optimization - Optimizing loop with few instructions(SSE2, SSE4) with TBB -

-

i have simple image processing related algorithm. briefly, image(mean) in float subtracted 8-bit image result save float image(dest)

this function written intrinsics.

i have tried optimize function tbb, parrallel_for, received no gain in speed penalty.

what should ? should use more low-level scheme such tbb task optimize code ?

float           *m, **m_data,                 *o, **o_data; unsigned char   *p, **src_data; register unsigned long len, i; unsigned long   nr,                 nc;  src_data    =   src->ubytedata;    // 2d array m_data      =   mean->floatdata;   // 2d array o_data      =   dest->floatdata;   // 2d array nr          =   src->rows; nc          =   src->cols;  __m128i xmm0;  for(i=0; i<nr; i++) {     m = m_data[i];     o = o_data[i];     p = src_data[i];     len = nc;         {         _mm_prefetch((const char *)(p + 16),  _mm_hint_nta);         _mm_prefetch((const char *)(m + 16),  _mm_hint_nta);          xmm0 = _mm_load_si128((__m128i *) (p));          _mm_stream_ps(                         o,                         _mm_sub_ps(                                     _mm_cvtepi32_ps(_mm_cvtepu8_epi32(_mm_srli_si128(xmm0, 0))),                                     _mm_load_ps(m + offset)                                 )                     );         _mm_stream_ps(                         o + 4,                         _mm_sub_ps(                                     _mm_cvtepi32_ps(_mm_cvtepu8_epi32(_mm_srli_si128(xmm0, 4))),                                     _mm_load_ps(m + offset + 4)                                 )                     );         _mm_stream_ps(                         o + 8,                         _mm_sub_ps(                                     _mm_cvtepi32_ps(_mm_cvtepu8_epi32(_mm_srli_si128(xmm0, 8))),                                     _mm_load_ps(m + offset + 8)                                 )                     );         _mm_stream_ps(                         o + 12,                         _mm_sub_ps(                                     _mm_cvtepi32_ps(_mm_cvtepu8_epi32(_mm_srli_si128(xmm0, 12))),                                     _mm_load_ps(m + offset + 12)                                 )                     );          p += 16;         m += 16;         o += 16;         len -= 16;     }     while(len); }

you doing no computation here, relative number of loads , stores, it's being limited memory bandwidth rather computation. explain why don't see improvement in throughput when optimise computation.

i rid of _mm_prefetch instructions though - not helping here , may hurting performance.

if possible should combine loop other operations doing before/after - way amortise cost of memory i/o on more computation.


Comments

Post a Comment

Popular posts from this blog

python - Scipy curvefit RuntimeError:Optimal parameters not found: Number of calls to function has reached maxfev = 1000 -

-
Image
i want make logharitmic fit. keep getting runtime error: optimal parameters not found: number of calls function has reached maxfev = 1000 i use following script. can tell me go wrong? use spyder still beginner. import math import matplotlib mpl scipy.optimize import curve_fit import numpy np #data f1=[735.0,696.0,690.0,683.0,680.0,678.0,679.0,675.0,671.0,669.0,668.0,664.0,664.0] t1=[1,90000.0,178200.0,421200.0,505800.0,592200.0,768600.0,1036800.0,1371600.0,1630800.0,1715400.0,2345400.0,2409012.0] f1n=np.array(f1) t1n=np.array(t1) plt.plot(t1,f1,'ro',label="original data") # curvefit def func(t,a,b): return a+b*np.log(t) t=np.linspace(0,3600*24*28,13) popt, pcov = curve_fit(func, t, f1n, maxfev=1000) plt.plot(t, func(t, *popt), label="fitted curve") plt.legend(loc='upper left') plt.show() your original data t1 , f1 . therefore curve_fit should given t1 second argument, not t . popt, pcov = curve_fit(func,
Read more

c# - How to add a new treeview at the selected node? -

-
i have treeview 4 levels; parent, child, grandchild, great-grandchild. selectednode @ grandchild level. what i'm trying create new "treeview" @ grandchild - no, dont wnat create new node "selectednode" (grandchild). should somelike this: parent child grandchild (new treeview) parent grandchild great-grandchild child great-grandchild child grandchild grandchild it similar parent table mother , father went off , had new childern different spouse other spouse of there existing children. private sub populaterootlevel() ' query find first round of parent populatenodes(dt, jcatreeview.nodes) end sub private sub populatenodes(byval dt datatable, byval nodes treenodecollection) each dr datarow in dt.rows dim tn new treenode() tn.text = dr("title").tostring()
Read more

java - netbeans "Please wait - classpath scanning in progress..." -

-
when try compile netbeans projects message/error : "please wait - classpath scanning in progress..." and nothing happens , have have wait enternity conmpile "hello world"... now first thing have moved new office building, laptop same. it's secure enviroment lot of "security", can't connect anywhere almost. have changed user dir in netbeans.conf save data in c:\program files\user any ideas how fix this? alrady tried reinstalling netbeans, changing save location between standart , newone... just ideas find problem look @ cpu - netbeans busy (100% on 1 core) or blocking (95% idle)? standard, ant or maven build? libraries , use absolute or relative paths? start simple "hello world" project (no dependencies), add 1 lib (log4j or similiar), located in project folder , - if works - move lib other locations (choose locations similiar libs on real classpath) there chance, don't have access rights folders/shares conta
Read more