/* * Copyright 2013 The LibYuv 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 "libyuv/row.h" #include "libyuv/scale_row.h" #ifdef __cplusplus namespace libyuv { extern "C" { #endif // This module is for GCC x86 and x64. #if !defined(LIBYUV_DISABLE_X86) && (defined(__x86_64__) || defined(__i386__)) // Offsets for source bytes 0 to 9 static const uvec8 kShuf0 = …; // Offsets for source bytes 11 to 20 with 8 subtracted = 3 to 12. static const uvec8 kShuf1 = …; // Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31. static const uvec8 kShuf2 = …; // Offsets for source bytes 0 to 10 static const uvec8 kShuf01 = …; // Offsets for source bytes 10 to 21 with 8 subtracted = 3 to 13. static const uvec8 kShuf11 = …; // Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31. static const uvec8 kShuf21 = …; // Coefficients for source bytes 0 to 10 static const uvec8 kMadd01 = …; // Coefficients for source bytes 10 to 21 static const uvec8 kMadd11 = …; // Coefficients for source bytes 21 to 31 static const uvec8 kMadd21 = …; // Coefficients for source bytes 21 to 31 static const vec16 kRound34 = …; static const uvec8 kShuf38a = …; static const uvec8 kShuf38b = …; // Arrange words 0,3,6 into 0,1,2 static const uvec8 kShufAc = …; // Arrange words 0,3,6 into 3,4,5 static const uvec8 kShufAc3 = …; // Scaling values for boxes of 3x3 and 2x3 static const uvec16 kScaleAc33 = …; // Arrange first value for pixels 0,1,2,3,4,5 static const uvec8 kShufAb0 = …; // Arrange second value for pixels 0,1,2,3,4,5 static const uvec8 kShufAb1 = …; // Arrange third value for pixels 0,1,2,3,4,5 static const uvec8 kShufAb2 = …; // Scaling values for boxes of 3x2 and 2x2 static const uvec16 kScaleAb2 = …; // GCC versions of row functions are verbatim conversions from Visual C. // Generated using gcc disassembly on Visual C object file: // objdump -D yuvscaler.obj >yuvscaler.txt void ScaleRowDown2_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown2Linear_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown2Box_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } #ifdef HAS_SCALEROWDOWN2_AVX2 void ScaleRowDown2_AVX2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown2Linear_AVX2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown2Box_AVX2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } #endif // HAS_SCALEROWDOWN2_AVX2 void ScaleRowDown4_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown4Box_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } #ifdef HAS_SCALEROWDOWN4_AVX2 void ScaleRowDown4_AVX2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown4Box_AVX2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } #endif // HAS_SCALEROWDOWN4_AVX2 void ScaleRowDown34_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown34_1_Box_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown34_0_Box_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown38_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown38_2_Box_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } void ScaleRowDown38_3_Box_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } static const uvec8 kLinearShuffleFar = …; static const uvec8 kLinearMadd31 = …; #ifdef HAS_SCALEROWUP2_LINEAR_SSE2 void ScaleRowUp2_Linear_SSE2(const uint8_t* src_ptr, uint8_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_BILINEAR_SSE2 void ScaleRowUp2_Bilinear_SSE2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_LINEAR_12_SSSE3 void ScaleRowUp2_Linear_12_SSSE3(const uint16_t* src_ptr, uint16_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_BILINEAR_12_SSSE3 void ScaleRowUp2_Bilinear_12_SSSE3(const uint16_t* src_ptr, ptrdiff_t src_stride, uint16_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_LINEAR_16_SSE2 void ScaleRowUp2_Linear_16_SSE2(const uint16_t* src_ptr, uint16_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_BILINEAR_16_SSE2 void ScaleRowUp2_Bilinear_16_SSE2(const uint16_t* src_ptr, ptrdiff_t src_stride, uint16_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_LINEAR_SSSE3 void ScaleRowUp2_Linear_SSSE3(const uint8_t* src_ptr, uint8_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_BILINEAR_SSSE3 void ScaleRowUp2_Bilinear_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_LINEAR_AVX2 void ScaleRowUp2_Linear_AVX2(const uint8_t* src_ptr, uint8_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_BILINEAR_AVX2 void ScaleRowUp2_Bilinear_AVX2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_LINEAR_12_AVX2 void ScaleRowUp2_Linear_12_AVX2(const uint16_t* src_ptr, uint16_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_BILINEAR_12_AVX2 void ScaleRowUp2_Bilinear_12_AVX2(const uint16_t* src_ptr, ptrdiff_t src_stride, uint16_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_LINEAR_16_AVX2 void ScaleRowUp2_Linear_16_AVX2(const uint16_t* src_ptr, uint16_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEROWUP2_BILINEAR_16_AVX2 void ScaleRowUp2_Bilinear_16_AVX2(const uint16_t* src_ptr, ptrdiff_t src_stride, uint16_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif // Reads 16xN bytes and produces 16 shorts at a time. void ScaleAddRow_SSE2(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width) { … } #ifdef HAS_SCALEADDROW_AVX2 // Reads 32 bytes and accumulates to 32 shorts at a time. void ScaleAddRow_AVX2(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width) { … } #endif // HAS_SCALEADDROW_AVX2 // Constant for making pixels signed to avoid pmaddubsw // saturation. static const uvec8 kFsub80 = …; // Constant for making pixels unsigned and adding .5 for rounding. static const uvec16 kFadd40 = …; // Bilinear column filtering. SSSE3 version. void ScaleFilterCols_SSSE3(uint8_t* dst_ptr, const uint8_t* src_ptr, int dst_width, int x, int dx) { … } // Reads 4 pixels, duplicates them and writes 8 pixels. // Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned. void ScaleColsUp2_SSE2(uint8_t* dst_ptr, const uint8_t* src_ptr, int dst_width, int x, int dx) { … } void ScaleARGBRowDown2_SSE2(const uint8_t* src_argb, ptrdiff_t src_stride, uint8_t* dst_argb, int dst_width) { … } void ScaleARGBRowDown2Linear_SSE2(const uint8_t* src_argb, ptrdiff_t src_stride, uint8_t* dst_argb, int dst_width) { … } void ScaleARGBRowDown2Box_SSE2(const uint8_t* src_argb, ptrdiff_t src_stride, uint8_t* dst_argb, int dst_width) { … } // Reads 4 pixels at a time. // Alignment requirement: dst_argb 16 byte aligned. void ScaleARGBRowDownEven_SSE2(const uint8_t* src_argb, ptrdiff_t src_stride, int src_stepx, uint8_t* dst_argb, int dst_width) { … } // Blends four 2x2 to 4x1. // Alignment requirement: dst_argb 16 byte aligned. void ScaleARGBRowDownEvenBox_SSE2(const uint8_t* src_argb, ptrdiff_t src_stride, int src_stepx, uint8_t* dst_argb, int dst_width) { … } void ScaleARGBCols_SSE2(uint8_t* dst_argb, const uint8_t* src_argb, int dst_width, int x, int dx) { … } // Reads 4 pixels, duplicates them and writes 8 pixels. // Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned. void ScaleARGBColsUp2_SSE2(uint8_t* dst_argb, const uint8_t* src_argb, int dst_width, int x, int dx) { … } // Shuffle table for arranging 2 pixels into pairs for pmaddubsw static const uvec8 kShuffleColARGB = …; // Shuffle table for duplicating 2 fractions into 8 bytes each static const uvec8 kShuffleFractions = …; // Bilinear row filtering combines 4x2 -> 4x1. SSSE3 version void ScaleARGBFilterCols_SSSE3(uint8_t* dst_argb, const uint8_t* src_argb, int dst_width, int x, int dx) { … } // Divide num by div and return as 16.16 fixed point result. int FixedDiv_X86(int num, int div) { … } // Divide num - 1 by div - 1 and return as 16.16 fixed point result. int FixedDiv1_X86(int num, int div) { … } #if defined(HAS_SCALEUVROWDOWN2BOX_SSSE3) || \ defined(HAS_SCALEUVROWDOWN2BOX_AVX2) // Shuffle table for splitting UV into upper and lower part of register. static const uvec8 kShuffleSplitUV = …; static const uvec8 kShuffleMergeUV = …; #endif #ifdef HAS_SCALEUVROWDOWN2BOX_SSSE3 void ScaleUVRowDown2Box_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } #endif // HAS_SCALEUVROWDOWN2BOX_SSSE3 #ifdef HAS_SCALEUVROWDOWN2BOX_AVX2 void ScaleUVRowDown2Box_AVX2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, int dst_width) { … } #endif // HAS_SCALEUVROWDOWN2BOX_AVX2 static const uvec8 kUVLinearMadd31 = …; #ifdef HAS_SCALEUVROWUP2_LINEAR_SSSE3 void ScaleUVRowUp2_Linear_SSSE3(const uint8_t* src_ptr, uint8_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEUVROWUP2_BILINEAR_SSSE3 void ScaleUVRowUp2_Bilinear_SSSE3(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEUVROWUP2_LINEAR_AVX2 void ScaleUVRowUp2_Linear_AVX2(const uint8_t* src_ptr, uint8_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEUVROWUP2_BILINEAR_AVX2 void ScaleUVRowUp2_Bilinear_AVX2(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEUVROWUP2_LINEAR_16_SSE41 void ScaleUVRowUp2_Linear_16_SSE41(const uint16_t* src_ptr, uint16_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEUVROWUP2_BILINEAR_16_SSE41 void ScaleUVRowUp2_Bilinear_16_SSE41(const uint16_t* src_ptr, ptrdiff_t src_stride, uint16_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #ifdef HAS_SCALEUVROWUP2_LINEAR_16_AVX2 void ScaleUVRowUp2_Linear_16_AVX2(const uint16_t* src_ptr, uint16_t* dst_ptr, int dst_width) { … } #endif #ifdef HAS_SCALEUVROWUP2_BILINEAR_16_AVX2 void ScaleUVRowUp2_Bilinear_16_AVX2(const uint16_t* src_ptr, ptrdiff_t src_stride, uint16_t* dst_ptr, ptrdiff_t dst_stride, int dst_width) { … } #endif #endif // defined(__x86_64__) || defined(__i386__) #ifdef __cplusplus } // extern "C" } // namespace libyuv #endif
Codebase Browser
chromium