potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-21 19:33:03 +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/dnn/src/layers/convolution_layer.cpp
+23 -1
View File
@@ -986,12 +986,13 @@ public:
bool useAVX2;
bool useAVX512;
bool useRVV;
bool useLASX;
int blk_size_cn;
ParallelConv()
: input_(0), weights_(0), output_(0), ngroups_(0), nstripes_(0),
biasvec_(0), reluslope_(0), activ_(0), is1x1_(false), useAVX(false), useAVX2(false), useAVX512(false), useRVV(false)
, blk_size_cn(0)
, useLASX(false), blk_size_cn(0)
{}
static void run( const Mat& input, Mat& output, const Mat& weights,
@@ -1049,6 +1050,7 @@ public:
p.useAVX2 = checkHardwareSupport(CPU_AVX2) && isConv2D;
p.useAVX512 = CV_CPU_HAS_SUPPORT_AVX512_SKX && isConv2D;
p.useRVV = checkHardwareSupport(CPU_RVV) && isConv2D;
p.useLASX = checkHardwareSupport(CPU_LASX) && isConv2D;
int kernel_d = isConv3D? kernel_size[0] : 1;
int kernel_h = isConv1D? 1 : kernel_size[kernel_size.size() - 2];
@@ -1256,6 +1258,13 @@ public:
stride_h, stride_w, dilation_h, dilation_w, pad_t, pad_l,
biasptr, relu, inptr_, height, width, outptr_, out_d, outH, outW);
else
#endif
#if CV_TRY_LASX
if(useLASX)
opt_LASX::fastDepthwiseConv(wptr, kernel_h, kernel_w,
stride_h, stride_w, dilation_h, dilation_w, pad_t, pad_l,
biasptr, relu, inptr_, height, width, outptr_, out_d, outH, outW);
else
#endif
{
const float w00_ = wptr[0], w01_ = wptr[1], w02_ = wptr[2],
@@ -1631,6 +1640,12 @@ public:
opt_RVV::fastConv(wptr, wstep, biasptr, rowbuf0, data_out0 + ofs0,
outShape, bsz, vsz, vsz_a, relu, cn0 == 0);
else
#endif
#if CV_TRY_LASX
if(useLASX)
opt_LASX::fastConv(wptr, wstep, biasptr, rowbuf0, data_out0 + ofs0,
outShape, bsz, vsz, vsz_a, relu, cn0 == 0);
else
#endif
for( int i = 0; i < outCn; i += 2 )
{
@@ -2437,6 +2452,7 @@ public:
useAVX2 = checkHardwareSupport(CPU_AVX2);
useAVX512 = CV_CPU_HAS_SUPPORT_AVX512_SKX;
useRVV = checkHardwareSupport(CPU_RVV);
useLASX = checkHardwareSupport(CPU_LASX);
}
void operator()(const Range& range_) const CV_OVERRIDE
@@ -2474,6 +2490,11 @@ public:
opt_RVV::fastGEMM( aptr, astep, bptr, bstep, cptr, cstep, mmax, kmax, nmax );
}
else
#endif
#if CV_TRY_LASX
if( useLASX )
opt_LASX::fastGEMM( aptr, astep, bptr, bstep, cptr, cstep, mmax, kmax, nmax );
else
#endif
for( m = 0; m < mmax; m += 2 )
{
@@ -2574,6 +2595,7 @@ public:
bool useAVX2;
bool useAVX512;
bool useRVV;
bool useLASX;
};
class Col2ImInvoker : public cv::ParallelLoopBody