  \
+  \
+  out3 = vec_pack(out3l, out3h);  \
+  out4 = vec_pack(out4l, out4h);  \
+}
+
+
+void
+jsimd_idct_islow_altivec (void * dct_table_, JCOEFPTR coef_block,
+                          JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  short *dct_table = (short *)dct_table_;
+  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
+    col0, col1, col2, col3, col4, col5, col6, col7,
+    quant0, quant1, quant2, quant3, quant4, quant5, quant6, quant7,
+    tmp0, tmp1, tmp2, tmp3, z3, z4,
+    z34l, z34h, col71l, col71h, col26l, col26h, col53l, col53h,
+    row71l, row71h, row26l, row26h, row53l, row53h,
+    out0, out1, out2, out3, out4, out5, out6, out7;
+  __vector int tmp0l, tmp0h, tmp1l, tmp1h, tmp2l, tmp2h, tmp3l, tmp3h,
+    tmp10l, tmp10h, tmp11l, tmp11h, tmp12l, tmp12h, tmp13l, tmp13h,
+    z3l, z3h, z4l, z4h,
+    out0l, out0h, out1l, out1h, out2l, out2h, out3l, out3h, out4l, out4h,
+    out5l, out5h, out6l, out6h, out7l, out7h;
+  __vector signed char outb;
+  int *outptr;
+
+  /* Constants */
+  __vector short zero16 = { __8X(0) },
+    pw_f130_f054 = { __4X2(F_0_541 + F_0_765, F_0_541) },
+    pw_f054_mf130 = { __4X2(F_0_541, F_0_541 - F_1_847) },
+    pw_mf078_f117 = { __4X2(F_1_175 - F_1_961, F_1_175) },
+    pw_f117_f078 = { __4X2(F_1_175, F_1_175 - F_0_390) },
+    pw_mf060_mf089 = { __4X2(F_0_298 - F_0_899, -F_0_899) },
+    pw_mf089_f060 = { __4X2(-F_0_899, F_1_501 - F_0_899) },
+    pw_mf050_mf256 = { __4X2(F_2_053 - F_2_562, -F_2_562) },
+    pw_mf256_f050 = { __4X2(-F_2_562, F_3_072 - F_2_562) };
+  __vector unsigned short pass1_bits = { __8X(PASS1_BITS) };
+  __vector int zero32 = { __4X(0) },
+    pd_descale_p1 = { __4X(1 << (DESCALE_P1 - 1)) },
+    pd_descale_p2 = { __4X(1 << (DESCALE_P2 - 1)) };
+  __vector unsigned int descale_p1 = { __4X(DESCALE_P1) },
+    descale_p2 = { __4X(DESCALE_P2) },
+    const_bits = { __4X(CONST_BITS) };
+  __vector signed char pb_centerjsamp = { __16X(CENTERJSAMPLE) };
+
+  /* Pass 1: process columns */
+
+  col0 = *(__vector short *)&coef_block[0];
+  col1 = *(__vector short *)&coef_block[8];
+  col2 = *(__vector short *)&coef_block[16];
+  col3 = *(__vector short *)&coef_block[24];
+  col4 = *(__vector short *)&coef_block[32];
+  col5 = *(__vector short *)&coef_block[40];
+  col6 = *(__vector short *)&coef_block[48];
+  col7 = *(__vector short *)&coef_block[56];
+
+  tmp1 = vec_or(col1, col2);
+  tmp2 = vec_or(col3, col4);
+  tmp1 = vec_or(tmp1, tmp2);
+  tmp3 = vec_or(col5, col6);
+  tmp3 = vec_or(tmp3, col7);
+  tmp1 = vec_or(tmp1, tmp3);
+
+  quant0 = *(__vector short *)&dct_table[0];
+  col0 = vec_mladd(col0, quant0, zero16);
+
+  if (vec_all_eq(tmp1, zero16)) {
+    /* AC terms all zero */
+
+    col0 = vec_sl(col0, pass1_bits);
+
+    row0 = vec_splat(col0, 0);
+    row1 = vec_splat(col0, 1);
+    row2 = vec_splat(col0, 2);
+    row3 = vec_splat(col0, 3);
+    row4 = vec_splat(col0, 4);
+    row5 = vec_splat(col0, 5);
+    row6 = vec_splat(col0, 6);
+    row7 = vec_splat(col0, 7);
+
+  } else {
+
+    quant1 = *(__vector short *)&dct_table[8];
+    quant2 = *(__vector short *)&dct_table[16];
+    quant3 = *(__vector short *)&dct_table[24];
+    quant4 = *(__vector short *)&dct_table[32];
+    quant5 = *(__vector short *)&dct_table[40];
+    quant6 = *(__vector short *)&dct_table[48];
+    quant7 = *(__vector short *)&dct_table[56];
+
+    col1 = vec_mladd(col1, quant1, zero16);
+    col2 = vec_mladd(col2, quant2, zero16);
+    col3 = vec_mladd(col3, quant3, zero16);
+    col4 = vec_mladd(col4, quant4, zero16);
+    col5 = vec_mladd(col5, quant5, zero16);
+    col6 = vec_mladd(col6, quant6, zero16);
+    col7 = vec_mladd(col7, quant7, zero16);
+
+    DO_IDCT(col, 1);
+
+    TRANSPOSE(out, row);
+  }
+
+  /* Pass 2: process rows */
+
+  DO_IDCT(row, 2);
+
+  TRANSPOSE(out, col);
+
+  outb = vec_packs(col0, col0);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[0] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col1, col1);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[1] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col2, col2);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[2] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col3, col3);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[3] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col4, col4);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[4] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col5, col5);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[5] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col6, col6);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[6] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+
+  outb = vec_packs(col7, col7);
+  outb = vec_add(outb, pb_centerjsamp);
+  outptr = (int *)(output_buf[7] + output_col);
+  vec_ste((__vector int)outb, 0, outptr);
+  vec_ste((__vector int)outb, 4, outptr);
+}

simd/jquanti-altivec.c

@@ -0,0 +1,236 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, D. R. Commander.
+ * All rights reserved.
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* INTEGER QUANTIZATION AND SAMPLE CONVERSION */
+
+#include "jsimd_altivec.h"
+
+
+/* NOTE: The address will either be aligned or offset by 8 bytes, so we can
+ * always get the data we want by using a single vector load (although we may
+ * have to permute the result.)
+ */
+#define LOAD_ROW(row) {  \
+  elemptr = sample_data[row] + start_col;  \
+  in##row = vec_ld(0, elemptr);  \
+  if ((size_t)elemptr & 15)  \
+    in##row = vec_perm(in##row, in##row, vec_lvsl(0, elemptr));  \
+}
+
+
+void
+jsimd_convsamp_altivec (JSAMPARRAY sample_data, JDIMENSION start_col,
+                        DCTELEM * workspace)
+{
+  JSAMPROW elemptr;
+  __vector unsigned char in0, in1, in2, in3, in4, in5, in6, in7;
+  __vector short out0, out1, out2, out3, out4, out5, out6, out7;
+
+  /* Constants */
+  __vector short pw_centerjsamp = { __8X(CENTERJSAMPLE) };
+  __vector unsigned char zero = { __16X(0) };
+
+  LOAD_ROW(0);
+  LOAD_ROW(1);
+  LOAD_ROW(2);
+  LOAD_ROW(3);
+  LOAD_ROW(4);
+  LOAD_ROW(5);
+  LOAD_ROW(6);
+  LOAD_ROW(7);
+
+  out0 = (__vector short)vec_mergeh(zero, in0);
+  out1 = (__vector short)vec_mergeh(zero, in1);
+  out2 = (__vector short)vec_mergeh(zero, in2);
+  out3 = (__vector short)vec_mergeh(zero, in3);
+  out4 = (__vector short)vec_mergeh(zero, in4);
+  out5 = (__vector short)vec_mergeh(zero, in5);
+  out6 = (__vector short)vec_mergeh(zero, in6);
+  out7 = (__vector short)vec_mergeh(zero, in7);
+
+  out0 = vec_sub(out0, pw_centerjsamp);
+  out1 = vec_sub(out1, pw_centerjsamp);
+  out2 = vec_sub(out2, pw_centerjsamp);
+  out3 = vec_sub(out3, pw_centerjsamp);
+  out4 = vec_sub(out4, pw_centerjsamp);
+  out5 = vec_sub(out5, pw_centerjsamp);
+  out6 = vec_sub(out6, pw_centerjsamp);
+  out7 = vec_sub(out7, pw_centerjsamp);
+
+  vec_st(out0, 0, workspace);
+  vec_st(out1, 16, workspace);
+  vec_st(out2, 32, workspace);
+  vec_st(out3, 48, workspace);
+  vec_st(out4, 64, workspace);
+  vec_st(out5, 80, workspace);
+  vec_st(out6, 96, workspace);
+  vec_st(out7, 112, workspace);
+}
+
+
+#define WORD_BIT 16
+
+/* There is no AltiVec unsigned multiply instruction, hence this. */
+
+#define MULTIPLY(vs0, vs1, out) {  \
+  tmpe = vec_mule((__vector unsigned short)vs0,  \
+                  (__vector unsigned short)vs1);  \
+  tmpo = vec_mulo((__vector unsigned short)vs0,  \
+                  (__vector unsigned short)vs1);  \
+  out = (__vector short)vec_perm((__vector unsigned short)tmpe,  \
+                                 (__vector unsigned short)tmpo,  \
+                                 shift_pack_index);  \
+}
+
+void
+jsimd_quantize_altivec (JCOEFPTR coef_block, DCTELEM * divisors,
+                        DCTELEM * workspace)
+{
+  __vector short row0, row1, row2, row3, row4, row5, row6, row7;
+  __vector short row0s, row1s, row2s, row3s, row4s, row5s, row6s, row7s;
+  __vector short corr0, corr1, corr2, corr3, corr4, corr5, corr6, corr7;
+  __vector short recip0, recip1, recip2, recip3, recip4, recip5, recip6,
+    recip7;
+  __vector short scale0, scale1, scale2, scale3, scale4, scale5, scale6,
+    scale7;
+  __vector unsigned int tmpe, tmpo;
+
+  /* Constants */
+  __vector unsigned short pw_word_bit_m1 = { __8X(WORD_BIT - 1) };
+  __vector unsigned char shift_pack_index =
+      { 0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29};
+
+  row0 = vec_ld(0, workspace);
+  row1 = vec_ld(16, workspace);
+  row2 = vec_ld(32, workspace);
+  row3 = vec_ld(48, workspace);
+  row4 = vec_ld(64, workspace);
+  row5 = vec_ld(80, workspace);
+  row6 = vec_ld(96, workspace);
+  row7 = vec_ld(112, workspace);
+
+  /* Branch-less absolute value */
+  row0s = vec_sra(row0, pw_word_bit_m1);
+  row1s = vec_sra(row1, pw_word_bit_m1);
+  row2s = vec_sra(row2, pw_word_bit_m1);
+  row3s = vec_sra(row3, pw_word_bit_m1);
+  row4s = vec_sra(row4, pw_word_bit_m1);
+  row5s = vec_sra(row5, pw_word_bit_m1);
+  row6s = vec_sra(row6, pw_word_bit_m1);
+  row7s = vec_sra(row7, pw_word_bit_m1);
+  row0 = vec_xor(row0, row0s);
+  row1 = vec_xor(row1, row1s);
+  row2 = vec_xor(row2, row2s);
+  row3 = vec_xor(row3, row3s);
+  row4 = vec_xor(row4, row4s);
+  row5 = vec_xor(row5, row5s);
+  row6 = vec_xor(row6, row6s);
+  row7 = vec_xor(row7, row7s);
+  row0 = vec_sub(row0, row0s);
+  row1 = vec_sub(row1, row1s);
+  row2 = vec_sub(row2, row2s);
+  row3 = vec_sub(row3, row3s);
+  row4 = vec_sub(row4, row4s);
+  row5 = vec_sub(row5, row5s);
+  row6 = vec_sub(row6, row6s);
+  row7 = vec_sub(row7, row7s);
+
+  corr0 = vec_ld(DCTSIZE2 * 2, divisors);
+  corr1 = vec_ld(DCTSIZE2 * 2 + 16, divisors);
+  corr2 = vec_ld(DCTSIZE2 * 2 + 32, divisors);
+  corr3 = vec_ld(DCTSIZE2 * 2 + 48, divisors);
+  corr4 = vec_ld(DCTSIZE2 * 2 + 64, divisors);
+  corr5 = vec_ld(DCTSIZE2 * 2 + 80, divisors);
+  corr6 = vec_ld(DCTSIZE2 * 2 + 96, divisors);
+  corr7 = vec_ld(DCTSIZE2 * 2 + 112, divisors);
+
+  row0 = vec_add(row0, corr0);
+  row1 = vec_add(row1, corr1);
+  row2 = vec_add(row2, corr2);
+  row3 = vec_add(row3, corr3);
+  row4 = vec_add(row4, corr4);
+  row5 = vec_add(row5, corr5);
+  row6 = vec_add(row6, corr6);
+  row7 = vec_add(row7, corr7);
+
+  recip0 = vec_ld(0, divisors);
+  recip1 = vec_ld(16, divisors);
+  recip2 = vec_ld(32, divisors);
+  recip3 = vec_ld(48, divisors);
+  recip4 = vec_ld(64, divisors);
+  recip5 = vec_ld(80, divisors);
+  recip6 = vec_ld(96, divisors);
+  recip7 = vec_ld(112, divisors);
+
+  MULTIPLY(row0, recip0, row0);
+  MULTIPLY(row1, recip1, row1);
+  MULTIPLY(row2, recip2, row2);
+  MULTIPLY(row3, recip3, row3);
+  MULTIPLY(row4, recip4, row4);
+  MULTIPLY(row5, recip5, row5);
+  MULTIPLY(row6, recip6, row6);
+  MULTIPLY(row7, recip7, row7);
+
+  scale0 = vec_ld(DCTSIZE2 * 4, divisors);
+  scale1 = vec_ld(DCTSIZE2 * 4 + 16, divisors);
+  scale2 = vec_ld(DCTSIZE2 * 4 + 32, divisors);
+  scale3 = vec_ld(DCTSIZE2 * 4 + 48, divisors);
+  scale4 = vec_ld(DCTSIZE2 * 4 + 64, divisors);
+  scale5 = vec_ld(DCTSIZE2 * 4 + 80, divisors);
+  scale6 = vec_ld(DCTSIZE2 * 4 + 96, divisors);
+  scale7 = vec_ld(DCTSIZE2 * 4 + 112, divisors);
+
+  MULTIPLY(row0, scale0, row0);
+  MULTIPLY(row1, scale1, row1);
+  MULTIPLY(row2, scale2, row2);
+  MULTIPLY(row3, scale3, row3);
+  MULTIPLY(row4, scale4, row4);
+  MULTIPLY(row5, scale5, row5);
+  MULTIPLY(row6, scale6, row6);
+  MULTIPLY(row7, scale7, row7);
+
+  row0 = vec_xor(row0, row0s);
+  row1 = vec_xor(row1, row1s);
+  row2 = vec_xor(row2, row2s);
+  row3 = vec_xor(row3, row3s);
+  row4 = vec_xor(row4, row4s);
+  row5 = vec_xor(row5, row5s);
+  row6 = vec_xor(row6, row6s);
+  row7 = vec_xor(row7, row7s);
+  row0 = vec_sub(row0, row0s);
+  row1 = vec_sub(row1, row1s);
+  row2 = vec_sub(row2, row2s);
+  row3 = vec_sub(row3, row3s);
+  row4 = vec_sub(row4, row4s);
+  row5 = vec_sub(row5, row5s);
+  row6 = vec_sub(row6, row6s);
+  row7 = vec_sub(row7, row7s);
+
+  vec_st(row0, 0, coef_block);
+  vec_st(row1, 16, coef_block);
+  vec_st(row2, 32, coef_block);
+  vec_st(row3, 48, coef_block);
+  vec_st(row4, 64, coef_block);
+  vec_st(row5, 80, coef_block);
+  vec_st(row6, 96, coef_block);
+  vec_st(row7, 112, coef_block);
+}

simd/jsimd.h

@@ -116,6 +116,28 @@ EXTERN(void) jsimd_extxrgb_ycc_convert_mips_dspr2
         (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
          JDIMENSION output_row, int num_rows);
 
+EXTERN(void) jsimd_rgb_ycc_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgb_ycc_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgbx_ycc_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgr_ycc_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgrx_ycc_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxbgr_ycc_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxrgb_ycc_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+
 /* RGB & extended RGB --> Grayscale Colorspace Conversion */
 EXTERN(void) jsimd_rgb_gray_convert_mmx
         (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
@@ -184,6 +206,28 @@ EXTERN(void) jsimd_extxrgb_gray_convert_mips_dspr2
         (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
          JDIMENSION output_row, int num_rows);
 
+EXTERN(void) jsimd_rgb_gray_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgb_gray_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgbx_gray_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgr_gray_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgrx_gray_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxbgr_gray_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxrgb_gray_convert_altivec
+        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+         JDIMENSION output_row, int num_rows);
+
 /* YCC --> RGB & extended RGB Colorspace Conversion */
 EXTERN(void) jsimd_ycc_rgb_convert_mmx
         (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
@@ -524,6 +568,9 @@ EXTERN(void) jsimd_convsamp_neon
 EXTERN(void) jsimd_convsamp_mips_dspr2
         (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM * workspace);
 
+EXTERN(void) jsimd_convsamp_altivec
+        (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM * workspace);
+
 /* Floating Point Sample Conversion */
 EXTERN(void) jsimd_convsamp_float_3dnow
         (JSAMPARRAY sample_data, JDIMENSION start_col, FAST_FLOAT * workspace);
@@ -545,6 +592,8 @@ EXTERN(void) jsimd_fdct_islow_sse2 (DCTELEM * data);
 
 EXTERN(void) jsimd_fdct_islow_mips_dspr2 (DCTELEM * data);
 
+EXTERN(void) jsimd_fdct_islow_altivec (DCTELEM * data);
+
 /* Fast Integer Forward DCT */
 EXTERN(void) jsimd_fdct_ifast_mmx (DCTELEM * data);
 
@@ -576,6 +625,9 @@ EXTERN(void) jsimd_quantize_neon
 EXTERN(void) jsimd_quantize_mips_dspr2
         (JCOEFPTR coef_block, DCTELEM * divisors, DCTELEM * workspace);
 
+EXTERN(void) jsimd_quantize_altivec
+        (JCOEFPTR coef_block, DCTELEM * divisors, DCTELEM * workspace);
+
 /* Floating Point Quantization */
 EXTERN(void) jsimd_quantize_float_3dnow
         (JCOEFPTR coef_block, FAST_FLOAT * divisors, FAST_FLOAT * workspace);
@@ -644,6 +696,10 @@ EXTERN(void) jsimd_idct_islow_mips_dspr2
         (void * dct_table, JCOEFPTR coef_block, int * output_buf,
          JSAMPLE * output_col);
 
+EXTERN(void) jsimd_idct_islow_altivec
+        (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
+         JDIMENSION output_col);
+
 /* Fast Integer Inverse DCT */
 EXTERN(void) jsimd_idct_ifast_mmx
         (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
@@ -665,6 +721,10 @@ EXTERN(void) jsimd_idct_ifast_rows_mips_dspr2
         (DCTELEM * wsptr, JSAMPARRAY output_buf, JDIMENSION output_col,
          const int * idct_coefs);
 
+EXTERN(void) jsimd_idct_ifast_altivec
+        (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
+         JDIMENSION output_col);
+
 /* Floating Point Inverse DCT */
 EXTERN(void) jsimd_idct_float_3dnow
         (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,

simd/jsimd_altivec.h

@@ -29,6 +29,14 @@
 #include "jsimd.h"
 #include <altivec.h>
 
+
+/* Common code */
+
+#define __4X(a) a, a, a, a
+#define __4X2(a, b) a, b, a, b, a, b, a, b
+#define __8X(a) __4X(a), __4X(a)
+#define __16X(a) __8X(a), __8X(a)
+
 #define TRANSPOSE(row, col)  \
 {  \
   __vector short row04l, row04h, row15l, row15h,  \
@@ -47,7 +55,7 @@
   row37h = vec_mergel(row##3, row##7); /* row37h=(34 74 35 75 36 76 37 77) */ \
   \
                                        /* transpose coefficients (phase 2) */ \
-  col01e = vec_mergeh(row04l, row26l); /* col01e=(00 20 40 60 01 21 41 61} */ \
+  col01e = vec_mergeh(row04l, row26l); /* col01e=(00 20 40 60 01 21 41 61) */ \
   col23e = vec_mergel(row04l, row26l); /* col23e=(02 22 42 62 03 23 43 63) */ \
   col45e = vec_mergeh(row04h, row26h); /* col45e=(04 24 44 64 05 25 45 65) */ \
   col67e = vec_mergel(row04h, row26h); /* col67e=(06 26 46 66 07 27 47 67) */ \
@@ -58,7 +66,7 @@
   \
                                        /* transpose coefficients (phase 3) */ \
   col##0 = vec_mergeh(col01e, col01o); /* col0=(00 10 20 30 40 50 60 70) */   \
-  col##1 = vec_mergel(col01e, col01o); /* col1=(01 11 21 31 41 51 61 71} */   \
+  col##1 = vec_mergel(col01e, col01o); /* col1=(01 11 21 31 41 51 61 71) */   \
   col##2 = vec_mergeh(col23e, col23o); /* col2=(02 12 22 32 42 52 62 72) */   \
   col##3 = vec_mergel(col23e, col23o); /* col3=(03 13 23 33 43 53 63 73) */   \
   col##4 = vec_mergeh(col45e, col45o); /* col4=(04 14 24 34 44 54 64 74) */   \
@@ -66,125 +74,3 @@
   col##6 = vec_mergeh(col67e, col67o); /* col6=(06 16 26 36 46 56 66 76) */   \
   col##7 = vec_mergel(col67e, col67o); /* col7=(07 17 27 37 47 57 67 77) */   \
 }
-
-static const __vector short constants __attribute__((aligned(16))) =
-{
-  98 << 5,   /* FIX(0.382683433) */
-  139 << 5,  /* FIX(0.541196100) */
-  181 << 5,  /* FIX(0.707106781) */
-  334 << 5   /* FIX(1.306562965) */
-};
-
-#define DO_DCT()  \
-{  \
-  /* Even part */  \
-  \
-  tmp10 = vec_add(tmp0, tmp3);  \
-  tmp13 = vec_sub(tmp0, tmp3);  \
-  tmp11 = vec_add(tmp1, tmp2);  \
-  tmp12 = vec_sub(tmp1, tmp2);  \
-  \
-  out0  = vec_add(tmp10, tmp11);  \
-  out4  = vec_sub(tmp10, tmp11);  \
-  \
-  z1 = vec_add(tmp12, tmp13);  \
-  z1 = vec_sl(z1, PRE_MULTIPLY_SCALE_BITS);  \
-  z1 = vec_madds(z1, PW_0707, zero);  \
-  \
-  out2 = vec_add(tmp13, z1);  \
-  out6 = vec_sub(tmp13, z1);  \
-  \
-  /* Odd part */  \
-  \
-  tmp10 = vec_add(tmp4, tmp5);  \
-  tmp11 = vec_add(tmp5, tmp6);  \
-  tmp12 = vec_add(tmp6, tmp7);  \
-  \
-  tmp10 = vec_sl(tmp10, PRE_MULTIPLY_SCALE_BITS);  \
-  tmp12 = vec_sl(tmp12, PRE_MULTIPLY_SCALE_BITS);  \
-  z5 = vec_sub(tmp10, tmp12);  \
-  z5 = vec_madds(z5, PW_0382, zero);  \
-  \
-  z2 = vec_madds(tmp10, PW_0541, zero);  \
-  z2 = vec_add(z2, z5);  \
-  \
-  z4 = vec_madds(tmp12, PW_1306, zero);  \
-  z4 = vec_add(z4, z5);  \
-  \
-  tmp11 = vec_sl(tmp11, PRE_MULTIPLY_SCALE_BITS);  \
-  z3 = vec_madds(tmp11, PW_0707, zero);  \
-  \
-  z11 = vec_add(tmp7, z3);  \
-  z13 = vec_sub(tmp7, z3);  \
-  \
-  out5 = vec_add(z13, z2);  \
-  out3 = vec_sub(z13, z2);  \
-  out1 = vec_add(z11, z4);  \
-  out7 = vec_sub(z11, z4);  \
-}
-
-void
-jsimd_fdct_ifast_altivec (DCTELEM *data)
-{
-  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
-    col0, col1, col2, col3, col4, col5, col6, col7,
-    tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp10, tmp11, tmp12, tmp13,
-    z1, z2, z3, z4, z5, z11, z13,
-    out0, out1, out2, out3, out4, out5, out6, out7;
-
-  /* Constants */
-  __vector short zero = vec_splat_s16(0),
-    PW_0382 = vec_splat(constants, 0),
-    PW_0541 = vec_splat(constants, 1),
-    PW_0707 = vec_splat(constants, 2),
-    PW_1306 = vec_splat(constants, 3);
-  __vector unsigned short PRE_MULTIPLY_SCALE_BITS = vec_splat_u16(2);
-
-  /* Pass 1: process rows. */
-
-  row0 = *(__vector short *)&data[0];
-  row1 = *(__vector short *)&data[8];
-  row2 = *(__vector short *)&data[16];
-  row3 = *(__vector short *)&data[24];
-  row4 = *(__vector short *)&data[32];
-  row5 = *(__vector short *)&data[40];
-  row6 = *(__vector short *)&data[48];
-  row7 = *(__vector short *)&data[56];
-
-  TRANSPOSE(row, col);
-
-  tmp0 = vec_add(col0, col7);
-  tmp7 = vec_sub(col0, col7);
-  tmp1 = vec_add(col1, col6);
-  tmp6 = vec_sub(col1, col6);
-  tmp2 = vec_add(col2, col5);
-  tmp5 = vec_sub(col2, col5);
-  tmp3 = vec_add(col3, col4);
-  tmp4 = vec_sub(col3, col4);
-
-  DO_DCT();
-
-  /* Pass 2: process columns. */
-
-  TRANSPOSE(out, row);
-
-  tmp0 = vec_add(row0, row7);
-  tmp7 = vec_sub(row0, row7);
-  tmp1 = vec_add(row1, row6);
-  tmp6 = vec_sub(row1, row6);
-  tmp2 = vec_add(row2, row5);
-  tmp5 = vec_sub(row2, row5);
-  tmp3 = vec_add(row3, row4);
-  tmp4 = vec_sub(row3, row4);
-
-  DO_DCT();
-
-  *(__vector short *)&data[0] = out0;
-  *(__vector short *)&data[8] = out1;
-  *(__vector short *)&data[16] = out2;
-  *(__vector short *)&data[24] = out3;
-  *(__vector short *)&data[32] = out4;
-  *(__vector short *)&data[40] = out5;
-  *(__vector short *)&data[48] = out6;
-  *(__vector short *)&data[56] = out7;
-}

simd/jsimd_arm64_neon.S

@@ -6,6 +6,7 @@
  * Author: Siarhei Siamashka <siarhei.siamashka@nokia.com>
  * Copyright (C) 2013-2014, Linaro Limited
  * Author: Ragesh Radhakrishnan <ragesh.r@linaro.org>
+ * Copyright (C) 2014, D. R. Commander.  All rights reserved.
  *
  * This software is provided 'as-is', without any express or implied
  * warranty.  In no event will the authors be held liable for any damages
@@ -197,21 +198,21 @@ _\fname:
     tmp13 = q1;                                                               \
 }
 
-#define XFIX_0_899976223                    v0.4h[0]
+#define XFIX_0_899976223                    v0.h[0]
-#define XFIX_0_541196100                    v0.4h[1]
+#define XFIX_0_541196100                    v0.h[1]
-#define XFIX_2_562915447                    v0.4h[2]
+#define XFIX_2_562915447                    v0.h[2]
-#define XFIX_0_298631336_MINUS_0_899976223  v0.4h[3]
+#define XFIX_0_298631336_MINUS_0_899976223  v0.h[3]
-#define XFIX_1_501321110_MINUS_0_899976223  v1.4h[0]
+#define XFIX_1_501321110_MINUS_0_899976223  v1.h[0]
-#define XFIX_2_053119869_MINUS_2_562915447  v1.4h[1]
+#define XFIX_2_053119869_MINUS_2_562915447  v1.h[1]
-#define XFIX_0_541196100_PLUS_0_765366865   v1.4h[2]
+#define XFIX_0_541196100_PLUS_0_765366865   v1.h[2]
-#define XFIX_1_175875602                    v1.4h[3]
+#define XFIX_1_175875602                    v1.h[3]
-#define XFIX_1_175875602_MINUS_0_390180644  v2.4h[0]
+#define XFIX_1_175875602_MINUS_0_390180644  v2.h[0]
-#define XFIX_0_541196100_MINUS_1_847759065  v2.4h[1]
+#define XFIX_0_541196100_MINUS_1_847759065  v2.h[1]
-#define XFIX_3_072711026_MINUS_2_562915447  v2.4h[2]
+#define XFIX_3_072711026_MINUS_2_562915447  v2.h[2]
-#define XFIX_1_175875602_MINUS_1_961570560  v2.4h[3]
+#define XFIX_1_175875602_MINUS_1_961570560  v2.h[3]
 
 .balign 16
-jsimd_idct_islow_neon_consts:
+Ljsimd_idct_islow_neon_consts:
     .short FIX_0_899976223                    /* d0[0] */
     .short FIX_0_541196100                    /* d0[1] */
     .short FIX_2_562915447                    /* d0[2] */
@@ -256,54 +257,54 @@ asm_function jsimd_idct_islow_neon
     /* Save all NEON registers and x15 (32 NEON registers * 8 bytes + 16) */
     sub             sp, sp, 272
     str             x15, [sp], 16
-    adr             x15, jsimd_idct_islow_neon_consts
+    adr             x15, Ljsimd_idct_islow_neon_consts
-    st1             {v0.8b - v3.8b}, [sp], 32
+    st1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    st1             {v20.8b - v23.8b}, [sp], 32
+    st1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    st1             {v24.8b - v27.8b}, [sp], 32
+    st1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    st1             {v28.8b - v31.8b}, [sp], 32
+    st1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
     ld1             {v16.4h, v17.4h, v18.4h, v19.4h}, [COEF_BLOCK], 32
     ld1             {v0.4h, v1.4h, v2.4h, v3.4h}, [DCT_TABLE], 32
     ld1             {v20.4h, v21.4h, v22.4h, v23.4h}, [COEF_BLOCK], 32
     mul             v16.4h, v16.4h, v0.4h
     mul             v17.4h, v17.4h, v1.4h
-    ins             v16.2d[1], v17.2d[0]  /* 128 bit q8 */
+    ins             v16.d[1], v17.d[0]  /* 128 bit q8 */
     ld1             {v4.4h, v5.4h, v6.4h, v7.4h}, [DCT_TABLE], 32
     mul             v18.4h, v18.4h, v2.4h
     mul             v19.4h, v19.4h, v3.4h
-    ins             v18.2d[1], v19.2d[0]  /* 128 bit q9 */
+    ins             v18.d[1], v19.d[0]  /* 128 bit q9 */
     ld1             {v24.4h, v25.4h, v26.4h, v27.4h}, [COEF_BLOCK], 32
     mul             v20.4h, v20.4h, v4.4h
     mul             v21.4h, v21.4h, v5.4h
-    ins             v20.2d[1], v21.2d[0]  /* 128 bit q10 */
+    ins             v20.d[1], v21.d[0]  /* 128 bit q10 */
     ld1             {v0.4h, v1.4h, v2.4h, v3.4h}, [DCT_TABLE], 32
     mul             v22.4h, v22.4h, v6.4h
     mul             v23.4h, v23.4h, v7.4h
-    ins             v22.2d[1], v23.2d[0]  /* 128 bit q11 */
+    ins             v22.d[1], v23.d[0]  /* 128 bit q11 */
     ld1             {v28.4h, v29.4h, v30.4h, v31.4h}, [COEF_BLOCK]
     mul             v24.4h, v24.4h, v0.4h
     mul             v25.4h, v25.4h, v1.4h
-    ins             v24.2d[1], v25.2d[0]  /* 128 bit q12 */
+    ins             v24.d[1], v25.d[0]  /* 128 bit q12 */
     ld1             {v4.4h, v5.4h, v6.4h, v7.4h}, [DCT_TABLE], 32
     mul             v28.4h, v28.4h, v4.4h
     mul             v29.4h, v29.4h, v5.4h
-    ins             v28.2d[1], v29.2d[0]  /* 128 bit q14 */
+    ins             v28.d[1], v29.d[0]  /* 128 bit q14 */
     mul             v26.4h, v26.4h, v2.4h
     mul             v27.4h, v27.4h, v3.4h
-    ins             v26.2d[1], v27.2d[0]  /* 128 bit q13 */
+    ins             v26.d[1], v27.d[0]  /* 128 bit q13 */
     ld1             {v0.4h, v1.4h, v2.4h, v3.4h}, [x15]  /* load constants */
     add             x15, x15, #16
     mul             v30.4h, v30.4h, v6.4h
     mul             v31.4h, v31.4h, v7.4h
-    ins             v30.2d[1], v31.2d[0]  /* 128 bit q15 */
+    ins             v30.d[1], v31.d[0]  /* 128 bit q15 */
     /* Go to the bottom of the stack */
     sub             sp, sp, 352
     stp             x4, x5, [sp], 16
-    st1             {v8.4h - v11.4h}, [sp], 32  /* save NEON registers */
+    st1             {v8.4h, v9.4h, v10.4h, v11.4h}, [sp], 32  /* save NEON registers */
-    st1             {v12.4h - v15.4h}, [sp], 32
+    st1             {v12.4h, v13.4h, v14.4h, v15.4h}, [sp], 32
     /* 1-D IDCT, pass 1, left 4x8 half */
     add             v4.4h,    ROW7L.4h, ROW3L.4h
     add             v5.4h,    ROW5L.4h, ROW1L.4h
@@ -378,7 +379,7 @@ asm_function jsimd_idct_islow_neon
     rshrn           ROW0L.4h, v12.4s,   #11
     rshrn           ROW4L.4h, v6.4s,    #11
 
-      beq             3f /* Go to do some special handling for the sparse right 4x8 half */
+      b.eq          3f /* Go to do some special handling for the sparse right 4x8 half */
 
     /* 1-D IDCT, pass 1, right 4x8 half */
     ld1             {v2.4h},  [x15]    /* reload constants */
@@ -553,33 +554,33 @@ asm_function jsimd_idct_islow_neon
     shrn            ROW4R.4h, v6.4s,    #16
 
 2:  /* Descale to 8-bit and range limit */
-    ins             v16.2d[1], v17.2d[0]
+    ins             v16.d[1], v17.d[0]
-    ins             v18.2d[1], v19.2d[0]
+    ins             v18.d[1], v19.d[0]
-    ins             v20.2d[1], v21.2d[0]
+    ins             v20.d[1], v21.d[0]
-    ins             v22.2d[1], v23.2d[0]
+    ins             v22.d[1], v23.d[0]
     sqrshrn         v16.8b,   v16.8h,   #2
     sqrshrn2        v16.16b,  v18.8h,   #2
     sqrshrn         v18.8b,   v20.8h,   #2
     sqrshrn2        v18.16b,  v22.8h,   #2
 
     /* vpop            {v8.4h - d15.4h} */ /* restore NEON registers */
-    ld1             {v8.4h - v11.4h}, [sp], 32
+    ld1             {v8.4h, v9.4h, v10.4h, v11.4h}, [sp], 32
-    ld1             {v12.4h - v15.4h}, [sp], 32
+    ld1             {v12.4h, v13.4h, v14.4h, v15.4h}, [sp], 32
-    ins             v24.2d[1], v25.2d[0]
+    ins             v24.d[1], v25.d[0]
 
     sqrshrn         v20.8b,   v24.8h,   #2
       /* Transpose the final 8-bit samples and do signed->unsigned conversion */
     /* trn1            v16.8h,    v16.8h,  v18.8h */
     transpose       v16, v18, v3, .16b, .8h
-    ins             v26.2d[1], v27.2d[0]
+    ins             v26.d[1], v27.d[0]
-    ins             v28.2d[1], v29.2d[0]
+    ins             v28.d[1], v29.d[0]
-    ins             v30.2d[1], v31.2d[0]
+    ins             v30.d[1], v31.d[0]
     sqrshrn2        v20.16b,  v26.8h,   #2
     sqrshrn         v22.8b,   v28.8h,   #2
     movi            v0.16b,   #(CENTERJSAMPLE)
     sqrshrn2        v22.16b,  v30.8h,   #2
-    transpose_single v16, v17, v3, .2d, .8b
+    transpose_single v16, v17, v3, .d, .8b
-    transpose_single v18, v19, v3, .2d, .8b
+    transpose_single v18, v19, v3, .d, .8b
     add             v16.8b,   v16.8b,   v0.8b
     add             v17.8b,   v17.8b,   v0.8b
     add             v18.8b,   v18.8b,   v0.8b
@@ -590,7 +591,7 @@ asm_function jsimd_idct_islow_neon
     add             TMP1,     TMP1,     OUTPUT_COL
     add             TMP2,     TMP2,     OUTPUT_COL
     st1             {v16.8b}, [TMP1]
-    transpose_single v20, v21, v3, .2d, .8b
+    transpose_single v20, v21, v3, .d, .8b
     st1             {v17.8b}, [TMP2]
     ldp             TMP1,     TMP2,     [OUTPUT_BUF], 16
     add             TMP1,     TMP1,     OUTPUT_COL
@@ -605,7 +606,7 @@ asm_function jsimd_idct_islow_neon
     add             TMP2,     TMP2,     OUTPUT_COL
     add             TMP3,     TMP3,     OUTPUT_COL
     add             TMP4,     TMP4,     OUTPUT_COL
-    transpose_single v22, v23, v3, .2d, .8b
+    transpose_single v22, v23, v3, .d, .8b
     st1             {v20.8b}, [TMP1]
     add             v22.8b,   v22.8b,   v0.8b
     add             v23.8b,   v23.8b,   v0.8b
@@ -613,14 +614,14 @@ asm_function jsimd_idct_islow_neon
     st1             {v22.8b}, [TMP3]
     st1             {v23.8b}, [TMP4]
     ldr             x15, [sp], 16
-    ld1             {v0.8b - v3.8b}, [sp], 32
+    ld1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    ld1             {v20.8b - v23.8b}, [sp], 32
+    ld1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    ld1             {v24.8b - v27.8b}, [sp], 32
+    ld1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    ld1             {v28.8b - v31.8b}, [sp], 32
+    ld1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
     blr             x30
 
 3:  /* Left 4x8 half is done, right 4x8 half contains mostly zeros */
@@ -636,17 +637,17 @@ asm_function jsimd_idct_islow_neon
     transpose       ROW0L, ROW2L, v3, .16b, .2s
     transpose       ROW5L, ROW7L, v3, .16b, .2s
     cmp             x0, #0
-    beq             4f /* Right 4x8 half has all zeros, go to 'sparse' second pass */
+    b.eq            4f /* Right 4x8 half has all zeros, go to 'sparse' second pass */
 
     /* Only row 0 is non-zero for the right 4x8 half  */
-    dup             ROW1R.4h, ROW0R.4h[1]
+    dup             ROW1R.4h, ROW0R.h[1]
-    dup             ROW2R.4h, ROW0R.4h[2]
+    dup             ROW2R.4h, ROW0R.h[2]
-    dup             ROW3R.4h, ROW0R.4h[3]
+    dup             ROW3R.4h, ROW0R.h[3]
-    dup             ROW4R.4h, ROW0R.4h[0]
+    dup             ROW4R.4h, ROW0R.h[0]
-    dup             ROW5R.4h, ROW0R.4h[1]
+    dup             ROW5R.4h, ROW0R.h[1]
-    dup             ROW6R.4h, ROW0R.4h[2]
+    dup             ROW6R.4h, ROW0R.h[2]
-    dup             ROW7R.4h, ROW0R.4h[3]
+    dup             ROW7R.4h, ROW0R.h[3]
-    dup             ROW0R.4h, ROW0R.4h[0]
+    dup             ROW0R.4h, ROW0R.h[0]
     b               1b /* Go to 'normal' second pass */
 
 4:  /* 1-D IDCT, pass 2 (sparse variant with zero rows 4-7), left 4x8 half */
@@ -770,13 +771,13 @@ asm_function jsimd_idct_islow_neon
  * per 1-D IDCT pass, totalling to 5 VQDMULH and 35 VADD/VSUB instructions.
  */
 
-#define XFIX_1_082392200 v0.4h[0]
+#define XFIX_1_082392200 v0.h[0]
-#define XFIX_1_414213562 v0.4h[1]
+#define XFIX_1_414213562 v0.h[1]
-#define XFIX_1_847759065 v0.4h[2]
+#define XFIX_1_847759065 v0.h[2]
-#define XFIX_2_613125930 v0.4h[3]
+#define XFIX_2_613125930 v0.h[3]
 
 .balign 16
-jsimd_idct_ifast_neon_consts:
+Ljsimd_idct_ifast_neon_consts:
     .short (277 * 128 - 256 * 128) /* XFIX_1_082392200 */
     .short (362 * 128 - 256 * 128) /* XFIX_1_414213562 */
     .short (473 * 128 - 256 * 128) /* XFIX_1_847759065 */
@@ -810,12 +811,12 @@ asm_function jsimd_idct_ifast_neon
     /* Save NEON registers used in fast IDCT */
     sub             sp, sp, #176
     stp             x22, x23, [sp], 16
-    adr             x23, jsimd_idct_ifast_neon_consts
+    adr             x23, Ljsimd_idct_ifast_neon_consts
-    st1             {v0.8b - v3.8b}, [sp], 32
+    st1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
     ld1             {v8.8h, v9.8h}, [COEF_BLOCK], 32
     ld1             {v0.8h, v1.8h}, [DCT_TABLE], 32
     ld1             {v10.8h, v11.8h}, [COEF_BLOCK], 32
@@ -909,24 +910,24 @@ asm_function jsimd_idct_ifast_neon
     trn2            v15.4s,   v18.4s,   v15.4s
     /* vswp            v14.4h,   v10-MSB.4h */
     umov            x22, v14.d[0]
-    ins             v14.2d[0], v10.2d[1]
+    ins             v14.d[0], v10.d[1]
-    ins             v10.2d[1], x22
+    ins             v10.d[1], x22
     /* vswp            v13.4h,   v9MSB.4h */
 
     umov            x22, v13.d[0]
-    ins             v13.2d[0], v9.2d[1]
+    ins             v13.d[0], v9.d[1]
-    ins             v9.2d[1], x22
+    ins             v9.d[1], x22
     /* 1-D IDCT, pass 2 */
     sub             v2.8h,    v10.8h,   v14.8h
     /* vswp            v15.4h,   v11MSB.4h */
     umov            x22, v15.d[0]
-    ins             v15.2d[0], v11.2d[1]
+    ins             v15.d[0], v11.d[1]
-    ins             v11.2d[1], x22
+    ins             v11.d[1], x22
     add             v14.8h,   v10.8h,   v14.8h
     /* vswp            v12.4h,   v8-MSB.4h */
     umov            x22, v12.d[0]
-    ins             v12.2d[0], v8.2d[1]
+    ins             v12.d[0], v8.d[1]
-    ins             v8.2d[1], x22
+    ins             v8.d[1],  x22
     sub             v1.8h,    v11.8h,   v13.8h
     add             v13.8h,   v11.8h,   v13.8h
     sub             v5.8h,    v9.8h,    v15.8h
@@ -997,13 +998,13 @@ asm_function jsimd_idct_ifast_neon
     trn1            v9.4s,    v9.4s,    v11.4s
     trn2            v11.4s,   v18.4s,   v11.4s
     /* make copy */
-    ins             v17.2d[0], v8.2d[1]
+    ins             v17.d[0], v8.d[1]
     /* Transpose  d16-d17-msb */
     mov             v18.16b,  v8.16b
     trn1            v8.8b,    v8.8b,    v17.8b
     trn2            v17.8b,   v18.8b,   v17.8b
     /* make copy */
-    ins             v19.2d[0], v9.2d[1]
+    ins             v19.d[0], v9.d[1]
     mov             v18.16b,  v9.16b
     trn1            v9.8b,    v9.8b,    v19.8b
     trn2            v19.8b,   v18.8b,   v19.8b
@@ -1018,7 +1019,7 @@ asm_function jsimd_idct_ifast_neon
     add             TMP2,     TMP2,     OUTPUT_COL
     st1             {v9.8b},  [TMP1]
     /* make copy */
-    ins             v7.2d[0], v10.2d[1]
+    ins             v7.d[0],  v10.d[1]
     mov             v18.16b,  v10.16b
     trn1            v10.8b,   v10.8b,   v7.8b
     trn2            v7.8b,    v18.8b,   v7.8b
@@ -1031,7 +1032,7 @@ asm_function jsimd_idct_ifast_neon
     add             TMP5,     TMP5,     OUTPUT_COL
     st1             {v10.8b}, [TMP1]
     /* make copy */
-    ins             v16.2d[0], v11.2d[1]
+    ins             v16.d[0], v11.d[1]
     mov             v18.16b,  v11.16b
     trn1            v11.8b,   v11.8b,   v16.8b
     trn2            v16.8b,   v18.8b,   v16.8b
@@ -1040,11 +1041,11 @@ asm_function jsimd_idct_ifast_neon
     st1             {v16.8b}, [TMP5]
     sub             sp, sp, #176
     ldp             x22, x23, [sp], 16
-    ld1             {v0.8b - v3.8b}, [sp], 32
+    ld1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
     blr             x30
 
     .unreq          DCT_TABLE
@@ -1095,38 +1096,38 @@ asm_function jsimd_idct_ifast_neon
 #define FIX_3_624509785  (29692) /* FIX(3.624509785) */
 
 .balign 16
-jsimd_idct_4x4_neon_consts:
+Ljsimd_idct_4x4_neon_consts:
-    .short     FIX_1_847759065     /* v0.4h[0] */
+    .short     FIX_1_847759065     /* v0.h[0] */
-    .short     -FIX_0_765366865    /* v0.4h[1] */
+    .short     -FIX_0_765366865    /* v0.h[1] */
-    .short     -FIX_0_211164243    /* v0.4h[2] */
+    .short     -FIX_0_211164243    /* v0.h[2] */
-    .short     FIX_1_451774981     /* v0.4h[3] */
+    .short     FIX_1_451774981     /* v0.h[3] */
     .short     -FIX_2_172734803    /* d1[0] */
     .short     FIX_1_061594337     /* d1[1] */
     .short     -FIX_0_509795579    /* d1[2] */
     .short     -FIX_0_601344887    /* d1[3] */
-    .short     FIX_0_899976223     /* v2.4h[0] */
+    .short     FIX_0_899976223     /* v2.h[0] */
-    .short     FIX_2_562915447     /* v2.4h[1] */
+    .short     FIX_2_562915447     /* v2.h[1] */
-    .short     1 << (CONST_BITS+1) /* v2.4h[2] */
+    .short     1 << (CONST_BITS+1) /* v2.h[2] */
-    .short     0                   /* v2.4h[3] */
+    .short     0                   /* v2.h[3] */
 
 .macro idct_helper x4, x6, x8, x10, x12, x14, x16, shift, y26, y27, y28, y29
-    smull           v28.4s, \x4,    v2.4h[2]
+    smull           v28.4s, \x4,    v2.h[2]
-    smlal           v28.4s, \x8,    v0.4h[0]
+    smlal           v28.4s, \x8,    v0.h[0]
-    smlal           v28.4s, \x14,   v0.4h[1]
+    smlal           v28.4s, \x14,   v0.h[1]
 
-    smull           v26.4s, \x16,   v1.4h[2]
+    smull           v26.4s, \x16,   v1.h[2]
-    smlal           v26.4s, \x12,   v1.4h[3]
+    smlal           v26.4s, \x12,   v1.h[3]
-    smlal           v26.4s, \x10,   v2.4h[0]
+    smlal           v26.4s, \x10,   v2.h[0]
-    smlal           v26.4s, \x6,    v2.4h[1]
+    smlal           v26.4s, \x6,    v2.h[1]
 
-    smull           v30.4s, \x4,    v2.4h[2]
+    smull           v30.4s, \x4,    v2.h[2]
-    smlsl           v30.4s, \x8,    v0.4h[0]
+    smlsl           v30.4s, \x8,    v0.h[0]
-    smlsl           v30.4s, \x14,   v0.4h[1]
+    smlsl           v30.4s, \x14,   v0.h[1]
 
-    smull           v24.4s, \x16,   v0.4h[2]
+    smull           v24.4s, \x16,   v0.h[2]
-    smlal           v24.4s, \x12,   v0.4h[3]
+    smlal           v24.4s, \x12,   v0.h[3]
-    smlal           v24.4s, \x10,   v1.4h[0]
+    smlal           v24.4s, \x10,   v1.h[0]
-    smlal           v24.4s, \x6,    v1.4h[1]
+    smlal           v24.4s, \x6,    v1.h[1]
 
     add             v20.4s, v28.4s, v26.4s
     sub             v28.4s, v28.4s, v26.4s
@@ -1171,15 +1172,15 @@ asm_function jsimd_idct_4x4_neon
     sub             sp, sp, 272
     str             x15, [sp], 16
     /* Load constants (v3.4h is just used for padding) */
-    adr             TMP4, jsimd_idct_4x4_neon_consts
+    adr             TMP4, Ljsimd_idct_4x4_neon_consts
-    st1             {v0.8b - v3.8b}, [sp], 32
+    st1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    st1             {v20.8b - v23.8b}, [sp], 32
+    st1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    st1             {v24.8b - v27.8b}, [sp], 32
+    st1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    st1             {v28.8b - v31.8b}, [sp], 32
+    st1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
     ld1             {v0.4h, v1.4h, v2.4h, v3.4h}, [TMP4]
 
     /* Load all COEF_BLOCK into NEON registers with the following allocation:
@@ -1203,45 +1204,45 @@ asm_function jsimd_idct_4x4_neon
     ld1             {v18.4h, v19.4h, v20.4h, v21.4h}, [DCT_TABLE], 32
     mul             v4.4h, v4.4h, v18.4h
     mul             v5.4h, v5.4h, v19.4h
-    ins             v4.2d[1], v5.2d[0]    /* 128 bit q4 */
+    ins             v4.d[1], v5.d[0]    /* 128 bit q4 */
     ld1             {v22.4h, v23.4h, v24.4h, v25.4h}, [DCT_TABLE], 32
     mul             v6.4h, v6.4h, v20.4h
     mul             v7.4h, v7.4h, v21.4h
-    ins             v6.2d[1], v7.2d[0]    /* 128 bit q6 */
+    ins             v6.d[1], v7.d[0]    /* 128 bit q6 */
     mul             v8.4h, v8.4h, v22.4h
     mul             v9.4h, v9.4h, v23.4h
-    ins             v8.2d[1], v9.2d[0]    /* 128 bit q8 */
+    ins             v8.d[1], v9.d[0]    /* 128 bit q8 */
     add             DCT_TABLE, DCT_TABLE, #16
     ld1             {v26.4h, v27.4h, v28.4h, v29.4h}, [DCT_TABLE], 32
     mul             v10.4h, v10.4h, v24.4h
     mul             v11.4h, v11.4h, v25.4h
-    ins             v10.2d[1], v11.2d[0]  /* 128 bit q10 */
+    ins             v10.d[1], v11.d[0]  /* 128 bit q10 */
     mul             v12.4h, v12.4h, v26.4h
     mul             v13.4h, v13.4h, v27.4h
-    ins             v12.2d[1], v13.2d[0]  /* 128 bit q12 */
+    ins             v12.d[1], v13.d[0]  /* 128 bit q12 */
     ld1             {v30.4h, v31.4h}, [DCT_TABLE], 16
     mul             v14.4h, v14.4h, v28.4h
     mul             v15.4h, v15.4h, v29.4h
-    ins             v14.2d[1], v15.2d[0]  /* 128 bit q14 */
+    ins             v14.d[1], v15.d[0]  /* 128 bit q14 */
     mul             v16.4h, v16.4h, v30.4h
     mul             v17.4h, v17.4h, v31.4h
-    ins             v16.2d[1], v17.2d[0]  /* 128 bit q16 */
+    ins             v16.d[1], v17.d[0]  /* 128 bit q16 */
 
     /* Pass 1 */
     idct_helper     v4.4h, v6.4h, v8.4h, v10.4h, v12.4h, v14.4h, v16.4h, 12, v4.4h, v6.4h, v8.4h, v10.4h
     transpose_4x4   v4, v6, v8, v10, v3
-    ins             v10.2d[1], v11.2d[0]
+    ins             v10.d[1], v11.d[0]
     idct_helper     v5.4h, v7.4h, v9.4h, v11.4h, v13.4h, v15.4h, v17.4h, 12, v5.4h, v7.4h, v9.4h, v11.4h
     transpose_4x4   v5, v7, v9, v11, v3
-    ins             v10.2d[1], v11.2d[0]
+    ins             v10.d[1], v11.d[0]
     /* Pass 2 */
     idct_helper     v4.4h, v6.4h, v8.4h, v10.4h, v7.4h, v9.4h, v11.4h, 19, v26.4h, v27.4h, v28.4h, v29.4h
     transpose_4x4   v26, v27, v28, v29, v3
 
     /* Range limit */
     movi            v30.8h, #0x80
-    ins             v26.2d[1], v27.2d[0]
+    ins             v26.d[1], v27.d[0]
-    ins             v28.2d[1], v29.2d[0]
+    ins             v28.d[1], v29.d[0]
     add             v26.8h, v26.8h, v30.8h
     add             v28.8h, v28.8h, v30.8h
     sqxtun          v26.8b, v26.8h
@@ -1286,14 +1287,14 @@ asm_function jsimd_idct_4x4_neon
     /* vpop            {v8.4h - v15.4h}    ;not available */
     sub             sp, sp, #272
     ldr             x15, [sp], 16
-    ld1             {v0.8b - v3.8b}, [sp], 32
+    ld1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    ld1             {v20.8b - v23.8b}, [sp], 32
+    ld1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    ld1             {v24.8b - v27.8b}, [sp], 32
+    ld1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    ld1             {v28.8b - v31.8b}, [sp], 32
+    ld1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
     blr             x30
 
     .unreq          DCT_TABLE
@@ -1325,7 +1326,7 @@ asm_function jsimd_idct_4x4_neon
  */
 
 .balign 8
-jsimd_idct_2x2_neon_consts:
+Ljsimd_idct_2x2_neon_consts:
     .short     -FIX_0_720959822    /* v14[0] */
     .short     FIX_0_850430095     /* v14[1] */
     .short     -FIX_1_272758580    /* v14[2] */
@@ -1333,10 +1334,10 @@ jsimd_idct_2x2_neon_consts:
 
 .macro idct_helper x4, x6, x10, x12, x16, shift, y26, y27
     sshll      v15.4s, \x4,    #15
-    smull      v26.4s, \x6,    v14.4h[3]
+    smull      v26.4s, \x6,    v14.h[3]
-    smlal      v26.4s, \x10,   v14.4h[2]
+    smlal      v26.4s, \x10,   v14.h[2]
-    smlal      v26.4s, \x12,   v14.4h[1]
+    smlal      v26.4s, \x12,   v14.h[1]
-    smlal      v26.4s, \x16,   v14.4h[0]
+    smlal      v26.4s, \x16,   v14.h[0]
 
     add        v20.4s, v15.4s, v26.4s
     sub        v15.4s, v15.4s, v26.4s
@@ -1367,14 +1368,14 @@ asm_function jsimd_idct_2x2_neon
     str             x15, [sp], 16
 
     /* Load constants */
-    adr             TMP2, jsimd_idct_2x2_neon_consts
+    adr             TMP2, Ljsimd_idct_2x2_neon_consts
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    st1             {v21.8b - v22.8b}, [sp], 16
+    st1             {v21.8b, v22.8b}, [sp], 16
-    st1             {v24.8b - v27.8b}, [sp], 32
+    st1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    st1             {v30.8b - v31.8b}, [sp], 16
+    st1             {v30.8b, v31.8b}, [sp], 16
     ld1             {v14.4h}, [TMP2]
 
     /* Load all COEF_BLOCK into NEON registers with the following allocation:
@@ -1400,25 +1401,25 @@ asm_function jsimd_idct_2x2_neon
     ld1             {v18.4h, v19.4h, v20.4h, v21.4h}, [DCT_TABLE], 32
     mul             v4.4h, v4.4h, v18.4h
     mul             v5.4h, v5.4h, v19.4h
-    ins             v4.2d[1], v5.2d[0]
+    ins             v4.d[1], v5.d[0]
     mul             v6.4h, v6.4h, v20.4h
     mul             v7.4h, v7.4h, v21.4h
-    ins             v6.2d[1], v7.2d[0]
+    ins             v6.d[1], v7.d[0]
     add             DCT_TABLE, DCT_TABLE, #16
     ld1             {v24.4h, v25.4h}, [DCT_TABLE], 16
     mul             v10.4h, v10.4h, v24.4h
     mul             v11.4h, v11.4h, v25.4h
-    ins             v10.2d[1], v11.2d[0]
+    ins             v10.d[1], v11.d[0]
     add             DCT_TABLE, DCT_TABLE, #16
     ld1             {v26.4h, v27.4h}, [DCT_TABLE], 16
     mul             v12.4h, v12.4h, v26.4h
     mul             v13.4h, v13.4h, v27.4h
-    ins             v12.2d[1], v13.2d[0]
+    ins             v12.d[1], v13.d[0]
     add             DCT_TABLE, DCT_TABLE, #16
     ld1             {v30.4h, v31.4h}, [DCT_TABLE], 16
     mul             v16.4h, v16.4h, v30.4h
     mul             v17.4h, v17.4h, v31.4h
-    ins             v16.2d[1], v17.2d[0]
+    ins             v16.d[1], v17.d[0]
 
     /* Pass 1 */
 #if 0
@@ -1427,14 +1428,14 @@ asm_function jsimd_idct_2x2_neon
     idct_helper     v5.4h, v7.4h, v11.4h, v13.4h, v17.4h, 13, v5.4h, v7.4h
     transpose_4x4   v5.4h, v7.4h, v9.4h,  v11.4h
 #else
-    smull           v26.4s, v6.4h,  v14.4h[3]
+    smull           v26.4s, v6.4h,  v14.h[3]
-    smlal           v26.4s, v10.4h, v14.4h[2]
+    smlal           v26.4s, v10.4h, v14.h[2]
-    smlal           v26.4s, v12.4h, v14.4h[1]
+    smlal           v26.4s, v12.4h, v14.h[1]
-    smlal           v26.4s, v16.4h, v14.4h[0]
+    smlal           v26.4s, v16.4h, v14.h[0]
-    smull           v24.4s, v7.4h,  v14.4h[3]
+    smull           v24.4s, v7.4h,  v14.h[3]
-    smlal           v24.4s, v11.4h, v14.4h[2]
+    smlal           v24.4s, v11.4h, v14.h[2]
-    smlal           v24.4s, v13.4h, v14.4h[1]
+    smlal           v24.4s, v13.4h, v14.h[1]
-    smlal           v24.4s, v17.4h, v14.4h[0]
+    smlal           v24.4s, v17.4h, v14.h[0]
     sshll           v15.4s, v4.4h,  #15
     sshll           v30.4s, v5.4h,  #15
     add             v20.4s, v15.4s, v26.4s
@@ -1445,12 +1446,12 @@ asm_function jsimd_idct_2x2_neon
     sub             v15.4s, v30.4s, v24.4s
     rshrn           v5.4h,  v20.4s, #13
     rshrn           v7.4h,  v15.4s, #13
-    ins             v4.2d[1], v5.2d[0]
+    ins             v4.d[1], v5.d[0]
-    ins             v6.2d[1], v7.2d[0]
+    ins             v6.d[1], v7.d[0]
     transpose       v4, v6, v3, .16b, .8h
     transpose       v6, v10, v3, .16b, .4s
-    ins             v11.2d[0], v10.2d[1]
+    ins             v11.d[0], v10.d[1]
-    ins             v7.2d[0], v6.2d[1]
+    ins             v7.d[0], v6.d[1]
 #endif
 
     /* Pass 2 */
@@ -1458,10 +1459,10 @@ asm_function jsimd_idct_2x2_neon
 
     /* Range limit */
     movi            v30.8h, #0x80
-    ins             v26.2d[1], v27.2d[0]
+    ins             v26.d[1], v27.d[0]
     add             v26.8h, v26.8h, v30.8h
     sqxtun          v30.8b, v26.8h
-    ins             v26.2d[0], v30.2d[0]
+    ins             v26.d[0], v30.d[0]
     sqxtun          v27.8b, v26.8h
 
     /* Store results to the output buffer */
@@ -1476,13 +1477,13 @@ asm_function jsimd_idct_2x2_neon
 
     sub             sp, sp, #208
     ldr             x15, [sp], 16
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    ld1             {v21.8b - v22.8b}, [sp], 16
+    ld1             {v21.8b, v22.8b}, [sp], 16
-    ld1             {v24.8b - v27.8b}, [sp], 32
+    ld1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    ld1             {v30.8b - v31.8b}, [sp], 16
+    ld1             {v30.8b, v31.8b}, [sp], 16
     blr             x30
 
     .unreq          DCT_TABLE
@@ -1514,9 +1515,9 @@ asm_function jsimd_idct_2x2_neon
         ld1  {v4.8b}, [U], 8
         ld1  {v5.8b}, [V], 8
         ld1  {v0.8b}, [Y], 8
-        prfm PLDL1KEEP, [U, #64]
+        prfm pldl1keep, [U, #64]
-        prfm PLDL1KEEP, [V, #64]
+        prfm pldl1keep, [V, #64]
-        prfm PLDL1KEEP, [Y, #64]
+        prfm pldl1keep, [Y, #64]
     .elseif \size == 4
         ld1  {v4.b}[0], [U], 1
         ld1  {v4.b}[1], [U], 1
@@ -1606,14 +1607,14 @@ asm_function jsimd_idct_2x2_neon
 .macro do_yuv_to_rgb_stage1
     uaddw        v6.8h, v2.8h, v4.8b     /* q3 = u - 128 */
     uaddw        v8.8h, v2.8h, v5.8b     /* q2 = v - 128 */
-    smull        v20.4s, v6.4h, v1.4h[1] /* multiply by -11277 */
+    smull        v20.4s, v6.4h, v1.h[1]  /* multiply by -11277 */
-    smlal        v20.4s, v8.4h, v1.4h[2] /* multiply by -23401 */
+    smlal        v20.4s, v8.4h, v1.h[2]  /* multiply by -23401 */
-    smull2       v22.4s, v6.8h, v1.4h[1] /* multiply by -11277 */
+    smull2       v22.4s, v6.8h, v1.h[1]  /* multiply by -11277 */
-    smlal2       v22.4s, v8.8h, v1.4h[2] /* multiply by -23401 */
+    smlal2       v22.4s, v8.8h, v1.h[2]  /* multiply by -23401 */
-    smull        v24.4s, v8.4h, v1.4h[0] /* multiply by 22971 */
+    smull        v24.4s, v8.4h, v1.h[0]  /* multiply by 22971 */
-    smull2       v26.4s, v8.8h, v1.4h[0] /* multiply by 22971 */
+    smull2       v26.4s, v8.8h, v1.h[0]  /* multiply by 22971 */
-    smull        v28.4s, v6.4h, v1.4h[3] /* multiply by 29033 */
+    smull        v28.4s, v6.4h, v1.h[3]  /* multiply by 29033 */
-    smull2       v30.4s, v6.8h, v1.4h[3] /* multiply by 29033 */
+    smull2       v30.4s, v6.8h, v1.h[3]  /* multiply by 29033 */
 .endm
 
 .macro do_yuv_to_rgb_stage2
@@ -1656,18 +1657,18 @@ asm_function jsimd_idct_2x2_neon
     sqxtun       v1\g_offs\defsize, v20.8h
     ld1          {v0.8b}, [Y], 8
     sqxtun       v1\r_offs\defsize, v24.8h
-    prfm         PLDL1KEEP, [U, #64]
+    prfm         pldl1keep, [U, #64]
-    prfm         PLDL1KEEP, [V, #64]
+    prfm         pldl1keep, [V, #64]
-    prfm         PLDL1KEEP, [Y, #64]
+    prfm         pldl1keep, [Y, #64]
     sqxtun       v1\b_offs\defsize, v28.8h
     uaddw        v6.8h, v2.8h, v4.8b     /* v6.16b = u - 128 */
     uaddw        v8.8h, v2.8h, v5.8b     /* q2 = v - 128 */
-    smull        v20.4s, v6.4h, v1.4h[1] /* multiply by -11277 */
+    smull        v20.4s, v6.4h, v1.h[1]  /* multiply by -11277 */
-    smlal        v20.4s, v8.4h, v1.4h[2] /* multiply by -23401 */
+    smlal        v20.4s, v8.4h, v1.h[2]  /* multiply by -23401 */
-    smull2       v22.4s, v6.8h, v1.4h[1] /* multiply by -11277 */
+    smull2       v22.4s, v6.8h, v1.h[1]  /* multiply by -11277 */
-    smlal2       v22.4s, v8.8h, v1.4h[2] /* multiply by -23401 */
+    smlal2       v22.4s, v8.8h, v1.h[2]  /* multiply by -23401 */
-    smull        v24.4s, v8.4h, v1.4h[0] /* multiply by 22971 */
+    smull        v24.4s, v8.4h, v1.h[0]  /* multiply by 22971 */
-    smull2       v26.4s, v8.8h, v1.4h[0] /* multiply by 22971 */
+    smull2       v26.4s, v8.8h, v1.h[0]  /* multiply by 22971 */
 .else /**************************** rgb565 ***********************************/
     sqshlu       v21.8h, v20.8h, #8
     sqshlu       v25.8h, v24.8h, #8
@@ -1675,21 +1676,21 @@ asm_function jsimd_idct_2x2_neon
     uaddw        v6.8h, v2.8h, v4.8b     /* v6.16b = u - 128 */
     uaddw        v8.8h, v2.8h, v5.8b     /* q2 = v - 128 */
     ld1          {v0.8b}, [Y], 8
-    smull        v20.4s, v6.4h, v1.4h[1] /* multiply by -11277 */
+    smull        v20.4s, v6.4h, v1.h[1]  /* multiply by -11277 */
-    smlal        v20.4s, v8.4h, v1.4h[2] /* multiply by -23401 */
+    smlal        v20.4s, v8.4h, v1.h[2]  /* multiply by -23401 */
-    smull2       v22.4s, v6.8h, v1.4h[1] /* multiply by -11277 */
+    smull2       v22.4s, v6.8h, v1.h[1]  /* multiply by -11277 */
-    smlal2       v22.4s, v8.8h, v1.4h[2] /* multiply by -23401 */
+    smlal2       v22.4s, v8.8h, v1.h[2]  /* multiply by -23401 */
     sri          v25.8h, v21.8h, #5
-    smull        v24.4s, v8.4h, v1.4h[0] /* multiply by 22971 */
+    smull        v24.4s, v8.4h, v1.h[0]  /* multiply by 22971 */
-    smull2       v26.4s, v8.8h, v1.4h[0] /* multiply by 22971 */
+    smull2       v26.4s, v8.8h, v1.h[0]  /* multiply by 22971 */
-    prfm         PLDL1KEEP, [U, #64]
+    prfm         pldl1keep, [U, #64]
-    prfm         PLDL1KEEP, [V, #64]
+    prfm         pldl1keep, [V, #64]
-    prfm         PLDL1KEEP, [Y, #64]
+    prfm         pldl1keep, [Y, #64]
     sri          v25.8h, v29.8h, #11
 .endif
     do_store     \bpp, 8
-    smull        v28.4s, v6.4h, v1.4h[3] /* multiply by 29033 */
+    smull        v28.4s, v6.4h, v1.h[3]  /* multiply by 29033 */
-    smull2       v30.4s, v6.8h, v1.4h[3] /* multiply by 29033 */
+    smull2       v30.4s, v6.8h, v1.h[3]  /* multiply by 29033 */
 .endm
 
 .macro do_yuv_to_rgb
@@ -1702,7 +1703,7 @@ asm_function jsimd_idct_2x2_neon
  */
 
 .balign 16
-jsimd_ycc_\colorid\()_neon_consts:
+Ljsimd_ycc_\colorid\()_neon_consts:
     .short          0,      0,     0,      0
     .short          22971, -11277, -23401, 29033
     .short          -128,  -128,   -128,   -128
@@ -1717,7 +1718,7 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
 
     INPUT_BUF0      .req x5
     INPUT_BUF1      .req x6
-    INPUT_BUF2      .req INPUT_BUF
+    INPUT_BUF2      .req x1
 
     RGB             .req x7
     Y               .req x8
@@ -1728,16 +1729,16 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
     sub             sp, sp, 336
     str             x15, [sp], 16
     /* Load constants to d1, d2, d3 (v0.4h is just used for padding) */
-    adr             x15, jsimd_ycc_\colorid\()_neon_consts
+    adr             x15, Ljsimd_ycc_\colorid\()_neon_consts
     /* Save NEON registers */
-    st1             {v0.8b - v3.8b}, [sp], 32
+    st1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    st1             {v20.8b - v23.8b}, [sp], 32
+    st1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    st1             {v24.8b - v27.8b}, [sp], 32
+    st1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    st1             {v28.8b - v31.8b}, [sp], 32
+    st1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
     ld1             {v0.4h, v1.4h}, [x15], 16
     ld1             {v2.8h}, [x15]
 
@@ -1748,8 +1749,8 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
     stp             x8, x9, [sp], 16
     stp             x10, x30, [sp], 16
     ldr             INPUT_BUF0, [INPUT_BUF]
-    ldr             INPUT_BUF1, [INPUT_BUF, 8]
+    ldr             INPUT_BUF1, [INPUT_BUF, #8]
-    ldr             INPUT_BUF2, [INPUT_BUF, 16]
+    ldr             INPUT_BUF2, [INPUT_BUF, #16]
     .unreq          INPUT_BUF
 
     /* Initially set v10, v11.4h, v12.8b, d13 to 0xFF */
@@ -1758,7 +1759,7 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
 
     /* Outer loop over scanlines */
     cmp             NUM_ROWS, #1
-    blt             9f
+    b.lt            9f
 0:
     lsl             x16, INPUT_ROW, #3
     ldr             Y, [INPUT_BUF0, x16]
@@ -1770,60 +1771,60 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
 
     /* Inner loop over pixels */
     subs            N, N, #8
-    blt             3f
+    b.lt            3f
     do_load         8
     do_yuv_to_rgb_stage1
     subs            N, N, #8
-    blt             2f
+    b.lt            2f
 1:
     do_yuv_to_rgb_stage2_store_load_stage1
     subs            N, N, #8
-    bge             1b
+    b.ge            1b
 2:
     do_yuv_to_rgb_stage2
     do_store        \bpp, 8
     tst             N, #7
-    beq             8f
+    b.eq            8f
 3:
     tst             N, #4
-    beq             3f
+    b.eq            3f
     do_load         4
 3:
     tst             N, #2
-    beq             4f
+    b.eq            4f
     do_load         2
 4:
     tst             N, #1
-    beq             5f
+    b.eq            5f
     do_load         1
 5:
     do_yuv_to_rgb
     tst             N, #4
-    beq             6f
+    b.eq            6f
     do_store        \bpp, 4
 6:
     tst             N, #2
-    beq             7f
+    b.eq            7f
     do_store        \bpp, 2
 7:
     tst             N, #1
-    beq             8f
+    b.eq            8f
     do_store        \bpp, 1
 8:
     subs            NUM_ROWS, NUM_ROWS, #1
-    bgt             0b
+    b.gt            0b
 9:
     /* Restore all registers and return */
     sub             sp, sp, #336
     ldr             x15, [sp], 16
-    ld1             {v0.8b - v3.8b}, [sp], 32
+    ld1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    ld1             {v20.8b - v23.8b}, [sp], 32
+    ld1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    ld1             {v24.8b - v27.8b}, [sp], 32
+    ld1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    ld1             {v28.8b - v31.8b}, [sp], 32
+    ld1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
     /* pop             {r4, r5, r6, r7, r8, r9, r10, pc} */
     ldp             x4, x5, [sp], 16
     ldp             x6, x7, [sp], 16

simd/jsimd_powerpc.c

@@ -1,5 +1,5 @@
 /*
- * jsimd_powerpc64.c
+ * jsimd_powerpc.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * Copyright 2009-2011, 2014 D. R. Commander
@@ -42,12 +42,38 @@ init_simd (void)
 GLOBAL(int)
 jsimd_can_rgb_ycc (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
 GLOBAL(int)
 jsimd_can_rgb_gray (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
@@ -68,6 +94,37 @@ jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
                        JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
                        JDIMENSION output_row, int num_rows)
 {
+  void (*altivecfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+
+  switch(cinfo->in_color_space) {
+    case JCS_EXT_RGB:
+      altivecfct=jsimd_extrgb_ycc_convert_altivec;
+      break;
+    case JCS_EXT_RGBX:
+    case JCS_EXT_RGBA:
+      altivecfct=jsimd_extrgbx_ycc_convert_altivec;
+      break;
+    case JCS_EXT_BGR:
+      altivecfct=jsimd_extbgr_ycc_convert_altivec;
+      break;
+    case JCS_EXT_BGRX:
+    case JCS_EXT_BGRA:
+      altivecfct=jsimd_extbgrx_ycc_convert_altivec;
+      break;
+    case JCS_EXT_XBGR:
+    case JCS_EXT_ABGR:
+      altivecfct=jsimd_extxbgr_ycc_convert_altivec;
+      break;
+    case JCS_EXT_XRGB:
+    case JCS_EXT_ARGB:
+      altivecfct=jsimd_extxrgb_ycc_convert_altivec;
+      break;
+    default:
+      altivecfct=jsimd_rgb_ycc_convert_altivec;
+      break;
+  }
+
+  altivecfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
 }
 
 GLOBAL(void)
@@ -75,6 +132,37 @@ jsimd_rgb_gray_convert (j_compress_ptr cinfo,
                         JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
                         JDIMENSION output_row, int num_rows)
 {
+  void (*altivecfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+
+  switch(cinfo->in_color_space) {
+    case JCS_EXT_RGB:
+      altivecfct=jsimd_extrgb_gray_convert_altivec;
+      break;
+    case JCS_EXT_RGBX:
+    case JCS_EXT_RGBA:
+      altivecfct=jsimd_extrgbx_gray_convert_altivec;
+      break;
+    case JCS_EXT_BGR:
+      altivecfct=jsimd_extbgr_gray_convert_altivec;
+      break;
+    case JCS_EXT_BGRX:
+    case JCS_EXT_BGRA:
+      altivecfct=jsimd_extbgrx_gray_convert_altivec;
+      break;
+    case JCS_EXT_XBGR:
+    case JCS_EXT_ABGR:
+      altivecfct=jsimd_extxbgr_gray_convert_altivec;
+      break;
+    case JCS_EXT_XRGB:
+    case JCS_EXT_ARGB:
+      altivecfct=jsimd_extxrgb_gray_convert_altivec;
+      break;
+    default:
+      altivecfct=jsimd_rgb_gray_convert_altivec;
+      break;
+  }
+
+  altivecfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
 }
 
 GLOBAL(void)
@@ -202,6 +290,21 @@ jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo,
 GLOBAL(int)
 jsimd_can_convsamp (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
@@ -215,6 +318,7 @@ GLOBAL(void)
 jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col,
                 DCTELEM * workspace)
 {
+  jsimd_convsamp_altivec(sample_data, start_col, workspace);
 }
 
 GLOBAL(void)
@@ -226,6 +330,17 @@ jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
 GLOBAL(int)
 jsimd_can_fdct_islow (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
@@ -255,6 +370,7 @@ jsimd_can_fdct_float (void)
 GLOBAL(void)
 jsimd_fdct_islow (DCTELEM * data)
 {
+  jsimd_fdct_islow_altivec(data);
 }
 
 GLOBAL(void)
@@ -271,6 +387,19 @@ jsimd_fdct_float (FAST_FLOAT * data)
 GLOBAL(int)
 jsimd_can_quantize (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
@@ -284,6 +413,7 @@ GLOBAL(void)
 jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors,
                 DCTELEM * workspace)
 {
+  jsimd_quantize_altivec(coef_block, divisors, workspace);
 }
 
 GLOBAL(void)
@@ -321,12 +451,34 @@ jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
 GLOBAL(int)
 jsimd_can_idct_islow (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
 GLOBAL(int)
 jsimd_can_idct_ifast (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
@@ -341,6 +493,8 @@ jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                   JCOEFPTR coef_block, JSAMPARRAY output_buf,
                   JDIMENSION output_col)
 {
+  jsimd_idct_islow_altivec(compptr->dct_table, coef_block, output_buf,
+                           output_col);
 }
 
 GLOBAL(void)
@@ -348,6 +502,8 @@ jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                   JCOEFPTR coef_block, JSAMPARRAY output_buf,
                   JDIMENSION output_col)
 {
+  jsimd_idct_ifast_altivec(compptr->dct_table, coef_block, output_buf,
+                           output_col);
 }
 
 GLOBAL(void)

turbojpeg-mapfile

@@ -50,4 +50,7 @@ TURBOJPEG_1.4
 		tjDecompressToYUVPlanes;
 		tjEncodeYUV3;
 		tjEncodeYUVPlanes;
+		tjPlaneHeight;
+		tjPlaneSizeYUV;
+		tjPlaneWidth;
 } TURBOJPEG_1.2;

turbojpeg-mapfile.jni

@@ -76,6 +76,9 @@ TURBOJPEG_1.4
 		tjDecompressToYUVPlanes;
 		tjEncodeYUV3;
 		tjEncodeYUVPlanes;
+		tjPlaneHeight;
+		tjPlaneSizeYUV;
+		tjPlaneWidth;
 		Java_org_libjpegturbo_turbojpeg_TJ_bufSizeYUV__IIII;
 		Java_org_libjpegturbo_turbojpeg_TJCompressor_compressFromYUV___3_3B_3II_3III_3BII;
 		Java_org_libjpegturbo_turbojpeg_TJCompressor_encodeYUV___3BIIIIII_3_3B_3I_3III;

turbojpeg.c

@@ -320,6 +320,14 @@ static int setDecompDefaults(struct jpeg_decompress_struct *dinfo,
 static int getSubsamp(j_decompress_ptr dinfo)
 {
 	int retval=-1, i, k;
+
+	/* The sampling factors actually have no meaning with grayscale JPEG files,
+	   and in fact it's possible to generate grayscale JPEGs with sampling
+	   factors > 1 (even though those sampling factors are ignored by the
+	   decompressor.)  Thus, we need to treat grayscale as a special case. */
+	if(dinfo->num_components==1 && dinfo->jpeg_color_space==JCS_GRAYSCALE)
+		return TJSAMP_GRAY;
+
 	for(i=0; i<NUMSUBOPT; i++)
 	{
 		if(dinfo->num_components==pixelsize[i]