Built-Ins

Warp built-in types and functions.

Built-in types and functions are available within Warp kernels and optionally also from the Warp Python runtime API.

Each built-in function is tagged to indicate where it can be used:

• Kernel - Can be called from inside a Warp kernel

• Python - Can be called at the Python scope

• Differentiable - Propagates gradients when used in reverse mode automatic differentiation

For a listing of the API that is exclusively intended to be used at the Python scope and run inside the CPython interpreter, see warp.

Scalar Math

abs	Return the absolute value of x.
acos	Return arccos of x in radians.
asin	Return arcsin of x in radians.
atan	Return the arctangent of x in radians.
atan2	Return the 2-argument arctangent, atan2, of the point (x, y) in radians.
cbrt	Return the cube root of x.
ceil	Return the smallest integer that is greater than or equal to x.
clamp	Clamp the value of x to the range [low, high].
cos	Return the cosine of x in radians.
cosh	Return the cosh of x.
degrees	Convert x from radians into degrees.
erf	Return the error function of x.
erfc	Return the complementary error function of x.
erfcinv	Return the inverse complementary error function of x.
erfinv	Return the inverse error function of x.
exp	Return the value of the exponential function \(e^x\).
floor	Return the largest integer that is less than or equal to x.
frac	Retrieve the fractional part of x.
isfinite	Return True if a is a finite number, otherwise return False.
isinf	Return True if a is positive or negative infinity, otherwise return False.
isnan	Return True if a is NaN, otherwise return False.
log	Return the natural logarithm (base-e) of x, where x is positive.
log2	Return the binary logarithm (base-2) of x, where x is positive.
log10	Return the common logarithm (base-10) of x, where x is positive.
max	Return the maximum of two scalars.
min	Return the minimum of two scalars.
nonzero	Return 1.0 if x is not equal to zero, return 0.0 otherwise.
pow	Return the result of x raised to power of y.
radians	Convert x from degrees into radians.
rint	Return the nearest integer value to x, rounding halfway cases to nearest even integer.
round	Return the nearest integer value to x, rounding halfway cases away from zero.
sign	Return -1 if x < 0, return 1 otherwise.
sin	Return the sine of x in radians.
sinh	Return the sinh of x.
sqrt	Return the square root of x, where x is positive.
step	Return 1.0 if x < 0.0, return 0.0 otherwise.
tan	Return the tangent of x in radians.
tanh	Return the tanh of x.
trunc	Return the nearest integer that is closer to zero than x.

Vector Math

argmax	Return the index of the maximum element of a vector a.
argmin	Return the index of the minimum element of a vector a.
cross	Compute the cross product of two 3D vectors.
cw_div	Component-wise division of two vectors.
cw_mul	Component-wise multiplication of two vectors.
ddot	Compute the double dot product between two matrices.
determinant	Return the determinant of a 2x2 matrix a.
diag	Returns a matrix with the components of the vector vec on the diagonal.
dot	Compute the dot product between two vectors.
eig3	Compute the eigendecomposition of a 3x3 matrix A.
get_diag	Returns a vector containing the diagonal elements of the square matrix mat.
identity	Create an identity matrix with shape=(n,n) with the type given by dtype.
inverse	Return the inverse of a 2x2 matrix a.
length	Compute the length of a floating-point vector a.
length_sq	Compute the squared length of a vector a.
matrix	Construct a matrix.
matrix_from_cols	Construct a matrix from column vectors.
matrix_from_rows	Construct a matrix from row vectors.
norm_huber	Computes the Huber norm of a vector v with a given delta.
norm_l1	Computes the L1 norm of a vector v.
norm_l2	Computes the L2 norm of a vector v.
norm_pseudo_huber	Computes the "pseudo" Huber norm of a vector v with a given delta.
normalize	Compute the normalized value of a.
outer	Compute the outer product a*b^T for two vectors.
qr3	Compute the QR decomposition of a 3x3 matrix A.
skew	Compute the skew-symmetric 3x3 matrix for a 3D vector vec.
smooth_normalize	Normalizes a vector using the pseudo-Huber norm.
svd2	Compute the SVD of a 2x2 matrix A.
svd3	Compute the SVD of a 3x3 matrix A.
trace	Return the trace of the matrix a.
transpose	Return the transpose of the matrix a.
vector	Construct a vector of given length and dtype.

Quaternion Math

quat_from_axis_angle	Construct a quaternion representing a rotation of angle radians around the given axis.
quat_from_matrix	Construct a quaternion from a 3x3 matrix.
quat_identity	Construct an identity quaternion with zero imaginary part and real part of 1.0
quat_inverse	Compute quaternion conjugate.
quat_rotate	Rotate a vector by a quaternion.
quat_rotate_inv	Rotate a vector by the inverse of a quaternion.
quat_rpy	Construct a quaternion representing a combined roll (z), pitch (x), yaw rotations (y) in radians.
quat_slerp	Linearly interpolate between two quaternions.
quat_to_axis_angle	Extract the rotation axis and angle radians a quaternion represents.
quat_to_matrix	Convert a quaternion to a 3x3 rotation matrix.
quaternion	Construct a zero-initialized quaternion.

Transformations

transform_compose	Compose a 4x4 transformation matrix from a 3D position, quaternion orientation, and 3D scale.
transform_decompose	Decompose a 4x4 transformation matrix into 3D position, quaternion orientation, and 3D scale.
transform_from_matrix	Construct a transformation from a 4x4 matrix.
transform_get_rotation	Return the rotational part of a transform xform.
transform_get_translation	Return the translational part of a transform xform.
transform_identity	Construct an identity transform with zero translation and identity rotation.
transform_inverse	Compute the inverse of the transformation xform.
transform_multiply	Multiply two rigid body transformations together.
transform_point	Apply the transform to a point point treating the homogeneous coordinate as w=1 (translation and rotation).
transform_set_rotation	Set the rotational part of a transform xform.
transform_set_translation	Set the translational part of a transform xform.
transform_to_matrix	Convert a transformation to a 4x4 matrix.
transform_vector	Apply the transform to a vector vec treating the homogeneous coordinate as w=0 (rotation only).
transformation	Construct a rigid-body transformation with translation part p and rotation q.

Spatial Math

spatial_adjoint	Construct a 6x6 spatial inertial matrix from two 3x3 diagonal blocks.
spatial_bottom	Return the bottom (second) part of a 6D screw vector.
spatial_cross	Compute the cross product of two 6D screw vectors.
spatial_cross_dual	Compute the dual cross product of two 6D screw vectors.
spatial_dot	Compute the dot product of two 6D screw vectors.
spatial_jacobian
spatial_mass
spatial_top	Return the top (first) part of a 6D screw vector.
spatial_vector	Zero-initialize a 6D screw vector.

Tile Primitives

tile	Construct a new tile from per-thread kernel values.
tile_arange	Generate a tile of linearly spaced elements.
tile_argmax	Cooperatively compute the index of the maximum element in the tile using all threads in the block.
tile_argmin	Cooperatively compute the index of the minimum element in the tile using all threads in the block.
tile_assign	Assign a tile to a subrange of a destination tile.
tile_astype	Return a new tile with the same data as the input tile, but with a different data type.
tile_atomic_add	Atomically add a tile onto the array a, each element will be updated atomically.
tile_atomic_add_indexed	Atomically add a tile to a global memory array, with storage along a specified axis mapped according to a 1D tile of indices.
tile_broadcast	Broadcast a tile.
tile_bvh_query_aabb	Construct an axis-aligned bounding box query against a BVH object for thread-block parallel traversal.
tile_bvh_query_next	Move to the next bound in a thread-block parallel BVH query and return results as a tile.
tile_bvh_query_ray	Construct a ray query against a BVH object for thread-block parallel traversal.
tile_cholesky	Compute the Cholesky factorization L of a matrix A.
tile_cholesky_inplace	Compute the Cholesky factorization L of a matrix A.
tile_cholesky_solve	With L such that LL^T = A, solve for x in Ax = y
tile_cholesky_solve_inplace	With L such that LL^T = A, solve for x in Ax = y by overwriting y with x
tile_diag_add	Add a square matrix and a diagonal matrix 'd' represented as a 1D tile
tile_extract	Extract a single element from the tile.
tile_fft	Compute the forward FFT along the second dimension of a 2D tile of data.
tile_full	Allocate a tile filled with the specified value.
tile_ifft	Compute the inverse FFT along the second dimension of a 2D tile of data.
tile_load	Loads a tile from a global memory array.
tile_load_indexed	Loads a tile from a global memory array, with loads along a specified axis mapped according to a 1D tile of indices.
tile_lower_solve	Solve for z in Lz = y, where L is a lower triangular matrix.
tile_lower_solve_inplace	Solve for z in Lz = y, where L is a lower triangular matrix by overwriting y with z.
tile_map	Apply a unary function onto the tile.
tile_matmul	Computes the matrix product and accumulates out = alpha * a*b + beta * out.
tile_max	Cooperatively compute the maximum of the tile elements using all threads in the block.
tile_mesh_query_aabb	Construct an axis-aligned bounding box query against a warp.Mesh for thread-block parallel traversal.
tile_mesh_query_aabb_next	Move to the next triangle in a thread-block parallel mesh AABB query and return results as a tile.
tile_min	Cooperatively compute the minimum of the tile elements using all threads in the block.
tile_ones	Allocate a tile of one-initialized items.
tile_randf	Generate a tile of random floats.
tile_randi	Generate a tile of random integers.
tile_reduce	Apply a custom reduction operator across the tile.
tile_reshape	Return a reshaped view of a tile with the same data.
tile_scan_exclusive	Exclusive scan (prefix sum) across the tile.
tile_scan_inclusive	Inclusive scan (prefix sum) across the tile.
tile_scan_max_inclusive	Inclusive max scan across the tile.
tile_scan_min_inclusive	Inclusive min scan across the tile.
tile_sort	Cooperatively sort the elements of two tiles in ascending order based on the keys, using all threads in the block.
tile_squeeze	Return a squeezed view of a tile with the same data.
tile_store	Store a tile to a global memory array.
tile_store_indexed	Store a tile to a global memory array, with storage along a specified axis mapped according to a 1D tile of indices.
tile_sum	Cooperatively compute the sum of the tile elements using all threads in the block.
tile_transpose	Transpose a tile.
tile_upper_solve	Solve for x in Ux = z, where U is an upper triangular matrix.
tile_upper_solve_inplace	Solve for x in Ux = z, where U is an upper triangular matrix by overwriting z with x.
tile_view	Return a slice of a given tile [offset, offset+shape], if shape is not specified it will be inferred from the unspecified offset dimensions.
tile_zeros	Allocate a tile of zero-initialized items.
untile	Convert a tile back to per-thread values.

Geometry

bvh_get_group_root	Get the root of a group in a BVH.
bvh_query_aabb	Construct an axis-aligned bounding box query against a BVH object.
bvh_query_aabb_tiled	Construct an axis-aligned bounding box query against a BVH object for thread-block parallel traversal.
bvh_query_next	Move to the next bound returned by the query.
bvh_query_next_tiled	Move to the next bound in a thread-block parallel BVH query and return results as a tile.
bvh_query_ray	Construct a ray query against a BVH object.
bvh_query_ray_tiled	Construct a ray query against a BVH object for thread-block parallel traversal.
closest_point_edge_edge	Finds the closest points between two edges.
hash_grid_point_id	Return the index of a point in the warp.HashGrid.
hash_grid_query	Construct a point query against a warp.HashGrid.
hash_grid_query_next	Move to the next point in the hash grid query.
intersect_tri_tri	Tests for intersection between two triangles (v0, v1, v2) and (u0, u1, u2) using Moller's method.
mesh_eval_face_normal	Evaluates the face normal the mesh given a face index.
mesh_eval_position	Evaluates the position on the warp.Mesh given a face index and barycentric coordinates.
mesh_eval_velocity	Evaluates the velocity on the warp.Mesh given a face index and barycentric coordinates.
mesh_get	Retrieves the mesh given its index.
mesh_get_group_root	Get the root of a group in a warp.Mesh.
mesh_get_index	Returns the point-index of the mesh given a face-vertex index.
mesh_get_point	Returns the point of the mesh given a index.
mesh_get_velocity	Returns the velocity of the mesh given a index.
mesh_query_aabb	Construct an axis-aligned bounding box query against a warp.Mesh.
mesh_query_aabb_next	Move to the next triangle whose bounding box overlaps the query bounding box.
mesh_query_aabb_next_tiled	Move to the next triangle in a thread-block parallel mesh AABB query and return results as a tile.
mesh_query_aabb_tiled	Construct an axis-aligned bounding box query against a warp.Mesh for thread-block parallel traversal.
mesh_query_furthest_point_no_sign	Computes the furthest point on the mesh with identifier id to the given point in space.
mesh_query_point	Computes the closest point on the warp.Mesh with identifier id to the given point in space.
mesh_query_point_no_sign	Computes the closest point on the warp.Mesh with identifier id to the given point in space.
mesh_query_point_sign_normal	Computes the closest point on the warp.Mesh with identifier id to the given point in space.
mesh_query_point_sign_parity	Computes the closest point on the warp.Mesh with identifier id to the given point in space.
mesh_query_point_sign_winding_number	Computes the closest point on the warp.Mesh with identifier id to the given point in space.
mesh_query_ray	Computes the closest ray hit on the warp.Mesh with identifier id, returns True if a hit < max_t is found.
mesh_query_ray_anyhit	Returns True immediately upon the first ray hit on the warp.Mesh with identifier id.
mesh_query_ray_count_intersections	Count the number of intersections between a ray and a warp.Mesh.

Utility

array
atomic_add	Atomically adds value onto arr[i] and returns the original value of arr[i].
atomic_and	Atomically performs a bitwise AND between value and arr[i], atomically update the array, and return the old value.
atomic_cas	Atomically compare and swap value with arr[i] if arr[i] equals compare, and return the old value.
atomic_exch	Atomically exchange value with arr[i] and return the old value.
atomic_max	Compute the maximum of value and arr[i], atomically update the array, and return the old value.
atomic_min	Compute the minimum of value and arr[i], atomically update the array, and return the old value.
atomic_or	Atomically performs a bitwise OR between value and arr[i], atomically update the array, and return the old value.
atomic_sub	Atomically subtracts value onto arr[i] and returns the original value of arr[i].
atomic_xor	Atomically performs a bitwise XOR between value and arr[i], atomically update the array, and return the old value.
block_dim	Returns the number of threads in the current block.
breakpoint	Debugger breakpoint
cast	Reinterpret a value as a different type while preserving its bit pattern.
expect_near	Prints an error to stdout if a and b are not closer than tolerance in magnitude
len	Return the number of elements in a vector.
lerp	Linearly interpolate two values a and b using factor t, computed as a*(1-t) + b*t
print	Print variable to stdout
printf	Allows printing formatted strings using C-style format specifiers.
reversed	Returns the range in reversed order.
select	Select between two arguments, if cond is False then return value_if_false, otherwise return value_if_true.
smoothstep	Smoothly interpolate between two values a and b using a factor x, and return a result between 0 and 1 using a cubic Hermite interpolation after clamping.
tid	Return the current thread index for a 1D kernel launch.
where	Select between two arguments, if cond is True then return value_if_true, otherwise return value_if_false.
zeros

Volumes

volume_index_to_world	Transform a point uvw defined in volume index space to world space given the volume's intrinsic affine transformation.
volume_index_to_world_dir	Transform a direction uvw defined in volume index space to world space given the volume's intrinsic affine transformation.
volume_lookup	Returns the value of voxel with coordinates i, j, k for a volume of type type dtype.
volume_lookup_f	Returns the value of voxel with coordinates i, j, k.
volume_lookup_i	Returns the int32 value of voxel with coordinates i, j, k.
volume_lookup_index	Returns the index associated to the voxel with coordinates i, j, k.
volume_lookup_v	Returns the vector value of voxel with coordinates i, j, k.
volume_sample	Sample the volume of type dtype given by id at the volume local-space point uvw.
volume_sample_f	Sample the volume given by id at the volume local-space point uvw.
volume_sample_grad	Sample the volume given by id and its gradient at the volume local-space point uvw.
volume_sample_grad_f	Sample the volume and its gradient given by id at the volume local-space point uvw.
volume_sample_grad_index	Sample the volume given by id and its gradient at the volume local-space point uvw.
volume_sample_i	Sample the int32 volume given by id at the volume local-space point uvw.
volume_sample_index	Sample the volume given by id at the volume local-space point uvw.
volume_sample_v	Sample the vector volume given by id at the volume local-space point uvw.
volume_store	Store value at the voxel with coordinates i, j, k.
volume_store_f	Store value at the voxel with coordinates i, j, k.
volume_store_i	Store value at the voxel with coordinates i, j, k.
volume_store_v	Store value at the voxel with coordinates i, j, k.
volume_world_to_index	Transform a point xyz defined in volume world space to the volume's index space given the volume's intrinsic affine transformation.
volume_world_to_index_dir	Transform a direction xyz defined in volume world space to the volume's index space given the volume's intrinsic affine transformation.

Random

curlnoise	Divergence-free vector field based on the gradient of a Perlin noise function.
noise	Non-periodic Perlin-style noise in 1D.
pnoise	Periodic Perlin-style noise in 1D.
poisson	Generate a random sample from a Poisson distribution.
rand_init	Initialize a new random number generator given a user-defined seed.
randf	Return a random float between [0.0, 1.0).
randi	Return a random integer in the range [-2^31, 2^31).
randn	Sample a normal (Gaussian) distribution of mean 0 and variance 1.
randu	Return a random unsigned integer in the range [0, 2^32).
sample_cdf	Inverse-transform sample a cumulative distribution function.
sample_triangle	Uniformly sample a triangle.
sample_unit_cube	Uniformly sample a unit cube.
sample_unit_disk	Uniformly sample a disk in the xy plane.
sample_unit_hemisphere	Uniformly sample a unit hemisphere.
sample_unit_hemisphere_surface	Uniformly sample a unit hemisphere surface.
sample_unit_ring	Uniformly sample a ring in the xy plane.
sample_unit_sphere	Uniformly sample a unit sphere.
sample_unit_sphere_surface	Uniformly sample a unit sphere surface.
sample_unit_square	Uniformly sample a unit square.

Other

lower_bound

Search a sorted array arr for the closest element greater than or equal to value.

Operators

add
bit_and
bit_or
bit_xor
div
floordiv
invert
lshift
mod	Modulo operation using truncated division.
mul
neg
pos
rshift
sub
unot

Code Generation

static

Evaluate a static Python expression and replaces it with its result.