Cytnx-dev
conda install cytnx MKL problem.
I followed the documentation to install Cytnx version 0.9.7.
When I do not import the cytnx package, all NumPy-dependent code runs correctly, as shown below:
(cytnx) [jerrychen@cn7 2d_new]$ python3 test_bug.py (1.0000000000000004+0j)
However, after importing cytnx, the code triggers an Intel MKL-related error:
(cytnx) [jerrychen@cn7 2d_new]$ python3 test_bug.py
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
Traceback (most recent call last):
File "/data2/home/jerrychen/Jerry_work/2d_new/test_bug.py", line 130, in
Since GitHub does not allow attaching .py files directly, I have renamed the uploaded script with a .txt extension.
Please kindly change the filename extension back to .py when testing.
Thank you very much for your help! test_bug.txt
This might be related to #244 and #348
Cytnx defaults to MKL ilp64 interface but the one from conda uses lp64 interface, thus causing the trouble.
Maybe we should also ship lp64 version?
Here is the code attached by the OP.
import sys
#import cytnx
import numpy as np
from ncon import ncon
def inner_MPS (mps1, mps2):
assert len(mps1) == len(mps2)
res = ncon([mps1[0], np.conj(mps2[0])], ((1,2,-1), (1,2,-2)))
for i in range(1,len(mps1)):
res = ncon([res,mps1[i],np.conj(mps2[i])], ((1,2), (1,3,-1), (2,3,-2)))
res = res.reshape(-1)
assert len(res) == 1
return res[0]
def compress_MPS (mps, D=sys.maxsize, cutoff=0.):
N = len(mps)
#
# 2 ---
# |
# R = o
# |
# 1 ---
#
Rs = [None for i in range(N+1)]
Rs[-1] = np.array([1.]).reshape((1,1))
for i in range(N-1, 0, -1):
Rs[i] = ncon([Rs[i+1],mps[i],np.conjugate(mps[i])], ((1,2), (-1,3,1), (-2,3,2)))
#
# 2
# |
# rho = o
# |
# 1
#
rho = ncon([Rs[1],mps[0],np.conjugate(mps[0])], ((1,2), (-1,-2,1), (-3,-4,2)))
rho = rho.reshape((rho.shape[1], rho.shape[3]))
# 1
# |
# 0 x--o-- 2
#
evals, U = np.linalg.eigh(rho)
U = U.reshape((1,*U.shape))
res = [U]
#
# ---- 2
# |
# L = |
# |
# ---- 1
#
L = np.array([1.]).reshape((1,1))
for i in range(1,N):
#
# 2---(U)-- -2
# | |
# L = (L) 3
# | |
# 1---(A)--- -1
#
L = ncon([L,mps[i-1],np.conjugate(U)], ((1,2), (1,3,-1), (2,3,-2)))
#
# -- -1
# |
# A = | -2
# | |
# (L)--1--(mps)--- -3
#
A = ncon([L,mps[i]], ((1,-1), (1,-2,-3)))
#
# -3 --(A)--2--
# | |
# -4 |
# rho = (R)
# -2 |
# | |
# -1 --(A)--1--
#
rho = ncon([Rs[i+1],A,np.conjugate(A)], ((1,2), (-1,-2,1), (-3,-4,2)))
d = rho.shape
rho = rho.reshape((d[0]*d[1], d[2]*d[3]))
# 1
# |
# 0 --o-- 2
#
evals, U = np.linalg.eigh(rho)
# truncate by dimension
DD = min(D, mps[i].shape[2])
U = U[:,-DD:]
evals = evals[-DD:]
# truncate by cutoff
iis = (evals > cutoff)
U = U[:,iis]
#
U = U.reshape((d[2],d[3],U.shape[1]))
res.append(U)
return res
def random_MPS(N, phydim, seed, vdim=1, dtype=np.complex128):
mps = []
np.random.seed(seed)
is_complex = np.issubdtype(dtype, np.complexfloating)
for i in range(N):
if i == 0:
arr = np.random.rand(1, phydim, vdim).astype(dtype)
elif i == N-1:
arr = np.random.rand(vdim, phydim, 1).astype(dtype)
else:
arr = np.random.rand(vdim, phydim, vdim).astype(dtype)
if is_complex:
if i == 0:
arr += 1j * np.random.rand(1, phydim, vdim).astype(dtype)
elif i == N-1:
arr += 1j * np.random.rand(vdim, phydim, 1).astype(dtype)
else:
arr += 1j * np.random.rand(vdim, phydim, vdim).astype(dtype)
mps.append(arr)
return mps
psi0 = random_MPS (12, 2, 15)
psi0 = compress_MPS (psi0, cutoff=1e-18)
print(inner_MPS(psi0,psi0))
I checked the numpy version in my current cytnx environment, and it seems that it is linked to OpenBLAS rather than MKL.
One strange point I noticed is that when installing Cytnx 0.9.7 via conda install -c kaihsinwu cytnx=0.9.7, the numpy that comes along is actually the OpenBLAS version, not the MKL version.
When I follow the instructions from the tenpy documentation to initially install the MKL version of numpy with conda install "libblas=*=*mkl" and then proceed to install cytnx, I noticed that cytnx replaces the previously installed numpy with a version that is dependent on cytnx. You can find the instructions I followed here:
https://tenpy.readthedocs.io/en/latest/install/conda.html
The key infomations of numpy shows below:
(cytnx) [jerrychen@cn7 ~]$ python Python 3.11.12 | packaged by conda-forge | (main, Apr 10 2025, 22:23:25) [GCC 13.3.0] on linux Type "help", "copyright", "credits" or "license" for more information.
import numpy as np np.show_config()
"Machine Information": { "host": { "cpu": "x86_64", "family": "x86_64", "endian": "little", "system": "linux" }, "build": { "cpu": "x86_64", "family": "x86_64", "endian": "little", "system": "linux" } }, "Build Dependencies": { "blas": { "name": "blas", "found": true, "version": "3.9.0", "detection method": "pkgconfig", "include directory": "/data2/home/jerrychen/anaconda3/envs/cytnx/include", "lib directory": "/data2/home/jerrychen/anaconda3/envs/cytnx/lib", "openblas configuration": "unknown", "pc file directory": "/data2/home/jerrychen/anaconda3/envs/cytnx/lib/pkgconfig" }, "lapack": { "name": "dep140567732183440", "found": true, "version": "1.26.4", "detection method": "internal", "include directory": "unknown", "lib directory": "unknown", "openblas configuration": "unknown", "pc file directory": "unknown" } }, "Python Information": { "path": "/data2/home/jerrychen/anaconda3/envs/cytnx/bin/python", "version": "3.11" },
@Heliumky Per offline discussion, could you provide the outputs of these two command on both your workable and not workable device/environment? They show what BLAS libs are actually linked by NumPy and Cyntx.
ldd $(python -c "import numpy._core._multiarray_umath as m; print(m.__file__)") \
| awk '/=>/ {
orig = $1 " => " $3;
cmd = "realpath " $3;
cmd | getline real;
close(cmd);
print orig " (real: " real ")"
}'
ldd $(python -c "import cytnx; print(cytnx.cytnx.__file__)") \
| awk '/=>/ {
orig = $1 " => " $3;
cmd = "realpath " $3;
cmd | getline real;
close(cmd);
print orig " (real: " real ")"
}'
I can reproduce the problem when NumPy links against BLIS and Cytnx links against the MKL Single Dynamic Library.
Here is the output of the minimal code to reproduce the problem when importing NumPy first.
import numpy as np
import cytnx
A = np.array([[1., -3.j], [3.j, 2.]], dtype=np.complex128)
print(np.linalg.matrix_rank(A, hermitian=True))
output:
cytnx/__init__.py:14: UserWarning: numpy and/or scipy are imported before cytnx. Please make sure it support ILP64.
warnings.warn("numpy and/or scipy are imported before cytnx. Please make sure it support ILP64.")
2
And here is the code and the output when importing Cytnx first.
import cytnx
import numpy as np
A = np.array([[1., -3.j], [3.j, 2.]], dtype=np.complex128)
print(np.linalg.matrix_rank(A, hermitian=True))
output:
Intel MKL ERROR: Parameter 3 was incorrect on entry to ZHEEVD.
1
If Cytnx links against OpenBLAS instead (NumPy still links against BLIS), there is no error and the outputs are correct regardless of the order of the import statements.
We need a smarter method to detect the validity of BLAS runtime. Currently, Cytnx only prints an warning when NumPy is imported before Cytnx. This is not a detection and is not enough either. Misalignment of BLAS runtime between Cytnx and other libraries using BLAS breaks other libraries and causes wrong calculation results even when Cytnx is imported first as the example in the previous comment.
Here is how NumPy detects the validity of BLAS runtime now. Obviously, it's not enough because it doesn't catch the case in the previous comment. However, we can adapt the similar strategy to do the detection.
https://github.com/numpy/numpy/blob/470c458aacda6ecd5207c7442cbe720833194c60/numpy/init.py#L430-L454
@Heliumky For make your script workable in your environment, please check which BLAS library that your NumPy actually links against. I think Cytnx, both version 0.9.7 and 1.0.1, ships the acceptable MKL variant on conda now.
@IvanaGyro
Thanks for your helping, I found the tamporary way to solve the partial bug( but still have the bugs) that I reverse the ordering of import cytnx and numpy, and ignoring the warm information in the unworked machine:
warnings.warn("numpy and/or scipy are imported before cytnx. Please make sure it support ILP64.")
Thanks for your minized the test_bug.txt. In the unworked machine, I also try the small script as:
import cytnx
import numpy as np
A = np.array([[1,0],[0,1]])
eigval, eigvec = np.linalg.eigh(A)
print(eigval)
firstly run, it outputs:
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to DSYEVD.
[-2.00000000e+000 -4.39018811e+293]
secondly run, it outputs：
Intel oneMKL ERROR: Parameter 3 was incorrect on entry to DSYEVD.
[-2.00000000e+00 1.34302774e-05]
The eigval of A are [1,1] obviously.
I run the command
import cytnx
cytnx.__cpp_flags__
Output:
'-I/home/jerrychen/anaconda3/include -DUNI_HPTT -fopenmp '
running
cytnx.__cpp_linkflags__
Output:
'/opt/intel/oneapi/mkl/2025.0/lib/libmkl_rt.so -lm -ldl -lm -ldl -Wl,-rpath,/opt/intel/oneapi/mkl/2025.0/lib '
keep running
import numpy as np
np.show_config()
I pick up the important info from Output:
"Build Dependencies": {
"blas": {
"name": "blas",
"found": true,
"version": "3.9.0",
"detection method": "pkgconfig",
"include directory": "/home/jerrychen/anaconda3/envs/cytnx/include",
"lib directory": "/home/jerrychen/anaconda3/envs/cytnx/lib",
"openblas configuration": "unknown",
"pc file directory": "/home/jerrychen/anaconda3/envs/cytnx/lib/pkgconfig"
},
"lapack": {
"name": "lapack",
"found": true,
"version": "3.9.0",
"detection method": "pkgconfig",
"include directory": "/home/jerrychen/anaconda3/envs/cytnx/include",
"lib directory": "/home/jerrychen/anaconda3/envs/cytnx/lib",
"openblas configuration": "unknown",
"pc file directory": "/home/jerrychen/anaconda3/envs/cytnx/lib/pkgconfig"
}
},
running
ldd $(python -c "import numpy._core._multiarray_umath as m; print(m.__file__)") \
| awk '/=>/ {
orig = $1 " => " $3;
cmd = "realpath " $3;
cmd | getline real;
close(cmd);
print orig " (real: " real ")"
}'
Output:
libcblas.so.3 => /home/jerrychen/anaconda3/envs/cytnx/lib/python3.11/site-packages/numpy/_core/../../../../libcblas.so.3 (real: /home/jerrychen/anaconda3/envs/cytnx/lib/libmkl_rt.so.2)
libstdc++.so.6 => /home/jerrychen/anaconda3/envs/cytnx/lib/python3.11/site-packages/numpy/_core/../../../../libstdc++.so.6 (real: /home/jerrychen/anaconda3/envs/cytnx/lib/libstdc++.so.6.0.33)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (real: /usr/lib/x86_64-linux-gnu/libm.so.6)
libgcc_s.so.1 => /home/jerrychen/anaconda3/envs/cytnx/lib/python3.11/site-packages/numpy/_core/../../../../libgcc_s.so.1 (real: /home/jerrychen/anaconda3/envs/cytnx/lib/libgcc_s.so.1)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (real: /usr/lib/x86_64-linux-gnu/libc.so.6)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (real: /usr/lib/x86_64-linux-gnu/libdl.so.2)
running
ldd $(python -c "import cytnx; print(cytnx.cytnx.__file__)") \
| awk '/=>/ {
orig = $1 " => " $3;
cmd = "realpath " $3;
cmd | getline real;
close(cmd);
print orig " (real: " real ")"
}'
output
libpython3.11.so.1.0 => not (real: /home/jerrychen/Downloads/QTTHydrogen-main/onebody/hydrogen/2d_new/not)
libmkl_rt.so.2 => /opt/intel/oneapi/mkl/2025.0/lib/libmkl_rt.so.2 (real: /opt/intel/oneapi/mkl/2025.0/lib/libmkl_rt.so.2)
libstdc++.so.6 => /lib/x86_64-linux-gnu/libstdc++.so.6 (real: /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.33)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (real: /usr/lib/x86_64-linux-gnu/libm.so.6)
libgomp.so.1 => /lib/x86_64-linux-gnu/libgomp.so.1 (real: /usr/lib/x86_64-linux-gnu/libgomp.so.1.0.0)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (real: /usr/lib/x86_64-linux-gnu/libgcc_s.so.1)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (real: /usr/lib/x86_64-linux-gnu/libc.so.6)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (real: /usr/lib/x86_64-linux-gnu/libdl.so.2)
In the worked machine: run:
np.config()
output:
Build Dependencies:
blas:
detection method: pkgconfig
found: true
include directory: /usr/local/include
lib directory: /usr/local/lib
name: openblas64
openblas configuration: USE_64BITINT=1 DYNAMIC_ARCH=1 DYNAMIC_OLDER= NO_CBLAS=
NO_LAPACK= NO_LAPACKE= NO_AFFINITY=1 USE_OPENMP= HASWELL MAX_THREADS=2
pc file directory: /usr/local/lib/pkgconfig
version: 0.3.23.dev
lapack:
detection method: internal
found: true
include directory: unknown
lib directory: unknown
name: dep140680530736656
openblas configuration: unknown
pc file directory: unknown
version: 1.26.1
Compilers:
c:
commands: cc
linker: ld.bfd
name: gcc
version: 10.2.1
c++:
commands: c++
linker: ld.bfd
name: gcc
version: 10.2.1
cython:
commands: cython
linker: cython
name: cython
version: 3.0.3
run
cytnx.__cpp_flags__
output
'-I/home/jerrychen/anaconda3/envs/cytnx/include -DUNI_HPTT -fvisibility-inlines-hidden -fmessage-length=0 -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/jerrychen/anaconda3/envs/cytnx/include -fdebug-prefix-map=/usr/share/miniconda3/envs/test/conda-bld/cytnx_1714963947811/work=/usr/local/src/conda/cytnx-0.9.7 -fdebug-prefix-map=/home/jerrychen/anaconda3/envs/cytnx=/usr/local/src/conda-prefix -fvisibility-inlines-hidden -fmessage-length=0 -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/jerrychen/anaconda3/envs/cytnx/include -fdebug-prefix-map=/usr/share/miniconda3/envs/test/conda-bld/cytnx_1714963947811/work=/usr/local/src/conda/cytnx-0.9.7 -fdebug-prefix-map=/home/jerrychen/anaconda3/envs/cytnx=/usr/local/src/conda-prefix -DVERSION_INFO=\\"0.9.7\\" -fopenmp '
run
cytnx.__cpp_linkflags__
output
'/home/jerrychen/anaconda3/envs/cytnx/lib/libmkl_rt.so -lpthread -lm -ldl -lpthread -lm -ldl -Wl,-rpath,/home/jerrychen/anaconda3/envs/cytnx/lib /home/jerrychen/anaconda3/envs/cytnx/lib/python3.10/site-packages/cytnx/hptt/lib/libhptt.a'
run
ldd $(python -c "import numpy._core._multiarray_umath as m; print(m.__file__)") \
| awk '/=>/ {
orig = $1 " => " $3;
cmd = "realpath " $3;
cmd | getline real;
close(cmd);
print orig " (real: " real ")"
}'
output
libopenblas64_p-r0-0cf96a72.3.23.dev.so => /home/jerrychen/.local/lib/python3.10/site-packages/numpy/core/../../numpy.libs/libopenblas64_p-r0-0cf96a72.3.23.dev.so (real: /home/jerrychen/.local/lib/python3.10/site-packages/numpy.libs/libopenblas64_p-r0-0cf96a72.3.23.dev.so)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (real: /usr/lib/x86_64-linux-gnu/libm.so.6)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (real: /usr/lib/x86_64-linux-gnu/libgcc_s.so.1)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (real: /usr/lib/x86_64-linux-gnu/libc.so.6)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (real: /usr/lib/x86_64-linux-gnu/libpthread.so.0)
libgfortran-040039e1.so.5.0.0 => /home/jerrychen/.local/lib/python3.10/site-packages/numpy/core/../../numpy.libs/libgfortran-040039e1.so.5.0.0 (real: /home/jerrychen/.local/lib/python3.10/site-packages/numpy.libs/libgfortran-040039e1.so.5.0.0)
libquadmath-96973f99.so.0.0.0 => /home/jerrychen/.local/lib/python3.10/site-packages/numpy/core/../../numpy.libs/libquadmath-96973f99.so.0.0.0 (real: /home/jerrychen/.local/lib/python3.10/site-packages/numpy.libs/libquadmath-96973f99.so.0.0.0)
libz.so.1 => /lib/x86_64-linux-gnu/libz.so.1 (real: /usr/lib/x86_64-linux-gnu/libz.so.1.2.11)
run
ldd $(python -c "import cytnx; print(cytnx.cytnx.__file__)") \
| awk '/=>/ {
orig = $1 " => " $3;
cmd = "realpath " $3;
cmd | getline real;
close(cmd);
print orig " (real: " real ")"
}'
output
libpython3.10.so.1.0 => /home/jerrychen/anaconda3/envs/cytnx/lib/python3.10/site-packages/cytnx/../../../libpython3.10.so.1.0 (real: /home/jerrychen/anaconda3/envs/cytnx/lib/libpython3.10.so.1.0)
libmkl_rt.so.2 => /home/jerrychen/anaconda3/envs/cytnx/lib/python3.10/site-packages/cytnx/../../../libmkl_rt.so.2 (real: /home/jerrychen/anaconda3/envs/cytnx/lib/libmkl_rt.so.2)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (real: /usr/lib/x86_64-linux-gnu/libpthread.so.0)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (real: /usr/lib/x86_64-linux-gnu/libm.so.6)
libgomp.so.1 => /home/jerrychen/anaconda3/envs/cytnx/lib/python3.10/site-packages/cytnx/../../../libgomp.so.1 (real: /home/jerrychen/anaconda3/envs/cytnx/lib/libomp.so)
libstdc++.so.6 => /home/jerrychen/anaconda3/envs/cytnx/lib/python3.10/site-packages/cytnx/../../../libstdc++.so.6 (real: /home/jerrychen/anaconda3/envs/cytnx/lib/libstdc++.so.6.0.33)
libgcc_s.so.1 => /home/jerrychen/anaconda3/envs/cytnx/lib/python3.10/site-packages/cytnx/../../../libgcc_s.so.1 (real: /home/jerrychen/anaconda3/envs/cytnx/lib/libgcc_s.so.1)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (real: /usr/lib/x86_64-linux-gnu/libc.so.6)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (real: /usr/lib/x86_64-linux-gnu/libdl.so.2)
libutil.so.1 => /lib/x86_64-linux-gnu/libutil.so.1 (real: /usr/lib/x86_64-linux-gnu/libutil.so.1)
librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (real: /usr/lib/x86_64-linux-gnu/librt.so.1)
Finally, I run my small script
(cytnx) jerrychen@jerrychen-X550VX:~$ python3 test.py
[1. 1.]
Thank you again.
For you unworkable machine, it seems that MKL 2024.2.2 is not compatible with MKL 2025 even if they are both MKL... And the other problem is that CMake doesn't adopt the MKL installed in your conda environment, instead, it linked to your system MKL.
For you workable machine, somehow it works even NumPy links against OpenBLAS and Cytnx links MKL. Did you get different outputs when reversing the import statements on your workable machine?
To make CMake use the MKL library installed in your conda environment, try
conda install cxx-compiler cmake
and
- Remove
build/folder cmake --preset mkl-cpucmake --build --preset mkl-cpu
I’ve found a reliable way to resolve the issue.
First, I followed the official documentation to install cytnx using conda.
Second, I used pip3 uninstall numpy to remove only the NumPy package. This avoids the problem with conda remove numpy, which would otherwise uninstall all packages that depend on numpy, including cytnx.
Third, I reinstalled numpy using pip3 install numpy. After that, running np.show_config() confirms that numpy is properly configured.
},
"Build Dependencies": {
"blas": {
"name": "scipy-openblas",
"found": true,
"version": "0.3.28",
"detection method": "pkgconfig",
"include directory": "/opt/_internal/cpython-3.11.10/lib/python3.11/site-packages/scipy_openblas64/include",
"lib directory": "/opt/_internal/cpython-3.11.10/lib/python3.11/site-packages/scipy_openblas64/lib",
"openblas configuration": "OpenBLAS 0.3.28 USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell MAX_THREADS=64",
"pc file directory": "/project/.openblas"
},
"lapack": {
"name": "scipy-openblas",
"found": true,
"version": "0.3.28",
"detection method": "pkgconfig",
"include directory": "/opt/_internal/cpython-3.11.10/lib/python3.11/site-packages/scipy_openblas64/include",
"lib directory": "/opt/_internal/cpython-3.11.10/lib/python3.11/site-packages/scipy_openblas64/lib",
"openblas configuration": "OpenBLAS 0.3.28 USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell MAX_THREADS=64",
"pc file directory": "/project/.openblas"
}
}
With this setup, my program now runs smoothly. Thank you for your helping.
I suggest we ship binaries with mkl_lp64 instead of mkl_ilp64 by default. Users can decide to compile the ilp64 version if that is real necessary.
I suggest we detect the validity of BLAS at runtime like what numpy does. We may add a testcase like this
A = np.array([[1., -3.j], [3.j, 2.]], dtype=np.complex128)
print(np.linalg.matrix_rank(A, hermitian=True))
but a Cytnx version.