c++ - Align double vs align float for AVX operations -


i want multiply 2 (float/double) vectors avx operators. in order that, need aligned memory. function float values is:

#define size 65536 float *g, *h, *j; g = (float*)aligned_alloc(32, sizeof(float)*size); h = (float*)aligned_alloc(32, sizeof(float)*size); j = (float*)aligned_alloc(32, sizeof(float)*size); //filling g , h data for(int = 0; < size/8; i++)     {         __m256 a_a, b_a, c_a;         a_a = _mm256_load_ps(g+8*i);         b_a = _mm256_load_ps(h+8*i);         c_a = _mm256_mul_ps(a_a, b_a);         _mm256_store_ps (j+i*8, c_a);     } free(g); free(h); free(j);

that works, when trying double values, memory access error (such if memory not aligned correctly):

double *g_d, *h_d, *i_d; g_d = (double*)aligned_alloc(32, sizeof(double)*size); h_d = (double*)aligned_alloc(32, sizeof(double)*size); i_d = (double*)aligned_alloc(32, sizeof(double)*size); for(int = 0; < size/4; i++) {     __m256d a_a, b_a, c_a;     a_a = _mm256_load_pd(g_d+4*i);     b_a = _mm256_load_pd(h_d+4*i);     c_a = _mm256_mul_pd(a_a, b_a);     _mm256_store_pd (i_d+i*4, c_a); } free(g_d); free(h_d); free(i_d);

why alignment not working double-values?

when running in gdb, get

program received signal sigsegv, segmentation fault. 0x0000000000401669 in _mm256_load_pd (__p=0x619f70) @ /usr/lib/gcc/x86_64-linux-gnu/5/include/avxintrin.h:836

edit: found mistake, copy/paste error former function, manifested in function. due not being helpful others (as assume), close question.

well, problem seems stem different data sizes.

  • in first snippet increment float loop size/8=8192. here i'm unsure why increase float array element size 4 8. i < 8192
  • in second snippet increment double loop size/4=16384. here i'm unsure why increase double array element size 8 4. i < 16384 --- ** opposite!**

the last element of double array may surpass memory boundaries!

in both cases increment loop i++. cases proceed follows:

first : (float (4)) j+i*8 (0 < < 8192 ) =>

0      4      8      12      16     20     24     28   v1     .      v2     .       v3     .      v4     .

second: (double(8)) j+i*4 (0 < < 16384) => v1/v2/v3/v4

0      4      8      12      16     20     24     28     32   v1(h)  v1(l)  v2(l)  v3(l)   v4(l)  v5(l)  v6(l)  v7(l)  v1(h)  v2(h)  v3(h)  v4(h)   v5(h)  v6(h)  v7(h)  v8(h)  v8(h) -------------------------------------------------------------- thing ... thing ... thing .. thing ...

in second snippet mix high parts(32-bit) , low parts(32-bit) of 64-bit double incrementing 4 (sizeof float) instead of 8 (sizeof double).

another problem _mm256_store_pd requires that...

when source or destination operand memory operand, operand must aligned on 32-byte boundary or general-protection exception (#gp) generated.

for(int = 0; < size/4; i++) doesn't fulfill requirement.

i wondering float version seems work, because _mm256_store_ps requires that...

when source or destination operand memory operand, operand must aligned on 16-byte boundary or general-protection exception (#gp) generated.

but have alignment of 8 bytes...

however, need fix 'scale' of i variable make work.


-

Comments

Post a Comment

Popular posts from this blog

PySide and Qt Properties: Connecting signals from Python to QML -

i trying run simple pyside appliccation design made in qml. call function in qml python in separated thread. i have following qml view: import qt 4.7 rectangle { visible: true width: 100; height: 100 color: "lightgrey" text { id: datestr objectname: "datestr" x: 10 y: 10 text: "init text" font.pixelsize: 12 function updatedata(text) { datestr.text = qstr(text) console.log("you said: " + text) } } } here application: some imports.... #!/usr/bin/python import sys import time pyside import qtdeclarative, qtcore pyside.qtgui import qapplication pyside.qtdeclarative import qdeclarativeview the qthread worker.... #worker thread class worker(qtcore.qthread): updateprogress = qtcore.signal(int) def __init__(self, child): qtcore.qthread.__init__(self) self.child = child def run(self):
Read more

c# - DevExpress.Wpf.Grid.InfiniteGridSizeException was unhandled -

Image
when using devexpress, see error: devexpress.wpf.grid.infinitegridsizeexception unhandled message="by default, infinite grid height not allowed since grid rows rendered , hence grid work slowly. fix issue, should place grid container give finite height grid, or should manually specify grid's height or maxheight. note can avoid exception setting gridcontrol.allowinfinitegridsize static property true, in case grid run slowly." the problem dxgrid has infinite height. to fix, set height non-infinite. snoop absolutely invaluable this: if "height" xaml element infinite (i.e. 0 or nan ), can set using one of following: option 1: height="{binding path=actualheight, relativesource={relativesource mode=findancestor, ancestortype=uielement}}" option 2: verticalalignment="stretch" option 3: height="auto" hint: use verticalalignment="stretch" if child of grid <rowdefinition height=
Read more

scala - 'wrong top statement declaration' when using slick in IntelliJ -

Image
i'm porting project scala 2.9 2.10, therefore have use slick instead of scalaquery. i'm using slick 2.1.0 since supports ms access. according this tutorial , upgrade guide changed robs object class , added val robs : why error message wrong top statement declaration , how rid of it? edit: i'm new scala... according this question seems can't put val outside of methods or classes, right? code above directly in packge. right approach slick then? should move code in class or trait? according this changed val robs = tablequery[robs] to object robs extends tablequery(new robs(_)) {} . no warnings or errors anymore. :)
Read more