Commit 6a82f0d7 authored by Johann's avatar Johann
Browse files

Move sub pixel variance to vpx_dsp

Change-Id: I66bf6720c396c89aa2d1fd26d5d52bf5d5e3dff1
parent 155b9416
This diff is collapsed.
This diff is collapsed.
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "vpx_config.h"
#include "./vp8_rtcd.h"
#include "./vpx_dsp_rtcd.h"
#include "vp8/common/variance.h"
#include "vp8/common/filter.h"
// TODO(johannkoenig): Move this to vpx_dsp or vp8/encoder
#if CONFIG_VP8_ENCODER
#if HAVE_MEDIA
#include "vp8/common/arm/bilinearfilter_arm.h"
unsigned int vp8_sub_pixel_variance8x8_armv6
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
)
{
unsigned short first_pass[10*8];
unsigned char second_pass[8*8];
const short *HFilter, *VFilter;
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
vp8_filter_block2d_bil_first_pass_armv6(src_ptr, first_pass,
src_pixels_per_line,
9, 8, HFilter);
vp8_filter_block2d_bil_second_pass_armv6(first_pass, second_pass,
8, 8, 8, VFilter);
return vpx_variance8x8_media(second_pass, 8, dst_ptr,
dst_pixels_per_line, sse);
}
unsigned int vp8_sub_pixel_variance16x16_armv6
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
)
{
unsigned short first_pass[36*16];
unsigned char second_pass[20*16];
const short *HFilter, *VFilter;
unsigned int var;
if (xoffset == 4 && yoffset == 0)
{
var = vp8_variance_halfpixvar16x16_h_armv6(src_ptr, src_pixels_per_line,
dst_ptr, dst_pixels_per_line, sse);
}
else if (xoffset == 0 && yoffset == 4)
{
var = vp8_variance_halfpixvar16x16_v_armv6(src_ptr, src_pixels_per_line,
dst_ptr, dst_pixels_per_line, sse);
}
else if (xoffset == 4 && yoffset == 4)
{
var = vp8_variance_halfpixvar16x16_hv_armv6(src_ptr, src_pixels_per_line,
dst_ptr, dst_pixels_per_line, sse);
}
else
{
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
vp8_filter_block2d_bil_first_pass_armv6(src_ptr, first_pass,
src_pixels_per_line,
17, 16, HFilter);
vp8_filter_block2d_bil_second_pass_armv6(first_pass, second_pass,
16, 16, 16, VFilter);
var = vpx_variance16x16_media(second_pass, 16, dst_ptr,
dst_pixels_per_line, sse);
}
return var;
}
#endif // HAVE_MEDIA
#if HAVE_NEON
extern unsigned int vp8_sub_pixel_variance16x16_neon_func
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
);
unsigned int vp8_sub_pixel_variance16x16_neon
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
)
{
if (xoffset == 4 && yoffset == 0)
return vp8_variance_halfpixvar16x16_h_neon(src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
else if (xoffset == 0 && yoffset == 4)
return vp8_variance_halfpixvar16x16_v_neon(src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
else if (xoffset == 4 && yoffset == 4)
return vp8_variance_halfpixvar16x16_hv_neon(src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
else
return vp8_sub_pixel_variance16x16_neon_func(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pixels_per_line, sse);
}
#endif // HAVE_NEON
#endif // CONFIG_VP8_ENCODER
......@@ -20,7 +20,7 @@
#include "./vp8_rtcd.h"
#include "./vpx_dsp_rtcd.h"
#include "vp8/common/postproc.h"
#include "vp8/common/variance.h"
#include "vpx_dsp/variance.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_scale/yv12config.h"
......
......@@ -237,47 +237,6 @@ add_proto qw/void vp8_bilinear_predict4x4/, "unsigned char *src, int src_pitch,
specialize qw/vp8_bilinear_predict4x4 mmx media/;
$vp8_bilinear_predict4x4_media=vp8_bilinear_predict4x4_armv6;
#
# Sub-pixel Variance
#
add_proto qw/unsigned int vp8_sub_pixel_variance4x4/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse";
specialize qw/vp8_sub_pixel_variance4x4 mmx sse2/;
$vp8_sub_pixel_variance4x4_sse2=vp8_sub_pixel_variance4x4_wmt;
add_proto qw/unsigned int vp8_sub_pixel_variance8x8/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse";
specialize qw/vp8_sub_pixel_variance8x8 mmx sse2 media/;
$vp8_sub_pixel_variance8x8_sse2=vp8_sub_pixel_variance8x8_wmt;
$vp8_sub_pixel_variance8x8_media=vp8_sub_pixel_variance8x8_armv6;
add_proto qw/unsigned int vp8_sub_pixel_variance8x16/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse";
specialize qw/vp8_sub_pixel_variance8x16 mmx sse2/;
$vp8_sub_pixel_variance8x16_sse2=vp8_sub_pixel_variance8x16_wmt;
add_proto qw/unsigned int vp8_sub_pixel_variance16x8/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse";
specialize qw/vp8_sub_pixel_variance16x8 mmx sse2 ssse3/;
$vp8_sub_pixel_variance16x8_sse2=vp8_sub_pixel_variance16x8_wmt;
add_proto qw/unsigned int vp8_sub_pixel_variance16x16/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse";
specialize qw/vp8_sub_pixel_variance16x16 mmx sse2 ssse3 media neon_asm/;
$vp8_sub_pixel_variance16x16_sse2=vp8_sub_pixel_variance16x16_wmt;
$vp8_sub_pixel_variance16x16_media=vp8_sub_pixel_variance16x16_armv6;
$vp8_sub_pixel_variance16x16_neon_asm=vp8_sub_pixel_variance16x16_neon;
add_proto qw/unsigned int vp8_variance_halfpixvar16x16_h/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse";
specialize qw/vp8_variance_halfpixvar16x16_h mmx sse2 media neon/;
$vp8_variance_halfpixvar16x16_h_sse2=vp8_variance_halfpixvar16x16_h_wmt;
$vp8_variance_halfpixvar16x16_h_media=vp8_variance_halfpixvar16x16_h_armv6;
add_proto qw/unsigned int vp8_variance_halfpixvar16x16_v/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse";
specialize qw/vp8_variance_halfpixvar16x16_v mmx sse2 media neon/;
$vp8_variance_halfpixvar16x16_v_sse2=vp8_variance_halfpixvar16x16_v_wmt;
$vp8_variance_halfpixvar16x16_v_media=vp8_variance_halfpixvar16x16_v_armv6;
add_proto qw/unsigned int vp8_variance_halfpixvar16x16_hv/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse";
specialize qw/vp8_variance_halfpixvar16x16_hv mmx sse2 media neon/;
$vp8_variance_halfpixvar16x16_hv_sse2=vp8_variance_halfpixvar16x16_hv_wmt;
$vp8_variance_halfpixvar16x16_hv_media=vp8_variance_halfpixvar16x16_hv_armv6;
#
# Encoder functions below this point.
#
......
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP8_COMMON_VARIANCE_H_
#define VP8_COMMON_VARIANCE_H_
#include "vpx_config.h"
#include "vpx/vpx_integer.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef unsigned int(*vpx_sad_fn_t)(
const uint8_t *src_ptr,
int source_stride,
const uint8_t *ref_ptr,
int ref_stride);
typedef void (*vp8_copy32xn_fn_t)(
const unsigned char *src_ptr,
int source_stride,
unsigned char *ref_ptr,
int ref_stride,
int n);
typedef void (*vpx_sad_multi_fn_t)(
const unsigned char *src_ptr,
int source_stride,
const unsigned char *ref_array,
int ref_stride,
unsigned int *sad_array);
typedef void (*vpx_sad_multi_d_fn_t)
(
const unsigned char *src_ptr,
int source_stride,
const unsigned char * const ref_array[],
int ref_stride,
unsigned int *sad_array
);
typedef unsigned int (*vpx_variance_fn_t)
(
const unsigned char *src_ptr,
int source_stride,
const unsigned char *ref_ptr,
int ref_stride,
unsigned int *sse
);
typedef unsigned int (*vp8_subpixvariance_fn_t)
(
const unsigned char *src_ptr,
int source_stride,
int xoffset,
int yoffset,
const unsigned char *ref_ptr,
int Refstride,
unsigned int *sse
);
typedef struct variance_vtable
{
vpx_sad_fn_t sdf;
vpx_variance_fn_t vf;
vp8_subpixvariance_fn_t svf;
vpx_variance_fn_t svf_halfpix_h;
vpx_variance_fn_t svf_halfpix_v;
vpx_variance_fn_t svf_halfpix_hv;
vpx_sad_multi_fn_t sdx3f;
vpx_sad_multi_fn_t sdx8f;
vpx_sad_multi_d_fn_t sdx4df;
#if ARCH_X86 || ARCH_X86_64
vp8_copy32xn_fn_t copymem;
#endif
} vp8_variance_fn_ptr_t;
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP8_COMMON_VARIANCE_H_
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "./vp8_rtcd.h"
#include "filter.h"
#include "variance.h"
/* This is a bad idea.
* ctz = count trailing zeros */
static int ctz(int a) {
int b = 0;
while (a != 1) {
a >>= 1;
b++;
}
return b;
}
static unsigned int variance(
const unsigned char *src_ptr,
int source_stride,
const unsigned char *ref_ptr,
int recon_stride,
int w,
int h,
unsigned int *sse)
{
int i, j;
int diff, sum;
sum = 0;
*sse = 0;
for (i = 0; i < h; i++)
{
for (j = 0; j < w; j++)
{
diff = src_ptr[j] - ref_ptr[j];
sum += diff;
*sse += diff * diff;
}
src_ptr += source_stride;
ref_ptr += recon_stride;
}
return (*sse - (((unsigned int)sum * sum) >> (int)((ctz(w) + ctz(h)))));
}
/****************************************************************************
*
* 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.
*
* OUTPUTS : INT32 *output_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.
*
* 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.
*
****************************************************************************/
static void var_filter_block2d_bil_first_pass
(
const 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 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[pixel_step] * 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;
}
}
/****************************************************************************
*
* 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.
*
* OUTPUTS : UINT16 *output_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.
*
* 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.
*
****************************************************************************/
static void var_filter_block2d_bil_second_pass
(
const unsigned short *src_ptr,
unsigned char *output_ptr,
unsigned int src_pixels_per_line,
unsigned int pixel_step,
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[pixel_step] * 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_width;
}
}
unsigned int vp8_sub_pixel_variance4x4_c
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
)
{
unsigned char temp2[20*16];
const short *HFilter, *VFilter;
unsigned short FData3[5*4]; /* Temp data bufffer used in filtering */
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
/* First filter 1d Horizontal */
var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 5, 4, HFilter);
/* Now filter Verticaly */
var_filter_block2d_bil_second_pass(FData3, temp2, 4, 4, 4, 4, VFilter);
return variance(temp2, 4, dst_ptr, dst_pixels_per_line, 4, 4, sse);
}
unsigned int vp8_sub_pixel_variance8x8_c
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
)
{
unsigned short FData3[9*8]; /* Temp data bufffer used in filtering */
unsigned char temp2[20*16];
const short *HFilter, *VFilter;
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 9, 8, HFilter);
var_filter_block2d_bil_second_pass(FData3, temp2, 8, 8, 8, 8, VFilter);
return variance(temp2, 8, dst_ptr, dst_pixels_per_line, 8, 8, sse);
}
unsigned int vp8_sub_pixel_variance16x16_c
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
)
{
unsigned short FData3[17*16]; /* Temp data bufffer used in filtering */
unsigned char temp2[20*16];
const short *HFilter, *VFilter;
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 17, 16, HFilter);
var_filter_block2d_bil_second_pass(FData3, temp2, 16, 16, 16, 16, VFilter);
return variance(temp2, 16, dst_ptr, dst_pixels_per_line, 16, 16, sse);
}
unsigned int vp8_variance_halfpixvar16x16_h_c(
const unsigned char *src_ptr,
int source_stride,
const unsigned char *ref_ptr,
int recon_stride,
unsigned int *sse)
{
return vp8_sub_pixel_variance16x16_c(src_ptr, source_stride, 4, 0,
ref_ptr, recon_stride, sse);
}
unsigned int vp8_variance_halfpixvar16x16_v_c(
const unsigned char *src_ptr,
int source_stride,
const unsigned char *ref_ptr,
int recon_stride,
unsigned int *sse)
{
return vp8_sub_pixel_variance16x16_c(src_ptr, source_stride, 0, 4,
ref_ptr, recon_stride, sse);
}
unsigned int vp8_variance_halfpixvar16x16_hv_c(
const unsigned char *src_ptr,
int source_stride,
const unsigned char *ref_ptr,
int recon_stride,
unsigned int *sse)
{
return vp8_sub_pixel_variance16x16_c(src_ptr, source_stride, 4, 4,
ref_ptr, recon_stride, sse);
}
unsigned int vp8_sub_pixel_variance16x8_c
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
)
{
unsigned short FData3[16*9]; /* Temp data bufffer used in filtering */
unsigned char temp2[20*16];
const short *HFilter, *VFilter;
HFilter = vp8_bilinear_filters[xoffset];
VFilter = vp8_bilinear_filters[yoffset];
var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 9, 16, HFilter);
var_filter_block2d_bil_second_pass(FData3, temp2, 16, 16, 8, 16, VFilter);
return variance(temp2, 16, dst_ptr, dst_pixels_per_line, 16, 8, sse);
}
unsigned int vp8_sub_pixel_variance8x16_c
(
const unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
const unsigned char *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse
)
{
unsigned short FData3[9*16]; /* Temp data bufffer used in filtering */
unsigned char temp2[20*16];
const short *HFilter,<