void aom_scaled_2d_ssse3(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
scaledconvolve2d(src, src_stride, dst, dst_stride, filters_x, x0_q4,
x_step_q4, filters_y, y0_q4, y_step_q4, w, h);
}
void aom_highbd_convolve8_add_src_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h, int bd) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
highbd_convolve_add_src(src, src_stride, dst, dst_stride, filters_x, x0_q4,
x_step_q4, filters_y, y0_q4, y_step_q4, w, h, bd);
}
void vp9_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {
const subpel_kernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
const subpel_kernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
convolve(src, src_stride, dst, dst_stride,
filters_x, x0_q4, x_step_q4,
filters_y, y0_q4, y_step_q4, w, h);
}
void vpx_highbd_convolve8_neon(const uint8_t *src8, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h, int bd) {
const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
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]);
const int intermediate_height =
(((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
/* Filter starting 3 lines back. The neon implementation will ignore the given
* height and filter a multiple of 4 lines. Since this goes in to the temp
* buffer which has lots of extra room and is subsequently discarded this is
* safe if somewhat less than ideal. */
vpx_highbd_convolve8_horiz_neon(CONVERT_TO_BYTEPTR(src - src_stride * 3),
src_stride, CONVERT_TO_BYTEPTR(temp), w,
filter_x, x_step_q4, filter_y, y_step_q4, w,
intermediate_height, bd);
/* Step into the temp buffer 3 lines to get the actual frame data */
vpx_highbd_convolve8_vert_neon(CONVERT_TO_BYTEPTR(temp + w * 3), w, dst,
dst_stride, filter_x, x_step_q4, filter_y,
y_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h, int bd) {
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
(void)filter_x;
(void)x_step_q4;
highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y,
y0_q4, y_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h, int bd) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
(void)filter_y;
(void)y_step_q4;
highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x,
x0_q4, x_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_avg_neon(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
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]);
const int intermediate_height =
(((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
/* This implementation has the same issues as above. In addition, we only want
* to average the values after both passes.
*/
vpx_highbd_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w,
filter_x, x_step_q4, filter_y, y_step_q4, w,
intermediate_height, bd);
vpx_highbd_convolve8_avg_vert_neon(temp + w * 3, w, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h, bd);
}
