michael@0: /*
michael@0:  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0:  *
michael@0:  *  Use of this source code is governed by a BSD-style license
michael@0:  *  that can be found in the LICENSE file in the root of the source
michael@0:  *  tree. An additional intellectual property rights grant can be found
michael@0:  *  in the file PATENTS.  All contributing project authors may
michael@0:  *  be found in the AUTHORS file in the root of the source tree.
michael@0:  */
michael@0: 
michael@0: #include "vpx_config.h"
michael@0: #include "vp8/common/variance.h"
michael@0: #include "vp8/common/pragmas.h"
michael@0: #include "vpx_ports/mem.h"
michael@0: #include "vp8/common/x86/filter_x86.h"
michael@0: 
michael@0: extern void filter_block1d_h6_mmx
michael@0: (
michael@0:     const unsigned char *src_ptr,
michael@0:     unsigned short *output_ptr,
michael@0:     unsigned int src_pixels_per_line,
michael@0:     unsigned int pixel_step,
michael@0:     unsigned int output_height,
michael@0:     unsigned int output_width,
michael@0:     short *filter
michael@0: );
michael@0: extern void filter_block1d_v6_mmx
michael@0: (
michael@0:     const short *src_ptr,
michael@0:     unsigned char *output_ptr,
michael@0:     unsigned int pixels_per_line,
michael@0:     unsigned int pixel_step,
michael@0:     unsigned int output_height,
michael@0:     unsigned int output_width,
michael@0:     short *filter
michael@0: );
michael@0: 
michael@0: extern unsigned int vp8_get_mb_ss_mmx(const short *src_ptr);
michael@0: extern unsigned int vp8_get8x8var_mmx
michael@0: (
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *SSE,
michael@0:     int *Sum
michael@0: );
michael@0: extern unsigned int vp8_get4x4var_mmx
michael@0: (
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *SSE,
michael@0:     int *Sum
michael@0: );
michael@0: extern void vp8_filter_block2d_bil4x4_var_mmx
michael@0: (
michael@0:     const unsigned char *ref_ptr,
michael@0:     int ref_pixels_per_line,
michael@0:     const unsigned char *src_ptr,
michael@0:     int src_pixels_per_line,
michael@0:     const short *HFilter,
michael@0:     const short *VFilter,
michael@0:     int *sum,
michael@0:     unsigned int *sumsquared
michael@0: );
michael@0: extern void vp8_filter_block2d_bil_var_mmx
michael@0: (
michael@0:     const unsigned char *ref_ptr,
michael@0:     int ref_pixels_per_line,
michael@0:     const unsigned char *src_ptr,
michael@0:     int src_pixels_per_line,
michael@0:     unsigned int Height,
michael@0:     const short *HFilter,
michael@0:     const short *VFilter,
michael@0:     int *sum,
michael@0:     unsigned int *sumsquared
michael@0: );
michael@0: 
michael@0: 
michael@0: unsigned int vp8_variance4x4_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     unsigned int var;
michael@0:     int avg;
michael@0: 
michael@0:     vp8_get4x4var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
michael@0:     *sse = var;
michael@0:     return (var - (((unsigned int)avg * avg) >> 4));
michael@0: 
michael@0: }
michael@0: 
michael@0: unsigned int vp8_variance8x8_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     unsigned int var;
michael@0:     int avg;
michael@0: 
michael@0:     vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
michael@0:     *sse = var;
michael@0: 
michael@0:     return (var - (((unsigned int)avg * avg) >> 6));
michael@0: 
michael@0: }
michael@0: 
michael@0: unsigned int vp8_mse16x16_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     unsigned int sse0, sse1, sse2, sse3, var;
michael@0:     int sum0, sum1, sum2, sum3;
michael@0: 
michael@0: 
michael@0:     vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
michael@0:     vp8_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
michael@0:     vp8_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2) ;
michael@0:     vp8_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride, ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3);
michael@0: 
michael@0:     var = sse0 + sse1 + sse2 + sse3;
michael@0:     *sse = var;
michael@0:     return var;
michael@0: }
michael@0: 
michael@0: 
michael@0: unsigned int vp8_variance16x16_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     unsigned int sse0, sse1, sse2, sse3, var;
michael@0:     int sum0, sum1, sum2, sum3, avg;
michael@0: 
michael@0: 
michael@0:     vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
michael@0:     vp8_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
michael@0:     vp8_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2) ;
michael@0:     vp8_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride, ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3);
michael@0: 
michael@0:     var = sse0 + sse1 + sse2 + sse3;
michael@0:     avg = sum0 + sum1 + sum2 + sum3;
michael@0:     *sse = var;
michael@0:     return (var - (((unsigned int)avg * avg) >> 8));
michael@0: }
michael@0: 
michael@0: unsigned int vp8_variance16x8_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     unsigned int sse0, sse1, var;
michael@0:     int sum0, sum1, avg;
michael@0: 
michael@0:     vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
michael@0:     vp8_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
michael@0: 
michael@0:     var = sse0 + sse1;
michael@0:     avg = sum0 + sum1;
michael@0:     *sse = var;
michael@0:     return (var - (((unsigned int)avg * avg) >> 7));
michael@0: 
michael@0: }
michael@0: 
michael@0: 
michael@0: unsigned int vp8_variance8x16_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     unsigned int sse0, sse1, var;
michael@0:     int sum0, sum1, avg;
michael@0: 
michael@0:     vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
michael@0:     vp8_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse1, &sum1) ;
michael@0: 
michael@0:     var = sse0 + sse1;
michael@0:     avg = sum0 + sum1;
michael@0:     *sse = var;
michael@0: 
michael@0:     return (var - (((unsigned int)avg * avg) >> 7));
michael@0: 
michael@0: }
michael@0: 
michael@0: 
michael@0: unsigned int vp8_sub_pixel_variance4x4_mmx
michael@0: (
michael@0:     const unsigned char  *src_ptr,
michael@0:     int  src_pixels_per_line,
michael@0:     int  xoffset,
michael@0:     int  yoffset,
michael@0:     const unsigned char *dst_ptr,
michael@0:     int dst_pixels_per_line,
michael@0:     unsigned int *sse)
michael@0: 
michael@0: {
michael@0:     int xsum;
michael@0:     unsigned int xxsum;
michael@0:     vp8_filter_block2d_bil4x4_var_mmx(
michael@0:         src_ptr, src_pixels_per_line,
michael@0:         dst_ptr, dst_pixels_per_line,
michael@0:         vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
michael@0:         &xsum, &xxsum
michael@0:     );
michael@0:     *sse = xxsum;
michael@0:     return (xxsum - (((unsigned int)xsum * xsum) >> 4));
michael@0: }
michael@0: 
michael@0: 
michael@0: unsigned int vp8_sub_pixel_variance8x8_mmx
michael@0: (
michael@0:     const unsigned char  *src_ptr,
michael@0:     int  src_pixels_per_line,
michael@0:     int  xoffset,
michael@0:     int  yoffset,
michael@0:     const unsigned char *dst_ptr,
michael@0:     int dst_pixels_per_line,
michael@0:     unsigned int *sse
michael@0: )
michael@0: {
michael@0: 
michael@0:     int xsum;
michael@0:     unsigned int xxsum;
michael@0:     vp8_filter_block2d_bil_var_mmx(
michael@0:         src_ptr, src_pixels_per_line,
michael@0:         dst_ptr, dst_pixels_per_line, 8,
michael@0:         vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
michael@0:         &xsum, &xxsum
michael@0:     );
michael@0:     *sse = xxsum;
michael@0:     return (xxsum - (((unsigned int)xsum * xsum) >> 6));
michael@0: }
michael@0: 
michael@0: unsigned int vp8_sub_pixel_variance16x16_mmx
michael@0: (
michael@0:     const unsigned char  *src_ptr,
michael@0:     int  src_pixels_per_line,
michael@0:     int  xoffset,
michael@0:     int  yoffset,
michael@0:     const unsigned char *dst_ptr,
michael@0:     int dst_pixels_per_line,
michael@0:     unsigned int *sse
michael@0: )
michael@0: {
michael@0: 
michael@0:     int xsum0, xsum1;
michael@0:     unsigned int xxsum0, xxsum1;
michael@0: 
michael@0: 
michael@0:     vp8_filter_block2d_bil_var_mmx(
michael@0:         src_ptr, src_pixels_per_line,
michael@0:         dst_ptr, dst_pixels_per_line, 16,
michael@0:         vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
michael@0:         &xsum0, &xxsum0
michael@0:     );
michael@0: 
michael@0: 
michael@0:     vp8_filter_block2d_bil_var_mmx(
michael@0:         src_ptr + 8, src_pixels_per_line,
michael@0:         dst_ptr + 8, dst_pixels_per_line, 16,
michael@0:         vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
michael@0:         &xsum1, &xxsum1
michael@0:     );
michael@0: 
michael@0:     xsum0 += xsum1;
michael@0:     xxsum0 += xxsum1;
michael@0: 
michael@0:     *sse = xxsum0;
michael@0:     return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
michael@0: 
michael@0: 
michael@0: }
michael@0: 
michael@0: unsigned int vp8_sub_pixel_mse16x16_mmx(
michael@0:     const unsigned char  *src_ptr,
michael@0:     int  src_pixels_per_line,
michael@0:     int  xoffset,
michael@0:     int  yoffset,
michael@0:     const unsigned char *dst_ptr,
michael@0:     int dst_pixels_per_line,
michael@0:     unsigned int *sse
michael@0: )
michael@0: {
michael@0:     vp8_sub_pixel_variance16x16_mmx(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pixels_per_line, sse);
michael@0:     return *sse;
michael@0: }
michael@0: 
michael@0: unsigned int vp8_sub_pixel_variance16x8_mmx
michael@0: (
michael@0:     const unsigned char  *src_ptr,
michael@0:     int  src_pixels_per_line,
michael@0:     int  xoffset,
michael@0:     int  yoffset,
michael@0:     const unsigned char *dst_ptr,
michael@0:     int dst_pixels_per_line,
michael@0:     unsigned int *sse
michael@0: )
michael@0: {
michael@0:     int xsum0, xsum1;
michael@0:     unsigned int xxsum0, xxsum1;
michael@0: 
michael@0: 
michael@0:     vp8_filter_block2d_bil_var_mmx(
michael@0:         src_ptr, src_pixels_per_line,
michael@0:         dst_ptr, dst_pixels_per_line, 8,
michael@0:         vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
michael@0:         &xsum0, &xxsum0
michael@0:     );
michael@0: 
michael@0: 
michael@0:     vp8_filter_block2d_bil_var_mmx(
michael@0:         src_ptr + 8, src_pixels_per_line,
michael@0:         dst_ptr + 8, dst_pixels_per_line, 8,
michael@0:         vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
michael@0:         &xsum1, &xxsum1
michael@0:     );
michael@0: 
michael@0:     xsum0 += xsum1;
michael@0:     xxsum0 += xxsum1;
michael@0: 
michael@0:     *sse = xxsum0;
michael@0:     return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
michael@0: }
michael@0: 
michael@0: unsigned int vp8_sub_pixel_variance8x16_mmx
michael@0: (
michael@0:     const unsigned char  *src_ptr,
michael@0:     int  src_pixels_per_line,
michael@0:     int  xoffset,
michael@0:     int  yoffset,
michael@0:     const unsigned char *dst_ptr,
michael@0:     int dst_pixels_per_line,
michael@0:     unsigned int *sse
michael@0: )
michael@0: {
michael@0:     int xsum;
michael@0:     unsigned int xxsum;
michael@0:     vp8_filter_block2d_bil_var_mmx(
michael@0:         src_ptr, src_pixels_per_line,
michael@0:         dst_ptr, dst_pixels_per_line, 16,
michael@0:         vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
michael@0:         &xsum, &xxsum
michael@0:     );
michael@0:     *sse = xxsum;
michael@0:     return (xxsum - (((unsigned int)xsum * xsum) >> 7));
michael@0: }
michael@0: 
michael@0: 
michael@0: unsigned int vp8_variance_halfpixvar16x16_h_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     return vp8_sub_pixel_variance16x16_mmx(src_ptr, source_stride, 4, 0,
michael@0:                                            ref_ptr, recon_stride, sse);
michael@0: }
michael@0: 
michael@0: 
michael@0: unsigned int vp8_variance_halfpixvar16x16_v_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     return vp8_sub_pixel_variance16x16_mmx(src_ptr, source_stride, 0, 4,
michael@0:                                            ref_ptr, recon_stride, sse);
michael@0: }
michael@0: 
michael@0: 
michael@0: unsigned int vp8_variance_halfpixvar16x16_hv_mmx(
michael@0:     const unsigned char *src_ptr,
michael@0:     int  source_stride,
michael@0:     const unsigned char *ref_ptr,
michael@0:     int  recon_stride,
michael@0:     unsigned int *sse)
michael@0: {
michael@0:     return vp8_sub_pixel_variance16x16_mmx(src_ptr, source_stride, 4, 4,
michael@0:                                            ref_ptr, recon_stride, sse);
michael@0: }