numba_vulkan.errors#
Exceptions raised by the Vulkan target.
Exceptions#
A kernel computes in float32 where its code says float64. |
|
A kernel computes with float64, which is slow on most GPUs. |
|
Vulkan's validation layer found a problem in how a device is used. |
|
A device lacks a capability that a kernel needs. |
Classes#
A kernel uses something Vulkan compute shaders cannot express. |
|
LLVM's SPIR-V backend could not translate a kernel. |
Module Contents#
- class numba_vulkan.errors.VulkanUnsupportedError#
A kernel uses something Vulkan compute shaders cannot express.
Raised while compiling, for example for slices, recursion or float64 transcendental functions. It is a numba.core.errors.NumbaError, so Numba adds the source location of the offending statement.
- class numba_vulkan.errors.SpirvCodegenError#
LLVM’s SPIR-V backend could not translate a kernel.
Raised when the backend aborts, when the generated module fails the built-in sanity check, or when
spirv-valrejects it (withNUMBA_VULKAN_VALIDATE=1). It usually points at a gap in this package rather than at a mistake in the kernel.
- exception numba_vulkan.errors.VulkanPrecisionWarning#
A kernel computes in float32 where its code says float64.
Issued once per kernel when it is compiled for a device without float64 support; see numba_vulkan.narrowing.
Initialize self. See help(type(self)) for accurate signature.
- exception numba_vulkan.errors.VulkanPerformanceWarning#
A kernel computes with float64, which is slow on most GPUs.
Issued once per kernel when it is compiled.
NUMBA_VULKAN_WARNINGS=0silences it, as doeswarnings.filterwarnings("ignore", category=nv.VulkanPerformanceWarning).Initialize self. See help(type(self)) for accurate signature.
- exception numba_vulkan.errors.VulkanValidationWarning#
Vulkan’s validation layer found a problem in how a device is used.
Issued only with
NUMBA_VULKAN_DEBUG=1, for every warning and error the layer reports; numba_vulkan.runtime.validation_messages returns them as well.Initialize self. See help(type(self)) for accurate signature.
- exception numba_vulkan.errors.VulkanSupportError#
A device lacks a capability that a kernel needs.
Raised when a kernel is launched, for example on a device without float64 support, or when no usable Vulkan device exists.
Initialize self. See help(type(self)) for accurate signature.