Hi,
I was writing a very simple class template so that we can use Fortran syntax (1-based index, column-first) to access the elements of 2D arrays in acc routines.
The problem and a workaround have been included in the comment of the following code sample.
Thanks,
stw
/*
$ pgc++ --version
pgc++ 19.4-0 LLVM 64-bit target on x86-64 Linux -tp penryn
PGI Compilers and Tools
Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
This will fail:
$ pgc++ -c -acc eg.cc -O2
nvvmCompileProgram error: 9.
Error: /tmp/pgaccgvQdsHV92JUn.gpu (85, 31): parse '%li99' defined with type 'i64'
PGCC-S-0155-Compiler failed to translate accelerator region (see -Minfo messages): Device compiler exited with error status code (eg.cc: 1)
PGCC/x86-64 Linux 19.4-0: compilation completed with severe errors
These won't fail:
$ pgc++ -c -acc eg.cc -O1
$ pgc++ -c -acc eg.cc -O2 -DWORKAROUND
$ pgc++ -c -acc eg.cc -O1 -DWORKAROUND
*/
template <class T, int FRow>
class FMat_2DImpl {
private:
T (*data_)[FRow];
public:
#pragma acc routine seq
FMat_2DImpl(T (*arr)[FRow]) : data_(arr) {}
#pragma acc routine seq
const T &operator()(int ifrow1, int ifcol1) const {
return data_[ifcol1 - 1][ifrow1 - 1];
}
};
template <class T, int FRow>
class FMat_1DImpl {
private:
T *data_;
public:
#pragma acc routine seq
FMat_1DImpl(T (*arr)[FRow]) : data_((T *)arr) {}
#pragma acc routine seq
const T &operator()(int ifrow1, int ifcol1) const {
return data_[(ifcol1 - 1) * FRow + (ifrow1 - 1)];
}
};
#pragma acc routine seq
void assign(float w, float *output) { output[0] = w; }
#pragma acc routine seq
void func(float *output) {
float x[3][3];
#ifndef WORKAROUND
FMat_2DImpl<float, 3> fx(x);
#else
FMat_1DImpl<float, 3> fx(x);
#endif
float w = fx(1, 1) + fx(2, 1) + fx(3, 1);
assign(w, output);
}