shithub: libvpx

Info  •  Files  •  Log  •  Branches

remove DECLARE_ALIGNED from neon code

Unlike x86 neon only requires type alignment when loading into vectors.

Change-Id: I7bbbe4d51f78776e499ce137578d8c0effdbc02f

--- a/vpx_dsp/arm/highbd_vpx_convolve_neon.c

+++ b/vpx_dsp/arm/highbd_vpx_convolve_neon.c

@@ -20,7 +20,7 @@

                                int h, int bd) {

   const int y0_q4 = get_filter_offset(filter_y, get_filter_base(filter_y));

   // + 1 to make it divisible by 4

-  DECLARE_ALIGNED(16, uint16_t, temp[64 * 136]);

+  uint16_t temp[64 * 136];

   const int intermediate_height =

       (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

@@ -44,7 +44,7 @@

                                    int w, int h, int bd) {

   const int y0_q4 = get_filter_offset(filter_y, get_filter_base(filter_y));

   // + 1 to make it divisible by 4

-  DECLARE_ALIGNED(16, uint16_t, temp[64 * 136]);

+  uint16_t temp[64 * 136];

   const int intermediate_height =

       (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

--- a/vpx_dsp/arm/idct_neon.h

+++ b/vpx_dsp/arm/idct_neon.h

@@ -18,7 +18,7 @@

 #include "vpx_dsp/txfm_common.h"

 #include "vpx_dsp/vpx_dsp_common.h"

-DECLARE_ALIGNED(16, static const int16_t, kCospi[16]) = {

+static const int16_t kCospi[16] = {

   16384 /*  cospi_0_64  */, 15137 /*  cospi_8_64  */,

   11585 /*  cospi_16_64 */, 6270 /*  cospi_24_64 */,

   16069 /*  cospi_4_64  */, 13623 /*  cospi_12_64 */,

@@ -29,7 +29,7 @@

   12665 /*  cospi_14_64 */, -10394 /* -cospi_18_64 */

 };

-DECLARE_ALIGNED(16, static const int32_t, kCospi32[16]) = {

+static const int32_t kCospi32[16] = {

   16384 /*  cospi_0_64  */, 15137 /*  cospi_8_64  */,

   11585 /*  cospi_16_64 */, 6270 /*  cospi_24_64 */,

   16069 /*  cospi_4_64  */, 13623 /*  cospi_12_64 */,

--- a/vpx_dsp/arm/vpx_convolve_neon.c

+++ b/vpx_dsp/arm/vpx_convolve_neon.c

@@ -21,7 +21,7 @@

   /* Given our constraints: w <= 64, h <= 64, taps == 8 we can reduce the

    * maximum buffer size to 64 * 64 + 7 (+ 1 to make it divisible by 4).

    */

-  DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);

+  uint8_t temp[64 * 72];

   // Account for the vertical phase needing 3 lines prior and 4 lines post

   const int intermediate_height = h + 7;

@@ -47,7 +47,7 @@

                             const int16_t *filter_x, int x_step_q4,

                             const int16_t *filter_y, int y_step_q4, int w,

                             int h) {

-  DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);

+  uint8_t temp[64 * 72];

   const int intermediate_height = h + 7;

   assert(y_step_q4 == 16);