michael@0: // Copyright (c) 2010 The Chromium Authors. All rights reserved.
michael@0: // Use of this source code is governed by a BSD-style license that can be
michael@0: // found in the LICENSE file.
michael@0: 
michael@0: #include <mmintrin.h>
michael@0: #include "yuv_row.h"
michael@0: 
michael@0: namespace mozilla {
michael@0: namespace gfx {
michael@0: 
michael@0: // FilterRows combines two rows of the image using linear interpolation.
michael@0: // MMX version does 8 pixels at a time.
michael@0: void FilterRows_MMX(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
michael@0:                     int source_width, int source_y_fraction) {
michael@0:   __m64 zero = _mm_setzero_si64();
michael@0:   __m64 y1_fraction = _mm_set1_pi16(source_y_fraction);
michael@0:   __m64 y0_fraction = _mm_set1_pi16(256 - source_y_fraction);
michael@0: 
michael@0:   const __m64* y0_ptr64 = reinterpret_cast<const __m64*>(y0_ptr);
michael@0:   const __m64* y1_ptr64 = reinterpret_cast<const __m64*>(y1_ptr);
michael@0:   __m64* dest64 = reinterpret_cast<__m64*>(ybuf);
michael@0:   __m64* end64 = reinterpret_cast<__m64*>(ybuf + source_width);
michael@0: 
michael@0:   do {
michael@0:     __m64 y0 = *y0_ptr64++;
michael@0:     __m64 y1 = *y1_ptr64++;
michael@0:     __m64 y2 = _mm_unpackhi_pi8(y0, zero);
michael@0:     __m64 y3 = _mm_unpackhi_pi8(y1, zero);
michael@0:     y0 = _mm_unpacklo_pi8(y0, zero);
michael@0:     y1 = _mm_unpacklo_pi8(y1, zero);
michael@0:     y0 = _mm_mullo_pi16(y0, y0_fraction);
michael@0:     y1 = _mm_mullo_pi16(y1, y1_fraction);
michael@0:     y2 = _mm_mullo_pi16(y2, y0_fraction);
michael@0:     y3 = _mm_mullo_pi16(y3, y1_fraction);
michael@0:     y0 = _mm_add_pi16(y0, y1);
michael@0:     y2 = _mm_add_pi16(y2, y3);
michael@0:     y0 = _mm_srli_pi16(y0, 8);
michael@0:     y2 = _mm_srli_pi16(y2, 8);
michael@0:     y0 = _mm_packs_pu16(y0, y2);
michael@0:     *dest64++ = y0;
michael@0:   } while (dest64 < end64);
michael@0: }
michael@0: 
michael@0: }
michael@0: }