michael@0: #include "precompiled.h"
michael@0: //
michael@0: // Copyright (c) 2002-2012 The ANGLE Project 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: 
michael@0: // ImageSSE2.cpp: Implements SSE2-based functions of rx::Image class. It's
michael@0: // in a separated file for GCC, which can enable SSE usage only per-file,
michael@0: // not for code blocks that use SSE2 explicitly.
michael@0: 
michael@0: #include "libGLESv2/Texture.h"
michael@0: #include "libGLESv2/renderer/Image.h"
michael@0: 
michael@0: namespace rx
michael@0: {
michael@0: 
michael@0: void Image::loadRGBAUByteDataToBGRASSE2(GLsizei width, GLsizei height,
michael@0:                                         int inputPitch, const void *input, size_t outputPitch, void *output)
michael@0: {
michael@0:     const unsigned int *source = NULL;
michael@0:     unsigned int *dest = NULL;
michael@0:     __m128i brMask = _mm_set1_epi32(0x00ff00ff);
michael@0: 
michael@0:     for (int y = 0; y < height; y++)
michael@0:     {
michael@0:         source = reinterpret_cast<const unsigned int*>(static_cast<const unsigned char*>(input) + y * inputPitch);
michael@0:         dest = reinterpret_cast<unsigned int*>(static_cast<unsigned char*>(output) + y * outputPitch);
michael@0:         int x = 0;
michael@0: 
michael@0:         // Make output writes aligned
michael@0:         for (x = 0; ((reinterpret_cast<intptr_t>(&dest[x]) & 15) != 0) && x < width; x++)
michael@0:         {
michael@0:             unsigned int rgba = source[x];
michael@0:             dest[x] = (_rotl(rgba, 16) & 0x00ff00ff) | (rgba & 0xff00ff00);
michael@0:         }
michael@0: 
michael@0:         for (; x + 3 < width; x += 4)
michael@0:         {
michael@0:             __m128i sourceData = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&source[x]));
michael@0:             // Mask out g and a, which don't change
michael@0:             __m128i gaComponents = _mm_andnot_si128(brMask, sourceData);
michael@0:             // Mask out b and r
michael@0:             __m128i brComponents = _mm_and_si128(sourceData, brMask);
michael@0:             // Swap b and r
michael@0:             __m128i brSwapped = _mm_shufflehi_epi16(_mm_shufflelo_epi16(brComponents, _MM_SHUFFLE(2, 3, 0, 1)), _MM_SHUFFLE(2, 3, 0, 1));
michael@0:             __m128i result = _mm_or_si128(gaComponents, brSwapped);
michael@0:             _mm_store_si128(reinterpret_cast<__m128i*>(&dest[x]), result);
michael@0:         }
michael@0: 
michael@0:         // Perform leftover writes
michael@0:         for (; x < width; x++)
michael@0:         {
michael@0:             unsigned int rgba = source[x];
michael@0:             dest[x] = (_rotl(rgba, 16) & 0x00ff00ff) | (rgba & 0xff00ff00);
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: void Image::loadAlphaDataToBGRASSE2(GLsizei width, GLsizei height,
michael@0:                                     int inputPitch, const void *input, size_t outputPitch, void *output)
michael@0: {
michael@0:     const unsigned char *source = NULL;
michael@0:     unsigned int *dest = NULL;
michael@0:     __m128i zeroWide = _mm_setzero_si128();
michael@0: 
michael@0:     for (int y = 0; y < height; y++)
michael@0:     {
michael@0:         source = static_cast<const unsigned char*>(input) + y * inputPitch;
michael@0:         dest = reinterpret_cast<unsigned int*>(static_cast<unsigned char*>(output) + y * outputPitch);
michael@0: 
michael@0:         int x;
michael@0:         // Make output writes aligned
michael@0:         for (x = 0; ((reinterpret_cast<intptr_t>(&dest[x]) & 0xF) != 0 && x < width); x++)
michael@0:         {
michael@0:             dest[x] = static_cast<unsigned int>(source[x]) << 24;
michael@0:         }
michael@0: 
michael@0:         for (; x + 7 < width; x += 8)
michael@0:         {
michael@0:             __m128i sourceData = _mm_loadl_epi64(reinterpret_cast<const __m128i*>(&source[x]));
michael@0:             // Interleave each byte to 16bit, make the lower byte to zero
michael@0:             sourceData = _mm_unpacklo_epi8(zeroWide, sourceData);
michael@0:             // Interleave each 16bit to 32bit, make the lower 16bit to zero
michael@0:             __m128i lo = _mm_unpacklo_epi16(zeroWide, sourceData);
michael@0:             __m128i hi = _mm_unpackhi_epi16(zeroWide, sourceData);
michael@0: 
michael@0:             _mm_store_si128(reinterpret_cast<__m128i*>(&dest[x]), lo);
michael@0:             _mm_store_si128(reinterpret_cast<__m128i*>(&dest[x + 4]), hi);
michael@0:         }
michael@0: 
michael@0:         // Handle the remainder
michael@0:         for (; x < width; x++)
michael@0:         {
michael@0:             dest[x] = static_cast<unsigned int>(source[x]) << 24;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: }