Built-Ins Reference#

This section lists the Warp types and functions available to use from Warp kernels and optionally also from the Warp Python runtime API. For a listing of the API that is exclusively intended to be used at the Python Scope and run inside the CPython interpreter, see the Python Reference section.

Scalar Types#

class warp.int8[source]#

class warp.uint8[source]#

class warp.int16[source]#

class warp.uint16[source]#

class warp.int32[source]#

class warp.uint32[source]#

class warp.int64[source]#

class warp.uint64[source]#

class warp.float16[source]#

class warp.float32[source]#

class warp.float64[source]#

class warp.bool[source]#

Vector Types#

class warp.vec2b[source]#

class warp.vec2ub[source]#

class warp.vec2s[source]#

class warp.vec2us[source]#

class warp.vec2i[source]#

class warp.vec2ui[source]#

class warp.vec2l[source]#

class warp.vec2ul[source]#

class warp.vec2h[source]#

class warp.vec2f[source]#

class warp.vec2d[source]#

class warp.vec3b[source]#

class warp.vec3ub[source]#

class warp.vec3s[source]#

class warp.vec3us[source]#

class warp.vec3i[source]#

class warp.vec3ui[source]#

class warp.vec3l[source]#

class warp.vec3ul[source]#

class warp.vec3h[source]#

class warp.vec3f[source]#

class warp.vec3d[source]#

class warp.vec4b[source]#

class warp.vec4ub[source]#

class warp.vec4s[source]#

class warp.vec4us[source]#

class warp.vec4i[source]#

class warp.vec4ui[source]#

class warp.vec4l[source]#

class warp.vec4ul[source]#

class warp.vec4h[source]#

class warp.vec4f[source]#

class warp.vec4d[source]#

class warp.mat22h[source]#

class warp.mat22f[source]#

class warp.mat22d[source]#

class warp.mat33h[source]#

class warp.mat33f[source]#

class warp.mat33d[source]#

class warp.mat44h[source]#

class warp.mat44f[source]#

class warp.mat44d[source]#

class warp.quath[source]#

class warp.quatf[source]#

class warp.quatd[source]#

class warp.transformh[source]#

class warp.transformf[source]#

class warp.transformd[source]#

class warp.spatial_vectorh[source]#

class warp.spatial_vectorf[source]#

class warp.spatial_vectord[source]#

class warp.spatial_matrixh[source]#

class warp.spatial_matrixf[source]#

class warp.spatial_matrixd[source]#

Generic Types#

class warp.Int#

class warp.Float#

class warp.Scalar#

class warp.Vector#

class warp.Matrix#

class warp.Quaternion#

class warp.Transformation#

class warp.Array#

Scalar Math#

warp.min(a: Scalar, b: Scalar) → Scalar#

Kernel

Python

Differentiable

Return the minimum of two scalars.

warp.min( a: Vector[Any, Scalar], b: Vector[Any, Scalar], ) → Vector[Any, Scalar]

Kernel

Python

Differentiable

Return the element-wise minimum of two vectors.

warp.min(a: Vector[Any, Scalar]) → Scalar

Kernel

Python

Differentiable

Return the minimum element of a vector a.

warp.max(a: Scalar, b: Scalar) → Scalar#

Kernel

Python

Differentiable

Return the maximum of two scalars.

warp.max( a: Vector[Any, Scalar], b: Vector[Any, Scalar], ) → Vector[Any, Scalar]

Kernel

Python

Differentiable

Return the element-wise maximum of two vectors.

warp.max(a: Vector[Any, Scalar]) → Scalar

Kernel

Python

Differentiable

Return the maximum element of a vector a.

warp.clamp(x: Scalar, low: Scalar, high: Scalar) → Scalar#

Kernel

Python

Differentiable

Clamp the value of x to the range [low, high].

warp.abs(x: Scalar) → Scalar#

Kernel

Python

Differentiable

Return the absolute value of x.

warp.abs(x: Vector[Any, Scalar]) → Vector[Any, Scalar]

Kernel

Python

Differentiable

Return the absolute values of the elements of x.

warp.sign(x: Scalar) → Scalar#

Kernel

Python

Differentiable

Return -1 if x < 0, return 1 otherwise.

warp.sign(x: Vector[Any, Scalar]) → Scalar

Kernel

Python

Differentiable

Return -1 for the negative elements of x, and 1 otherwise.

warp.step(x: Scalar) → Scalar#

Kernel

Python

Differentiable

Return 1.0 if x < 0.0, return 0.0 otherwise.

warp.nonzero(x: Scalar) → Scalar#

Kernel

Python

Differentiable

Return 1.0 if x is not equal to zero, return 0.0 otherwise.

warp.sin(x: Float) → Float#

Kernel

Python

Differentiable

Return the sine of x in radians.

warp.cos(x: Float) → Float#

Kernel

Python

Differentiable

Return the cosine of x in radians.

warp.acos(x: Float) → Float#

Kernel

Python

Differentiable

Return arccos of x in radians. Inputs are automatically clamped to [-1.0, 1.0].

warp.asin(x: Float) → Float#

Kernel

Python

Differentiable

Return arcsin of x in radians. Inputs are automatically clamped to [-1.0, 1.0].

warp.sqrt(x: Float) → Float#

Kernel

Python

Differentiable

Return the square root of x, where x is positive.

warp.cbrt(x: Float) → Float#

Kernel

Python

Differentiable

Return the cube root of x.

warp.tan(x: Float) → Float#

Kernel

Python

Differentiable

Return the tangent of x in radians.

warp.atan(x: Float) → Float#

Kernel

Python

Differentiable

Return the arctangent of x in radians.

warp.atan2(y: Float, x: Float) → Float#

Kernel

Python

Differentiable

Return the 2-argument arctangent, atan2, of the point (x, y) in radians.

warp.sinh(x: Float) → Float#

Kernel

Python

Differentiable

Return the sinh of x.

warp.cosh(x: Float) → Float#

Kernel

Python

Differentiable

Return the cosh of x.

warp.tanh(x: Float) → Float#

Kernel

Python

Differentiable

Return the tanh of x.

warp.degrees(x: Float) → Float#

Kernel

Python

Differentiable

Convert x from radians into degrees.

warp.radians(x: Float) → Float#

Kernel

Python

Differentiable

Convert x from degrees into radians.

warp.log(x: Float) → Float#

Kernel

Python

Differentiable

Return the natural logarithm (base-e) of x, where x is positive.

warp.log2(x: Float) → Float#

Kernel

Python

Differentiable

Return the binary logarithm (base-2) of x, where x is positive.

warp.log10(x: Float) → Float#

Kernel

Python

Differentiable

Return the common logarithm (base-10) of x, where x is positive.

warp.exp(x: Float) → Float#

Kernel

Python

Differentiable

Return the value of the exponential function \(e^x\).

warp.pow(x: Float, y: Float) → Float#

Kernel

Python

Differentiable

Return the result of x raised to power of y.

warp.round(x: Float) → Float#

Kernel

Python

Differentiable

Return the nearest integer value to x, rounding halfway cases away from zero.

This is the most intuitive form of rounding in the colloquial sense, but can be slower than other options like warp.rint(). Differs from numpy.round(), which behaves the same way as numpy.rint().

warp.rint(x: Float) → Float#

Kernel

Python

Differentiable

Return the nearest integer value to x, rounding halfway cases to nearest even integer.

It is generally faster than warp.round(). Equivalent to numpy.rint().

warp.trunc(x: Float) → Float#

Kernel

Python

Differentiable

Return the nearest integer that is closer to zero than x.

In other words, it discards the fractional part of x. It is similar to casting float(int(a)), but preserves the negative sign when x is in the range [-0.0, -1.0). Equivalent to numpy.trunc() and numpy.fix().

warp.floor(x: Float) → Float#

Kernel

Python

Differentiable

Return the largest integer that is less than or equal to x.

warp.ceil(x: Float) → Float#

Kernel

Python

Differentiable

Return the smallest integer that is greater than or equal to x.

warp.frac(x: Float) → Float#

Kernel

Python

Differentiable

Retrieve the fractional part of x.

In other words, it discards the integer part of x and is equivalent to x - trunc(x).

warp.isfinite(a: Scalar) → bool#

Kernel

Python

Differentiable

Return True if a is a finite number, otherwise return False.

warp.isfinite(a: Vector[Any, Scalar]) → bool

Kernel

Python

Differentiable

Return True if all elements of the vector a are finite, otherwise return False.

warp.isfinite(a: Quaternion[Scalar]) → bool

Kernel

Python

Differentiable

Return True if all elements of the quaternion a are finite, otherwise return False.

warp.isfinite(a: Matrix[Any, Any, Scalar]) → bool

Kernel

Python

Differentiable

Return True if all elements of the matrix a are finite, otherwise return False.

warp.isnan(a: Scalar) → bool#

Kernel

Python

Differentiable

Return True if a is NaN, otherwise return False.

warp.isnan(a: Vector[Any, Scalar]) → bool

Kernel

Python

Differentiable

Return True if any element of the vector a is NaN, otherwise return False.

warp.isnan(a: Quaternion[Scalar]) → bool

Kernel

Python

Differentiable

Return True if any element of the quaternion a is NaN, otherwise return False.

warp.isnan(a: Matrix[Any, Any, Scalar]) → bool

Kernel

Python

Differentiable

Return True if any element of the matrix a is NaN, otherwise return False.

warp.isinf(a: Scalar) → bool#

Kernel

Python

Differentiable

Return True if a is positive or negative infinity, otherwise return False.

warp.isinf(a: Vector[Any, Scalar]) → bool

Kernel

Python

Differentiable

Return True if any element of the vector a is positive or negative infinity, otherwise return False.

warp.isinf(a: Quaternion[Scalar]) → bool

Kernel

Python

Differentiable

Return True if any element of the quaternion a is positive or negative infinity, otherwise return False.

warp.isinf(a: Matrix[Any, Any, Scalar]) → bool

Kernel

Python

Differentiable

Return True if any element of the matrix a is positive or negative infinity, otherwise return False.

Vector Math#

warp.dot(a: Vector[Any, Scalar], b: Vector[Any, Scalar]) → Scalar#

Kernel

Python

Differentiable

Compute the dot product between two vectors.

warp.dot(a: Quaternion[Float], b: Quaternion[Float]) → Float

Kernel

Python

Differentiable

Compute the dot product between two quaternions.

warp.ddot(a: Matrix[Any, Any, Scalar], b: Matrix[Any, Any, Scalar]) → Scalar#

Kernel

Python

Differentiable

Compute the double dot product between two matrices.

warp.argmin(a: Vector[Any, Scalar]) → uint32#

Kernel

Python

Return the index of the minimum element of a vector a.

warp.argmax(a: Vector[Any, Scalar]) → uint32#

Kernel

Python

Return the index of the maximum element of a vector a.

warp.outer( a: Vector[Any, Scalar], b: Vector[Any, Scalar], ) → Matrix[Any, Any, Scalar]#

Kernel

Python

Differentiable

Compute the outer product a*b^T for two vectors.

warp.cross(a: Vector[3, Scalar], b: Vector[3, Scalar]) → Vector[3, Scalar]#

Kernel

Python

Differentiable

Compute the cross product of two 3D vectors.

warp.skew(vec: Vector[3, Scalar]) → Matrix[3, 3, Scalar]#

Kernel

Python

Differentiable

Compute the skew-symmetric 3x3 matrix for a 3D vector vec.

warp.length(a: Vector[Any, Float]) → Float#

Kernel

Python

Differentiable

Compute the length of a floating-point vector a.

warp.length(a: Quaternion[Float]) → Float

Kernel

Python

Differentiable

Compute the length of a quaternion a.

warp.length_sq(a: Vector[Any, Scalar]) → Scalar#

Kernel

Python

Differentiable

Compute the squared length of a vector a.

warp.length_sq(a: Quaternion[Scalar]) → Scalar

Kernel

Python

Differentiable

Compute the squared length of a quaternion a.

warp.normalize(a: Vector[Any, Float]) → Vector[Any, Float]#

Kernel

Python

Differentiable

Compute the normalized value of a. If length(a) is 0 then the zero vector is returned.

warp.normalize(a: Quaternion[Float]) → Quaternion[Float]

Kernel

Python

Differentiable

Compute the normalized value of a. If length(a) is 0, then the zero quaternion is returned.

warp.transpose(a: Matrix[Any, Any, Scalar]) → Matrix[Any, Any, Scalar]#

Kernel

Python

Differentiable

Return the transpose of the matrix a.

warp.inverse(a: Matrix[2, 2, Float]) → Matrix[Any, Any, Float]#

Kernel

Python

Differentiable

Return the inverse of a 2x2 matrix a.

warp.inverse(a: Matrix[3, 3, Float]) → Matrix[Any, Any, Float]

Kernel

Python

Differentiable

Return the inverse of a 3x3 matrix a.

warp.inverse(a: Matrix[4, 4, Float]) → Matrix[Any, Any, Float]

Kernel

Python

Differentiable

Return the inverse of a 4x4 matrix a.

warp.determinant(a: Matrix[2, 2, Float]) → Float#

Kernel

Python

Differentiable

Return the determinant of a 2x2 matrix a.

warp.determinant(a: Matrix[3, 3, Float]) → Float

Kernel

Python

Differentiable

Return the determinant of a 3x3 matrix a.

warp.determinant(a: Matrix[4, 4, Float]) → Float

Kernel

Python

Differentiable

Return the determinant of a 4x4 matrix a.

warp.trace(a: Matrix[Any, Any, Scalar]) → Scalar#

Kernel

Python

Differentiable

Return the trace of the matrix a.

warp.diag(vec: Vector[Any, Scalar]) → Matrix[Any, Any, Scalar]#

Kernel

Python

Differentiable

Returns a matrix with the components of the vector vec on the diagonal.

warp.get_diag(mat: Matrix[Any, Any, Scalar]) → Vector[Any, Scalar]#

Kernel

Python

Differentiable

Returns a vector containing the diagonal elements of the square matrix mat.

warp.cw_mul( a: Vector[Any, Scalar], b: Vector[Any, Scalar], ) → Vector[Any, Scalar]#

Kernel

Python

Differentiable

Component-wise multiplication of two vectors.

warp.cw_mul( a: Matrix[Any, Any, Scalar], b: Matrix[Any, Any, Scalar], ) → Matrix[Any, Any, Scalar]

Kernel

Python

Differentiable

Component-wise multiplication of two matrices.

warp.cw_div( a: Vector[Any, Scalar], b: Vector[Any, Scalar], ) → Vector[Any, Scalar]#

Kernel

Python

Differentiable

Component-wise division of two vectors.

warp.cw_div( a: Matrix[Any, Any, Scalar], b: Matrix[Any, Any, Scalar], ) → Matrix[Any, Any, Scalar]

Kernel

Python

Differentiable

Component-wise division of two matrices.

warp.vector( *args: Scalar, length: int32, dtype: Scalar, ) → Vector[Any, Scalar]#

Kernel

Differentiable

Construct a vector of given length and dtype.

warp.matrix( pos: Vector[3, Float], rot: Quaternion[Float], scale: Vector[3, Float], dtype: Float, ) → Matrix[4, 4, Float]#

Kernel

Differentiable

Construct a 4x4 transformation matrix that applies the transformations as Translation(pos)*Rotation(rot)*Scaling(scale) when applied to column vectors, i.e.: y = (TRS)*x

warp.matrix( *args: Scalar, shape: Tuple[int, int], dtype: Scalar, ) → Matrix[Any, Any, Scalar]

Kernel

Differentiable

Construct a matrix. If the positional arg_types are not given, then matrix will be zero-initialized.

warp.matrix_from_cols(*args: Vector[Any, Scalar]) → Matrix[Any, Any, Scalar]#

Kernel

Differentiable

Construct a matrix from column vectors.

warp.matrix_from_rows(*args: Vector[Any, Scalar]) → Matrix[Any, Any, Scalar]#

Kernel

Differentiable

Construct a matrix from row vectors.

warp.identity(n: int32, dtype: Scalar) → Matrix[Any, Any, Scalar]#

Kernel

Differentiable

Create an identity matrix with shape=(n,n) with the type given by dtype.

warp.svd3( A: Matrix[3, 3, Float], U: Matrix[3, 3, Float], sigma: Vector[3, Float], V: Matrix[3, 3, Scalar], ) → None#

Kernel

Differentiable

Compute the SVD of a 3x3 matrix A. The singular values are returned in sigma, while the left and right basis vectors are returned in U and V.

warp.svd2( A: Matrix[2, 2, Float], U: Matrix[2, 2, Float], sigma: Vector[2, Float], V: Matrix[2, 2, Scalar], ) → None#

Kernel

Differentiable

Compute the SVD of a 2x2 matrix A. The singular values are returned in sigma, while the left and right basis vectors are returned in U and V.

warp.qr3( A: Matrix[3, 3, Float], Q: Matrix[3, 3, Float], R: Matrix[3, 3, Float], ) → None#

Kernel

Differentiable

Compute the QR decomposition of a 3x3 matrix A. The orthogonal matrix is returned in Q, while the upper triangular matrix is returned in R.

warp.eig3( A: Matrix[3, 3, Float], Q: Matrix[3, 3, Float], d: Vector[3, Float], ) → None#

Kernel

Differentiable

Compute the eigendecomposition of a 3x3 matrix A. The eigenvectors are returned as the columns of Q, while the corresponding eigenvalues are returned in d.

warp.math.norm_l1(v)#

Kernel

Python

Differentiable

Computes the L1 norm of a vector v.

\[\|v\|_1 = \sum_i |v_i|\]

Parameters:: v (Vector[Any,Float]) – The vector to compute the L1 norm of.
Returns:: The L1 norm of the vector.
Return type:: float

warp.math.norm_l2(v)#

Kernel

Python

Differentiable

Computes the L2 norm of a vector v.

\[\|v\|_2 = \sqrt{\sum_i v_i^2}\]

Parameters:: v (Vector[Any,Float]) – The vector to compute the L2 norm of.
Returns:: The L2 norm of the vector.
Return type:: float

warp.math.norm_huber(v, delta=1.0)#

Kernel

Python

Differentiable

Computes the Huber norm of a vector v with a given delta.

\[\begin{split}H(v) = \begin{cases} \frac{1}{2} \|v\|^2 & \text{if } \|v\| \leq \delta \\ \delta(\|v\| - \frac{1}{2}\delta) & \text{otherwise} \end{cases}\end{split}\]

Parameters:

v (Vector[Any,Float]) – The vector to compute the Huber norm of.
delta (float) – The threshold value, defaults to 1.0.

Returns:

The Huber norm of the vector.

Return type:

float

warp.math.norm_pseudo_huber(v, delta=1.0)#

Kernel

Python

Differentiable

Computes the “pseudo” Huber norm of a vector v with a given delta.

\[H^\prime(v) = \delta \sqrt{1 + \frac{\|v\|^2}{\delta^2}}\]

Parameters:

v (Vector[Any,Float]) – The vector to compute the Huber norm of.
delta (float) – The threshold value, defaults to 1.0.

Returns:

The Huber norm of the vector.

Return type:

float

warp.math.smooth_normalize(v, delta=1.0)#

Kernel

Python

Differentiable

Normalizes a vector using the pseudo-Huber norm.

See norm_pseudo_huber().

\[\frac{v}{H^\prime(v)}\]

Parameters:

v (Vector[Any,Float]) – The vector to normalize.
delta (float) – The threshold value, defaults to 1.0.

Returns:

The normalized vector.

Return type:

Vector[Any,Float]

Quaternion Math#

warp.quaternion(dtype: Float) → Quaternion[Float]#

Kernel

Differentiable

Construct a zero-initialized quaternion. Quaternions are laid out as [ix, iy, iz, r], where ix, iy, iz are the imaginary part, and r the real part.

warp.quaternion( x: Float, y: Float, z: Float, w: Float, dtype: Scalar, ) → Quaternion[Float]

Kernel

Differentiable

Create a quaternion using the supplied components (type inferred from component type).

warp.quaternion( ijk: Vector[3, Float], real: Float, dtype: Float, ) → Quaternion[Float]

Kernel

Differentiable

Create a quaternion using the supplied vector/scalar (type inferred from scalar type).

warp.quaternion(quat: Quaternion[Float], dtype: Float) → Quaternion[Float]

Kernel

Differentiable

Construct a quaternion of type dtype from another quaternion of a different dtype.

warp.quat_identity(dtype: Float) → quatf#

Kernel

Python

Differentiable

Construct an identity quaternion with zero imaginary part and real part of 1.0

warp.quat_from_axis_angle( axis: Vector[3, Float], angle: Float, ) → Quaternion[Float]#

Kernel

Python

Differentiable

Construct a quaternion representing a rotation of angle radians around the given axis.

warp.quat_to_axis_angle( quat: Quaternion[Float], axis: Vector[3, Float], angle: Float, ) → None#

Kernel

Differentiable

Extract the rotation axis and angle radians a quaternion represents.

warp.quat_from_matrix(mat: Matrix[3, 3, Float]) → Quaternion[Float]#

Kernel

Python

Differentiable

Construct a quaternion from a 3x3 matrix.

If the matrix is not a pure rotation, but for example includes scaling or skewing, the result is undefined.

warp.quat_from_matrix(mat: Matrix[4, 4, Float]) → Quaternion[Float]

Kernel

Python

Differentiable

Construct a quaternion from a 4x4 matrix.

If the top-left 3x3 block of the matrix is not a pure rotation, but for example includes scaling or skewing, the result is undefined.

warp.quat_rpy(roll: Float, pitch: Float, yaw: Float) → Quaternion[Float]#

Kernel

Python

Differentiable

Construct a quaternion representing a combined roll (z), pitch (x), yaw rotations (y) in radians.

warp.quat_inverse(quat: Quaternion[Float]) → Quaternion[Float]#

Kernel

Python

Differentiable

Compute quaternion conjugate.

warp.quat_rotate( quat: Quaternion[Float], vec: Vector[3, Float], ) → Vector[3, Float]#

Kernel

Python

Differentiable

Rotate a vector by a quaternion.

warp.quat_rotate_inv( quat: Quaternion[Float], vec: Vector[3, Float], ) → Vector[3, Float]#

Kernel

Python

Differentiable

Rotate a vector by the inverse of a quaternion.

warp.quat_slerp( a: Quaternion[Float], b: Quaternion[Float], t: Float, ) → Quaternion[Float]#

Kernel

Python

Differentiable

Linearly interpolate between two quaternions.

warp.quat_to_matrix(quat: Quaternion[Float]) → Matrix[3, 3, Float]#

Kernel

Python

Differentiable

Convert a quaternion to a 3x3 rotation matrix.

Transformations#

warp.transformation( pos: Vector[3, Float], rot: Quaternion[Float], dtype: Float, ) → Transformation[Float]#

Kernel

Differentiable

Construct a rigid-body transformation with translation part pos and rotation rot.

warp.transform_identity(dtype: Float) → transformf#

Kernel

Python

Differentiable

Construct an identity transform with zero translation and identity rotation.

warp.transform_get_translation( xform: Transformation[Float], ) → Vector[3, Float]#

Kernel

Python

Differentiable

Return the translational part of a transform xform.

warp.transform_get_rotation( xform: Transformation[Float], ) → Quaternion[Float]#

Kernel

Python

Differentiable

Return the rotational part of a transform xform.

warp.transform_multiply( a: Transformation[Float], b: Transformation[Float], ) → Transformation[Float]#

Kernel

Python

Differentiable

Multiply two rigid body transformations together.

warp.transform_point( xform: Transformation[Float], point: Vector[3, Float], ) → Vector[3, Float]#

Kernel

Python

Differentiable

Apply the transform to a point point treating the homogeneous coordinate as w=1 (translation and rotation).

warp.transform_point( mat: Matrix[4, 4, Float], point: Vector[3, Float], ) → Vector[3, Float]

Kernel

Python

Differentiable

Apply the transform to a point point treating the homogeneous coordinate as w=1.

The transformation is applied treating point as a column vector, e.g.: y = mat*point.

This is in contrast to some libraries, notably USD, which applies transforms to row vectors, y^T = point^T*mat^T. If the transform is coming from a library that uses row-vectors, then users should transpose the transformation matrix before calling this method.

warp.transform_vector( xform: Transformation[Float], vec: Vector[3, Float], ) → Vector[3, Float]#

Kernel

Python

Differentiable

Apply the transform to a vector vec treating the homogeneous coordinate as w=0 (rotation only).

warp.transform_vector( mat: Matrix[4, 4, Float], vec: Vector[3, Float], ) → Vector[3, Float]

Kernel

Python

Differentiable

Apply the transform to a vector vec treating the homogeneous coordinate as w=0.

The transformation is applied treating vec as a column vector, e.g.: y = mat*vec.

This is in contrast to some libraries, notably USD, which applies transforms to row vectors, y^T = vec^T*mat^T. If the transform is coming from a library that uses row-vectors, then users should transpose the transformation matrix before calling this method.

warp.transform_inverse( xform: Transformation[Float], ) → Transformation[Float]#

Kernel

Python

Differentiable

Compute the inverse of the transformation xform.

warp.math.transform_from_matrix(mat)#

Kernel

Python

Differentiable

Construct a transformation from a 4x4 matrix.

Parameters:: mat (Matrix[4, 4, Float]) – Matrix to convert.
Returns:: The transformation.
Return type:: Transformation[Float]

warp.math.transform_to_matrix(xform)#

Kernel

Python

Differentiable

Convert a transformation to a 4x4 matrix.

Parameters:: xform (Transformation[Float]) – Transformation to convert.
Returns:: The matrix.
Return type:: Matrix[4, 4, Float]

Spatial Math#

warp.spatial_vector(dtype: Float) → Vector[6, Float][source]#

Kernel

Differentiable

Zero-initialize a 6D screw vector.

warp.spatial_vector( w: Vector[3, Float], v: Vector[3, Float], dtype: Float, ) → Vector[6, Float][source]

Kernel

Differentiable

Construct a 6D screw vector from two 3D vectors.

warp.spatial_vector( wx: Float, wy: Float, wz: Float, vx: Float, vy: Float, vz: Float, dtype: Float, ) → Vector[6, Float][source]

Kernel

Differentiable

Construct a 6D screw vector from six values.

warp.spatial_adjoint( r: Matrix[3, 3, Float], s: Matrix[3, 3, Float], ) → Matrix[6, 6, Float]#

Kernel

Differentiable

Construct a 6x6 spatial inertial matrix from two 3x3 diagonal blocks.

warp.spatial_dot(a: Vector[6, Float], b: Vector[6, Float]) → Float#

Kernel

Python

Differentiable

Compute the dot product of two 6D screw vectors.

warp.spatial_cross( a: Vector[6, Float], b: Vector[6, Float], ) → Vector[6, Float]#

Kernel

Python

Differentiable

Compute the cross product of two 6D screw vectors.

warp.spatial_cross_dual( a: Vector[6, Float], b: Vector[6, Float], ) → Vector[6, Float]#

Kernel

Python

Differentiable

Compute the dual cross product of two 6D screw vectors.

warp.spatial_top(svec: Vector[6, Float]) → Vector[3, Float]#

Kernel

Python

Differentiable

Return the top (first) part of a 6D screw vector.

warp.spatial_bottom(svec: Vector[6, Float]) → Vector[3, Float]#

Kernel

Python

Differentiable

Return the bottom (second) part of a 6D screw vector.

warp.spatial_jacobian( S: Array[Vector[6, Float]], joint_parents: Array[int32], joint_qd_start: Array[int32], joint_start: int32, joint_count: int32, J_start: int32, J_out: Array[Float], ) → None#

Kernel

Differentiable

warp.spatial_mass( I_s: Array[Matrix[6, 6, Float]], joint_start: int32, joint_count: int32, M_start: int32, M: Array[Float], ) → None#

Kernel

Differentiable

Tile Primitives#

warp.tile_zeros(shape: Tuple[int, ...], dtype: Any, storage: str) → Tile#

Kernel

Allocate a tile of zero-initialized items.

Parameters:

shape – Shape of the output tile
dtype – Data type of output tile’s elements (default float)
storage – The storage location for the tile: "register" for registers (default) or "shared" for shared memory.

Returns:

A zero-initialized tile with shape and data type as specified

warp.tile_ones(shape: Tuple[int, ...], dtype: Any, storage: str) → Tile#

Kernel

Allocate a tile of one-initialized items.

Parameters:

shape – Shape of the output tile
dtype – Data type of output tile’s elements
storage – The storage location for the tile: "register" for registers (default) or "shared" for shared memory.

Returns:

A one-initialized tile with shape and data type as specified

warp.tile_arange(*args: Scalar, dtype: Any, storage: str) → Tile#

Kernel

Generate a tile of linearly spaced elements.

Parameters:

args –
Variable-length positional arguments, interpreted as:
- (stop,): Generates values from 0 to stop - 1
- (start, stop): Generates values from start to stop - 1
- (start, stop, step): Generates values from start to stop - 1 with a step size
dtype – Data type of output tile’s elements (optional, default: float)
storage – The storage location for the tile: "register" for registers (default) or "shared" for shared memory.

Returns:

A tile with shape=(n) with linearly spaced elements of specified data type

warp.tile_load( a: Array[Any], shape: Tuple[int, ...], offset: Tuple[int, ...], storage: str, ) → Array[Scalar]#

Kernel

Differentiable

Loads a tile from a global memory array.

This method will cooperatively load a tile from global memory using all threads in the block.

Parameters:

a – The source array in global memory
shape – Shape of the tile to load, must have the same number of dimensions as a
offset – Offset in the source array to begin reading from (optional)
storage – The storage location for the tile: "register" for registers (default) or "shared" for shared memory.

Returns:

A tile with shape as specified and data type the same as the source array

warp.tile_store(a: Array[Any], t: Tile, offset: Tuple[int, ...]) → None#

Kernel

Differentiable

Store a tile to a global memory array.

This method will cooperatively store a tile to global memory using all threads in the block.

Parameters:

a – The destination array in global memory
t – The source tile to store data from, must have the same data type and number of dimensions as the destination array
offset – Offset in the destination array (optional)

warp.tile_atomic_add( a: Array[Any], t: Tile, offset: Tuple[int, ...], ) → Tile#

Kernel

Differentiable

Atomically add a 1D tile to the array a, each element will be updated atomically.

Parameters:

a – Array in global memory, should have the same dtype as the input tile
t – Source tile to add to the destination array
offset – Offset in the destination array (optional)

Returns:

A tile with the same dimensions and data type as the source tile, holding the original value of the destination elements

warp.tile_view( t: Tile, offset: Tuple[int, ...], shape: Tuple[int, ...], ) → Tile#

Kernel

Return a slice of a given tile [offset, offset+shape], if shape is not specified it will be inferred from the unspecified offset dimensions.

Parameters:

t – Input tile to extract a subrange from
offset – Offset in the source tile
shape – Shape of the returned slice

Returns:

A tile with dimensions given by the specified shape or the remaining source tile dimensions

warp.tile_assign(dst: Tile, src: Tile, offset: Tuple[int, ...]) → None#

Kernel

Differentiable

Assign a tile to a subrange of a destination tile.

Parameters:

dst – The destination tile to assign to
src – The source tile to read values from
offset – Offset in the destination tile to write to

warp.tile(x: Any) → Tile#

Kernel

Differentiable

Construct a new tile from per-thread kernel values.

This function converts values computed using scalar kernel code to a tile representation for input into collective operations.

If the input value is a scalar, then the resulting tile has shape=(block_dim,)
If the input value is a vector, then the resulting tile has shape=(length(vector), block_dim)

Parameters:: x – A per-thread local value, e.g. scalar, vector, or matrix.
Returns:: A tile with first dimension according to the value type length and a second dimension equal to block_dim

This example shows how to create a linear sequence from thread variables:

@wp.kernel
def compute():
    i = wp.tid()
    t = wp.tile(i*2)
    print(t)

wp.launch(compute, dim=16, inputs=[], block_dim=16)

Prints:

[0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30] = tile(shape=(16), storage=register)

warp.untile(a: Tile) → Scalar#

Kernel

Differentiable

Convert a tile back to per-thread values.

This function converts a block-wide tile back to per-thread values.

If the input tile is 1D, then the resulting value will be a per-thread scalar
If the input tile is 2D, then the resulting value will be a per-thread vector of length M

Parameters:: a – A tile with dimensions shape=(M, block_dim)
Returns:: A single value per-thread with the same data type as the tile

This example shows how to create a linear sequence from thread variables:

@wp.kernel
def compute():
    i = wp.tid()

    # create block-wide tile
    t = wp.tile(i)*2

    # convert back to per-thread values
    s = wp.untile(t)

    print(s)

wp.launch(compute, dim=16, inputs=[], block_dim=16)

Prints:

0
2
4
6
8
...

warp.tile_transpose(a: Tile) → Tile#

Kernel

Differentiable

Transpose a tile.

For shared memory tiles, this operation will alias the input tile. Register tiles will first be transferred to shared memory before transposition.

Parameters:: a – Tile to transpose with shape=(M,N)
Returns:: Tile with shape=(N,M)

warp.tile_broadcast(a: Tile, shape: Tuple[int, ...]) → Tile#

Kernel

Differentiable

Broadcast a tile.

Broadcasts the input tile a to the destination shape. Broadcasting follows NumPy broadcast rules.

Parameters:

a – Tile to broadcast
shape – The shape to broadcast to

Returns:

Tile with broadcast shape

warp.tile_sum(a: Tile) → Tile#

Kernel

Differentiable

Cooperatively compute the sum of the tile elements using all threads in the block.

Parameters:: a – The tile to compute the sum of
Returns:: A single-element tile holding the sum

Example:

@wp.kernel
def compute():

    t = wp.tile_ones(dtype=float, shape=(16, 16))
    s = wp.tile_sum(t)

    print(s)

wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=64)

Prints:

[256] = tile(shape=(1), storage=register)

warp.tile_min(a: Tile) → Tile#

Kernel

Differentiable

Cooperatively compute the minimum of the tile elements using all threads in the block.

Parameters:: a – The tile to compute the minimum of
Returns:: A single-element tile holding the minimum value

Example:

@wp.kernel
def compute():

    t = wp.tile_arange(64, 128)
    s = wp.tile_min(t)

    print(s)


wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=64)

Prints:

[64] = tile(shape=(1), storage=register)

warp.tile_max(a: Tile) → Tile#

Kernel

Differentiable

Cooperatively compute the maximum of the tile elements using all threads in the block.

Parameters:: a – The tile to compute the maximum from
Returns:: A single-element tile holding the maximum value

Example:

@wp.kernel
def compute():

    t = wp.tile_arange(64, 128)
    s = wp.tile_max(t)

    print(s)

wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=64)

Prints:

[127] = tile(shape=(1), storage=register)

warp.tile_reduce(op: Callable, a: Tile) → Tile#

Kernel

Differentiable

Apply a custom reduction operator across the tile.

This function cooperatively performs a reduction using the provided operator across the tile.

Parameters:

op – A callable function that accepts two arguments and returns one argument, may be a user function or builtin
a – The input tile, the operator (or one of its overloads) must be able to accept the tile’s data type

Returns:

A single-element tile with the same data type as the input tile.

Example:

@wp.kernel
def factorial():

    t = wp.tile_arange(1, 10, dtype=int)
    s = wp.tile_reduce(wp.mul, t)

    print(s)

wp.launch_tiled(factorial, dim=[1], inputs=[], block_dim=16)

Prints:

[362880] = tile(shape=(1), storage=register)

warp.tile_map(op: Callable, a: Tile) → Tile#

Kernel

Differentiable

Apply a unary function onto the tile.

This function cooperatively applies a unary function to each element of the tile using all threads in the block.

Parameters:

op – A callable function that accepts one argument and returns one argument, may be a user function or builtin
a – The input tile, the operator (or one of its overloads) must be able to accept the tile’s data type

Returns:

A tile with the same dimensions and data type as the input tile.

Example:

@wp.kernel
def compute():

    t = wp.tile_arange(0.0, 1.0, 0.1, dtype=float)
    s = wp.tile_map(wp.sin, t)

    print(s)

wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=16)

Prints:

[0 0.0998334 0.198669 0.29552 0.389418 0.479426 0.564642 0.644218 0.717356 0.783327] = tile(shape=(10), storage=register)

warp.tile_map(op: Callable, a: Tile, b: Tile) → Tile

Kernel

Differentiable

Apply a binary function onto the tile.

This function cooperatively applies a binary function to each element of the tiles using all threads in the block. Both input tiles must have the same dimensions and datatype.

Parameters:

op – A callable function that accepts two arguments and returns one argument, all of the same type, may be a user function or builtin
a – The first input tile, the operator (or one of its overloads) must be able to accept the tile’s dtype
b – The second input tile, the operator (or one of its overloads) must be able to accept the tile’s dtype

Returns:

A tile with the same dimensions and datatype as the input tiles.

Example:

@wp.kernel
def compute():

    a = wp.tile_arange(0.0, 1.0, 0.1, dtype=float)
    b = wp.tile_ones(shape=10, dtype=float)

    s = wp.tile_map(wp.add, a, b)

    print(s)

wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=16)

Prints:

[1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9] = tile(shape=(10), storage=register)

warp.tile_diag_add(a: Tile, d: Tile) → Tile#

Kernel

Differentiable

Add a square matrix and a diagonal matrix ‘d’ represented as a 1D tile

warp.tile_matmul(a: Tile, b: Tile, out: Tile) → Tile#

Kernel

Differentiable

Computes the matrix product and accumulates out += a*b.

Supported datatypes are:

fp16, fp32, fp64 (real)
vec2h, vec2f, vec2d (complex)

All input and output tiles must have the same datatype. Tile data will automatically be migrated to shared memory if necessary and will use TensorCore operations when available.

Parameters:

a – A tile with shape=(M, K)
b – A tile with shape=(K, N)
out – A tile with shape=(M, N)

warp.tile_matmul(a: Tile, b: Tile) → Tile

Kernel

Differentiable

Computes the matrix product out = a*b.

Supported datatypes are:

fp16, fp32, fp64 (real)
vec2h, vec2f, vec2d (complex)

Both input tiles must have the same datatype. Tile data will automatically be migrated to shared memory if necessary and will use TensorCore operations when available.

Parameters:

a – A tile with shape=(M, K)
b – A tile with shape=(K, N)

Returns:

A tile with shape=(M, N)

warp.tile_fft(inout: Tile) → Tile#

Kernel

Differentiable

Compute the forward FFT along the second dimension of a 2D tile of data.

This function cooperatively computes the forward FFT on a tile of data inplace, treating each row individually.

Note that computing the adjoint is not yet supported.

Supported datatypes are:

vec2f, vec2d

Parameters:: inout – The input/output tile

warp.tile_ifft(inout: Tile) → Tile#

Kernel

Differentiable

Compute the inverse FFT along the second dimension of a 2D tile of data.

This function cooperatively computes the inverse FFT on a tile of data inplace, treating each row individually.

Note that computing the adjoint is not yet supported.

Supported datatypes are:

vec2f, vec2d

Parameters:: inout – The input/output tile

warp.tile_cholesky(A: Tile) → Tile#

Kernel

Differentiable

Compute the Cholesky factorization L of a matrix A. L is lower triangular and satisfies LL^T = A.

Note that computing the adjoint is not yet supported.

Supported datatypes are:

float32
float64

Parameters:: A – A square, symmetric positive-definite, matrix.
Returns L:: A square, lower triangular, matrix, such that LL^T = A

warp.tile_cholesky_solve(L: Tile, x: Tile) → None#

Kernel

Differentiable

With L such that LL^T = A, solve for x in Ax = y

Note that computing the adjoint is not yet supported.

Supported datatypes are:

float32
float64

Parameters:

L – A square, lower triangular, matrix, such that LL^T = A
x – An 1D tile of length M

Returns y:

An 1D tile of length M such that LL^T y = x

Utility#

warp.mlp( weights: Array[float32], bias: Array[float32], activation: Callable, index: int32, x: Array[float32], out: Array[float32], ) → None#

Kernel

Differentiable

Evaluate a multi-layer perceptron (MLP) layer in the form: out = act(weights*x + bias).

Deprecated since version 1.6: Use tile primitives instead.

Parameters:

weights – A layer’s network weights with dimensions (m, n).
bias – An array with dimensions (n).
activation – A wp.func function that takes a single scalar float as input and returns a scalar float as output
index – The batch item to process, typically each thread will process one item in the batch, in which case index should be wp.tid()
x – The feature matrix with dimensions (n, b)
out – The network output with dimensions (m, b)

Note:

Feature and output matrices are transposed compared to some other frameworks such as PyTorch. All matrices are assumed to be stored in flattened row-major memory layout (NumPy default).

warp.reversed(range: range_t) → range_t#

Kernel

Differentiable

Returns the range in reversed order.

warp.printf(fmt: str, *args: Any) → None#

Kernel

Differentiable

Allows printing formatted strings using C-style format specifiers.

warp.print(value: Any) → None#

Kernel

Differentiable

Print variable to stdout

warp.breakpoint() → None#

Kernel

Differentiable

Debugger breakpoint

warp.tid() → int#

Kernel

Differentiable

Return the current thread index for a 1D kernel launch.

Note that this is the global index of the thread in the range [0, dim) where dim is the parameter passed to kernel launch.

This function may not be called from user-defined Warp functions.

warp.tid() → Tuple[int, int]

Kernel

Differentiable

Return the current thread indices for a 2D kernel launch.

Use i,j = wp.tid() syntax to retrieve the coordinates inside the kernel thread grid.

This function may not be called from user-defined Warp functions.

warp.tid() → Tuple[int, int, int]

Kernel

Differentiable

Return the current thread indices for a 3D kernel launch.

Use i,j,k = wp.tid() syntax to retrieve the coordinates inside the kernel thread grid.

This function may not be called from user-defined Warp functions.

warp.tid() → Tuple[int, int, int, int]

Kernel

Differentiable

Return the current thread indices for a 4D kernel launch.

Use i,j,k,l = wp.tid() syntax to retrieve the coordinates inside the kernel thread grid.

This function may not be called from user-defined Warp functions.

warp.select(cond: bool, value_if_false: Any, value_if_true: Any) → Any#

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select(cond: int8, value_if_false: Any, value_if_true: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select(cond: uint8, value_if_false: Any, value_if_true: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select(cond: int16, value_if_false: Any, value_if_true: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select(cond: uint16, value_if_false: Any, value_if_true: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select(cond: int32, value_if_false: Any, value_if_true: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select(cond: uint32, value_if_false: Any, value_if_true: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select(cond: int64, value_if_false: Any, value_if_true: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select(cond: uint64, value_if_false: Any, value_if_true: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(cond, value_if_true, value_if_false).

warp.select( arr: Array[Any], value_if_false: Any, value_if_true: Any, ) → Any

Kernel

Differentiable

Select between two arguments, if arr is null then return value_if_false, otherwise return value_if_true.

Deprecated since version 1.7: Use where() instead, which has the more intuitive argument order: where(arr, value_if_true, value_if_false).

warp.where(cond: bool, value_if_true: Any, value_if_false: Any) → Any#

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(cond: int8, value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(cond: uint8, value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(cond: int16, value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(cond: uint16, value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(cond: int32, value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(cond: uint32, value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(cond: int64, value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(cond: uint64, value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.

warp.where(arr: Array[Any], value_if_true: Any, value_if_false: Any) → Any

Kernel

Differentiable

Select between two arguments, if arr is not null then return value_if_true, otherwise return value_if_false.

warp.atomic_add(arr: Array[Any], i: Int, value: Any) → Any#

Kernel

Differentiable

Atomically adds value onto arr[i] and returns the original value of arr[i].: This function is automatically invoked when using the syntax arr[i] += value.

warp.atomic_add(arr: Array[Any], i: Int, j: Int, value: Any) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j] and returns the original value of arr[i,j].: This function is automatically invoked when using the syntax arr[i,j] += value.

warp.atomic_add(arr: Array[Any], i: Int, j: Int, k: Int, value: Any) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j,k] and returns the original value of arr[i,j,k].: This function is automatically invoked when using the syntax arr[i,j,k] += value.

warp.atomic_add( arr: Array[Any], i: Int, j: Int, k: Int, l: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j,k,l] and returns the original value of arr[i,j,k,l].: This function is automatically invoked when using the syntax arr[i,j,k,l] += value.

warp.atomic_add(arr: FabricArray[Any], i: Int, value: Any) → Any

Kernel

Differentiable

Atomically adds value onto arr[i] and returns the original value of arr[i].: This function is automatically invoked when using the syntax arr[i] += value.

warp.atomic_add(arr: FabricArray[Any], i: Int, j: Int, value: Any) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j] and returns the original value of arr[i,j].: This function is automatically invoked when using the syntax arr[i,j] += value.

warp.atomic_add( arr: FabricArray[Any], i: Int, j: Int, k: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j,k] and returns the original value of arr[i,j,k].: This function is automatically invoked when using the syntax arr[i,j,k] += value.

warp.atomic_add( arr: FabricArray[Any], i: Int, j: Int, k: Int, l: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j,k,l] and returns the original value of arr[i,j,k,l].: This function is automatically invoked when using the syntax arr[i,j,k,l] += value.

warp.atomic_add(arr: IndexedFabricArray[Any], i: Int, value: Any) → Any

Kernel

Differentiable

Atomically adds value onto arr[i] and returns the original value of arr[i].: This function is automatically invoked when using the syntax arr[i] += value.

warp.atomic_add( arr: IndexedFabricArray[Any], i: Int, j: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j] and returns the original value of arr[i,j].: This function is automatically invoked when using the syntax arr[i,j] += value.

warp.atomic_add( arr: IndexedFabricArray[Any], i: Int, j: Int, k: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j,k] and returns the original value of arr[i,j,k].: This function is automatically invoked when using the syntax arr[i,j,k] += value.

warp.atomic_add( arr: IndexedFabricArray[Any], i: Int, j: Int, k: Int, l: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically adds value onto arr[i,j,k,l] and returns the original value of arr[i,j,k,l].: This function is automatically invoked when using the syntax arr[i,j,k,l] += value.

warp.atomic_sub(arr: Array[Any], i: Int, value: Any) → Any#

Kernel

Differentiable

Atomically subtracts value onto arr[i] and returns the original value of arr[i].: This function is automatically invoked when using the syntax arr[i] -= value.

warp.atomic_sub(arr: Array[Any], i: Int, j: Int, value: Any) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j] and returns the original value of arr[i,j].: This function is automatically invoked when using the syntax arr[i,j] -= value.

warp.atomic_sub(arr: Array[Any], i: Int, j: Int, k: Int, value: Any) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j,k] and returns the original value of arr[i,j,k].: This function is automatically invoked when using the syntax arr[i,j,k] -= value.

warp.atomic_sub( arr: Array[Any], i: Int, j: Int, k: Int, l: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j,k,l] and returns the original value of arr[i,j,k,l].: This function is automatically invoked when using the syntax arr[i,j,k,l] -= value.

warp.atomic_sub(arr: FabricArray[Any], i: Int, value: Any) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i] and returns the original value of arr[i].: This function is automatically invoked when using the syntax arr[i] -= value.

warp.atomic_sub(arr: FabricArray[Any], i: Int, j: Int, value: Any) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j] and returns the original value of arr[i,j].: This function is automatically invoked when using the syntax arr[i,j] -= value.

warp.atomic_sub( arr: FabricArray[Any], i: Int, j: Int, k: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j,k] and returns the original value of arr[i,j,k].: This function is automatically invoked when using the syntax arr[i,j,k] -= value.

warp.atomic_sub( arr: FabricArray[Any], i: Int, j: Int, k: Int, l: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j,k,l] and returns the original value of arr[i,j,k,l].: This function is automatically invoked when using the syntax arr[i,j,k,l] -= value.

warp.atomic_sub(arr: IndexedFabricArray[Any], i: Int, value: Any) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i] and returns the original value of arr[i].: This function is automatically invoked when using the syntax arr[i] -= value.

warp.atomic_sub( arr: IndexedFabricArray[Any], i: Int, j: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j] and returns the original value of arr[i,j].: This function is automatically invoked when using the syntax arr[i,j] -= value.

warp.atomic_sub( arr: IndexedFabricArray[Any], i: Int, j: Int, k: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j,k] and returns the original value of arr[i,j,k].: This function is automatically invoked when using the syntax arr[i,j,k] -= value.

warp.atomic_sub( arr: IndexedFabricArray[Any], i: Int, j: Int, k: Int, l: Int, value: Any, ) → Any

Kernel

Differentiable

Atomically subtracts value onto arr[i,j,k,l] and returns the original value of arr[i,j,k,l].: This function is automatically invoked when using the syntax arr[i,j,k,l] -= value.

warp.atomic_min(arr: Array[Any], i: Int, value: Any) → Any#

Kernel

Differentiable

Compute the minimum of value and arr[i], atomically update the array, and return the old value.