_mm_cvtss_f32 support for "new[]" and "delete

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

dear experts,

I’ve tried to use Eigen-2.0.15, compiled by pgcpp version 10.9 (Win32)

http://eigen.tuxfamily.org/index.php?title=Main_Page

with the most simple code (test.cpp) and optimization:
’ –fastsse -Minform --gnu_version 4.3.4 --gnu_extensions ’

#include <Eigen/Core>

// import most common Eigen types 
USING_PART_OF_NAMESPACE_EIGEN

int main(int, char *[])
{
  Matrix3f m3;
  m3 << 1, 2, 3, 4, 5, 6, 7, 8, 9;
  Matrix4f m4 = Matrix4f::Identity();
  Vector4i v4(1, 2, 3, 4);

  std::cout << "m3\n" << m3 << "\nm4:\n"
    << m4 << "\nv4:\n" << v4 << std::endl;
}



However such errors appeared:

PGI$ . /d/Ubuntu/compile.sh > compError
“.\Eigen/src/Core/arch/SSE/PacketMath.h”, line 125: error: identifier
“_mm_cvtss_f32” is undefined
template<> EIGEN_STRONG_INLINE float ei_pfirst<__m128>(const __m128& a) { float x = _mm_cvtss_f32(a); return x; }
^

“.\Eigen/src/Core/Matrix.h”, line 144: error: support for “new[]” and
“delete[]” is disabled
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
^

2 errors detected in the compilation of “test.cpp”.


I cross-checked with g++ 4.2.4 (Ubuntu) as well, and there was no compile error and the test was okay. However,


For error (1):
I’ve looked around
xx:\Program Files\PGI\win32\10.9\include\emmintrin.h
xx:\Program Files\PGI\win32\10.9\include\xmmintrin.h

and PGI User Guide, it seems that function _mm_cvtss_f32 does not exist at all.

Will it be possible to emulate it as :

xxx _mm_cvtss_f32(__m128 _a){
	__m128d t;
	t = _mm_cvtps_pd(_a);	
	float  x=0.0;
	x = (double)_mm_cvtsd_f64(t);
	return x; 
}

For error (2):

Although I’ve used flags “–gnu_version 4.3.4 --gnu_extensions”, the compilation error remains.
Your helps are very appreciated…

best regards,
LamCo