Debian Patches
Status for libgpuarray/0.7.6-13
| Patch | Description | Author | Forwarded | Bugs | Origin | Last update |
|---|---|---|---|---|---|---|
| typos.patch | Fix spelling | Rebecca N. Palmer <rebecca_palmer@zoho.com> | yes | lintian (mostly) | ||
| 581_ulong_sizet.patch | Don't assume size_t and cl_ulong are the same type Fixes the error pygpu.gpuarray.GpuArrayException: b'clGetDeviceInfo(id, CL_DEVICE_LOCAL_MEM_SIZE, sizeof(sz), &sz, NULL): Invalid value' on accessing some memory size parameters on 32 bit systems. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | yes | upstream | ||
| fp64_errors.patch | More user-friendly "no double support" errors Messages that actually say that that's the problem, and the right error type for the tests to recognize as 'skip' not 'fail'. Add 'skip if unsupported' to the double-using tests that don't already have it. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | yes | |||
| cache_exception_safety.patch | Don't corrupt lru_cache on exceptions If user_function raises an exception, avoid adding a key to last_use and not cache, as such keys trigger KeyError when purged. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | yes | |||
| gemmBatch.patch | Fix stack smashing crash in *gemmBatch Calling ARRAY_INIT more than 3 times without resetting num_ev runs off the end of evl. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | yes | |||
| 920497_ger_workaround.patch | Use a whole-block size for test_ger test_ger crashes (not just fails) under pocl with the default size. This bug appears to be further down the stack (see linked report), not in libgpuarray itself. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | not-needed | debian | ||
| clblas_error_reporting.patch | Actually use all the clblasStatus-to-string table Error code -1024 is the lowest (most negative), not the highest |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | yes | |||
| tri_check_itemsize.patch | Fail unsupported tril/u instead of returning nonsense The extract_tri kernel assumes 4 bytes per entry when deciding which memory addresses to zero out: if the actual data type is another size, it will "succeed" but produce nonsense. There are real fixes for this upstream ( https://github.com/Theano/libgpuarray/commit/07cd4ad56054c279442ee28413b26939f4c03632 plus https://github.com/Theano/libgpuarray/pull/584), but I am not sure whether those are appropriate as a Debian patch. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | not-needed | |||
| test_dot_doubleref.patch | Fix intermittent test_dot failure test_dot sometimes fails with a slightly out-of-tolerance result, and comparing to the double precision result suggests most of the rounding error is in the CPU sdot used as a reference, not libgpuarray's result. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | yes | |||
| 877316_gemm_workaround.patch | Allow running the tests on beignet-opencl-icd The gemm tests crash (not just fail) on beignet due to the linked clblas bug. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | not-needed | debian | ||
| use_python3.patch | Use Python 3 for build scripts | Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | debian | ||
| typos2.patch | Fix spelling and grammar | Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | |||
| clinfo.patch | Add note on clinfo | Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | |||
| disable_out_of_order.patch | Stop using out-of-order queues, as they break clblas Avoids wrong results in gemm(batch) and possibly other clblas functions. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | debian | ||
| 992673_gemv_workaround.patch | 0*NaN is not 0 clblas gemv skips the +beta*Y if beta is 0, but clblast gemv doesn't. This matters if Y initially contains NaNs, which is likely if it is uninitialized. (My reading of the spec is that this is a bug in clblast, but I'm not certain of that.) |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | debian | ||
| sphinx42_compat.patch | Don't use add_stylesheet | Graham Inggs | no | debian | ||
| pytest.patch | Use pytest instead of nose | Rebecca N. Palmer <rebecca_palmer@zoho.com> | invalid | debian | ||
| 992673_ger_ignorecrash.patch | Ignore test_ger/test_gemv exception on arm64 This is an OpenCL compile failure from inside clblast, so I consider it likely that the underlying issue is in pocl or clblast, not here. This patch intentionally only ignores exceptions, not wrong answers. |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | not-needed | debian | ||
| pytest_yield.patch | Stop using yield tests, which pytest no longer has pytest.parameterized would be a better replacement, but this should at least run them |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | debian | ||
| setuptools_compat.patch | Import setuptools before distutils This is necessary because current setuptools replaces distutils |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | debian | ||
| 1026345_updateifcopy.patch | Don't try to read numpy's updateifcopy (it no longer exists) |
Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | debian | ||
| 102635_strides0size0.patch | Allow numpy strides of an 0-elements array to be 0 | Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | debian | ||
| numpy1p24.patch | Don't use no-longer-existing np.bool | Rebecca N. Palmer <rebecca_palmer@zoho.com> | no | debian |
All known versions for source package 'libgpuarray'
- 0.7.6-13 (bookworm)