Opened 8 months ago

Last modified 11 days ago

#26349 needs_work defect

Memory leaks with matrix multiplication over GF(2)

Reported by: gh-antoine06 Owned by:
Priority: major Milestone: sage-8.8
Component: linear algebra Keywords: memory leaks
Cc: Merged in:
Authors: Reviewers: Travis Scrimshaw
Report Upstream: N/A Work issues:
Branch: Commit:
Dependencies: Stopgaps:

Description (last modified by tmonteil)

As reported on this ask question:

n = 8
X = zero_vector(GF(2), n)
M = zero_matrix(GF(2), n, n)

for _ in range(10000000):
    Y = M * X

leads to:

m4ri_mm_malloc: malloc returned NULL
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-1-04da27075a1c> in <module>()
      4 
      5 for _ in range(Integer(10000000)):
----> 6     Y = M * X
      7 

/opt/sagemath/sage-source/local/lib/python2.7/site-packages/sage/structure/element.pyx in sage.structure.element.Matrix.__mul__ (build/cythonized/sage/structure/element.c:23537)()
   3666             return (<Matrix>left)._matrix_times_matrix_(<Matrix>right)
   3667         if BOTH_ARE_ELEMENT(cl):
-> 3668             return coercion_model.bin_op(left, right, mul)
   3669 
   3670         cdef long value

/opt/sagemath/sage-source/local/lib/python2.7/site-packages/sage/structure/coerce.pyx in sage.structure.coerce.CoercionModel_cache_maps.bin_op (build/cythonized/sage/structure/coerce.c:9757)()
   1173             action = self.get_action(xp, yp, op, x, y)
   1174             if action is not None:
-> 1175                 return (<Action>action)._call_(x, y)
   1176 
   1177         try:

/opt/sagemath/sage-source/local/lib/python2.7/site-packages/sage/matrix/action.pyx in sage.matrix.action.MatrixVectorAction._call_ (build/cythonized/sage/matrix/action.c:5997)()
    325             else:
    326                 v = v.dense_vector()
--> 327         return A._matrix_times_vector_(v)
    328 
    329 

/opt/sagemath/sage-source/local/lib/python2.7/site-packages/sage/matrix/matrix_mod2_dense.pyx in sage.matrix.matrix_mod2_dense.Matrix_mod2_dense._matrix_times_vector_ (build/cythonized/sage/matrix/matrix_mod2_dense.c:6487)()
    573             return VS.zero()
    574         cdef Vector_mod2_dense c = Vector_mod2_dense.__new__(Vector_mod2_dense)
--> 575         sig_str("matrix allocation failed")
    576         c._init(self._nrows, VS)
    577         c._entries = mzd_init(1, self._nrows)

RuntimeError: matrix allocation failed

Change History (5)

comment:1 Changed 8 months ago by tmonteil

  • Description modified (diff)

comment:2 Changed 8 months ago by nbruin

It would seem the memory management on these objects is rather fundamentally broken. Also with lines like Y= M+M or Y=X+X I am seeing similar leaking behaviour.

comment:3 Changed 2 weeks ago by gh-black-stones

  • Milestone changed from sage-8.4 to sage-duplicate/invalid/wontfix
  • Status changed from new to needs_review

Tested this on Sage 8.7, macOS Mojave and Ubuntu 19.04, and the problem seems to be fixed.

If someone else could confirm this on another OS (the person on ask used debian buster) then we can probably close this as wontfix.

comment:4 Changed 2 weeks ago by tscrim

  • Reviewers set to Travis Scrimshaw
  • Status changed from needs_review to positive_review

Confirmed on Ubuntu 16.04 that I am not seeing a leak.

comment:5 Changed 11 days ago by nbruin

  • Milestone changed from sage-duplicate/invalid/wontfix to sage-8.8
  • Status changed from positive_review to needs_work

I'm still seeing suspicious behaviour on sage 8.8.beta3 / python3, on Fedora:

If I set a vmem limit of 10 gigabyte and execute

n = 1000
X = zero_vector(GF(2), n)
M = zero_matrix(GF(2), n, n)

for _ in range(10000000):
    Y = M + M

sage still crashes for me (whereas a small number of iterations does run without problem).

Note: See TracTickets for help on using tickets.