C++博客-hwliu11

GFORTRAN 64位openmp问题

隔壁老刘 — Sat, 04 Feb 2017 02:59:00 GMT

GFORTRAN在64位时候设置线程数目使用OMP_SET_NUM_THREADS无法生效，查看glibc代码，可以看出OMP_SET_NUM_THREADS是用C语言实现的，fortran参数传过来默认为地址，而OMP_SET_NUM_THREADS把传过来地址默认处理为32位地址，所以调用失败，64位下要使用专门处理64位地址的对应函数OMP_SET_NUM_THREADS_8_，上述两个函数输入参数不需要确认输入的integer是32位还是64位，这个与Intel Fortran不一样，Intel Fortran只是需要确认OMP_SET_NUM_THREADS的参数是32位还是64位

隔壁老刘 2017-02-04 10:59 发表评论

C/C++位运算的陷阱

隔壁老刘 — Tue, 12 Apr 2016 02:41:00 GMT

昨天用C/C++写Base64编码，结果老是不对，
比如

unsigned char A =;

unsigned char B = ;

unsigned char C = ;
unsigned char a, b, c, d;

a = A >> 2;

b=(A << 6) >> 2|(B>>4);

c = ((B << 4)) >> 2) | (C >> 6);

d = (C << 2) >> 2;
结果发现bcd值都不对
但是写成

b = A<< 6;

b = b >> 2;
b=b|B>>4;
就可以，查看b计算两种方式汇编
第一种

00BACB66 movzx eax,byte ptr [ebp-1F9h]

00BACB6D shl eax,6

00BACB70 sar eax,2

00BACB73 mov byte ptr [ebp-1C9h],al

第二种

00BACB79 movzx eax,byte ptr [ebp-1F9h]

00BACB80 shl eax,6

00BACB83 mov byte ptr [ebp-1EDh],al

00BACB89 movzx eax,byte ptr [ebp-1EDh]

00BACB90 sar eax,2

00BACB93 mov byte ptr [ebp-1EDh],al
问题就在指令sar，这是算数右移位指令，会保留最高位，而第二种可以是因为movzx 将高位已经置零
找到了原因，可以写为

a = A >> 2;

b=((unsigned char)(A << 6)) >> 2|(B>>4);

c = (((unsigned char)(B << 4)) >> 2) | (C >> 6);

d = ((unsigned char)(C << 2)) >> 2;
只要涉及到右移，保证操作数是无符号就可以了。

隔壁老刘 2016-04-12 10:41 发表评论

Intel OpenMP运行时错误 OMP: Error #133 解决办法

隔壁老刘 — Sat, 27 Feb 2016 04:13:00 GMT

OMP: Error #133: Inconsistent THREADPRIVATE common block declarations are non-conforming and are unsupported. Either all threadprivate common blocks must be declared identically, or the largest instance of each threadprivate common block must be referenced first during the run.

发生以上错误确实很麻烦，而且可参考文档基本没有。Fortran代码中的Common块数据大小可以不一样大（相同名称的Common），但是在OpenMP中确要求，要么大小一样，要么占用内存最大的先调用，变量如果是隐式声明，一定要注意，变量名称不同，可能会编译为不同长度的内存

隔壁老刘 2016-02-27 12:13 发表评论

OpenCASCADE 6.9.1 Bug

隔壁老刘 — Thu, 12 Nov 2015 07:20:00 GMT

OpenCASCADE 6.9.1（估计从6.9.0开始），使用新的屏幕选择算法，我使用6.8开发的程序，升级到6.9.1之后，使用IVTK部分模型无法选中几何元素，经过调试发现6.9.1代码有BUG，导致数据丢失，无法判断选中的元素是否在选择范围，BUG出现的文件为SelectMgr_RectangularFrustum.cxx中的函数ScaleAndTransform

NCollection_Handle SelectMgr_RectangularFrustum::ScaleAndTransform (const Standard_Integer theScaleFactor,
const gp_Trsf& theTrsf)
{
Standard_ASSERT_RAISE (theScaleFactor > 0,
"Error! Pixel tolerance for selection should be greater than zero");

SelectMgr_RectangularFrustum* aRes = new SelectMgr_RectangularFrustum();
const Standard_Boolean isToScale = theScaleFactor != 1;
const Standard_Boolean isToTrsf = theTrsf.Form() != gp_Identity;

if (!isToScale && !isToTrsf)
return aRes ;//这里，当部分模型满足该条件，返回一个空的数据，使用该数据无法判断模型是否在区域

aRes->myIsOrthographic = myIsOrthographic;
SelectMgr_RectangularFrustum* aRef = this;

if (isToScale)
{
    aRes->myNearPickedPnt = myNearPickedPnt;
    aRes->myFarPickedPnt = myFarPickedPnt;
    aRes->myViewRayDir    = myViewRayDir;

    const gp_Pnt2d aMinPnt (myMousePos.X() - theScaleFactor * 0.5,
                            myMousePos.Y() - theScaleFactor * 0.5);
    const gp_Pnt2d aMaxPnt (myMousePos.X() + theScaleFactor * 0.5,
                            myMousePos.Y() + theScaleFactor * 0.5);

// recompute base frustum characteristics from scratch
computeFrustum (aMinPnt, aMaxPnt, myBuilder, aRes->myVertices, aRes->myEdgeDirs);

aRef = aRes;
}

if (isToTrsf)
{
    aRes->myNearPickedPnt = aRef->myNearPickedPnt.Transformed (theTrsf);
    aRes->myFarPickedPnt = aRef->myFarPickedPnt.Transformed (theTrsf);
    aRes->myViewRayDir    = aRes->myFarPickedPnt.XYZ() - aRes->myNearPickedPnt.XYZ();

      // LeftTopNear
    aRes->myVertices[0] = aRef->myVertices[0].Transformed (theTrsf);
    // LeftTopFar
    aRes->myVertices[1] = aRef->myVertices[1].Transformed (theTrsf);
    // LeftBottomNear
    aRes->myVertices[2] = aRef->myVertices[2].Transformed (theTrsf);
    // LeftBottomFar
    aRes->myVertices[3] = aRef->myVertices[3].Transformed (theTrsf);
    // RightTopNear
    aRes->myVertices[4] = aRef->myVertices[4].Transformed (theTrsf);
    // RightTopFar
    aRes->myVertices[5] = aRef->myVertices[5].Transformed (theTrsf);
    // RightBottomNear
    aRes->myVertices[6] = aRef->myVertices[6].Transformed (theTrsf);
    // RightBottomFar
    aRes->myVertices[7] = aRef->myVertices[7].Transformed (theTrsf);

    // Horizontal
    aRes->myEdgeDirs[0] = aRes->myVertices[4].XYZ() - aRes->myVertices[0].XYZ();
    // Vertical
    aRes->myEdgeDirs[1] = aRes->myVertices[2].XYZ() - aRes->myVertices[0].XYZ();
    // LeftLower
    aRes->myEdgeDirs[2] = aRes->myVertices[2].XYZ() - aRes->myVertices[3].XYZ();
    // RightLower
    aRes->myEdgeDirs[3] = aRes->myVertices[6].XYZ() - aRes->myVertices[7].XYZ();
    // LeftUpper
    aRes->myEdgeDirs[4] = aRes->myVertices[0].XYZ() - aRes->myVertices[1].XYZ();
    // RightUpper
    aRes->myEdgeDirs[5] = aRes->myVertices[4].XYZ() - aRes->myVertices[5].XYZ();
}

// compute frustum normals
computeNormals (aRes->myEdgeDirs, aRes->myPlanes);

cacheVertexProjections (aRes);

return NCollection_Handle (aRes);
}

修改办法：

if (!isToScale && !isToTrsf)
{
aRes->myPixelTolerance=myPixelTolerance;
aRes->myIsOrthographic=myIsOrthographic;

aRes->myBuilder= myBuilder;

   aRes->myIsOrthographic = myIsOrthographic;
   aRes->myNearPickedPnt = myNearPickedPnt;
   aRes->myFarPickedPnt = myFarPickedPnt;
   aRes->myViewRayDir = myViewRayDir;
   aRes->myMousePos = myMousePos;

   for (int i = 0; i <6;++i)
   {
    aRes->myPlanes[i] = myPlanes[i];
   }
   for (int i = 0; i < 6; ++i)
   {
    aRes->myVertices[i] = myVertices[i];
   }
   for (int i = 0; i < 6; ++i)
   {
    aRes->myMaxVertsProjections[i] = myMaxVertsProjections[i];
   }
   for (int i = 0; i < 6; ++i)
   {
    aRes->myMinVertsProjections[i] = myMinVertsProjections[i];
   }
   for (int i = 0; i <3; ++i)
   {
    aRes->myMaxOrthoVertsProjections[i] = myMaxOrthoVertsProjections[i];
   }
   for (int i = 0; i <3; ++i)
   {
    aRes->myMinOrthoVertsProjections[i] = myMinOrthoVertsProjections[i];
   }
   for (int i = 0; i < 6; ++i)
   {
    aRes->myEdgeDirs[i] = myEdgeDirs[i];
   }
   return NCollection_Handle(aRes);
}

隔壁老刘 2015-11-12 15:20 发表评论

C++智能指针类用不好害死人

隔壁老刘 — Sat, 07 Nov 2015 13:43:00 GMT

C++智能指针类使用起来方便，但是不维护好引用计数，不能随时控制对象的释放，反而造成内存耗尽，这样说来，反而不能随时控制对象的生命周期，所以使用要谨慎啊。

隔壁老刘 2015-11-07 21:43 发表评论