Commit e5aaac24 authored by Johann's avatar Johann
Browse files

clean up bilinear filter

make reference version of bilinear_filters short.
use reference versions of bilinear_filters and sub_pel_filters when
possible.

recognize that Width was being passed into
filter_block2d_bil_first_pass multiple times. ARM version had already
fixed this. propegate to C.

change references to src_pixels_per_line to src_pitch and standardize on
src/dst (instead of input/output).

recognize that first_pass is only run in the verticle and second_pass
only horizontal. ARM version had already fixed this. propegate to C

Change-Id: I292d376d239a9a7ca37ec2bf03cc0720606983e2
parent cc17629f
......@@ -15,19 +15,19 @@
AREA |.text|, CODE, READONLY ; name this block of code
;-------------------------------------
; r0 unsigned char *src_ptr,
; r1 unsigned short *output_ptr,
; r2 unsigned int src_pixels_per_line,
; r3 unsigned int output_height,
; stack unsigned int output_width,
; stack const short *vp8_filter
; r0 unsigned char *src_ptr,
; r1 unsigned short *dst_ptr,
; r2 unsigned int src_pitch,
; r3 unsigned int height,
; stack unsigned int width,
; stack const short *vp8_filter
;-------------------------------------
; The output is transposed stroed in output array to make it easy for second pass filtering.
|vp8_filter_block2d_bil_first_pass_armv6| PROC
stmdb sp!, {r4 - r11, lr}
ldr r11, [sp, #40] ; vp8_filter address
ldr r4, [sp, #36] ; output width
ldr r4, [sp, #36] ; width
mov r12, r3 ; outer-loop counter
sub r2, r2, r4 ; src increment for height loop
......@@ -38,10 +38,10 @@
ldr r5, [r11] ; load up filter coefficients
mov r3, r3, lsl #1 ; output_height*2
mov r3, r3, lsl #1 ; height*2
add r3, r3, #2 ; plus 2 to make output buffer 4-bit aligned since height is actually (height+1)
mov r11, r1 ; save output_ptr for each row
mov r11, r1 ; save dst_ptr for each row
cmp r5, #128 ; if filter coef = 128, then skip the filter
beq bil_null_1st_filter
......@@ -140,17 +140,17 @@
;---------------------------------
; r0 unsigned short *src_ptr,
; r1 unsigned char *output_ptr,
; r2 int output_pitch,
; r3 unsigned int output_height,
; stack unsigned int output_width,
; stack const short *vp8_filter
; r1 unsigned char *dst_ptr,
; r2 int dst_pitch,
; r3 unsigned int height,
; stack unsigned int width,
; stack const short *vp8_filter
;---------------------------------
|vp8_filter_block2d_bil_second_pass_armv6| PROC
stmdb sp!, {r4 - r11, lr}
ldr r11, [sp, #40] ; vp8_filter address
ldr r4, [sp, #36] ; output width
ldr r4, [sp, #36] ; width
ldr r5, [r11] ; load up filter coefficients
mov r12, r4 ; outer-loop counter = width, since we work on transposed data matrix
......
......@@ -10,128 +10,48 @@
#include <math.h>
#include "filter.h"
#include "subpixel.h"
#define BLOCK_HEIGHT_WIDTH 4
#define VP8_FILTER_WEIGHT 128
#define VP8_FILTER_SHIFT 7
static const short bilinear_filters[8][2] =
{
{ 128, 0 },
{ 112, 16 },
{ 96, 32 },
{ 80, 48 },
{ 64, 64 },
{ 48, 80 },
{ 32, 96 },
{ 16, 112 }
};
extern void vp8_filter_block2d_bil_first_pass_armv6
(
unsigned char *src_ptr,
unsigned short *output_ptr,
unsigned int src_pixels_per_line,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
unsigned char *src_ptr,
unsigned short *dst_ptr,
unsigned int src_pitch,
unsigned int height,
unsigned int width,
const short *vp8_filter
);
extern void vp8_filter_block2d_bil_second_pass_armv6
(
unsigned short *src_ptr,
unsigned char *output_ptr,
int output_pitch,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
unsigned char *dst_ptr,
int dst_pitch,
unsigned int height,
unsigned int width,
const short *vp8_filter
);
#if 0
void vp8_filter_block2d_bil_first_pass_6
(
unsigned char *src_ptr,
unsigned short *output_ptr,
unsigned int src_pixels_per_line,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
)
{
unsigned int i, j;
for ( i=0; i<output_height; i++ )
{
for ( j=0; j<output_width; j++ )
{
/* Apply bilinear filter */
output_ptr[j] = ( ( (int)src_ptr[0] * vp8_filter[0]) +
((int)src_ptr[1] * vp8_filter[1]) +
(VP8_FILTER_WEIGHT/2) ) >> VP8_FILTER_SHIFT;
src_ptr++;
}
/* Next row... */
src_ptr += src_pixels_per_line - output_width;
output_ptr += output_width;
}
}
void vp8_filter_block2d_bil_second_pass_6
(
unsigned short *src_ptr,
unsigned char *output_ptr,
int output_pitch,
unsigned int output_height,
unsigned int output_width,
const short *vp8_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] * vp8_filter[0]) +
((int)src_ptr[output_width] * vp8_filter[1]) +
(VP8_FILTER_WEIGHT/2);
output_ptr[j] = (unsigned int)(Temp >> VP8_FILTER_SHIFT);
src_ptr++;
}
/* Next row... */
/*src_ptr += src_pixels_per_line - output_width;*/
output_ptr += output_pitch;
}
}
#endif
void vp8_filter_block2d_bil_armv6
(
unsigned char *src_ptr,
unsigned char *output_ptr,
unsigned int src_pixels_per_line,
unsigned char *dst_ptr,
unsigned int src_pitch,
unsigned int dst_pitch,
const short *HFilter,
const short *VFilter,
const short *HFilter,
const short *VFilter,
int Width,
int Height
)
{
unsigned short FData[36*16]; /* Temp data bufffer used in filtering */
unsigned short FData[36*16]; /* Temp data buffer used in filtering */
/* First filter 1-D horizontally... */
/* pixel_step = 1; */
vp8_filter_block2d_bil_first_pass_armv6(src_ptr, FData, src_pixels_per_line, Height + 1, Width, HFilter);
vp8_filter_block2d_bil_first_pass_armv6(src_ptr, FData, src_pitch, Height + 1, Width, HFilter);
/* then 1-D vertically... */
vp8_filter_block2d_bil_second_pass_armv6(FData, output_ptr, dst_pitch, Height, Width, VFilter);
vp8_filter_block2d_bil_second_pass_armv6(FData, dst_ptr, dst_pitch, Height, Width, VFilter);
}
......@@ -148,8 +68,8 @@ void vp8_bilinear_predict4x4_armv6
const short *HFilter;
const short *VFilter;
HFilter = bilinear_filters[xoffset];
VFilter = bilinear_filters[yoffset];
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
vp8_filter_block2d_bil_armv6(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter, 4, 4);
}
......@@ -167,8 +87,8 @@ void vp8_bilinear_predict8x8_armv6
const short *HFilter;
const short *VFilter;
HFilter = bilinear_filters[xoffset];
VFilter = bilinear_filters[yoffset];
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
vp8_filter_block2d_bil_armv6(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter, 8, 8);
}
......@@ -186,8 +106,8 @@ void vp8_bilinear_predict8x4_armv6
const short *HFilter;
const short *VFilter;
HFilter = bilinear_filters[xoffset];
VFilter = bilinear_filters[yoffset];
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
vp8_filter_block2d_bil_armv6(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter, 8, 4);
}
......@@ -205,8 +125,8 @@ void vp8_bilinear_predict16x16_armv6
const short *HFilter;
const short *VFilter;
HFilter = bilinear_filters[xoffset];
VFilter = bilinear_filters[yoffset];
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
vp8_filter_block2d_bil_armv6(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter, 16, 16);
}
......@@ -11,26 +11,10 @@
#include "vpx_ports/config.h"
#include <math.h>
#include "filter.h"
#include "subpixel.h"
#include "vpx_ports/mem.h"
#define BLOCK_HEIGHT_WIDTH 4
#define VP8_FILTER_WEIGHT 128
#define VP8_FILTER_SHIFT 7
DECLARE_ALIGNED(16, static const short, sub_pel_filters[8][6]) =
{
{ 0, 0, 128, 0, 0, 0 }, /* note that 1/8 pel positions are just as per alpha -0.5 bicubic */
{ 0, -6, 123, 12, -1, 0 },
{ 2, -11, 108, 36, -8, 1 }, /* New 1/4 pel 6 tap filter */
{ 0, -9, 93, 50, -6, 0 },
{ 3, -16, 77, 77, -16, 3 }, /* New 1/2 pel 6 tap filter */
{ 0, -6, 50, 93, -9, 0 },
{ 1, -8, 36, 108, -11, 2 }, /* New 1/4 pel 6 tap filter */
{ 0, -1, 12, 123, -6, 0 },
};
extern void vp8_filter_block2d_first_pass_armv6
(
unsigned char *src_ptr,
......@@ -93,11 +77,11 @@ void vp8_sixtap_predict_armv6
{
const short *HFilter;
const short *VFilter;
DECLARE_ALIGNED_ARRAY(4, short, FData, 12*4); /* Temp data bufffer used in filtering */
DECLARE_ALIGNED_ARRAY(4, short, FData, 12*4); /* Temp data buffer used in filtering */
HFilter = sub_pel_filters[xoffset]; /* 6 tap */
VFilter = sub_pel_filters[yoffset]; /* 6 tap */
HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */
VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */
/* Vfilter is null. First pass only */
if (xoffset && !yoffset)
......@@ -129,47 +113,6 @@ void vp8_sixtap_predict_armv6
}
}
#if 0
void vp8_sixtap_predict8x4_armv6
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
const short *HFilter;
const short *VFilter;
DECLARE_ALIGNED_ARRAY(4, short, FData, 16*8); /* Temp data bufffer used in filtering */
HFilter = sub_pel_filters[xoffset]; /* 6 tap */
VFilter = sub_pel_filters[yoffset]; /* 6 tap */
/*if (xoffset && !yoffset)
{
vp8_filter_block2d_first_pass_only_armv6 ( src_ptr, dst_ptr, src_pixels_per_line, 8, dst_pitch, HFilter );
}*/
/* Hfilter is null. Second pass only */
/*else if (!xoffset && yoffset)
{
vp8_filter_block2d_second_pass_only_armv6 ( src_ptr, dst_ptr, src_pixels_per_line, 8, dst_pitch, VFilter );
}
else
{
if (yoffset & 0x1)
vp8_filter_block2d_first_pass_armv6 ( src_ptr-src_pixels_per_line, FData+1, src_pixels_per_line, 8, 7, HFilter );
else*/
vp8_filter_block2d_first_pass_armv6 ( src_ptr-(2*src_pixels_per_line), FData, src_pixels_per_line, 8, 9, HFilter );
vp8_filter_block2d_second_pass_armv6 ( FData+2, dst_ptr, dst_pitch, 4, 8, VFilter );
/*}*/
}
#endif
void vp8_sixtap_predict8x8_armv6
(
unsigned char *src_ptr,
......@@ -182,10 +125,10 @@ void vp8_sixtap_predict8x8_armv6
{
const short *HFilter;
const short *VFilter;
DECLARE_ALIGNED_ARRAY(4, short, FData, 16*8); /* Temp data bufffer used in filtering */
DECLARE_ALIGNED_ARRAY(4, short, FData, 16*8); /* Temp data buffer used in filtering */
HFilter = sub_pel_filters[xoffset]; /* 6 tap */
VFilter = sub_pel_filters[yoffset]; /* 6 tap */
HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */
VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */
if (xoffset && !yoffset)
{
......@@ -224,10 +167,10 @@ void vp8_sixtap_predict16x16_armv6
{
const short *HFilter;
const short *VFilter;
DECLARE_ALIGNED_ARRAY(4, short, FData, 24*16); /* Temp data bufffer used in filtering */
DECLARE_ALIGNED_ARRAY(4, short, FData, 24*16); /* Temp data buffer used in filtering */
HFilter = sub_pel_filters[xoffset]; /* 6 tap */
VFilter = sub_pel_filters[yoffset]; /* 6 tap */
HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */
VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */
if (xoffset && !yoffset)
{
......
......@@ -10,13 +10,10 @@
#include <stdlib.h>
#include "filter.h"
#include "vpx_ports/mem.h"
#define BLOCK_HEIGHT_WIDTH 4
#define VP8_FILTER_WEIGHT 128
#define VP8_FILTER_SHIFT 7
static const int bilinear_filters[8][2] =
DECLARE_ALIGNED(16, const short, vp8_bilinear_filters[8][2]) =
{
{ 128, 0 },
{ 112, 16 },
......@@ -28,8 +25,7 @@ static const int bilinear_filters[8][2] =
{ 16, 112 }
};
static const short sub_pel_filters[8][6] =
DECLARE_ALIGNED(16, const short, vp8_sub_pel_filters[8][6]) =
{
{ 0, 0, 128, 0, 0, 0 }, /* note that 1/8 pel positions are just as per alpha -0.5 bicubic */
......@@ -40,9 +36,6 @@ static const short sub_pel_filters[8][6] =
{ 0, -6, 50, 93, -9, 0 },
{ 1, -8, 36, 108, -11, 2 }, /* New 1/4 pel 6 tap filter */
{ 0, -1, 12, 123, -6, 0 },
};
void vp8_filter_block2d_first_pass
......@@ -146,7 +139,7 @@ void vp8_filter_block2d
const short *VFilter
)
{
int FData[9*4]; /* Temp data bufffer used in filtering */
int FData[9*4]; /* Temp data buffer used in filtering */
/* First filter 1-D horizontally... */
vp8_filter_block2d_first_pass(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 1, 9, 4, HFilter);
......@@ -195,8 +188,8 @@ void vp8_sixtap_predict_c
const short *HFilter;
const short *VFilter;
HFilter = sub_pel_filters[xoffset]; /* 6 tap */
VFilter = sub_pel_filters[yoffset]; /* 6 tap */
HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */
VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */
vp8_filter_block2d(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter);
}
......@@ -212,10 +205,10 @@ void vp8_sixtap_predict8x8_c
{
const short *HFilter;
const short *VFilter;
int FData[13*16]; /* Temp data bufffer used in filtering */
int FData[13*16]; /* Temp data buffer used in filtering */
HFilter = sub_pel_filters[xoffset]; /* 6 tap */
VFilter = sub_pel_filters[yoffset]; /* 6 tap */
HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */
VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */
/* First filter 1-D horizontally... */
vp8_filter_block2d_first_pass(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 1, 13, 8, HFilter);
......@@ -238,10 +231,10 @@ void vp8_sixtap_predict8x4_c
{
const short *HFilter;
const short *VFilter;
int FData[13*16]; /* Temp data bufffer used in filtering */
int FData[13*16]; /* Temp data buffer used in filtering */
HFilter = sub_pel_filters[xoffset]; /* 6 tap */
VFilter = sub_pel_filters[yoffset]; /* 6 tap */
HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */
VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */
/* First filter 1-D horizontally... */
vp8_filter_block2d_first_pass(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 1, 9, 8, HFilter);
......@@ -264,11 +257,11 @@ void vp8_sixtap_predict16x16_c
{
const short *HFilter;
const short *VFilter;
int FData[21*24]; /* Temp data bufffer used in filtering */
int FData[21*24]; /* Temp data buffer used in filtering */
HFilter = sub_pel_filters[xoffset]; /* 6 tap */
VFilter = sub_pel_filters[yoffset]; /* 6 tap */
HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */
VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */
/* First filter 1-D horizontally... */
vp8_filter_block2d_first_pass(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 1, 21, 16, HFilter);
......@@ -283,57 +276,50 @@ void vp8_sixtap_predict16x16_c
*
* ROUTINE : filter_block2d_bil_first_pass
*
* INPUTS : UINT8 *src_ptr : Pointer to source block.
* UINT32 src_pixels_per_line : Stride of input block.
* UINT32 pixel_step : Offset between filter input samples (see notes).
* UINT32 output_height : Input block height.
* UINT32 output_width : Input block width.
* INT32 *vp8_filter : Array of 2 bi-linear filter taps.
* INPUTS : UINT8 *src_ptr : Pointer to source block.
* UINT32 src_stride : Stride of source block.
* UINT32 height : Block height.
* UINT32 width : Block width.
* INT32 *vp8_filter : Array of 2 bi-linear filter taps.
*
* OUTPUTS : INT32 *output_ptr : Pointer to filtered block.
* OUTPUTS : INT32 *dst_ptr : Pointer to filtered block.
*
* RETURNS : void
*
* FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
* either horizontal or vertical direction to produce the
* filtered output block. Used to implement first-pass
* of 2-D separable filter.
* FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block
* in the horizontal direction to produce the filtered output
* block. Used to implement first-pass of 2-D separable filter.
*
* SPECIAL NOTES : Produces INT32 output to retain precision for next pass.
* Two filter taps should sum to VP8_FILTER_WEIGHT.
* pixel_step defines whether the filter is applied
* horizontally (pixel_step=1) or vertically (pixel_step=stride).
* It defines the offset required to move from one input
* to the next.
*
****************************************************************************/
void vp8_filter_block2d_bil_first_pass
(
unsigned char *src_ptr,
unsigned short *output_ptr,
unsigned int src_pixels_per_line,
int pixel_step,
unsigned int output_height,
unsigned int output_width,
const int *vp8_filter
unsigned char *src_ptr,
unsigned short *dst_ptr,
unsigned int src_stride,
unsigned int height,
unsigned int width,
const short *vp8_filter
)
{
unsigned int i, j;
for (i = 0; i < output_height; i++)
for (i = 0; i < height; i++)
{
for (j = 0; j < output_width; j++)
for (j = 0; j < width; j++)
{
/* Apply bilinear filter */
output_ptr[j] = (((int)src_ptr[0] * vp8_filter[0]) +
((int)src_ptr[pixel_step] * vp8_filter[1]) +
(VP8_FILTER_WEIGHT / 2)) >> VP8_FILTER_SHIFT;
dst_ptr[j] = (((int)src_ptr[0] * vp8_filter[0]) +
((int)src_ptr[1] * vp8_filter[1]) +
(VP8_FILTER_WEIGHT / 2)) >> VP8_FILTER_SHIFT;
src_ptr++;
}
/* Next row... */
src_ptr += src_pixels_per_line - output_width;
output_ptr += output_width;
src_ptr += src_stride - width;
dst_ptr += width;
}
}
......@@ -341,60 +327,51 @@ void vp8_filter_block2d_bil_first_pass
*
* ROUTINE : filter_block2d_bil_second_pass
*
* INPUTS : INT32 *src_ptr : Pointer to source block.
* UINT32 src_pixels_per_line : Stride of input block.
* UINT32 pixel_step : Offset between filter input samples (see notes).
* UINT32 output_height : Input block height.
* UINT32 output_width : Input block width.
* INT32 *vp8_filter : Array of 2 bi-linear filter taps.
* INPUTS : INT32 *src_ptr : Pointer to source block.
* UINT32 dst_pitch : Destination block pitch.
* UINT32 height : Block height.
* UINT32 width : Block width.
* INT32 *vp8_filter : Array of 2 bi-linear filter taps.
*
* OUTPUTS : UINT16 *output_ptr : Pointer to filtered block.
* OUTPUTS : UINT16 *dst_ptr : Pointer to filtered block.
*
* RETURNS : void
*
* FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
* either horizontal or vertical direction to produce the
* filtered output block. Used to implement second-pass
* of 2-D separable filter.
* FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block
* in the vertical direction to produce the filtered output
* block. Used to implement second-pass of 2-D separable filter.
*
* SPECIAL NOTES : Requires 32-bit input as produced by filter_block2d_bil_first_pass.
* Two filter taps should sum to VP8_FILTER_WEIGHT.
* pixel_step defines whether the filter is applied
* horizontally (pixel_step=1) or vertically (pixel_step=stride).
* It defines the offset required to move from one input
* to the next.
*
****************************************************************************/
void vp8_filter_block2d_bil_second_pass
(
unsigned short *src_ptr,
unsigned char *output_ptr,
int output_pitch,
unsigned int src_pixels_per_line,
unsigned int pixel_step,
unsigned int output_height,
unsigned int output_width,
const int *vp8_filter
unsigned char *dst_ptr,
int dst_pitch,
unsigned int height,
unsigned int width,
const short *vp8_filter
)
{
unsigned int i, j;
int Temp;
for (i = 0; i < output_height; i++)
for (i = 0; i < height; i++)
{
for (j = 0; j < output_width; j++)
for (j = 0; j < width; j++)
{
/* Apply filter */
Temp = ((int)src_ptr[0] * vp8_filter[0]) +
((int)src_ptr[pixel_step] * vp8_filter[1]) +
Temp = ((int)src_ptr[0] * vp8_filter[0]) +
((int)src_ptr[width] * vp8_filter[1]) +
(VP8_FILTER_WEIGHT / 2);
output_ptr[j] = (unsigned int)(Temp >> VP8_FILTER_SHIFT);
dst_ptr[j] = (unsigned int)(Temp >> VP8_FILTER_SHIFT);
src_ptr++;
}
/* Next row... */
src_ptr += src_pixels_per_line - output_width;