xobjects 0.6.2__tar.gz → 0.6.4__tar.gz

{xobjects-0.6.2/xobjects.egg-info → xobjects-0.6.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: xobjects
-Version: 0.6.2
+Version: 0.6.4
 Summary: In-memory serialization and code generator for CPU and GPU
 Home-page: https://xsuite.readthedocs.io/
 Download-URL: https://pypi.python.org/pypi/xobjects

{xobjects-0.6.2 → xobjects-0.6.4}/tests/test_kernel.py

@@ -4,7 +4,9 @@
 # ########################################### #
 
 import cffi
+import pytest
 import sysconfig
+from pathlib import Path
 
 import numpy as np
 

xobjects-0.6.4/xobjects/_version.py

@@ -0,0 +1 @@
+__version__ = "0.6.4"

{xobjects-0.6.2 → xobjects-0.6.4}/xobjects/context_cpu.py

@@ -150,8 +150,7 @@ class ContextCpu(XContext):
         """
         super().__init__()
         self.omp_num_threads = omp_num_threads
-        if omp_num_threads == 0:
-            self.allow_prebuilt_kernels = True
+        self.allow_prebuilt_kernels = True
 
     def __str__(self):
         if not self.openmp_enabled:
@@ -415,8 +414,18 @@ class ContextCpu(XContext):
                 log.debug(f"cffi def {pyname} {signature}")
 
         if self.openmp_enabled:
-            ffi_interface.cdef("void omp_set_num_threads(int);")
-            ffi_interface.cdef("int omp_get_max_threads();")
+            # The wrapper is needed to ensure that the omp functions are linked
+            ffi_interface.cdef("void xo_omp_set_num_threads(int);")
+            ffi_interface.cdef("int xo_omp_get_max_threads();")
+            specialized_source += """
+                void xo_omp_set_num_threads(int num_threads) {
+                    omp_set_num_threads(num_threads);
+                }
+
+                int xo_omp_get_max_threads(void) {
+                    return omp_get_max_threads();
+                }
+                """
 
         # Compile
         xtr_compile_args = ["-std=c99", "-DXO_CONTEXT_CPU"]
@@ -528,8 +537,8 @@ class ContextCpu(XContext):
         spec.loader.exec_module(module)
 
         if self.openmp_enabled:
-            self.omp_set_num_threads = module.lib.omp_set_num_threads
-            self.omp_get_max_threads = module.lib.omp_get_max_threads
+            self.omp_set_num_threads = module.lib.xo_omp_set_num_threads
+            self.omp_get_max_threads = module.lib.xo_omp_get_max_threads
 
         return module
 
@@ -768,10 +777,8 @@ class BufferNumpy(XBuffer):
         value = nplike_to_numpy(value)
         if dest_dtype != value.dtype:
             value = value.astype(dtype=dest_dtype)  # make a copy
-        src = value.view("int8")
-        self.buffer[offset : offset + src.nbytes] = value.flatten().view(
-            "int8"
-        )
+        src = value.flatten().view("int8")
+        self.buffer[offset : offset + src.nbytes] = src
 
     def to_bytearray(self, offset, nbytes):
         """copy in byte array: used in update_from_xbuffer"""

{xobjects-0.6.2 → xobjects-0.6.4}/xobjects/context_cupy.py

@@ -4,6 +4,7 @@
 # ########################################### #
 
 import logging
+import warnings
 from typing import Dict, List, Tuple
 
 import numpy as np
@@ -32,6 +33,7 @@ try:
     import cupyx.scipy.special
     import cupyx.scipy.stats
     from cupyx.scipy import fftpack as cufftp
+    from cupy_backends.cuda.libs import nvrtc
 
     _enabled = True
 except ImportError:
@@ -40,6 +42,7 @@ except ImportError:
         message=("cupy is not installed. " "ContextCupy is not available!")
     )
     cufftp = cupy
+    nvrtc = None
     _enabled = False
 
 if _enabled:
@@ -466,12 +469,14 @@ class ContextCupy(XContext):
             *extra_compile_args,
             *include_flags,
             "-DXO_CONTEXT_CUDA",
-            # Skip heavy optimizations (e.g. involving cloning),
-            # which for us don't translate to a lot of runtime gains,
-            # but consume a lot of compile time and memory:
-            "--Ofast-compile=min",
         )
 
+        if nvrtc and nvrtc.getVersion() >= (12, 9):
+            # If supported, skip prohibitively heavy optimisations (e.g.
+            # involving cloning). This it at the expense of <20%
+            # runtime performance, but gain of a lot of compile time and memory.
+            extra_compile_args += ("--Ofast-compile=min",)
+
         module = cupy.RawModule(
             code=specialized_source, options=extra_compile_args
         )

{xobjects-0.6.2 → xobjects-0.6.4}/xobjects/struct.py

@@ -518,7 +518,6 @@ class Struct(metaclass=MetaStruct):
         extra_classes=(),
         extra_compile_args=(),
     ):
-
         if context.allow_prebuilt_kernels:
             _print_state = Print.suppress
             Print.suppress = True
@@ -532,6 +531,7 @@ class Struct(metaclass=MetaStruct):
                     config={},
                     tracker_element_classes=[],
                     classes=list(extra_classes) + [cls],
+                    context=context,
                 )
             except ImportError:
                 kernel_info = None

{xobjects-0.6.2 → xobjects-0.6.4/xobjects.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: xobjects
-Version: 0.6.2
+Version: 0.6.4
 Summary: In-memory serialization and code generator for CPU and GPU
 Home-page: https://xsuite.readthedocs.io/
 Download-URL: https://pypi.python.org/pypi/xobjects

xobjects-0.6.2/xobjects/_version.py

@@ -1 +0,0 @@
-__version__ = "0.6.2"