warp-lang 0.11.0__py3-none-manylinux2014_x86_64.whl → 1.0.0__py3-none-manylinux2014_x86_64.whl

warp/tests/test_verify_fp.py

@@ -0,0 +1,65 @@
+# Copyright (c) 2024 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 unittest
+
+import warp as wp
+from warp.tests.unittest_utils import *
+
+wp.init()
+
+
+def setUpModule():
+    wp.config.verify_fp = True  # Enable checking floating-point values to be finite
+
+
+def tearDownModule():
+    wp.config.verify_fp = False
+
+
+@wp.struct
+class TestStruct:
+    field: wp.float32
+
+
+@wp.kernel
+def finite_kernel(foos: wp.array(dtype=TestStruct)):
+    i = wp.tid()
+    foos[i].field += wp.float32(1.0)
+
+
+def test_finite(test, device):
+    foos = wp.zeros((10,), dtype=TestStruct, device=device)
+
+    wp.launch(
+        kernel=finite_kernel,
+        dim=(10,),
+        inputs=[foos],
+        device=device,
+    )
+    wp.synchronize()
+
+    expected = TestStruct()
+    expected.field = 1.0
+    for f in foos.list():
+        if f.field != expected.field:
+            raise AssertionError(f"Unexpected result, got: {f} expected: {expected}")
+
+
+devices = get_test_devices()
+
+
+class TestVerifyFP(unittest.TestCase):
+    pass
+
+
+add_function_test(TestVerifyFP, "test_finite", test_finite, devices=devices)
+
+
+if __name__ == "__main__":
+    wp.build.clear_kernel_cache()
+    unittest.main(verbosity=2)

warp/tests/test_volume.py

@@ -627,7 +627,7 @@ def test_volume_from_numpy(test, device):
 
     sphere_vdb_array = sphere_vdb.array()
     test.assertEqual(sphere_vdb_array.dtype, wp.uint8)
-    test.assertFalse(sphere_vdb_array.owner)
+    test.assertIsNone(sphere_vdb_array.deleter)
 
 
 class TestVolume(unittest.TestCase):

warp/tests/unittest_serial.py

@@ -8,6 +8,8 @@
 import warp as wp
 from warp.tests.unittest_utils import TeamCityTestRunner
 
+wp.init()
+
 
 def run_suite() -> bool:
     """Run a test suite"""

warp/tests/unittest_suites.py

@@ -91,6 +91,7 @@ def default_suite(test_loader: unittest.TestLoader):
     from warp.tests.test_arithmetic import TestArithmetic
     from warp.tests.test_array import TestArray
     from warp.tests.test_array_reduce import TestArrayReduce
+    from warp.tests.test_async import TestAsync
     from warp.tests.test_atomic import TestAtomic
     from warp.tests.test_bool import TestBool
     from warp.tests.test_builtins_resolution import TestBuiltinsResolution
@@ -117,6 +118,7 @@ def default_suite(test_loader: unittest.TestLoader):
     from warp.tests.test_import import TestImport
     from warp.tests.test_indexedarray import TestIndexedArray
     from warp.tests.test_intersect import TestIntersect
+    from warp.tests.test_jax import TestJax
     from warp.tests.test_large import TestLarge
     from warp.tests.test_launch import TestLaunch
     from warp.tests.test_lerp import TestLerp
@@ -129,6 +131,7 @@ def default_suite(test_loader: unittest.TestLoader):
     from warp.tests.test_math import TestMath
     from warp.tests.test_matmul import TestMatmul
     from warp.tests.test_matmul_lite import TestMatmulLite
+    from warp.tests.test_mempool import TestMempool
     from warp.tests.test_mesh import TestMesh
     from warp.tests.test_mesh_query_aabb import TestMeshQueryAABBMethods
     from warp.tests.test_mesh_query_point import TestMeshQueryPoint
@@ -140,6 +143,7 @@ def default_suite(test_loader: unittest.TestLoader):
     from warp.tests.test_noise import TestNoise
     from warp.tests.test_operators import TestOperators
     from warp.tests.test_options import TestOptions
+    from warp.tests.test_peer import TestPeer
     from warp.tests.test_pinned import TestPinned
     from warp.tests.test_print import TestPrint
     from warp.tests.test_quat import TestQuat
@@ -147,6 +151,7 @@ def default_suite(test_loader: unittest.TestLoader):
     from warp.tests.test_reload import TestReload
     from warp.tests.test_rounding import TestRounding
     from warp.tests.test_runlength_encode import TestRunlengthEncode
+    from warp.tests.test_sim_kinematics import TestSimKinematics
     from warp.tests.test_smoothstep import TestSmoothstep
     from warp.tests.test_snippet import TestSnippets
     from warp.tests.test_sparse import TestSparse
@@ -161,6 +166,7 @@ def default_suite(test_loader: unittest.TestLoader):
     from warp.tests.test_vec import TestVec
     from warp.tests.test_vec_lite import TestVecLite
     from warp.tests.test_vec_scalar_ops import TestVecScalarOps
+    from warp.tests.test_verify_fp import TestVerifyFP
     from warp.tests.test_volume import TestVolume
     from warp.tests.test_volume_write import TestVolumeWrite
 
@@ -169,6 +175,7 @@ def default_suite(test_loader: unittest.TestLoader):
         TestArithmetic,
         TestArray,
         TestArrayReduce,
+        TestAsync,
         TestAtomic,
         TestBool,
         TestBuiltinsResolution,
@@ -198,6 +205,7 @@ def default_suite(test_loader: unittest.TestLoader):
         TestImport,
         TestIndexedArray,
         TestIntersect,
+        TestJax,
         TestLarge,
         TestLaunch,
         TestLerp,
@@ -210,6 +218,7 @@ def default_suite(test_loader: unittest.TestLoader):
         TestMath,
         TestMatmul,
         TestMatmulLite,
+        TestMempool,
         TestMesh,
         TestMeshQueryAABBMethods,
         TestMeshQueryPoint,
@@ -221,6 +230,7 @@ def default_suite(test_loader: unittest.TestLoader):
         TestNoise,
         TestOperators,
         TestOptions,
+        TestPeer,
         TestPinned,
         TestPrint,
         TestQuat,
@@ -228,6 +238,7 @@ def default_suite(test_loader: unittest.TestLoader):
         TestReload,
         TestRounding,
         TestRunlengthEncode,
+        TestSimKinematics,
         TestSmoothstep,
         TestSparse,
         TestSnippets,
@@ -242,6 +253,7 @@ def default_suite(test_loader: unittest.TestLoader):
         TestVec,
         TestVecLite,
         TestVecScalarOps,
+        TestVerifyFP,
         TestVolume,
         TestVolumeWrite,
     ]

warp/tests/unittest_utils.py

@@ -7,6 +7,7 @@
 
 import ctypes
 import ctypes.util
+import math
 import os
 import sys
 import time
@@ -101,6 +102,11 @@ def get_test_devices(mode=None):
     return devices
 
 
+def get_cuda_test_devices(mode=None):
+    devices = get_test_devices(mode=mode)
+    return [d for d in devices if d.is_cuda]
+
+
 # redirects and captures all stdout output (including from C-libs)
 class StdOutCapture:
     def begin(self):
@@ -127,11 +133,12 @@ class StdOutCapture:
         sys.stdout = self.tempfile
 
     def end(self):
-        if sys.platform == "win32":
-            # Workaround for what seems to be a Windows-specific bug where
-            # the output of CUDA's `printf` is not being immediately flushed
-            # despite the context synchronisation.
-            time.sleep(0.01)
+        # The following sleep doesn't seem to fix the test_print failure on Windows
+        # if sys.platform == "win32":
+        #    # Workaround for what seems to be a Windows-specific bug where
+        #    # the output of CUDA's `printf` is not being immediately flushed
+        #    # despite the context synchronisation.
+        #    time.sleep(0.01)
 
         os.dup2(self.target, self.saved.fileno())
         os.close(self.target)
@@ -185,9 +192,9 @@ def assert_np_equal(result, expect, tol=0.0):
     else:
         delta = a - b
         err = np.max(np.abs(delta))
-        if err > tol:
+        if err > tol or math.isnan(err):
             raise AssertionError(
-                f"Maximum expected error exceeds tolerance got: {a}, expected: {b}, with err: {err} > {tol}"
+                f"Maximum expected error exceeds absolute tolerance got: {a}, expected: {b}, with err: {err} > {tol}"
             )
 
 

warp/thirdparty/unittest_parallel.py

@@ -53,7 +53,16 @@ def main(argv=None):
     """
 
     # Command line arguments
-    parser = argparse.ArgumentParser(prog="unittest-parallel")
+    parser = argparse.ArgumentParser(
+        prog="unittest-parallel",
+        # NVIDIA Modifications follow:
+        formatter_class=argparse.RawTextHelpFormatter,
+        epilog="""Example usage:
+        python -m warp.tests -s autodetect -p 'test_a*.py'
+        python -m warp.tests -s kit
+        python -m warp.tests -k 'mgpu' -k 'cuda'
+        """,
+    )
     # parser.add_argument("-v", "--verbose", action="store_const", const=2, default=1, help="Verbose output")
     parser.add_argument("-q", "--quiet", dest="verbose", action="store_const", const=0, default=2, help="Quiet output")
     parser.add_argument("-f", "--failfast", action="store_true", default=False, help="Stop on first fail or error")
@@ -238,6 +247,8 @@ def main(argv=None):
 
             import warp as wp
 
+            wp.init()
+
             # force rebuild of all kernels
             wp.build.clear_kernel_cache()
             print("Cleared Warp kernel cache")
@@ -449,6 +460,7 @@ class ParallelTestManager:
             # Clean up kernel cache (NVIDIA modification)
             import warp as wp
 
+            wp.init()
             wp.build.clear_kernel_cache()
 
             # Return (test_count, errors, failures, skipped_count, expected_failure_count, unexpected_success_count)
@@ -537,11 +549,11 @@ def set_worker_cache(args, temp_dir):
             cache_root_dir = os.path.join(os.getenv("WARP_CACHE_ROOT"), f"{wp.config.version}-{pid}")
         else:
             cache_root_dir = appdirs.user_cache_dir(
-                appname="warp", appauthor="NVIDIA Corporation", version=f"{wp.config.version}-{pid}"
+                appname="warp", appauthor="NVIDIA", version=f"{wp.config.version}-{pid}"
             )
 
         wp.config.kernel_cache_dir = cache_root_dir
-
+        wp.init()
         wp.build.clear_kernel_cache()
 
 

warp/torch.py

@@ -121,6 +121,7 @@ def from_torch(t, dtype=None, requires_grad=None, grad=None):
 
     shape = tuple(t.shape)
     strides = tuple(s * ctype_size for s in t.stride())
+    device = device_from_torch(t.device)
 
     # if target is a vector or matrix type
     # then check if trailing dimensions match
@@ -157,20 +158,21 @@ def from_torch(t, dtype=None, requires_grad=None, grad=None):
     elif requires_grad:
         # wrap the tensor gradient, allocate if necessary
         if t.grad is None:
-            # allocate a zero-filled gradient tensor if it doesn't exist
-            import torch
-
-            t.grad = torch.zeros_like(t, requires_grad=False)
-        grad = from_torch(t.grad, dtype=dtype)
+            # allocate a zero-filled gradient if it doesn't exist
+            # Note: we use Warp to allocate the shared gradient with compatible strides
+            grad = warp.zeros(dtype=dtype, shape=shape, strides=strides, device=device)
+            t.grad = to_torch(grad, requires_grad=False)
+        else:
+            # TODO: this will fail if the strides are incompatible
+            grad = from_torch(t.grad, dtype=dtype)
 
-    a = warp.types.array(
+    a = warp.array(
         ptr=t.data_ptr(),
         dtype=dtype,
         shape=shape,
         strides=strides,
-        device=device_from_torch(t.device),
+        device=device,
         copy=False,
-        owner=False,
         grad=grad,
         requires_grad=requires_grad,
     )