potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 23:33:05 +04:00
Code Packages Releases Wiki Activity

Add Loongson Advanced SIMD Extension support: -DCPU_BASELINE=LASX

Browse Source 
* Add Loongson Advanced SIMD Extension support: -DCPU_BASELINE=LASX
* Add resize.lasx.cpp for Loongson SIMD acceleration
* Add imgwarp.lasx.cpp for Loongson SIMD acceleration
* Add LASX acceleration support for dnn/conv
* Add CV_PAUSE(v) for Loongarch
* Set LASX by default on Loongarch64
* LoongArch: tune test threshold for Core/HAL.mat_decomp/15

Co-authored-by: shengwenxue <shengwenxue@loongson.cn>
This commit is contained in:
wxsheng
2022-09-10 14:39:43 +08:00
committed by GitHub
parent 866191478f
commit 4154bd0667
24 changed files with 5071 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/imgproc/src/imgwarp.cpp
+7
View File
@@ -2178,6 +2178,9 @@ public:
#if CV_TRY_SSE4_1
bool useSSE4_1 = CV_CPU_HAS_SUPPORT_SSE4_1;
#endif
#if CV_TRY_LASX
bool useLASX = CV_CPU_HAS_SUPPORT_LASX;
#endif
int bh0 = std::min(BLOCK_SZ/2, dst.rows);
int bw0 = std::min(BLOCK_SZ*BLOCK_SZ/bh0, dst.cols);
@@ -2241,6 +2244,10 @@ public:
if ( useAVX2 )
x1 = opt_AVX2::warpAffineBlockline(adelta + x, bdelta + x, xy, alpha, X0, Y0, bw);
#endif
#if CV_TRY_LASX
if ( useLASX )
x1 = opt_LASX::warpAffineBlockline(adelta + x, bdelta + x, xy, alpha, X0, Y0, bw);
#endif
#if CV_SIMD128
{
v_int32x4 v__X0 = v_setall_s32(X0), v__Y0 = v_setall_s32(Y0);
modules/imgproc/src/imgwarp.hpp
+7
View File
@@ -61,6 +61,13 @@ int warpAffineBlockline(int *adelta, int *bdelta, short* xy, short* alpha, int X
#endif
}
namespace opt_LASX
{
#if CV_TRY_LASX
int warpAffineBlockline(int *adelta, int *bdelta, short* xy, short* alpha, int X0, int Y0, int bw);
#endif
}
namespace opt_SSE4_1
{
#if CV_TRY_SSE4_1
modules/imgproc/src/imgwarp.lasx.cpp
+98
View File
@@ -0,0 +1,98 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 2014-2015, Itseez Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
/* ////////////////////////////////////////////////////////////////////
//
// Geometrical transforms on images and matrices: rotation, zoom etc.
//
// */
#include "precomp.hpp"
#include "imgwarp.hpp"
#include "opencv2/core/hal/intrin.hpp"
namespace cv
{
namespace opt_LASX
{
int warpAffineBlockline(int *adelta, int *bdelta, short* xy, short* alpha, int X0, int Y0, int bw)
{
const int AB_BITS = MAX(10, (int)INTER_BITS);
int x1 = 0;
__m256i fxy_mask = _v256_setall_w(INTER_TAB_SIZE - 1);
__m256i XX = _v256_setall_w(X0), YY = _v256_setall_w(Y0);
for (; x1 <= bw - 16; x1 += 16)
{
__m256i tx0, tx1, ty0, ty1;
tx0 = __lasx_xvadd_w(__lasx_xvld((const __m256i*)(adelta + x1), 0), XX);
ty0 = __lasx_xvadd_w(__lasx_xvld((const __m256i*)(bdelta + x1), 0), YY);
tx1 = __lasx_xvadd_w(__lasx_xvld((const __m256i*)(adelta + x1), 8*4), XX);
ty1 = __lasx_xvadd_w(__lasx_xvld((const __m256i*)(bdelta + x1), 8*4), YY);
tx0 = __lasx_xvsrai_w(tx0, AB_BITS - INTER_BITS);
ty0 = __lasx_xvsrai_w(ty0, AB_BITS - INTER_BITS);
tx1 = __lasx_xvsrai_w(tx1, AB_BITS - INTER_BITS);
ty1 = __lasx_xvsrai_w(ty1, AB_BITS - INTER_BITS);
__m256i fx_ = _lasx_packs_w(__lasx_xvand_v(tx0, fxy_mask),
__lasx_xvand_v(tx1, fxy_mask));
__m256i fy_ = _lasx_packs_w(__lasx_xvand_v(ty0, fxy_mask),
__lasx_xvand_v(ty1, fxy_mask));
tx0 = _lasx_packs_w(__lasx_xvsrai_w(tx0, INTER_BITS),
__lasx_xvsrai_w(tx1, INTER_BITS));
ty0 = _lasx_packs_w(__lasx_xvsrai_w(ty0, INTER_BITS),
__lasx_xvsrai_w(ty1, INTER_BITS));
fx_ = __lasx_xvsadd_h(fx_, __lasx_xvslli_h(fy_, INTER_BITS));
fx_ = __lasx_xvpermi_d(fx_, (3 << 6) + (1 << 4) + (2 << 2) + 0);
__lasx_xvst(__lasx_xvilvl_h(ty0, tx0), (__m256i*)(xy + x1 * 2), 0);
__lasx_xvst(__lasx_xvilvh_h(ty0, tx0), (__m256i*)(xy + x1 * 2), 16*2);
__lasx_xvst(fx_, (__m256i*)(alpha + x1), 0);
}
return x1;
}
}
}
/* End of file. */
modules/imgproc/src/resize.cpp
+10
View File
@@ -1098,6 +1098,16 @@ resizeNN( const Mat& src, Mat& dst, double fx, double fy )
opt_SSE4_1::resizeNN4_SSE4_1(range, src, dst, x_ofs, ify);
}
else
#endif
#if CV_TRY_LASX
if(CV_CPU_HAS_SUPPORT_LASX && ((pix_size == 2) || (pix_size == 4)))
{
if(pix_size == 2)
opt_LASX::resizeNN2_LASX(range, src, dst, x_ofs, ify);
else
opt_LASX::resizeNN4_LASX(range, src, dst, x_ofs, ify);
}
else
#endif
{
resizeNNInvoker invoker(src, dst, x_ofs, ify);
modules/imgproc/src/resize.hpp
+9
View File
@@ -70,6 +70,15 @@ void resizeNN4_SSE4_1(const Range&, const Mat&, Mat&, int*, double);
int VResizeLanczos4Vec_32f16u_SSE41(const float** src, ushort* dst, const float* beta, int width);
#endif
}
namespace opt_LASX
{
#if CV_TRY_LASX
void resizeNN2_LASX(const Range&, const Mat&, Mat&, int*, double);
void resizeNN4_LASX(const Range&, const Mat&, Mat&, int*, double);
#endif
}
}
#endif
/* End of file. */
modules/imgproc/src/resize.lasx.cpp
+249
View File
@@ -0,0 +1,249 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 2014-2015, Itseez Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
/* ////////////////////////////////////////////////////////////////////
//
// Geometrical transforms on images and matrices: rotation, zoom etc.
//
// */
#include "precomp.hpp"
#include "resize.hpp"
#include "opencv2/core/hal/intrin.hpp"
namespace cv
{
namespace opt_LASX
{
class resizeNNInvokerLASX4 CV_FINAL :
public ParallelLoopBody
{
public:
resizeNNInvokerLASX4(const Mat& _src, Mat &_dst, int *_x_ofs, double _ify) :
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs),
ify(_ify)
{
}
virtual void operator() (const Range& range) const CV_OVERRIDE
{
Size ssize = src.size(), dsize = dst.size();
int y, x;
int width = dsize.width;
int avxWidth = width - (width & 0x7);
if(((int64)(dst.data + dst.step) & 0x1f) == 0)
{
for(y = range.start; y < range.end; y++)
{
uchar* D = dst.data + dst.step*y;
uchar* Dstart = D;
int sy = std::min(cvFloor(y*ify), ssize.height-1);
const uchar* S = src.data + sy*src.step;
#ifdef CV_ICC
#pragma unroll(4)
#endif
for(x = 0; x < avxWidth; x += 8)
{
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
__m256i CV_DECL_ALIGNED(64) pixels = v256_lut_quads((schar *)S, (int *)addr).val;
__lasx_xvst(pixels, (int*)D, 0);
D += 32;
}
for(; x < width; x++)
{
*(int*)(Dstart + x*4) = *(int*)(S + x_ofs[x]);
}
}
}
else
{
for(y = range.start; y < range.end; y++)
{
uchar* D = dst.data + dst.step*y;
uchar* Dstart = D;
int sy = std::min(cvFloor(y*ify), ssize.height-1);
const uchar* S = src.data + sy*src.step;
#ifdef CV_ICC
#pragma unroll(4)
#endif
for(x = 0; x < avxWidth; x += 8)
{
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
__m256i CV_DECL_ALIGNED(64) pixels = v256_lut_quads((schar *)S, (int *)addr).val;
__lasx_xvst(pixels, (int*)D, 0);
D += 32;
}
for(; x < width; x++)
{
*(int*)(Dstart + x*4) = *(int*)(S + x_ofs[x]);
}
}
}
}
private:
const Mat& src;
Mat& dst;
int* x_ofs;
double ify;
resizeNNInvokerLASX4(const resizeNNInvokerLASX4&);
resizeNNInvokerLASX4& operator=(const resizeNNInvokerLASX4&);
};
class resizeNNInvokerLASX2 CV_FINAL :
public ParallelLoopBody
{
public:
resizeNNInvokerLASX2(const Mat& _src, Mat &_dst, int *_x_ofs, double _ify) :
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs),
ify(_ify)
{
}
virtual void operator() (const Range& range) const CV_OVERRIDE
{
Size ssize = src.size(), dsize = dst.size();
int y, x;
int width = dsize.width;
int avxWidth = width - (width & 0xf);
const __m256i CV_DECL_ALIGNED(64) shuffle_mask = _v256_set_b(15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0,
15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0);
const __m256i CV_DECL_ALIGNED(64) permute_mask = _v256_set_w(7, 5, 3, 1, 6, 4, 2, 0);
if(((int64)(dst.data + dst.step) & 0x1f) == 0)
{
for(y = range.start; y < range.end; y++)
{
uchar* D = dst.data + dst.step*y;
uchar* Dstart = D;
int sy = std::min(cvFloor(y*ify), ssize.height-1);
const uchar* S = src.data + sy*src.step;
const uchar* S2 = S - 2;
#ifdef CV_ICC
#pragma unroll(4)
#endif
for(x = 0; x < avxWidth; x += 16)
{
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
__m256i CV_DECL_ALIGNED(64) pixels1 = v256_lut_quads((schar *)S, (int *)addr).val;
const __m256i CV_DECL_ALIGNED(64) *addr2 = (__m256i*)(x_ofs + x + 8);
__m256i CV_DECL_ALIGNED(64) pixels2 = v256_lut_quads((schar *)S2, (int *)addr2).val;
const __m256i h_mask = __lasx_xvreplgr2vr_w(0xFFFF0000);
__m256i CV_DECL_ALIGNED(64) unpacked = __lasx_xvbitsel_v(pixels1, pixels2, h_mask);
__m256i CV_DECL_ALIGNED(64) bytes_shuffled = __lasx_xvshuf_b(unpacked, unpacked, shuffle_mask);
__m256i CV_DECL_ALIGNED(64) ints_permuted = __lasx_xvperm_w(bytes_shuffled, permute_mask);
__lasx_xvst(ints_permuted, (int*)D, 0);
D += 32;
}
for(; x < width; x++)
{
*(ushort*)(Dstart + x*2) = *(ushort*)(S + x_ofs[x]);
}
}
}
else
{
for(y = range.start; y < range.end; y++)
{
uchar* D = dst.data + dst.step*y;
uchar* Dstart = D;
int sy = std::min(cvFloor(y*ify), ssize.height-1);
const uchar* S = src.data + sy*src.step;
const uchar* S2 = S - 2;
#ifdef CV_ICC
#pragma unroll(4)
#endif
for(x = 0; x < avxWidth; x += 16)
{
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
__m256i CV_DECL_ALIGNED(64) pixels1 = v256_lut_quads((schar *)S, (int *)addr).val;
const __m256i CV_DECL_ALIGNED(64) *addr2 = (__m256i*)(x_ofs + x + 8);
__m256i CV_DECL_ALIGNED(64) pixels2 = v256_lut_quads((schar *)S2, (int *)addr2).val;
const __m256i h_mask = __lasx_xvreplgr2vr_w(0xFFFF0000);
__m256i CV_DECL_ALIGNED(64) unpacked = __lasx_xvbitsel_v(pixels1, pixels2, h_mask);
__m256i CV_DECL_ALIGNED(64) bytes_shuffled = __lasx_xvshuf_b(unpacked, unpacked, shuffle_mask);
__m256i CV_DECL_ALIGNED(64) ints_permuted = __lasx_xvperm_w(bytes_shuffled, permute_mask);
__lasx_xvst(ints_permuted, (int*)D, 0);
D += 32;
}
for(; x < width; x++)
{
*(ushort*)(Dstart + x*2) = *(ushort*)(S + x_ofs[x]);
}
}
}
}
private:
const Mat& src;
Mat& dst;
int* x_ofs;
double ify;
resizeNNInvokerLASX2(const resizeNNInvokerLASX2&);
resizeNNInvokerLASX2& operator=(const resizeNNInvokerLASX2&);
};
void resizeNN2_LASX(const Range& range, const Mat& src, Mat &dst, int *x_ofs, double ify)
{
resizeNNInvokerLASX2 invoker(src, dst, x_ofs, ify);
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
}
void resizeNN4_LASX(const Range& range, const Mat& src, Mat &dst, int *x_ofs, double ify)
{
resizeNNInvokerLASX4 invoker(src, dst, x_ofs, ify);
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
}
}
}
/* End of file. */