1 #include "precompiled.h" 2 // 3 // Copyright (c) 2002-2012 The ANGLE Project Authors. All rights reserved. 4 // Use of this source code is governed by a BSD-style license that can be 5 // found in the LICENSE file. 6 // 7 8 // loadimage.cpp: Defines image loading functions. It's 9 // in a separated file for GCC, which can enable SSE usage only per-file, 10 // not for code blocks that use SSE2 explicitly. 11 12 #include "libGLESv2/renderer/loadimage.h" 13 14 namespace rx 15 { 16 17 void loadAlphaDataToBGRASSE2(int width, int height, int depth, 18 const void *input, unsigned int inputRowPitch, unsigned int inputDepthPitch, 19 void *output, unsigned int outputRowPitch, unsigned int outputDepthPitch) 20 { 21 const unsigned char *source = NULL; 22 unsigned int *dest = NULL; 23 __m128i zeroWide = _mm_setzero_si128(); 24 25 for (int z = 0; z < depth; z++) 26 { 27 for (int y = 0; y < height; y++) 28 { 29 source = static_cast<const unsigned char*>(input) + y * inputRowPitch + z * inputDepthPitch; 30 dest = reinterpret_cast<unsigned int*>(static_cast<unsigned char*>(output) + y * outputRowPitch + z * outputDepthPitch); 31 32 int x; 33 // Make output writes aligned 34 for (x = 0; ((reinterpret_cast<intptr_t>(&dest[x]) & 0xF) != 0 && x < width); x++) 35 { 36 dest[x] = static_cast<unsigned int>(source[x]) << 24; 37 } 38 39 for (; x + 7 < width; x += 8) 40 { 41 __m128i sourceData = _mm_loadl_epi64(reinterpret_cast<const __m128i*>(&source[x])); 42 // Interleave each byte to 16bit, make the lower byte to zero 43 sourceData = _mm_unpacklo_epi8(zeroWide, sourceData); 44 // Interleave each 16bit to 32bit, make the lower 16bit to zero 45 __m128i lo = _mm_unpacklo_epi16(zeroWide, sourceData); 46 __m128i hi = _mm_unpackhi_epi16(zeroWide, sourceData); 47 48 _mm_store_si128(reinterpret_cast<__m128i*>(&dest[x]), lo); 49 _mm_store_si128(reinterpret_cast<__m128i*>(&dest[x + 4]), hi); 50 } 51 52 // Handle the remainder 53 for (; x < width; x++) 54 { 55 dest[x] = static_cast<unsigned int>(source[x]) << 24; 56 } 57 } 58 } 59 } 60 61 void loadRGBAUByteDataToBGRASSE2(int width, int height, int depth, 62 const void *input, unsigned int inputRowPitch, unsigned int inputDepthPitch, 63 void *output, unsigned int outputRowPitch, unsigned int outputDepthPitch) 64 { 65 const unsigned int *source = NULL; 66 unsigned int *dest = NULL; 67 __m128i brMask = _mm_set1_epi32(0x00ff00ff); 68 69 for (int z = 0; z < depth; z++) 70 { 71 for (int y = 0; y < height; y++) 72 { 73 source = reinterpret_cast<const unsigned int*>(static_cast<const unsigned char*>(input) + y * inputRowPitch + z * inputDepthPitch); 74 dest = reinterpret_cast<unsigned int*>(static_cast<unsigned char*>(output) + y * outputRowPitch + z * outputDepthPitch); 75 int x = 0; 76 77 // Make output writes aligned 78 for (x = 0; ((reinterpret_cast<intptr_t>(&dest[x]) & 15) != 0) && x < width; x++) 79 { 80 unsigned int rgba = source[x]; 81 dest[x] = (_rotl(rgba, 16) & 0x00ff00ff) | (rgba & 0xff00ff00); 82 } 83 84 for (; x + 3 < width; x += 4) 85 { 86 __m128i sourceData = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&source[x])); 87 // Mask out g and a, which don't change 88 __m128i gaComponents = _mm_andnot_si128(brMask, sourceData); 89 // Mask out b and r 90 __m128i brComponents = _mm_and_si128(sourceData, brMask); 91 // Swap b and r 92 __m128i brSwapped = _mm_shufflehi_epi16(_mm_shufflelo_epi16(brComponents, _MM_SHUFFLE(2, 3, 0, 1)), _MM_SHUFFLE(2, 3, 0, 1)); 93 __m128i result = _mm_or_si128(gaComponents, brSwapped); 94 _mm_store_si128(reinterpret_cast<__m128i*>(&dest[x]), result); 95 } 96 97 // Perform leftover writes 98 for (; x < width; x++) 99 { 100 unsigned int rgba = source[x]; 101 dest[x] = (_rotl(rgba, 16) & 0x00ff00ff) | (rgba & 0xff00ff00); 102 } 103 } 104 } 105 } 106 107 } 108