Project comparisons#

Here is a list of NumPy APIs and corresponding cuNumeric implementations.

A dot in the cunumeric column denotes that cuNumeric implementation is not provided yet. We welcome contributions for these functions.

NumPy vs cuNumeric APIs#

Module-Level#

NumPy	cunumeric	single-GPU/CPU	multi-GPU/CPU
`numpy.all`	`cunumeric.all`	✓	✓
`numpy.allclose`	`cunumeric.allclose`	✓	✓
`numpy.amax`	`cunumeric.amax`	✓	✓
`numpy.amin`	`cunumeric.amin`	✓	✓
`numpy.angle`
`numpy.any`	`cunumeric.any`	✓	✓
`numpy.append`	`cunumeric.append`	✓	✓
`numpy.apply_along_axis`
`numpy.apply_over_axes`
`numpy.arange`	`cunumeric.arange`	✓	✓
`numpy.argmax`	`cunumeric.argmax`	✓	✓
`numpy.argmin`	`cunumeric.argmin`	✓	✓
`numpy.argpartition`	`cunumeric.argpartition`	✓	✓
`numpy.argsort`	`cunumeric.argsort`	✓	✓
`numpy.argwhere`	`cunumeric.argwhere`	✓	✓
`numpy.around`
`numpy.array`	`cunumeric.array`	✓	✓
`numpy.array2string`
`numpy.array_equal`	`cunumeric.array_equal`	✓	✓
`numpy.array_equiv`
`numpy.array_repr`
`numpy.array_split`	`cunumeric.array_split`	✓	✓
`numpy.array_str`
`numpy.asanyarray`
`numpy.asarray`	`cunumeric.asarray`	✓	✓
`numpy.asarray_chkfinite`
`numpy.ascontiguousarray`
`numpy.asfarray`
`numpy.asfortranarray`
`numpy.asmatrix`
`numpy.atleast_1d`	`cunumeric.atleast_1d`	✓	✓
`numpy.atleast_2d`	`cunumeric.atleast_2d`	✓	✓
`numpy.atleast_3d`	`cunumeric.atleast_3d`	✓	✓
`numpy.average`
`numpy.bartlett`	`cunumeric.bartlett`	✓	✓
`numpy.base_repr`
`numpy.binary_repr`
`numpy.bincount`	`cunumeric.bincount`	✓	✓
`numpy.blackman`	`cunumeric.blackman`	✓	✓
`numpy.block`	`cunumeric.block`	✓	✓
`numpy.bmat`
`numpy.broadcast_arrays`
`numpy.broadcast_shapes`
`numpy.broadcast_to`
`numpy.busday_count`
`numpy.busday_offset`
`numpy.byte_bounds`
`numpy.can_cast`
`numpy.choose`	`cunumeric.choose`	✓	✓
`numpy.clip`	`cunumeric.clip`	✓	✓
`numpy.column_stack`	`cunumeric.column_stack`	✓	✓
`numpy.common_type`
`numpy.compress`	`cunumeric.compress`	✓	✓
`numpy.concatenate`	`cunumeric.concatenate`	✓	✓
`numpy.convolve`	`cunumeric.convolve`	✓	✓
`numpy.copy`	`cunumeric.copy`	✓	✓
`numpy.copyto`
`numpy.corrcoef`
`numpy.correlate`
`numpy.count_nonzero`	`cunumeric.count_nonzero`	✓	✓
`numpy.cov`
`numpy.cross`
`numpy.cumprod`	`cunumeric.cumprod`	✓	✓
`numpy.cumsum`	`cunumeric.cumsum`	✓	✓
`numpy.datetime_as_string`
`numpy.datetime_data`
`numpy.delete`
`numpy.deprecate`
`numpy.deprecate_with_doc`
`numpy.diag`	`cunumeric.diag`	✓	✓
`numpy.diag_indices`	`cunumeric.diag_indices`	✓	✓
`numpy.diag_indices_from`	`cunumeric.diag_indices_from`	✓	✓
`numpy.diagflat`
`numpy.diagonal`	`cunumeric.diagonal`	✓	✓
`numpy.diff`
`numpy.digitize`
`numpy.disp`
`numpy.dot`	`cunumeric.dot`	✓	✓
`numpy.dsplit`	`cunumeric.dsplit`	✓	✓
`numpy.dstack`	`cunumeric.dstack`	✓	✓
`numpy.ediff1d`
`numpy.einsum`	`cunumeric.einsum`	✓	✓
`numpy.einsum_path`	`cunumeric.einsum_path`	✓	✓
`numpy.empty`	`cunumeric.empty`	✓	✓
`numpy.empty_like`	`cunumeric.empty_like`	✓	✓
`numpy.expand_dims`
`numpy.extract`	`cunumeric.extract`	✓	✓
`numpy.eye`	`cunumeric.eye`	✓	✓
`numpy.fill_diagonal`	`cunumeric.fill_diagonal`	✓	✓
`numpy.find_common_type`
`numpy.fix`
`numpy.flatnonzero`	`cunumeric.flatnonzero`	✓	✓
`numpy.flip`	`cunumeric.flip`	✓	❌
`numpy.fliplr`
`numpy.flipud`
`numpy.format_float_positional`
`numpy.format_float_scientific`
`numpy.from_dlpack`
`numpy.frombuffer`
`numpy.fromfile`
`numpy.fromfunction`
`numpy.fromiter`
`numpy.frompyfunc`
`numpy.fromregex`
`numpy.fromstring`
`numpy.full`	`cunumeric.full`	✓	✓
`numpy.full_like`	`cunumeric.full_like`	✓	✓
`numpy.genfromtxt`
`numpy.geomspace`
`numpy.get_include`
`numpy.get_printoptions`
`numpy.getbufsize`
`numpy.geterr`
`numpy.geterrcall`
`numpy.geterrobj`
`numpy.gradient`
`numpy.hamming`	`cunumeric.hamming`	✓	✓
`numpy.hanning`	`cunumeric.hanning`	✓	✓
`numpy.histogram`
`numpy.histogram2d`
`numpy.histogram_bin_edges`
`numpy.histogramdd`
`numpy.hsplit`	`cunumeric.hsplit`	✓	✓
`numpy.hstack`	`cunumeric.hstack`	✓	✓
`numpy.i0`
`numpy.identity`	`cunumeric.identity`	✓	✓
`numpy.imag`	`cunumeric.imag`	✓	✓
`numpy.in1d`
`numpy.indices`	`cunumeric.indices`	✓	✓
`numpy.info`
`numpy.inner`	`cunumeric.inner`	✓	✓
`numpy.insert`
`numpy.interp`
`numpy.intersect1d`
`numpy.is_busday`
`numpy.isclose`	`cunumeric.isclose`	✓	✓
`numpy.iscomplex`	`cunumeric.iscomplex`	✓	✓
`numpy.iscomplexobj`	`cunumeric.iscomplexobj`	✓	✓
`numpy.isfortran`
`numpy.isin`
`numpy.isneginf`	`cunumeric.isneginf`	✓	✓
`numpy.isposinf`	`cunumeric.isposinf`	✓	✓
`numpy.isreal`	`cunumeric.isreal`	✓	✓
`numpy.isrealobj`	`cunumeric.isrealobj`	✓	✓
`numpy.isscalar`	`cunumeric.isscalar`	✓	✓
`numpy.issctype`
`numpy.issubclass_`
`numpy.issubdtype`
`numpy.issubsctype`
`numpy.ix_`
`numpy.kaiser`	`cunumeric.kaiser`	✓	✓
`numpy.kron`
`numpy.lexsort`
`numpy.linspace`	`cunumeric.linspace`	✓	✓
`numpy.load`
`numpy.loadtxt`
`numpy.logspace`
`numpy.lookfor`
`numpy.mask_indices`	`cunumeric.mask_indices`	✓	✓
`numpy.mat`
`numpy.maximum_sctype`
`numpy.may_share_memory`
`numpy.mean`	`cunumeric.mean`	✓	✓
`numpy.median`
`numpy.meshgrid`
`numpy.min_scalar_type`
`numpy.mintypecode`
`numpy.moveaxis`	`cunumeric.moveaxis`	✓	✓
`numpy.msort`	`cunumeric.msort`	✓	✓
`numpy.nan_to_num`
`numpy.nanargmax`
`numpy.nanargmin`
`numpy.nancumprod`	`cunumeric.nancumprod`	✓	✓
`numpy.nancumsum`	`cunumeric.nancumsum`	✓	✓
`numpy.nanmax`
`numpy.nanmean`
`numpy.nanmedian`
`numpy.nanmin`
`numpy.nanpercentile`
`numpy.nanprod`
`numpy.nanquantile`
`numpy.nanstd`
`numpy.nansum`
`numpy.nanvar`
`numpy.ndim`	`cunumeric.ndim`	✓	✓
`numpy.nested_iters`
`numpy.nonzero`	`cunumeric.nonzero`	✓	✓
`numpy.obj2sctype`
`numpy.ones`	`cunumeric.ones`	✓	✓
`numpy.ones_like`	`cunumeric.ones_like`	✓	✓
`numpy.outer`	`cunumeric.outer`	✓	✓
`numpy.packbits`	`cunumeric.packbits`	✓	✓
`numpy.pad`
`numpy.partition`	`cunumeric.partition`	✓	✓
`numpy.percentile`
`numpy.piecewise`
`numpy.place`	`cunumeric.place`	✓	✓
`numpy.poly`
`numpy.polyadd`
`numpy.polyder`
`numpy.polydiv`
`numpy.polyfit`
`numpy.polyint`
`numpy.polymul`
`numpy.polysub`
`numpy.polyval`
`numpy.printoptions`
`numpy.prod`	`cunumeric.prod`	✓	✓
`numpy.promote_types`
`numpy.ptp`
`numpy.put`	`cunumeric.put`	✓	✓
`numpy.put_along_axis`	`cunumeric.put_along_axis`	✓	✓
`numpy.putmask`	`cunumeric.putmask`	✓	✓
`numpy.quantile`
`numpy.ravel`	`cunumeric.ravel`	✓	✓
`numpy.ravel_multi_index`
`numpy.real`	`cunumeric.real`	✓	✓
`numpy.real_if_close`
`numpy.repeat`	`cunumeric.repeat`	✓	✓
`numpy.require`
`numpy.reshape`	`cunumeric.reshape`	✓	✓
`numpy.resize`
`numpy.result_type`
`numpy.roll`
`numpy.rollaxis`
`numpy.roots`
`numpy.rot90`
`numpy.round_`
`numpy.row_stack`	`cunumeric.row_stack`	✓	✓
`numpy.save`
`numpy.savetxt`
`numpy.savez`
`numpy.savez_compressed`
`numpy.sctype2char`
`numpy.searchsorted`	`cunumeric.searchsorted`	✓	✓
`numpy.select`
`numpy.set_printoptions`
`numpy.set_string_function`
`numpy.setbufsize`
`numpy.setdiff1d`
`numpy.seterr`
`numpy.seterrcall`
`numpy.seterrobj`
`numpy.setxor1d`
`numpy.shape`	`cunumeric.shape`	✓	✓
`numpy.shares_memory`
`numpy.show_config`
`numpy.show_runtime`
`numpy.sinc`
`numpy.sort`	`cunumeric.sort`	✓	✓
`numpy.sort_complex`	`cunumeric.sort_complex`	✓	✓
`numpy.source`
`numpy.split`	`cunumeric.split`	✓	✓
`numpy.squeeze`	`cunumeric.squeeze`	✓	✓
`numpy.stack`	`cunumeric.stack`	✓	✓
`numpy.std`
`numpy.sum`	`cunumeric.sum`	✓	✓
`numpy.swapaxes`	`cunumeric.swapaxes`	✓	✓
`numpy.take`	`cunumeric.take`	✓	✓
`numpy.take_along_axis`	`cunumeric.take_along_axis`	✓	✓
`numpy.tensordot`	`cunumeric.tensordot`	✓	✓
`numpy.tile`	`cunumeric.tile`	✓	✓
`numpy.trace`	`cunumeric.trace`	✓	✓
`numpy.transpose`	`cunumeric.transpose`	✓	✓
`numpy.trapz`
`numpy.tri`	`cunumeric.tri`	✓	✓
`numpy.tril`	`cunumeric.tril`	✓	✓
`numpy.tril_indices`	`cunumeric.tril_indices`	✓	✓
`numpy.tril_indices_from`	`cunumeric.tril_indices_from`	✓	✓
`numpy.trim_zeros`
`numpy.triu`	`cunumeric.triu`	✓	✓
`numpy.triu_indices`	`cunumeric.triu_indices`	✓	✓
`numpy.triu_indices_from`	`cunumeric.triu_indices_from`	✓	✓
`numpy.typename`
`numpy.union1d`
`numpy.unique`	`cunumeric.unique`	✓	✓
`numpy.unpackbits`	`cunumeric.unpackbits`	✓	✓
`numpy.unravel_index`
`numpy.unwrap`
`numpy.vander`
`numpy.var`
`numpy.vdot`	`cunumeric.vdot`	✓	✓
`numpy.vsplit`	`cunumeric.vsplit`	✓	✓
`numpy.vstack`	`cunumeric.vstack`	✓	✓
`numpy.where`	`cunumeric.where`	✓	✓
`numpy.who`
`numpy.zeros`	`cunumeric.zeros`	✓	✓
`numpy.zeros_like`	`cunumeric.zeros_like`	✓	✓

Summary

Number of NumPy functions: 288

Number of functions covered by cunumeric: 122

Ufuncs#

NumPy	cunumeric	single-GPU/CPU	multi-GPU/CPU
`numpy.absolute`	`cunumeric.absolute`	✓	✓
`numpy.add`	`cunumeric.add`	✓	✓
`numpy.arccos`	`cunumeric.arccos`	✓	✓
`numpy.arccosh`	`cunumeric.arccosh`	✓	✓
`numpy.arcsin`	`cunumeric.arcsin`	✓	✓
`numpy.arcsinh`	`cunumeric.arcsinh`	✓	✓
`numpy.arctan`	`cunumeric.arctan`	✓	✓
`numpy.arctan2`	`cunumeric.arctan2`	✓	✓
`numpy.arctanh`	`cunumeric.arctanh`	✓	✓
`numpy.bitwise_and`	`cunumeric.bitwise_and`	✓	✓
`numpy.bitwise_or`	`cunumeric.bitwise_or`	✓	✓
`numpy.bitwise_xor`	`cunumeric.bitwise_xor`	✓	✓
`numpy.cbrt`	`cunumeric.cbrt`	✓	✓
`numpy.ceil`	`cunumeric.ceil`	✓	✓
`numpy.conj`	`cunumeric.conj`	✓	✓
`numpy.conjugate`	`cunumeric.conjugate`	✓	✓
`numpy.copysign`	`cunumeric.copysign`	✓	✓
`numpy.cos`	`cunumeric.cos`	✓	✓
`numpy.cosh`	`cunumeric.cosh`	✓	✓
`numpy.deg2rad`	`cunumeric.deg2rad`	✓	✓
`numpy.degrees`	`cunumeric.degrees`	✓	✓
`numpy.divide`	`cunumeric.divide`	✓	✓
`numpy.divmod`
`numpy.equal`	`cunumeric.equal`	✓	✓
`numpy.exp`	`cunumeric.exp`	✓	✓
`numpy.exp2`	`cunumeric.exp2`	✓	✓
`numpy.expm1`	`cunumeric.expm1`	✓	✓
`numpy.fabs`	`cunumeric.fabs`	✓	✓
`numpy.float_power`	`cunumeric.float_power`	✓	✓
`numpy.floor`	`cunumeric.floor`	✓	✓
`numpy.floor_divide`	`cunumeric.floor_divide`	✓	✓
`numpy.fmax`	`cunumeric.fmax`	✓	✓
`numpy.fmin`	`cunumeric.fmin`	✓	✓
`numpy.fmod`	`cunumeric.fmod`	✓	✓
`numpy.frexp`	`cunumeric.frexp`	✓	✓
`numpy.gcd`	`cunumeric.gcd`	✓	✓
`numpy.greater`	`cunumeric.greater`	✓	✓
`numpy.greater_equal`	`cunumeric.greater_equal`	✓	✓
`numpy.heaviside`
`numpy.hypot`	`cunumeric.hypot`	✓	✓
`numpy.invert`	`cunumeric.invert`	✓	✓
`numpy.isfinite`	`cunumeric.isfinite`	✓	✓
`numpy.isinf`	`cunumeric.isinf`	✓	✓
`numpy.isnan`	`cunumeric.isnan`	✓	✓
`numpy.isnat`
`numpy.lcm`	`cunumeric.lcm`	✓	✓
`numpy.ldexp`	`cunumeric.ldexp`	✓	✓
`numpy.left_shift`	`cunumeric.left_shift`	✓	✓
`numpy.less`	`cunumeric.less`	✓	✓
`numpy.less_equal`	`cunumeric.less_equal`	✓	✓
`numpy.log`	`cunumeric.log`	✓	✓
`numpy.log10`	`cunumeric.log10`	✓	✓
`numpy.log1p`	`cunumeric.log1p`	✓	✓
`numpy.log2`	`cunumeric.log2`	✓	✓
`numpy.logaddexp`	`cunumeric.logaddexp`	✓	✓
`numpy.logaddexp2`	`cunumeric.logaddexp2`	✓	✓
`numpy.logical_and`	`cunumeric.logical_and`	✓	✓
`numpy.logical_not`	`cunumeric.logical_not`	✓	✓
`numpy.logical_or`	`cunumeric.logical_or`	✓	✓
`numpy.logical_xor`	`cunumeric.logical_xor`	✓	✓
`numpy.matmul`	`cunumeric.matmul`	✓	✓
`numpy.maximum`	`cunumeric.maximum`	✓	✓
`numpy.minimum`	`cunumeric.minimum`	✓	✓
`numpy.mod`	`cunumeric.mod`	✓	✓
`numpy.modf`	`cunumeric.modf`	✓	✓
`numpy.multiply`	`cunumeric.multiply`	✓	✓
`numpy.negative`	`cunumeric.negative`	✓	✓
`numpy.nextafter`	`cunumeric.nextafter`	✓	✓
`numpy.not_equal`	`cunumeric.not_equal`	✓	✓
`numpy.positive`	`cunumeric.positive`	✓	✓
`numpy.power`	`cunumeric.power`	✓	✓
`numpy.rad2deg`	`cunumeric.rad2deg`	✓	✓
`numpy.radians`	`cunumeric.radians`	✓	✓
`numpy.reciprocal`	`cunumeric.reciprocal`	✓	✓
`numpy.remainder`	`cunumeric.remainder`	✓	✓
`numpy.right_shift`	`cunumeric.right_shift`	✓	✓
`numpy.rint`	`cunumeric.rint`	✓	✓
`numpy.sign`	`cunumeric.sign`	✓	✓
`numpy.signbit`	`cunumeric.signbit`	✓	✓
`numpy.sin`	`cunumeric.sin`	✓	✓
`numpy.sinh`	`cunumeric.sinh`	✓	✓
`numpy.spacing`
`numpy.sqrt`	`cunumeric.sqrt`	✓	✓
`numpy.square`	`cunumeric.square`	✓	✓
`numpy.subtract`	`cunumeric.subtract`	✓	✓
`numpy.tan`	`cunumeric.tan`	✓	✓
`numpy.tanh`	`cunumeric.tanh`	✓	✓
`numpy.true_divide`	`cunumeric.true_divide`	✓	✓
`numpy.trunc`	`cunumeric.trunc`	✓	✓

Summary

Number of NumPy functions: 89

Number of functions covered by cunumeric: 85

Multi-Dimensional Array#

NumPy	cunumeric	single-GPU/CPU	multi-GPU/CPU
`numpy.ndarray.all()`	`cunumeric.ndarray.all()`	✓	✓
`numpy.ndarray.any()`	`cunumeric.ndarray.any()`	✓	✓
`numpy.ndarray.argmax()`	`cunumeric.ndarray.argmax()`	✓	✓
`numpy.ndarray.argmin()`	`cunumeric.ndarray.argmin()`	✓	✓
`numpy.ndarray.argpartition()`	`cunumeric.ndarray.argpartition()`	✓	✓
`numpy.ndarray.argsort()`	`cunumeric.ndarray.argsort()`	✓	✓
`numpy.ndarray.astype()`	`cunumeric.ndarray.astype()`	✓	✓
`numpy.ndarray.byteswap()`
`numpy.ndarray.choose()`	`cunumeric.ndarray.choose()`	✓	✓
`numpy.ndarray.clip()`	`cunumeric.ndarray.clip()`	✓	✓
`numpy.ndarray.compress()`	`cunumeric.ndarray.compress()`	✓	✓
`numpy.ndarray.conj()`	`cunumeric.ndarray.conj()`	✓	✓
`numpy.ndarray.conjugate()`	`cunumeric.ndarray.conjugate()`	✓	✓
`numpy.ndarray.copy()`	`cunumeric.ndarray.copy()`	✓	✓
`numpy.ndarray.cumprod()`	`cunumeric.ndarray.cumprod()`	❌	❌
`numpy.ndarray.cumsum()`	`cunumeric.ndarray.cumsum()`	❌	❌
`numpy.ndarray.diagonal()`	`cunumeric.ndarray.diagonal()`	✓	✓
`numpy.ndarray.dot()`	`cunumeric.ndarray.dot()`	✓	✓
`numpy.ndarray.dump()`	`cunumeric.ndarray.dump()`	❌	❌
`numpy.ndarray.dumps()`	`cunumeric.ndarray.dumps()`	✓	✓
`numpy.ndarray.fill()`	`cunumeric.ndarray.fill()`	✓	✓
`numpy.ndarray.flatten()`	`cunumeric.ndarray.flatten()`	✓	✓
`numpy.ndarray.getfield()`	`cunumeric.ndarray.getfield()`	❌	❌
`numpy.ndarray.item()`	`cunumeric.ndarray.item()`	✓	✓
`numpy.ndarray.itemset()`	`cunumeric.ndarray.itemset()`	✓	✓
`numpy.ndarray.mean()`	`cunumeric.ndarray.mean()`	✓	✓
`numpy.ndarray.newbyteorder()`
`numpy.ndarray.nonzero()`	`cunumeric.ndarray.nonzero()`	✓	✓
`numpy.ndarray.partition()`	`cunumeric.ndarray.partition()`	✓	✓
`numpy.ndarray.prod()`	`cunumeric.ndarray.prod()`	✓	✓
`numpy.ndarray.ptp()`
`numpy.ndarray.put()`	`cunumeric.ndarray.put()`	✓	✓
`numpy.ndarray.ravel()`	`cunumeric.ndarray.ravel()`	✓	✓
`numpy.ndarray.repeat()`
`numpy.ndarray.reshape()`	`cunumeric.ndarray.reshape()`	✓	✓
`numpy.ndarray.resize()`
`numpy.ndarray.searchsorted()`	`cunumeric.ndarray.searchsorted()`	✓	✓
`numpy.ndarray.setfield()`	`cunumeric.ndarray.setfield()`	❌	❌
`numpy.ndarray.setflags()`	`cunumeric.ndarray.setflags()`	✓	✓
`numpy.ndarray.sort()`	`cunumeric.ndarray.sort()`	✓	✓
`numpy.ndarray.squeeze()`	`cunumeric.ndarray.squeeze()`	✓	✓
`numpy.ndarray.std()`
`numpy.ndarray.sum()`	`cunumeric.ndarray.sum()`	✓	✓
`numpy.ndarray.swapaxes()`	`cunumeric.ndarray.swapaxes()`	✓	✓
`numpy.ndarray.take()`	`cunumeric.ndarray.take()`	✓	✓
`numpy.ndarray.tobytes()`	`cunumeric.ndarray.tobytes()`	✓	✓
`numpy.ndarray.tofile()`	`cunumeric.ndarray.tofile()`	❌	❌
`numpy.ndarray.tolist()`	`cunumeric.ndarray.tolist()`	✓	✓
`numpy.ndarray.tostring()`	`cunumeric.ndarray.tostring()`	✓	✓
`numpy.ndarray.trace()`	`cunumeric.ndarray.trace()`	✓	✓
`numpy.ndarray.transpose()`	`cunumeric.ndarray.transpose()`	✓	✓
`numpy.ndarray.var()`
`numpy.ndarray.view()`	`cunumeric.ndarray.view()`	✓	✓

Summary

Number of NumPy functions: 53

Number of functions covered by cunumeric: 46

Linear Algebra#

NumPy	cunumeric	single-GPU/CPU	multi-GPU/CPU
`numpy.linalg.cholesky`	`cunumeric.linalg.cholesky`	✓	✓
`numpy.linalg.cond`
`numpy.linalg.det`
`numpy.linalg.eig`
`numpy.linalg.eigh`
`numpy.linalg.eigvals`
`numpy.linalg.eigvalsh`
`numpy.linalg.inv`
`numpy.linalg.lstsq`
`numpy.linalg.matrix_power`	`cunumeric.linalg.matrix_power`	✓	✓
`numpy.linalg.matrix_rank`
`numpy.linalg.multi_dot`	`cunumeric.linalg.multi_dot`	✓	✓
`numpy.linalg.norm`	`cunumeric.linalg.norm`	✓	✓
`numpy.linalg.pinv`
`numpy.linalg.qr`
`numpy.linalg.slogdet`
`numpy.linalg.solve`	`cunumeric.linalg.solve`	✓	❌
`numpy.linalg.svd`
`numpy.linalg.tensorinv`
`numpy.linalg.tensorsolve`

Summary

Number of NumPy functions: 20

Number of functions covered by cunumeric: 5

Discrete Fourier Transform#

NumPy	cunumeric	single-GPU/CPU	multi-GPU/CPU
`numpy.fft.fft`	`cunumeric.fft.fft`	✓	❌
`numpy.fft.fft2`	`cunumeric.fft.fft2`	✓	❌
`numpy.fft.fftfreq`
`numpy.fft.fftn`	`cunumeric.fft.fftn`	✓	❌
`numpy.fft.fftshift`
`numpy.fft.hfft`	`cunumeric.fft.hfft`	✓	❌
`numpy.fft.ifft`	`cunumeric.fft.ifft`	✓	❌
`numpy.fft.ifft2`	`cunumeric.fft.ifft2`	✓	❌
`numpy.fft.ifftn`	`cunumeric.fft.ifftn`	✓	❌
`numpy.fft.ifftshift`
`numpy.fft.ihfft`	`cunumeric.fft.ihfft`	✓	❌
`numpy.fft.irfft`	`cunumeric.fft.irfft`	✓	❌
`numpy.fft.irfft2`	`cunumeric.fft.irfft2`	✓	❌
`numpy.fft.irfftn`	`cunumeric.fft.irfftn`	✓	❌
`numpy.fft.rfft`	`cunumeric.fft.rfft`	✓	❌
`numpy.fft.rfft2`	`cunumeric.fft.rfft2`	✓	❌
`numpy.fft.rfftfreq`
`numpy.fft.rfftn`	`cunumeric.fft.rfftn`	✓	❌

Summary

Number of NumPy functions: 18

Number of functions covered by cunumeric: 14

Random Sampling#

NumPy	cunumeric	single-GPU/CPU	multi-GPU/CPU
`numpy.random.beta`	`cunumeric.random.beta`	✓	✓
`numpy.random.binomial`	`cunumeric.random.binomial`	✓	✓
`numpy.random.bytes`	`cunumeric.random.bytes`	✓	✓
`numpy.random.chisquare`	`cunumeric.random.chisquare`	✓	✓
`numpy.random.choice`
`numpy.random.default_rng`	`cunumeric.random.default_rng`	✓	✓
`numpy.random.dirichlet`
`numpy.random.exponential`	`cunumeric.random.exponential`	✓	✓
`numpy.random.f`	`cunumeric.random.f`	✓	✓
`numpy.random.gamma`	`cunumeric.random.gamma`	✓	✓
`numpy.random.geometric`	`cunumeric.random.geometric`	✓	✓
`numpy.random.get_bit_generator`
`numpy.random.get_state`
`numpy.random.gumbel`	`cunumeric.random.gumbel`	✓	✓
`numpy.random.hypergeometric`	`cunumeric.random.hypergeometric`	✓	✓
`numpy.random.laplace`	`cunumeric.random.laplace`	✓	✓
`numpy.random.logistic`	`cunumeric.random.logistic`	✓	✓
`numpy.random.lognormal`	`cunumeric.random.lognormal`	✓	✓
`numpy.random.logseries`	`cunumeric.random.logseries`	✓	✓
`numpy.random.multinomial`
`numpy.random.multivariate_normal`
`numpy.random.negative_binomial`	`cunumeric.random.negative_binomial`	✓	✓
`numpy.random.noncentral_chisquare`	`cunumeric.random.noncentral_chisquare`	✓	✓
`numpy.random.noncentral_f`	`cunumeric.random.noncentral_f`	✓	✓
`numpy.random.normal`	`cunumeric.random.normal`	✓	✓
`numpy.random.pareto`	`cunumeric.random.pareto`	✓	✓
`numpy.random.permutation`
`numpy.random.poisson`	`cunumeric.random.poisson`	✓	✓
`numpy.random.power`	`cunumeric.random.power`	✓	✓
`numpy.random.rand`	`cunumeric.random.rand`	✓	✓
`numpy.random.randint`	`cunumeric.random.randint`	✓	✓
`numpy.random.randn`	`cunumeric.random.randn`	✓	✓
`numpy.random.random`	`cunumeric.random.random`	✓	✓
`numpy.random.random_integers`	`cunumeric.random.random_integers`	✓	✓
`numpy.random.random_sample`	`cunumeric.random.random_sample`	✓	✓
`numpy.random.ranf`	`cunumeric.random.ranf`	✓	✓
`numpy.random.rayleigh`	`cunumeric.random.rayleigh`	✓	✓
`numpy.random.sample`	`cunumeric.random.sample`	✓	✓
`numpy.random.seed`	`cunumeric.random.seed`	✓	✓
`numpy.random.set_bit_generator`
`numpy.random.set_state`
`numpy.random.shuffle`
`numpy.random.standard_cauchy`	`cunumeric.random.standard_cauchy`	✓	✓
`numpy.random.standard_exponential`	`cunumeric.random.standard_exponential`	✓	✓
`numpy.random.standard_gamma`	`cunumeric.random.standard_gamma`	✓	✓
`numpy.random.standard_normal`
`numpy.random.standard_t`	`cunumeric.random.standard_t`	✓	✓
`numpy.random.triangular`	`cunumeric.random.triangular`	✓	✓
`numpy.random.uniform`	`cunumeric.random.uniform`	✓	✓
`numpy.random.vonmises`	`cunumeric.random.vonmises`	✓	✓
`numpy.random.wald`	`cunumeric.random.wald`	✓	✓
`numpy.random.weibull`	`cunumeric.random.weibull`	✓	✓
`numpy.random.zipf`	`cunumeric.random.zipf`	✓	✓

Summary

Number of NumPy functions: 53

Number of functions covered by cunumeric: 42

Measuring API coverage#

When running applications that use cunumeric, various command line options may be used to generate coverage reports.

Overall coverage report#

The environment variable CUNUMERIC_REPORT_COVERAGE may be used to print an overall percentage of cunumeric coverage:

CUNUMERIC_REPORT_COVERAGE=1 legate test.py

After execution completes, the percentage of NumPy API calls that were handled by cunumeric is printed:

cuNumeric API coverage: 26/26 (100.0%)

Detailed coverage report#

The environment variable CUNUMERIC_REPORT_DUMP_CSV may be used to save a detailed coverage report:

CUNUMERIC_REPORT_COVERAGE=1 CUNUMERIC_REPORT_DUMP_CSV="out.csv" legate test.py

After execution completes, a CSV file will be saved to the specified location (in this case out.csv). The file shows exactly what NumPy API functions were called, whether the are implemented by cunumeric, and the location of the call site:

function_name,location,implemented
numpy.array,tests/dot.py:27,True
numpy.ndarray.__init__,tests/dot.py:27,True
numpy.array,tests/dot.py:28,True
numpy.ndarray.__init__,tests/dot.py:28,True
numpy.ndarray.dot,tests/dot.py:31,True
numpy.ndarray.__init__,tests/dot.py:31,True
numpy.allclose,tests/dot.py:33,True
numpy.ndarray.__init__,tests/dot.py:33,True

Call stack reporting#

The environment variable CUNUMERIC_REPORT_DUMP_CALLSTACK may be added to include full call stack information in a CSV report:

CUNUMERIC_REPORT_COVERAGE=1 CUNUMERIC_REPORT_DUMP_CALLSTACK=1 CUNUMERIC_REPORT_DUMP_CALLSTACK=1 legate test.py

After execution completes, the CSV output file have full call stack information in the location column, with individual stack frames separated by pipe (|) characters: