warp-lang 1.0.2__py3-none-win_amd64.whl → 1.2.0__py3-none-win_amd64.whl

warp/__init__.py

@@ -1,97 +1,108 @@
-# Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
-# NVIDIA CORPORATION and its licensors retain all intellectual property
-# and proprietary rights in and to this software, related documentation
-# and any modifications thereto.  Any use, reproduction, disclosure or
-# distribution of this software and related documentation without an express
-# license agreement from NVIDIA CORPORATION is strictly prohibited.
-
-# for autocomplete on builtins
-# from warp.stubs import *
-
-from warp.types import array, array1d, array2d, array3d, array4d, constant
-from warp.types import indexedarray, indexedarray1d, indexedarray2d, indexedarray3d, indexedarray4d
-from warp.fabric import fabricarray, fabricarrayarray, indexedfabricarray, indexedfabricarrayarray
-
-from warp.types import bool, int8, uint8, int16, uint16, int32, uint32, int64, uint64, float16, float32, float64
-from warp.types import vec2, vec2b, vec2ub, vec2s, vec2us, vec2i, vec2ui, vec2l, vec2ul, vec2h, vec2f, vec2d
-from warp.types import vec3, vec3b, vec3ub, vec3s, vec3us, vec3i, vec3ui, vec3l, vec3ul, vec3h, vec3f, vec3d
-from warp.types import vec4, vec4b, vec4ub, vec4s, vec4us, vec4i, vec4ui, vec4l, vec4ul, vec4h, vec4f, vec4d
-from warp.types import mat22, mat22h, mat22f, mat22d
-from warp.types import mat33, mat33h, mat33f, mat33d
-from warp.types import mat44, mat44h, mat44f, mat44d
-from warp.types import quat, quath, quatf, quatd
-from warp.types import transform, transformh, transformf, transformd
-from warp.types import spatial_vector, spatial_vectorh, spatial_vectorf, spatial_vectord
-from warp.types import spatial_matrix, spatial_matrixh, spatial_matrixf, spatial_matrixd
-
-# geometry types
-from warp.types import Bvh, Mesh, HashGrid, Volume, MarchingCubes
-from warp.types import bvh_query_t, hash_grid_query_t, mesh_query_aabb_t, mesh_query_point_t, mesh_query_ray_t
-
-# device-wide gemms
-from warp.types import matmul, adj_matmul, batched_matmul, adj_batched_matmul, from_ptr
-
-# deprecated
-from warp.types import vector as vec
-from warp.types import matrix as mat
-
-# numpy interop
-from warp.types import dtype_from_numpy, dtype_to_numpy
-
-from warp.context import init, func, func_grad, func_replay, func_native, kernel, struct, overload
-from warp.context import is_cpu_available, is_cuda_available, is_device_available
-from warp.context import get_devices, get_preferred_device
-from warp.context import get_cuda_devices, get_cuda_device_count, get_cuda_device, map_cuda_device, unmap_cuda_device
-from warp.context import get_device, set_device, synchronize_device
-from warp.context import (
-    zeros,
-    zeros_like,
-    ones,
-    ones_like,
-    full,
-    full_like,
-    clone,
-    empty,
-    empty_like,
-    copy,
-    from_numpy,
-    launch,
-    synchronize,
-    force_load,
-    load_module,
-)
-from warp.context import set_module_options, get_module_options, get_module
-from warp.context import capture_begin, capture_end, capture_launch
-from warp.context import Kernel, Function, Launch
-from warp.context import Stream, get_stream, set_stream, wait_stream, synchronize_stream
-from warp.context import Event, record_event, wait_event, synchronize_event
-from warp.context import RegisteredGLBuffer
-from warp.context import is_mempool_supported, is_mempool_enabled, set_mempool_enabled
-from warp.context import set_mempool_release_threshold, get_mempool_release_threshold
-from warp.context import is_mempool_access_supported, is_mempool_access_enabled, set_mempool_access_enabled
-from warp.context import is_peer_access_supported, is_peer_access_enabled, set_peer_access_enabled
-
-from warp.tape import Tape
-from warp.utils import ScopedTimer, ScopedDevice, ScopedStream
-from warp.utils import ScopedMempool, ScopedMempoolAccess, ScopedPeerAccess
-from warp.utils import ScopedCapture
-from warp.utils import transform_expand, quat_between_vectors
-
-from warp.torch import from_torch, to_torch
-from warp.torch import dtype_from_torch, dtype_to_torch
-from warp.torch import device_from_torch, device_to_torch
-from warp.torch import stream_from_torch, stream_to_torch
-
-from warp.jax import from_jax, to_jax
-from warp.jax import dtype_from_jax, dtype_to_jax
-from warp.jax import device_from_jax, device_to_jax
-
-from warp.dlpack import from_dlpack, to_dlpack
-
-from warp.constants import *
-
-from . import builtins
-
-import warp.config
-
-__version__ = warp.config.version
+# Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
+# NVIDIA CORPORATION and its licensors retain all intellectual property
+# and proprietary rights in and to this software, related documentation
+# and any modifications thereto.  Any use, reproduction, disclosure or
+# distribution of this software and related documentation without an express
+# license agreement from NVIDIA CORPORATION is strictly prohibited.
+
+# isort: skip_file
+
+# for autocomplete on builtins
+# from warp.stubs import *
+
+from warp.types import array, array1d, array2d, array3d, array4d, constant
+from warp.types import indexedarray, indexedarray1d, indexedarray2d, indexedarray3d, indexedarray4d
+from warp.fabric import fabricarray, fabricarrayarray, indexedfabricarray, indexedfabricarrayarray
+
+from warp.types import bool, int8, uint8, int16, uint16, int32, uint32, int64, uint64, float16, float32, float64
+from warp.types import vec2, vec2b, vec2ub, vec2s, vec2us, vec2i, vec2ui, vec2l, vec2ul, vec2h, vec2f, vec2d
+from warp.types import vec3, vec3b, vec3ub, vec3s, vec3us, vec3i, vec3ui, vec3l, vec3ul, vec3h, vec3f, vec3d
+from warp.types import vec4, vec4b, vec4ub, vec4s, vec4us, vec4i, vec4ui, vec4l, vec4ul, vec4h, vec4f, vec4d
+from warp.types import mat22, mat22h, mat22f, mat22d
+from warp.types import mat33, mat33h, mat33f, mat33d
+from warp.types import mat44, mat44h, mat44f, mat44d
+from warp.types import quat, quath, quatf, quatd
+from warp.types import transform, transformh, transformf, transformd
+from warp.types import spatial_vector, spatial_vectorh, spatial_vectorf, spatial_vectord
+from warp.types import spatial_matrix, spatial_matrixh, spatial_matrixf, spatial_matrixd
+
+# geometry types
+from warp.types import Bvh, Mesh, HashGrid, Volume, MarchingCubes
+from warp.types import bvh_query_t, hash_grid_query_t, mesh_query_aabb_t, mesh_query_point_t, mesh_query_ray_t
+
+# device-wide gemms
+from warp.types import matmul, adj_matmul, batched_matmul, adj_batched_matmul, from_ptr
+
+# deprecated
+from warp.types import vector as vec
+from warp.types import matrix as mat
+
+# numpy interop
+from warp.types import dtype_from_numpy, dtype_to_numpy
+
+from warp.context import init, func, func_grad, func_replay, func_native, kernel, struct, overload
+from warp.context import is_cpu_available, is_cuda_available, is_device_available
+from warp.context import get_devices, get_preferred_device
+from warp.context import get_cuda_devices, get_cuda_device_count, get_cuda_device, map_cuda_device, unmap_cuda_device
+from warp.context import get_device, set_device, synchronize_device
+from warp.context import (
+    zeros,
+    zeros_like,
+    ones,
+    ones_like,
+    full,
+    full_like,
+    clone,
+    empty,
+    empty_like,
+    copy,
+    from_numpy,
+    launch,
+    synchronize,
+    force_load,
+    load_module,
+)
+from warp.context import set_module_options, get_module_options, get_module
+from warp.context import capture_begin, capture_end, capture_launch
+from warp.context import Kernel, Function, Launch
+from warp.context import Stream, get_stream, set_stream, wait_stream, synchronize_stream
+from warp.context import Event, record_event, wait_event, synchronize_event, get_event_elapsed_time
+from warp.context import RegisteredGLBuffer
+from warp.context import is_mempool_supported, is_mempool_enabled, set_mempool_enabled
+from warp.context import set_mempool_release_threshold, get_mempool_release_threshold
+from warp.context import is_mempool_access_supported, is_mempool_access_enabled, set_mempool_access_enabled
+from warp.context import is_peer_access_supported, is_peer_access_enabled, set_peer_access_enabled
+
+from warp.tape import Tape
+from warp.utils import ScopedTimer, ScopedDevice, ScopedStream
+from warp.utils import ScopedMempool, ScopedMempoolAccess, ScopedPeerAccess
+from warp.utils import ScopedCapture
+from warp.utils import transform_expand, quat_between_vectors
+from warp.utils import TimingResult, timing_begin, timing_end, timing_print
+from warp.utils import (
+    TIMING_KERNEL,
+    TIMING_KERNEL_BUILTIN,
+    TIMING_MEMCPY,
+    TIMING_MEMSET,
+    TIMING_GRAPH,
+    TIMING_ALL,
+)
+
+from warp.torch import from_torch, to_torch
+from warp.torch import dtype_from_torch, dtype_to_torch
+from warp.torch import device_from_torch, device_to_torch
+from warp.torch import stream_from_torch, stream_to_torch
+
+from warp.jax import from_jax, to_jax
+from warp.jax import dtype_from_jax, dtype_to_jax
+from warp.jax import device_from_jax, device_to_jax
+
+from warp.dlpack import from_dlpack, to_dlpack
+
+from warp.constants import *
+
+from . import builtins
+
+import warp.config
+
+__version__ = warp.config.version

warp/__init__.pyi

@@ -1 +1 @@
-from .stubs import *
+from .stubs import *

warp/build.py

@@ -1,113 +1,88 @@
-# Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
-# NVIDIA CORPORATION and its licensors retain all intellectual property
-# and proprietary rights in and to this software, related documentation
-# and any modifications thereto.  Any use, reproduction, disclosure or
-# distribution of this software and related documentation without an express
-# license agreement from NVIDIA CORPORATION is strictly prohibited.
-
-import os
-
-import warp.config
-from warp.thirdparty import appdirs
-
-
-# builds cuda source to PTX or CUBIN using NVRTC (output type determined by output_path extension)
-def build_cuda(cu_path, arch, output_path, config="release", verify_fp=False, fast_math=False):
-    with open(cu_path, "rb") as src_file:
-        src = src_file.read()
-        cu_path = cu_path.encode("utf-8")
-        inc_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "native").encode("utf-8")
-        output_path = output_path.encode("utf-8")
-
-        if warp.config.llvm_cuda:
-            warp.context.runtime.llvm.compile_cuda(src, cu_path, inc_path, output_path, False)
-
-        else:
-            err = warp.context.runtime.core.cuda_compile_program(
-                src, arch, inc_path, config == "debug", warp.config.verbose, verify_fp, fast_math, output_path
-            )
-            if err != 0:
-                raise Exception(f"CUDA kernel build failed with error code {err}")
-
-
-# load PTX or CUBIN as a CUDA runtime module (input type determined by input_path extension)
-def load_cuda(input_path, device):
-    if not device.is_cuda:
-        raise ("Not a CUDA device")
-
-    return warp.context.runtime.core.cuda_load_module(device.context, input_path.encode("utf-8"))
-
-
-def build_cpu(obj_path, cpp_path, mode="release", verify_fp=False, fast_math=False):
-    with open(cpp_path, "rb") as cpp:
-        src = cpp.read()
-        cpp_path = cpp_path.encode("utf-8")
-        inc_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "native").encode("utf-8")
-        obj_path = obj_path.encode("utf-8")
-
-        err = warp.context.runtime.llvm.compile_cpp(src, cpp_path, inc_path, obj_path, mode == "debug", verify_fp)
-        if err != 0:
-            raise Exception(f"CPU kernel build failed with error code {err}")
-
-
-kernel_bin_dir = None
-kernel_gen_dir = None
-
-
-def init_kernel_cache(path=None):
-    """Initialize kernel cache directory.
-
-    This function is used during Warp initialization, but it can also be called directly to change the cache location.
-    If the path is not explicitly specified, a default location will be chosen based on OS-specific conventions.
-
-    To change the default cache location, set warp.config.kernel_cache_dir before calling warp.init().
-    """
-
-    if path is not None:
-        cache_root_dir = os.path.realpath(path)
-    else:
-        cache_root_dir = appdirs.user_cache_dir(appname="warp", appauthor="NVIDIA", version=warp.config.version)
-
-    cache_bin_dir = os.path.join(cache_root_dir, "bin")
-    cache_gen_dir = os.path.join(cache_root_dir, "gen")
-
-    if not os.path.isdir(cache_root_dir):
-        # print("Creating cache directory '%s'" % cache_root_dir)
-        os.makedirs(cache_root_dir, exist_ok=True)
-
-    if not os.path.isdir(cache_gen_dir):
-        # print("Creating codegen directory '%s'" % cache_gen_dir)
-        os.makedirs(cache_gen_dir, exist_ok=True)
-
-    if not os.path.isdir(cache_bin_dir):
-        # print("Creating binary directory '%s'" % cache_bin_dir)
-        os.makedirs(cache_bin_dir, exist_ok=True)
-
-    warp.config.kernel_cache_dir = cache_root_dir
-
-    global kernel_bin_dir, kernel_gen_dir
-    kernel_bin_dir = cache_bin_dir
-    kernel_gen_dir = cache_gen_dir
-
-
-def clear_kernel_cache():
-    """Clear the kernel cache."""
-
-    is_intialized = kernel_bin_dir is not None and kernel_gen_dir is not None
-    assert is_intialized, "The kernel cache directory is not configured; wp.init() has not been called yet or failed."
-
-    import glob
-
-    paths = []
-
-    if os.path.isdir(kernel_bin_dir):
-        pattern = os.path.join(kernel_bin_dir, "wp_*")
-        paths += glob.glob(pattern)
-
-    if os.path.isdir(kernel_gen_dir):
-        pattern = os.path.join(kernel_gen_dir, "wp_*")
-        paths += glob.glob(pattern)
-
-    for p in paths:
-        if os.path.isfile(p):
-            os.remove(p)
+# Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
+# NVIDIA CORPORATION and its licensors retain all intellectual property
+# and proprietary rights in and to this software, related documentation
+# and any modifications thereto.  Any use, reproduction, disclosure or
+# distribution of this software and related documentation without an express
+# license agreement from NVIDIA CORPORATION is strictly prohibited.
+
+import os
+
+import warp.config
+from warp.thirdparty import appdirs
+
+
+# builds cuda source to PTX or CUBIN using NVRTC (output type determined by output_path extension)
+def build_cuda(cu_path, arch, output_path, config="release", verify_fp=False, fast_math=False):
+    with open(cu_path, "rb") as src_file:
+        src = src_file.read()
+        cu_path = cu_path.encode("utf-8")
+        inc_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "native").encode("utf-8")
+        output_path = output_path.encode("utf-8")
+
+        if warp.config.llvm_cuda:
+            warp.context.runtime.llvm.compile_cuda(src, cu_path, inc_path, output_path, False)
+
+        else:
+            err = warp.context.runtime.core.cuda_compile_program(
+                src, arch, inc_path, config == "debug", warp.config.verbose, verify_fp, fast_math, output_path
+            )
+            if err != 0:
+                raise Exception(f"CUDA kernel build failed with error code {err}")
+
+
+# load PTX or CUBIN as a CUDA runtime module (input type determined by input_path extension)
+def load_cuda(input_path, device):
+    if not device.is_cuda:
+        raise RuntimeError("Not a CUDA device")
+
+    return warp.context.runtime.core.cuda_load_module(device.context, input_path.encode("utf-8"))
+
+
+def build_cpu(obj_path, cpp_path, mode="release", verify_fp=False, fast_math=False):
+    with open(cpp_path, "rb") as cpp:
+        src = cpp.read()
+        cpp_path = cpp_path.encode("utf-8")
+        inc_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "native").encode("utf-8")
+        obj_path = obj_path.encode("utf-8")
+
+        err = warp.context.runtime.llvm.compile_cpp(src, cpp_path, inc_path, obj_path, mode == "debug", verify_fp)
+        if err != 0:
+            raise Exception(f"CPU kernel build failed with error code {err}")
+
+
+def init_kernel_cache(path=None):
+    """Initialize kernel cache directory.
+
+    This function is used during Warp initialization, but it can also be called directly to change the cache location.
+    If the path is not explicitly specified, a default location will be chosen based on OS-specific conventions.
+
+    To change the default cache location, set warp.config.kernel_cache_dir before calling warp.init().
+    """
+
+    if path is not None:
+        cache_root_dir = os.path.realpath(path)
+    elif "WARP_CACHE_PATH" in os.environ:
+        cache_root_dir = os.path.realpath(os.environ.get("WARP_CACHE_PATH"))
+    else:
+        cache_root_dir = appdirs.user_cache_dir(appname="warp", appauthor="NVIDIA", version=warp.config.version)
+
+    warp.config.kernel_cache_dir = cache_root_dir
+
+    os.makedirs(warp.config.kernel_cache_dir, exist_ok=True)
+
+
+def clear_kernel_cache():
+    """Clear the kernel cache."""
+
+    warp.context.init()
+
+    import shutil
+
+    is_intialized = warp.context.runtime is not None
+    assert is_intialized, "The kernel cache directory is not configured; wp.init() has not been called yet or failed."
+
+    for item in os.listdir(warp.config.kernel_cache_dir):
+        item_path = os.path.join(warp.config.kernel_cache_dir, item)
+        if os.path.isdir(item_path) and item.startswith("wp_"):
+            # Remove the directory and its contents
+            shutil.rmtree(item_path, ignore_errors=True)