matmul_r8_avx2：失败。使用 python f2py 增加维度

Question

我正在使用 Fortran 包装程序

f2py

。它以前与

n=5~30

一起用于矩阵乘法（f2py 无法转换为 C/Fortran 数组）。

当我增加到

50

时，它告诉我问题

python: /home/rdonnelly/mc/conda-bld/compilers_linux-64_1534627447954/work/.build/x86_64-conda_cos6-linux-gnu/src/gcc/libgfortran/generated/matmul_r8.c:642: matmul_r8_avx2: Assertion 

((a)->dtype & 0x07) == 2 || ((b)->dtype & 0x07) == 2' 失败。 中止（核心转储）`.

这里是python部分

test-f2py.py

import numpy as np
import test

dim_n = 50
A = np.random.random((dim_n, dim_n))
B = np.random.random((dim_n, dim_n))
C = np.zeros((dim_n, dim_n))

A = np.asfortranarray(A)
B = np.asfortranarray(B)
C = np.asfortranarray(C)

test.wrap(A, B, C)
print("Fortran result:")
#print(C)

# Cross-check with numpy.einsum
C_check = np.einsum('ij,jk->ik', A, B)
print("Numpy einsum result:")
#print(C_check)

# Compare the two results
print("Are the results equal?")
print(np.allclose(C, C_check))

fortran 部分，

test.f90

，由

python -m numpy.f2py -c  --f90flags='-O3'  -m test test.f90

运行

!f2py dp selected_real_kind(15, 307)
subroutine wrap(A, B, C)
    real(kind=8), intent(in) :: A(:, :), B(:, :)
    real(kind=8), intent(inout) :: C(:, :)
    integer :: m, n, p

    m = size(A, 1)
    n = size(A, 2)
    p = size(B, 2)
    
    if (n /= size(B, 1)) then
        print *, "Error: Incompatible dimensions for matrix multiplication"
        return
    end if

    C = matmul(A, B)
end subroutine wrap

如果我运行 python 代码，例如，

python test-f2py.py

我得到了

python: /home/rdonnelly/mc/conda-bld/compilers_linux-64_1534627447954/work/.build/x86_64-conda_cos6-linux-gnu/src/gcc/libgfortran/generated/matmul_r8.c:642: matmul_r8_avx2: Assertion `((a)->dtype & 0x07) == 2 || ((b)->dtype & 0x07) == 2' failed.
Aborted (core dumped)

Answer 1

升级

numpy

和

gfortran

没有帮助。我想到了四处走走。如果我使用

f2py

而不是

numpy.f2py

，它会告诉我

/usr/include/python3.8/pyconfig.h:3:12: fatal error: x86_64-linux-gnu/python3.8/pyconfig.h: No such file or directory  #  include <x86_64-linux-gnu/python3.8/pyconfig.h>

然后我在 coda 中切换到 python 3.9 env，它有效