summaryrefslogtreecommitdiffstats
path: root/third_party/aom/aom_dsp/variance.c
diff options
context:
space:
mode:
authortrav90 <travawine@palemoon.org>2018-10-18 06:04:57 -0500
committertrav90 <travawine@palemoon.org>2018-10-18 06:04:57 -0500
commit7369c7d7a5eed32963d8af37658286617919f91c (patch)
tree5397ce7ee9bca1641118fdc3187bd9e2b24fdc9c /third_party/aom/aom_dsp/variance.c
parent77887af9c4ad1420bbdb33984af4f74b55ca59db (diff)
downloadUXP-7369c7d7a5eed32963d8af37658286617919f91c.tar
UXP-7369c7d7a5eed32963d8af37658286617919f91c.tar.gz
UXP-7369c7d7a5eed32963d8af37658286617919f91c.tar.lz
UXP-7369c7d7a5eed32963d8af37658286617919f91c.tar.xz
UXP-7369c7d7a5eed32963d8af37658286617919f91c.zip
Update aom to commit id f5bdeac22930ff4c6b219be49c843db35970b918
Diffstat (limited to 'third_party/aom/aom_dsp/variance.c')
-rw-r--r--third_party/aom/aom_dsp/variance.c218
1 files changed, 171 insertions, 47 deletions
diff --git a/third_party/aom/aom_dsp/variance.c b/third_party/aom/aom_dsp/variance.c
index 79677c92f..a4c3616e7 100644
--- a/third_party/aom/aom_dsp/variance.c
+++ b/third_party/aom/aom_dsp/variance.c
@@ -9,6 +9,8 @@
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
@@ -20,6 +22,9 @@
#include "aom_dsp/aom_filter.h"
#include "aom_dsp/blend.h"
+#include "./av1_rtcd.h"
+#include "av1/common/filter.h"
+
uint32_t aom_get4x4sse_cs_c(const uint8_t *a, int a_stride, const uint8_t *b,
int b_stride) {
int distortion = 0;
@@ -246,6 +251,13 @@ VARIANCES(4, 2)
VARIANCES(2, 4)
VARIANCES(2, 2)
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+VARIANCES(4, 16)
+VARIANCES(16, 4)
+VARIANCES(8, 32)
+VARIANCES(32, 8)
+#endif
+
GET_VAR(16, 16)
GET_VAR(8, 8)
@@ -271,33 +283,66 @@ void aom_comp_avg_pred_c(uint8_t *comp_pred, const uint8_t *pred, int width,
// Get pred block from up-sampled reference.
void aom_upsampled_pred_c(uint8_t *comp_pred, int width, int height,
- const uint8_t *ref, int ref_stride) {
- int i, j, k;
- int stride = ref_stride << 3;
-
- for (i = 0; i < height; i++) {
- for (j = 0, k = 0; j < width; j++, k += 8) {
- comp_pred[j] = ref[k];
+ int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
+ int ref_stride) {
+ if (!subpel_x_q3 && !subpel_y_q3) {
+ int i;
+ for (i = 0; i < height; i++) {
+ memcpy(comp_pred, ref, width * sizeof(*comp_pred));
+ comp_pred += width;
+ ref += ref_stride;
+ }
+ } else {
+ InterpFilterParams filter;
+ filter = av1_get_interp_filter_params(EIGHTTAP_REGULAR);
+ if (!subpel_y_q3) {
+ const int16_t *kernel;
+ kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+ /*Directly call C version to allow this to work for small (2x2) sizes.*/
+ aom_convolve8_horiz_c(ref, ref_stride, comp_pred, width, kernel, 16, NULL,
+ -1, width, height);
+ } else if (!subpel_x_q3) {
+ const int16_t *kernel;
+ kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+ /*Directly call C version to allow this to work for small (2x2) sizes.*/
+ aom_convolve8_vert_c(ref, ref_stride, comp_pred, width, NULL, -1, kernel,
+ 16, width, height);
+ } else {
+ DECLARE_ALIGNED(16, uint8_t,
+ temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
+ const int16_t *kernel_x;
+ const int16_t *kernel_y;
+ int intermediate_height;
+ kernel_x = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+ kernel_y = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+ intermediate_height =
+ (((height - 1) * 8 + subpel_y_q3) >> 3) + filter.taps;
+ assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
+ /*Directly call C versions to allow this to work for small (2x2) sizes.*/
+ aom_convolve8_horiz_c(ref - ref_stride * ((filter.taps >> 1) - 1),
+ ref_stride, temp, MAX_SB_SIZE, kernel_x, 16, NULL,
+ -1, width, intermediate_height);
+ aom_convolve8_vert_c(temp + MAX_SB_SIZE * ((filter.taps >> 1) - 1),
+ MAX_SB_SIZE, comp_pred, width, NULL, -1, kernel_y,
+ 16, width, height);
}
- comp_pred += width;
- ref += stride;
}
}
void aom_comp_avg_upsampled_pred_c(uint8_t *comp_pred, const uint8_t *pred,
- int width, int height, const uint8_t *ref,
+ int width, int height, int subpel_x_q3,
+ int subpel_y_q3, const uint8_t *ref,
int ref_stride) {
int i, j;
- int stride = ref_stride << 3;
+ aom_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3, ref,
+ ref_stride);
for (i = 0; i < height; i++) {
for (j = 0; j < width; j++) {
- const int tmp = ref[(j << 3)] + pred[j];
- comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
+ comp_pred[j] = ROUND_POWER_OF_TWO(comp_pred[j] + pred[j], 1);
}
comp_pred += width;
pred += width;
- ref += stride;
}
}
@@ -611,6 +656,13 @@ HIGHBD_VARIANCES(4, 2)
HIGHBD_VARIANCES(2, 4)
HIGHBD_VARIANCES(2, 2)
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+HIGHBD_VARIANCES(4, 16)
+HIGHBD_VARIANCES(16, 4)
+HIGHBD_VARIANCES(8, 32)
+HIGHBD_VARIANCES(32, 8)
+#endif
+
HIGHBD_GET_VAR(8)
HIGHBD_GET_VAR(16)
@@ -637,37 +689,74 @@ void aom_highbd_comp_avg_pred_c(uint16_t *comp_pred, const uint8_t *pred8,
}
void aom_highbd_upsampled_pred_c(uint16_t *comp_pred, int width, int height,
- const uint8_t *ref8, int ref_stride) {
- int i, j;
- int stride = ref_stride << 3;
-
- uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
- for (i = 0; i < height; ++i) {
- for (j = 0; j < width; ++j) {
- comp_pred[j] = ref[(j << 3)];
+ int subpel_x_q3, int subpel_y_q3,
+ const uint8_t *ref8, int ref_stride, int bd) {
+ if (!subpel_x_q3 && !subpel_y_q3) {
+ const uint16_t *ref;
+ int i;
+ ref = CONVERT_TO_SHORTPTR(ref8);
+ for (i = 0; i < height; i++) {
+ memcpy(comp_pred, ref, width * sizeof(*comp_pred));
+ comp_pred += width;
+ ref += ref_stride;
+ }
+ } else {
+ InterpFilterParams filter;
+ filter = av1_get_interp_filter_params(EIGHTTAP_REGULAR);
+ if (!subpel_y_q3) {
+ const int16_t *kernel;
+ kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+ /*Directly call C version to allow this to work for small (2x2) sizes.*/
+ aom_highbd_convolve8_horiz_c(ref8, ref_stride,
+ CONVERT_TO_BYTEPTR(comp_pred), width, kernel,
+ 16, NULL, -1, width, height, bd);
+ } else if (!subpel_x_q3) {
+ const int16_t *kernel;
+ kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+ /*Directly call C version to allow this to work for small (2x2) sizes.*/
+ aom_highbd_convolve8_vert_c(ref8, ref_stride,
+ CONVERT_TO_BYTEPTR(comp_pred), width, NULL,
+ -1, kernel, 16, width, height, bd);
+ } else {
+ DECLARE_ALIGNED(16, uint16_t,
+ temp[((MAX_SB_SIZE + 16) + 16) * MAX_SB_SIZE]);
+ const int16_t *kernel_x;
+ const int16_t *kernel_y;
+ int intermediate_height;
+ kernel_x = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+ kernel_y = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+ intermediate_height =
+ (((height - 1) * 8 + subpel_y_q3) >> 3) + filter.taps;
+ assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
+ /*Directly call C versions to allow this to work for small (2x2) sizes.*/
+ aom_highbd_convolve8_horiz_c(ref8 - ref_stride * ((filter.taps >> 1) - 1),
+ ref_stride, CONVERT_TO_BYTEPTR(temp),
+ MAX_SB_SIZE, kernel_x, 16, NULL, -1, width,
+ intermediate_height, bd);
+ aom_highbd_convolve8_vert_c(
+ CONVERT_TO_BYTEPTR(temp + MAX_SB_SIZE * ((filter.taps >> 1) - 1)),
+ MAX_SB_SIZE, CONVERT_TO_BYTEPTR(comp_pred), width, NULL, -1, kernel_y,
+ 16, width, height, bd);
}
- comp_pred += width;
- ref += stride;
}
}
void aom_highbd_comp_avg_upsampled_pred_c(uint16_t *comp_pred,
const uint8_t *pred8, int width,
- int height, const uint8_t *ref8,
- int ref_stride) {
+ int height, int subpel_x_q3,
+ int subpel_y_q3, const uint8_t *ref8,
+ int ref_stride, int bd) {
int i, j;
- int stride = ref_stride << 3;
- uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
- uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+ const uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
+ aom_highbd_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3,
+ ref8, ref_stride, bd);
for (i = 0; i < height; ++i) {
for (j = 0; j < width; ++j) {
- const int tmp = pred[j] + ref[(j << 3)];
- comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
+ comp_pred[j] = ROUND_POWER_OF_TWO(pred[j] + comp_pred[j], 1);
}
comp_pred += width;
pred += width;
- ref += stride;
}
}
#endif // CONFIG_HIGHBITDEPTH
@@ -694,22 +783,23 @@ void aom_comp_mask_pred_c(uint8_t *comp_pred, const uint8_t *pred, int width,
}
void aom_comp_mask_upsampled_pred_c(uint8_t *comp_pred, const uint8_t *pred,
- int width, int height, const uint8_t *ref,
+ int width, int height, int subpel_x_q3,
+ int subpel_y_q3, const uint8_t *ref,
int ref_stride, const uint8_t *mask,
int mask_stride, int invert_mask) {
int i, j;
- int stride = ref_stride << 3;
+ aom_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3, ref,
+ ref_stride);
for (i = 0; i < height; i++) {
for (j = 0; j < width; j++) {
if (!invert_mask)
- comp_pred[j] = AOM_BLEND_A64(mask[j], ref[(j << 3)], pred[j]);
+ comp_pred[j] = AOM_BLEND_A64(mask[j], comp_pred[j], pred[j]);
else
- comp_pred[j] = AOM_BLEND_A64(mask[j], pred[j], ref[(j << 3)]);
+ comp_pred[j] = AOM_BLEND_A64(mask[j], pred[j], comp_pred[j]);
}
comp_pred += width;
pred += width;
- ref += stride;
mask += mask_stride;
}
}
@@ -753,6 +843,13 @@ MASK_SUBPIX_VAR(128, 64)
MASK_SUBPIX_VAR(128, 128)
#endif // CONFIG_EXT_PARTITION
+#if CONFIG_EXT_PARTITION_TYPES
+MASK_SUBPIX_VAR(4, 16)
+MASK_SUBPIX_VAR(16, 4)
+MASK_SUBPIX_VAR(8, 32)
+MASK_SUBPIX_VAR(32, 8)
+#endif
+
#if CONFIG_HIGHBITDEPTH
void aom_highbd_comp_mask_pred_c(uint16_t *comp_pred, const uint8_t *pred8,
int width, int height, const uint8_t *ref8,
@@ -775,26 +872,24 @@ void aom_highbd_comp_mask_pred_c(uint16_t *comp_pred, const uint8_t *pred8,
}
}
-void aom_highbd_comp_mask_upsampled_pred_c(uint16_t *comp_pred,
- const uint8_t *pred8, int width,
- int height, const uint8_t *ref8,
- int ref_stride, const uint8_t *mask,
- int mask_stride, int invert_mask) {
+void aom_highbd_comp_mask_upsampled_pred_c(
+ uint16_t *comp_pred, const uint8_t *pred8, int width, int height,
+ int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8, int ref_stride,
+ const uint8_t *mask, int mask_stride, int invert_mask, int bd) {
int i, j;
- int stride = ref_stride << 3;
uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
- uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+ aom_highbd_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3,
+ ref8, ref_stride, bd);
for (i = 0; i < height; ++i) {
for (j = 0; j < width; ++j) {
if (!invert_mask)
- comp_pred[j] = AOM_BLEND_A64(mask[j], ref[j << 3], pred[j]);
+ comp_pred[j] = AOM_BLEND_A64(mask[j], comp_pred[j], pred[j]);
else
- comp_pred[j] = AOM_BLEND_A64(mask[j], pred[j], ref[j << 3]);
+ comp_pred[j] = AOM_BLEND_A64(mask[j], pred[j], comp_pred[j]);
}
comp_pred += width;
pred += width;
- ref += stride;
mask += mask_stride;
}
}
@@ -884,6 +979,13 @@ HIGHBD_MASK_SUBPIX_VAR(64, 128)
HIGHBD_MASK_SUBPIX_VAR(128, 64)
HIGHBD_MASK_SUBPIX_VAR(128, 128)
#endif // CONFIG_EXT_PARTITION
+
+#if CONFIG_EXT_PARTITION_TYPES
+HIGHBD_MASK_SUBPIX_VAR(4, 16)
+HIGHBD_MASK_SUBPIX_VAR(16, 4)
+HIGHBD_MASK_SUBPIX_VAR(8, 32)
+HIGHBD_MASK_SUBPIX_VAR(32, 8)
+#endif
#endif // CONFIG_HIGHBITDEPTH
#endif // CONFIG_AV1 && CONFIG_EXT_INTER
@@ -983,6 +1085,17 @@ OBMC_VAR(128, 128)
OBMC_SUBPIX_VAR(128, 128)
#endif // CONFIG_EXT_PARTITION
+#if CONFIG_EXT_PARTITION_TYPES
+OBMC_VAR(4, 16)
+OBMC_SUBPIX_VAR(4, 16)
+OBMC_VAR(16, 4)
+OBMC_SUBPIX_VAR(16, 4)
+OBMC_VAR(8, 32)
+OBMC_SUBPIX_VAR(8, 32)
+OBMC_VAR(32, 8)
+OBMC_SUBPIX_VAR(32, 8)
+#endif
+
#if CONFIG_HIGHBITDEPTH
static INLINE void highbd_obmc_variance64(const uint8_t *pre8, int pre_stride,
const int32_t *wsrc,
@@ -1164,5 +1277,16 @@ HIGHBD_OBMC_SUBPIX_VAR(128, 64)
HIGHBD_OBMC_VAR(128, 128)
HIGHBD_OBMC_SUBPIX_VAR(128, 128)
#endif // CONFIG_EXT_PARTITION
+
+#if CONFIG_EXT_PARTITION_TYPES
+HIGHBD_OBMC_VAR(4, 16)
+HIGHBD_OBMC_SUBPIX_VAR(4, 16)
+HIGHBD_OBMC_VAR(16, 4)
+HIGHBD_OBMC_SUBPIX_VAR(16, 4)
+HIGHBD_OBMC_VAR(8, 32)
+HIGHBD_OBMC_SUBPIX_VAR(8, 32)
+HIGHBD_OBMC_VAR(32, 8)
+HIGHBD_OBMC_SUBPIX_VAR(32, 8)
+#endif
#endif // CONFIG_HIGHBITDEPTH
#endif // CONFIG_AV1 && CONFIG_MOTION_VAR