ref: 2612b99cc7a64d5377f59de4775a308e6cdd5706
parent: c4048dbdd39dbd4763852a6d0a27f183677c4225
author: Dmitry Kovalev <[email protected]>
date: Mon Aug 19 20:42:25 EDT 2013
Adding VP9_FILTER_BITS constant. Removing VP9_FILTER_WEIGHT, VP9_FILTER_SHIFT, BLOCK_WIDTH_HEIGHT constants. Using ROUND_POWER_OF_TWO for rounding. Change-Id: I2e8d6858dcd600a87096138209731137d7decc24
--- a/vp9/common/vp9_convolve.c
+++ b/vp9/common/vp9_convolve.c
@@ -17,9 +17,6 @@
#include "vpx/vpx_integer.h"
#include "vpx_ports/mem.h"
-#define VP9_FILTER_WEIGHT 128
-#define VP9_FILTER_SHIFT 7
-
/* Assume a bank of 16 filters to choose from. There are two implementations
* for filter wrapping behavior, since we want to be able to pick which filter
* to start with. We could either:
@@ -43,7 +40,7 @@
const int16_t *filter_x0, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h, int taps) {- int x, y, k, sum;
+ int x, y, k;
const int16_t *filter_x_base = filter_x0;
#if ALIGN_FILTERS_256
@@ -64,13 +61,13 @@
for (x = 0; x < w; ++x) {/* Per-pixel src offset */
int src_x = (x_q4 - x0_q4) >> 4;
+ int sum = 0;
- for (sum = 0, k = 0; k < taps; ++k) {+ for (k = 0; k < taps; ++k)
sum += src[src_x + k] * filter_x[k];
- }
- sum += (VP9_FILTER_WEIGHT >> 1);
- dst[x] = clip_pixel(sum >> VP9_FILTER_SHIFT);
+ dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, VP9_FILTER_BITS));
+
/* Adjust source and filter to use for the next pixel */
x_q4 += x_step_q4;
filter_x = filter_x_base + (x_q4 & 0xf) * taps;
@@ -85,7 +82,7 @@
const int16_t *filter_x0, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h, int taps) {- int x, y, k, sum;
+ int x, y, k;
const int16_t *filter_x_base = filter_x0;
#if ALIGN_FILTERS_256
@@ -106,13 +103,14 @@
for (x = 0; x < w; ++x) {/* Per-pixel src offset */
int src_x = (x_q4 - x0_q4) >> 4;
+ int sum = 0;
- for (sum = 0, k = 0; k < taps; ++k) {+ for (k = 0; k < taps; ++k)
sum += src[src_x + k] * filter_x[k];
- }
- sum += (VP9_FILTER_WEIGHT >> 1);
- dst[x] = (dst[x] + clip_pixel(sum >> VP9_FILTER_SHIFT) + 1) >> 1;
+ dst[x] = ROUND_POWER_OF_TWO(dst[x] +
+ clip_pixel(ROUND_POWER_OF_TWO(sum, VP9_FILTER_BITS)), 1);
+
/* Adjust source and filter to use for the next pixel */
x_q4 += x_step_q4;
filter_x = filter_x_base + (x_q4 & 0xf) * taps;
@@ -127,7 +125,7 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y0, int y_step_q4,
int w, int h, int taps) {- int x, y, k, sum;
+ int x, y, k;
const int16_t *filter_y_base = filter_y0;
@@ -148,13 +146,14 @@
for (y = 0; y < h; ++y) {/* Per-pixel src offset */
int src_y = (y_q4 - y0_q4) >> 4;
+ int sum = 0;
- for (sum = 0, k = 0; k < taps; ++k) {+ for (k = 0; k < taps; ++k)
sum += src[(src_y + k) * src_stride] * filter_y[k];
- }
- sum += (VP9_FILTER_WEIGHT >> 1);
- dst[y * dst_stride] = clip_pixel(sum >> VP9_FILTER_SHIFT);
+ dst[y * dst_stride] =
+ clip_pixel(ROUND_POWER_OF_TWO(sum, VP9_FILTER_BITS));
+
/* Adjust source and filter to use for the next pixel */
y_q4 += y_step_q4;
filter_y = filter_y_base + (y_q4 & 0xf) * taps;
@@ -169,7 +168,7 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y0, int y_step_q4,
int w, int h, int taps) {- int x, y, k, sum;
+ int x, y, k;
const int16_t *filter_y_base = filter_y0;
@@ -190,13 +189,13 @@
for (y = 0; y < h; ++y) {/* Per-pixel src offset */
int src_y = (y_q4 - y0_q4) >> 4;
+ int sum = 0;
- for (sum = 0, k = 0; k < taps; ++k) {+ for (k = 0; k < taps; ++k)
sum += src[(src_y + k) * src_stride] * filter_y[k];
- }
- sum += (VP9_FILTER_WEIGHT >> 1);
- dst[y * dst_stride] =
- (dst[y * dst_stride] + clip_pixel(sum >> VP9_FILTER_SHIFT) + 1) >> 1;
+
+ dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] +
+ clip_pixel(ROUND_POWER_OF_TWO(sum, VP9_FILTER_BITS)), 1);
/* Adjust source and filter to use for the next pixel */
y_q4 += y_step_q4;
--- a/vp9/common/vp9_convolve.h
+++ b/vp9/common/vp9_convolve.h
@@ -13,6 +13,8 @@
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
+#define VP9_FILTER_BITS 7
+
typedef void (*convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
--- a/vp9/common/vp9_filter.h
+++ b/vp9/common/vp9_filter.h
@@ -12,12 +12,7 @@
#define VP9_COMMON_VP9_FILTER_H_
#include "vpx_config.h"
-#include "vpx_scale/yv12config.h"
#include "vpx/vpx_integer.h"
-
-#define BLOCK_HEIGHT_WIDTH 4
-#define VP9_FILTER_WEIGHT 128
-#define VP9_FILTER_SHIFT 7
#define SUBPEL_BITS 4
#define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1)
--- a/vp9/common/vp9_subpelvar.h
+++ b/vp9/common/vp9_subpelvar.h
@@ -11,7 +11,8 @@
#ifndef VP9_COMMON_VP9_SUBPELVAR_H_
#define VP9_COMMON_VP9_SUBPELVAR_H_
-#include "vp9/common/vp9_filter.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_convolve.h"
static void variance(const uint8_t *src_ptr,
int source_stride,
@@ -78,10 +79,10 @@
for (i = 0; i < output_height; i++) { for (j = 0; j < output_width; j++) {- // Apply bilinear filter
- output_ptr[j] = (((int)src_ptr[0] * vp9_filter[0]) +
- ((int)src_ptr[pixel_step] * vp9_filter[1]) +
- (VP9_FILTER_WEIGHT / 2)) >> VP9_FILTER_SHIFT;
+ output_ptr[j] = ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
+ (int)src_ptr[pixel_step] * vp9_filter[1],
+ VP9_FILTER_BITS);
+
src_ptr++;
}
@@ -127,20 +128,16 @@
unsigned int output_width,
const int16_t *vp9_filter) {unsigned int i, j;
- int Temp;
for (i = 0; i < output_height; i++) { for (j = 0; j < output_width; j++) {- // Apply filter
- Temp = ((int)src_ptr[0] * vp9_filter[0]) +
- ((int)src_ptr[pixel_step] * vp9_filter[1]) +
- (VP9_FILTER_WEIGHT / 2);
- output_ptr[j] = (unsigned int)(Temp >> VP9_FILTER_SHIFT);
+ output_ptr[j] = ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
+ (int)src_ptr[pixel_step] * vp9_filter[1],
+ VP9_FILTER_BITS);
src_ptr++;
}
- // Next row...
- src_ptr += src_pixels_per_line - output_width;
+ src_ptr += src_pixels_per_line - output_width;
output_ptr += output_width;
}
}