Add compiler-independent alignment macro

Some variables and structures need memory alignment and the syntax
for declaring it differs when using MSVC compared to GCC and Clang.

This commit adds a compiler-independent alignment macro and uses it
to apply 16-byte alignment for constant pools and temporary buffers
used in Arm NEON SIMD code.

Bug: 922430
Change-Id: Ic2beb7bae88504ba7f3e86e33ef31bf388774403

10 files changed, 28 insertions, 12 deletions

diff --git a/README.chromium b/README.chromium
index 5e1b166..1ae3e1a 100644
--- a/README.chromium
+++ b/README.chromium
@@ -75,6 +75,7 @@ following changes which are not merged to upstream:
   - Add Arm NEON implementation of h2v2_downsample
   - Implement RGB->YCbCr using Arm NEON intrinsics
   - Add Arm NEON implementation of RGB->Grayscale
+  - Add compiler-independent alignment macro
 * Patches to enable running the upstream unit tests through gtest.
   The upstream unit tests are defined here under the section 'TESTS':
   https://github.com/libjpeg-turbo/libjpeg-turbo/blob/master/CMakeLists.txt
diff --git a/jconfigint.h b/jconfigint.h
index 10bcac7..974890d 100644
--- a/jconfigint.h
+++ b/jconfigint.h
@@ -52,3 +52,12 @@
 #define HAVEBITSCANFORWARD
 #endif
 #endif
+
+/* How to obtain memory alignment for structures and variables. */
+#if defined(_MSC_VER)
+#define ALIGN(ALIGNMENT) __declspec(align((ALIGNMENT)))
+#elif defined(__clang__) || defined(__GNUC__)
+#define ALIGN(ALIGNMENT) __attribute__((aligned(ALIGNMENT)))
+#else
+#error "Unknown compiler"
+#endif
diff --git a/simd/arm/arm/jccolext-neon.c b/simd/arm/arm/jccolext-neon.c
index 1e631b7..4f22e1f 100644
--- a/simd/arm/arm/jccolext-neon.c
+++ b/simd/arm/arm/jccolext-neon.c
@@ -78,7 +78,7 @@ void jsimd_rgb_ycc_convert_neon(JDIMENSION image_width,
       /* last (image_width % 8) columns of data are first memcopied to a */
       /* temporary buffer large enough to accommodate the vector load. */
       if (cols_remaining < 8) {
-        uint8_t __attribute__((aligned(8))) tmp_buf[8 * RGB_PIXELSIZE];
+        ALIGN(16) uint8_t tmp_buf[8 * RGB_PIXELSIZE];
         memcpy(tmp_buf, inptr, cols_remaining * RGB_PIXELSIZE);
         inptr = tmp_buf;
       }
diff --git a/simd/arm/arm64/jccolext-neon.c b/simd/arm/arm64/jccolext-neon.c
index 5c642fb..89f520a 100644
--- a/simd/arm/arm64/jccolext-neon.c
+++ b/simd/arm/arm64/jccolext-neon.c
@@ -161,7 +161,7 @@ void jsimd_rgb_ycc_convert_neon(JDIMENSION image_width,
       /* To prevent buffer overread by the vector load instructions, the */
       /* last (image_width % 16) columns of data are first memcopied to a */
       /* temporary buffer large enough to accommodate the vector load. */
-      uint8_t __attribute__((aligned(16))) tmp_buf[16 * RGB_PIXELSIZE];
+      ALIGN(16) uint8_t tmp_buf[16 * RGB_PIXELSIZE];
       memcpy(tmp_buf, inptr, cols_remaining * RGB_PIXELSIZE);
       inptr = tmp_buf;
 
@@ -252,7 +252,7 @@ void jsimd_rgb_ycc_convert_neon(JDIMENSION image_width,
       /* To prevent buffer overread by the vector load instructions, the */
       /* last (image_width % 8) columns of data are first memcopied to a */
       /* temporary buffer large enough to accommodate the vector load. */
-      uint8_t __attribute__((aligned(8))) tmp_buf[8 * RGB_PIXELSIZE];
+      ALIGN(16) uint8_t tmp_buf[8 * RGB_PIXELSIZE];
       memcpy(tmp_buf, inptr, cols_remaining * RGB_PIXELSIZE);
       inptr = tmp_buf;
 
diff --git a/simd/arm/common/jccolor-neon.c b/simd/arm/common/jccolor-neon.c
index 2ec1636..f87c8d9 100644
--- a/simd/arm/common/jccolor-neon.c
+++ b/simd/arm/common/jccolor-neon.c
@@ -21,6 +21,7 @@
  */
 
 #define JPEG_INTERNALS
+#include "../../../jconfigint.h"
 #include "../../../jinclude.h"
 #include "../../../jpeglib.h"
 #include "../../../jsimd.h"
@@ -41,11 +42,12 @@
 #define F_0_418  27439
 #define F_0_081  5329
 
-const static uint16_t jsimd_rgb_ycc_neon_consts[] = { F_0_298, F_0_587,
-                                                      F_0_113, F_0_168,
-                                                      F_0_331, F_0_500,
-                                                      F_0_418, F_0_081
-                                                    };
+ALIGN(16) static const uint16_t jsimd_rgb_ycc_neon_consts[] = {
+                                            F_0_298, F_0_587,
+                                            F_0_113, F_0_168,
+                                            F_0_331, F_0_500,
+                                            F_0_418, F_0_081
+                                          };
 
 /* Include inline routines for colorspace extensions. */
 
diff --git a/simd/arm/common/jcgray-neon.c b/simd/arm/common/jcgray-neon.c
index 4cba0ef..39d903f 100644
--- a/simd/arm/common/jcgray-neon.c
+++ b/simd/arm/common/jcgray-neon.c
@@ -21,6 +21,7 @@
  */
 
 #define JPEG_INTERNALS
+#include "../../../jconfigint.h"
 #include "../../../jinclude.h"
 #include "../../../jpeglib.h"
 #include "../../../jsimd.h"
diff --git a/simd/arm/common/jcgryext-neon.c b/simd/arm/common/jcgryext-neon.c
index 35ba0ac..69ea67f 100644
--- a/simd/arm/common/jcgryext-neon.c
+++ b/simd/arm/common/jcgryext-neon.c
@@ -58,7 +58,7 @@ void jsimd_rgb_gray_convert_neon(JDIMENSION image_width,
       /* last (image_width % 16) columns of data are first memcopied to a */
       /* temporary buffer large enough to accommodate the vector load. */
       if (cols_remaining < 16) {
-        uint8_t __attribute__((aligned(8))) tmp_buf[16 * RGB_PIXELSIZE];
+        ALIGN(16) uint8_t tmp_buf[16 * RGB_PIXELSIZE];
         memcpy(tmp_buf, inptr, cols_remaining * RGB_PIXELSIZE);
         inptr = tmp_buf;
       }
diff --git a/simd/arm/common/jcsample-neon.c b/simd/arm/common/jcsample-neon.c
index ff989dc..a5ddf16 100644
--- a/simd/arm/common/jcsample-neon.c
+++ b/simd/arm/common/jcsample-neon.c
@@ -21,6 +21,7 @@
  */
 
 #define JPEG_INTERNALS
+#include "../../../jconfigint.h"
 #include "../../../jinclude.h"
 #include "../../../jpeglib.h"
 #include "../../../jsimd.h"
@@ -31,7 +32,7 @@
 #include <arm_neon.h>
 
 
-static const uint8_t jsimd_h2_downsample_consts[] = {
+ALIGN(16) static const uint8_t jsimd_h2_downsample_consts[] = {
   0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,   /* Pad 0 */
   0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F,
   0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,   /* Pad 1 */
diff --git a/simd/arm/common/jidctint-neon.c b/simd/arm/common/jidctint-neon.c
index 11076a0..4ee9eb0 100644
--- a/simd/arm/common/jidctint-neon.c
+++ b/simd/arm/common/jidctint-neon.c
@@ -21,6 +21,7 @@
  */
 
 #define JPEG_INTERNALS
+#include "../../../jconfigint.h"
 #include "../../../jinclude.h"
 #include "../../../jpeglib.h"
 #include "../../../jsimd.h"
@@ -75,7 +76,7 @@
 #define F_2_053_MINUS_2_562  (F_2_053 - F_2_562)
 #define F_0_541_PLUS_0_765   (F_0_541 + F_0_765)
 
-__attribute__ ((aligned(8))) static int16_t jsimd_idct_islow_neon_consts[] = {
+ALIGN(16) static const int16_t jsimd_idct_islow_neon_consts[] = {
                               F_0_899,             F_0_541,
                               F_2_562,             F_0_298_MINUS_0_899,
                               F_1_501_MINUS_0_899, F_2_053_MINUS_2_562,
diff --git a/simd/arm/common/jidctred-neon.c b/simd/arm/common/jidctred-neon.c
index 7e95bf3..3c9393b 100644
--- a/simd/arm/common/jidctred-neon.c
+++ b/simd/arm/common/jidctred-neon.c
@@ -21,6 +21,7 @@
  */
 
 #define JPEG_INTERNALS
+#include "../../../jconfigint.h"
 #include "../../../jinclude.h"
 #include "../../../jpeglib.h"
 #include "../../../jsimd.h"
@@ -183,7 +184,7 @@ void jsimd_idct_2x2_neon(void *dct_table,
  *       exact compatibility with jpeg-6b.
  */
 
-__attribute__ ((aligned(8))) static int16_t jsimd_idct_4x4_neon_consts[] = {
+ALIGN(16) static const int16_t jsimd_idct_4x4_neon_consts[] = {
                                         F_1_847, -F_0_765, -F_0_211,  F_1_451,
                                        -F_2_172,  F_1_061, -F_0_509, -F_0_601,
                                         F_0_899,  F_2_562,        0,        0