Add setup.py and conda-recipe (#5)

* Add setup.py

It is initial draft version.

* Replace numba.dppl with numba_dppy

Now numba.dppl is placed in separate package numba_dppy.

* Import numba_dppy as dppl

Previousply it was `from numba import dppl`.
Now it is `import numba_dppy as dppl`.

* Import numba_dppy with its name and as dppl

There are many places in code which access to `numba_dppy`.
Previously it was `numba.dppl`. It was accessible because
`from numba import dppl`.

* Add version, url, author and classifiers to setup.py

This parameters are necessary for PyPI.

* Add conda-recipe

* Use dppy in dispatcher_registry and TargetDispatcher

`dispatcher_registry.ondemand['dppl'] = init_jit` can be used only one
time. So `dppl` was renamed to `dppy`.
Also in class TargetDispatcher field `target_dppl = 'dppl'` renamed to
`dppy`.

* Use --single-version-externally-managed

This flag configure installation of `numba_dppy` as folder in
`site-packages`. The folder is required for installing compiled spirv
code to it.

* Add compilation of SPIRV code

* Add dependency to numba and dpctl in conda recipe

* Add MANIFEST.in

MANIFEST.in

@@ -0,0 +1,5 @@
+include MANIFEST.in
+include README.md setup.py LICENSE
+
+recursive-include numba_dppy *.cl
+

conda-recipe/bld.bat

@@ -0,0 +1,16 @@
+%PYTHON% setup.py install --single-version-externally-managed --record=record.txt
+if errorlevel 1 exit 1
+
+echo "Activating oneAPI compiler environment..."
+call "%ONEAPI_ROOT%\compiler\latest\env\vars.bat"
+if errorlevel 1 exit 1
+REM conda uses %ERRORLEVEL% but FPGA scripts can set it. So it should be reseted.
+set ERRORLEVEL=
+
+echo on
+
+set "CC=clang.exe"
+
+%CC% -flto -target spir64-unknown-unknown -c -x cl -emit-llvm -cl-std=CL2.0 -Xclang -finclude-default-header numba_dppy/ocl/atomics/atomic_ops.cl -o numba_dppy/ocl/atomics/atomic_ops.bc
+llvm-spirv -o numba_dppy/ocl/atomics/atomic_ops.spir numba_dppy/ocl/atomics/atomic_ops.bc
+xcopy numba_dppy\ocl\atomics\atomic_ops.spir %SP_DIR%\numba_dppy\ocl\atomics /E /Y

conda-recipe/build.sh

@@ -0,0 +1,15 @@
+#!/bin/bash
+
+${PYTHON} setup.py install --single-version-externally-managed --record=record.txt
+
+
+if [ ! -z "${ONEAPI_ROOT}" ]; then
+    source ${ONEAPI_ROOT}/compiler/latest/env/vars.sh
+    export CC=clang
+else
+    echo "DPCPP is needed to build OpenCL kernel. Abort!"
+fi
+
+${CC} -flto -target spir64-unknown-unknown -c -x cl -emit-llvm -cl-std=CL2.0 -Xclang -finclude-default-header numba_dppy/ocl/atomics/atomic_ops.cl -o numba_dppy/ocl/atomics/atomic_ops.bc
+llvm-spirv -o numba_dppy/ocl/atomics/atomic_ops.spir numba_dppy/ocl/atomics/atomic_ops.bc
+cp numba_dppy/ocl/atomics/atomic_ops.spir ${SP_DIR}/numba_dppy/ocl/atomics/

conda-recipe/meta.yaml

@@ -0,0 +1,48 @@
+package:
+    name: numba-dppy
+    version: {{ GIT_DESCRIBE_TAG }}
+
+source:
+    path: ..
+
+build:
+    number: {{ GIT_DESCRIBE_NUMBER }}
+    script_env:
+        - ONEAPI_ROOT
+
+requirements:
+    build:
+        - {{ compiler('cxx') }}
+    host:
+        - python
+        - setuptools
+        - cython
+        - llvm-spirv
+        - numba
+        - dpctl
+    run:
+        - python
+        - numba >=0.51
+        - dpctl
+        - spirv-tools
+        - llvm-spirv
+        - dpnp
+
+about:
+    home: https://github.com/IntelPython/numba-dppy
+    license: BSD-2-Clause
+    license_file: LICENSE
+    summary: "Numba extension for Intel CPU and GPU backend"
+    description: |
+        <strong>LEGAL NOTICE: Use of this software package is subject to the
+        software license agreement (as set forth above, in the license section of
+        the installed Conda package and/or the README file) and all notices,
+        disclaimers or license terms for third party or open source software
+        included in or with the software.</strong>
+        <br/><br/>
+        EULA: <a href="https://opensource.org/licenses/BSD-2-Clause" target="_blank">BSD-2-Clause</a>
+        <br/><br/>
+
+extra:
+    recipe-maintainers:
+        - PokhodenkoSA

conda-recipe/run_test.bat

@@ -0,0 +1,4 @@
+python -m numba.runtests -b -v -m -- numba_dppy.tests
+IF %ERRORLEVEL% NEQ 0 exit /B 1
+
+exit /B 0

conda-recipe/run_test.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+set -ex
+
+python -m numba.runtests -b -v -m -- numba_dppy.tests
+
+exit 0

numba_dppy/__init__.py

@@ -164,7 +164,7 @@ def foo(in, out):
 
     ---------------------------------------------------------------------------
     import numpy as np
-    from numba import dppl
+    import numba_dppy, numba_dppy as dppl
     import dpctl
 
     @dppl.kernel
@@ -516,4 +516,4 @@ def test(*args, **kwargs):
     if not dppl_present and not is_available():
         dppl_error()
 
-    return numba.testing.test("numba.dppl.tests", *args, **kwargs)
+    return numba.testing.test("numba_dppy.tests", *args, **kwargs)

numba_dppy/codegen.py

@@ -41,7 +41,7 @@ def _finalize_specific(self):
 
     def get_asm_str(self):
         # Return nothing: we can only dump assembler code when it is later
-        # generated (in numba.dppl.compiler).
+        # generated (in numba_dppy.compiler).
         return None
 
 

numba_dppy/dispatcher.py

@@ -4,8 +4,8 @@
 
 #from numba.targets.descriptors import TargetDescriptor
 #from numba.targets.options import TargetOptions
-#from numba import dppl
-from numba.dppl import kernel, autojit
+#import numba_dppy, numba_dppy as dppl
+from numba_dppy import kernel, autojit
 from .descriptor import dppl_target
 #from numba.npyufunc.deviceufunc import (UFuncMechanism, GenerializedUFunc,
  #                                       GUFuncCallSteps)

numba_dppy/dppl_lowerer.py

@@ -12,7 +12,7 @@
 from numba.core import (compiler, ir, types, sigutils, lowering,
                 funcdesc, config)
 from numba.parfors import parfor
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 from numba.core.ir_utils import (add_offset_to_labels,
                             replace_var_names,
                             remove_dels,
@@ -40,7 +40,7 @@
 from .dufunc_inliner import dufunc_inliner
 from . import dppl_host_fn_call_gen as dppl_call_gen
 import dpctl
-from numba.dppl.target import DPPLTargetContext
+from numba_dppy.target import DPPLTargetContext
 
 
 def _print_block(block):
@@ -329,7 +329,7 @@ def addrspace_from(params, def_addr):
         return addrspaces
 
     addrspaces = addrspace_from(parfor_params,
-                                numba.dppl.target.SPIR_GLOBAL_ADDRSPACE)
+                                numba_dppy.target.SPIR_GLOBAL_ADDRSPACE)
 
     if config.DEBUG_ARRAY_OPT >= 1:
         print("parfor_params = ", parfor_params, type(parfor_params))
@@ -607,7 +607,7 @@ def print_arg_with_addrspaces(args):
         print('after DUFunc inline'.center(80, '-'))
         gufunc_ir.dump()
 
-    kernel_func = numba.dppl.compiler.compile_kernel_parfor(
+    kernel_func = numba_dppy.compiler.compile_kernel_parfor(
         dpctl.get_current_queue(),
         gufunc_ir,
         gufunc_param_types,
@@ -975,7 +975,7 @@ def relatively_deep_copy(obj, memo):
     from numba.core.types.functions import Function, Dispatcher
     from numba.core.bytecode import FunctionIdentity
     from numba.core.typing.templates import Signature
-    from numba.dppl.compiler import DPPLFunctionTemplate
+    from numba_dppy.compiler import DPPLFunctionTemplate
     from numba.core.compiler import CompileResult
     from numba.np.ufunc.dufunc import DUFunc
     from ctypes import _CFuncPtr
@@ -1169,7 +1169,7 @@ def lower(self):
             self.base_lower = self.gpu_lower
             lowering.lower_extensions[parfor.Parfor].pop()
         except Exception as e:
-            if numba.dppl.compiler.DEBUG:
+            if numba_dppy.compiler.DEBUG:
                 print("Failed to lower parfor on DPPL-device. Due to:\n", e)
             lowering.lower_extensions[parfor.Parfor].pop()
             if (lowering.lower_extensions[parfor.Parfor][-1] == numba.parfors.parfor_lowering._lower_parfor_parallel):
@@ -1195,7 +1195,7 @@ def copy_block(block):
 def lower_parfor_rollback(lowerer, parfor):
     try:
         _lower_parfor_gufunc(lowerer, parfor)
-        if numba.dppl.compiler.DEBUG:
+        if numba_dppy.compiler.DEBUG:
             msg = "Parfor lowered on DPPL-device"
             print(msg, parfor.loc)
     except Exception as e:

numba_dppy/dppl_offload_dispatcher.py

@@ -8,7 +8,7 @@ class DpplOffloadDispatcher(dispatcher.Dispatcher):
 
     def __init__(self, py_func, locals={}, targetoptions={}, impl_kind='direct', pipeline_class=compiler.Compiler):
         if dppl_config.dppl_present:
-            from numba.dppl.compiler import DPPLCompiler
+            from numba_dppy.compiler import DPPLCompiler
             targetoptions['parallel'] = True
             dispatcher.Dispatcher.__init__(self, py_func, locals=locals,
                     targetoptions=targetoptions, impl_kind=impl_kind, pipeline_class=DPPLCompiler)

numba_dppy/dppl_passes.py

@@ -33,7 +33,7 @@
 def dpnp_available():
     try:
        # import dpnp
-        from numba.dppl.dpnp_glue import dpnp_fptr_interface as dpnp_glue
+        from numba_dppy.dpnp_glue import dpnp_fptr_interface as dpnp_glue
         return True
     except:
         return False

numba_dppy/examples/dppl_func.py

@@ -1,6 +1,6 @@
 import sys
 import numpy as np
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 import math
 
 import dpctl

numba_dppy/examples/dppl_with_context.py

@@ -1,5 +1,6 @@
 import numpy as np
-from numba import dppl, njit, prange
+from numba import njit, prange
+import numba_dppy, numba_dppy as dppl
 import dpctl
 
 @njit

numba_dppy/examples/matmul.py

@@ -4,7 +4,7 @@
 
 import sys
 import numpy as np
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 import dpctl
 
 

numba_dppy/examples/pairwise_distance.py

@@ -6,7 +6,7 @@
 import argparse
 import timeit
 
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 import dpctl
 import dpctl._memory as dpctl_mem
 

numba_dppy/examples/sum-hybrid.py

@@ -4,7 +4,7 @@
 
 import sys
 import numpy as np
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 import dpctl
 
 

numba_dppy/examples/sum.py

@@ -4,7 +4,7 @@
 
 import sys
 import numpy as np
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 import dpctl
 
 

numba_dppy/examples/sum2D.py

@@ -4,7 +4,7 @@
 
 import sys
 import numpy as np
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 import dpctl
 
 

numba_dppy/examples/sum_ndarray.py

@@ -4,7 +4,7 @@
 
 import sys
 import numpy as np
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 import dpctl
 
 

numba_dppy/examples/sum_reduction.py

@@ -4,7 +4,7 @@
 import math
 import time
 
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 import dpctl
 
 

numba_dppy/examples/sum_reduction_ocl.py

@@ -1,6 +1,7 @@
 import sys
 import numpy as np
-from numba import dppl, int32
+from numba import int32
+import numba_dppy, numba_dppy as dppl
 import math
 
 import dpctl

numba_dppy/examples/sum_reduction_recursive_ocl.py

@@ -1,6 +1,7 @@
 import sys
 import numpy as np
-from numba import dppl, int32
+from numba import int32
+import numba_dppy, numba_dppy as dppl
 import math
 
 import dpctl

numba_dppy/initialize.py

@@ -5,12 +5,12 @@
 
 
 def init_jit():
-    from numba.dppl.dispatcher import DPPLDispatcher
+    from numba_dppy.dispatcher import DPPLDispatcher
     return DPPLDispatcher
 
 def initialize_all():
     from numba.core.registry import dispatcher_registry
-    dispatcher_registry.ondemand['dppl'] = init_jit
+    dispatcher_registry.ondemand['dppy'] = init_jit
 
     import dpctl
     import glob

numba_dppy/ocl/ocldecl.py

@@ -4,7 +4,7 @@
 from numba.core.typing.templates import (AttributeTemplate, ConcreteTemplate,
                                         AbstractTemplate, MacroTemplate,
                                         signature, Registry)
-from numba import dppl
+import numba_dppy, numba_dppy as dppl
 
 registry = Registry()
 intrinsic = registry.register

numba_dppy/ocl/oclimpl.py

@@ -11,9 +11,9 @@
 from numba.core.imputils import Registry
 from numba.core import cgutils, types
 from numba.core.itanium_mangler import mangle_c, mangle, mangle_type
-from numba.dppl import target
+from numba_dppy import target
 from . import stubs
-from numba.dppl.codegen import SPIR_DATA_LAYOUT
+from numba_dppy.codegen import SPIR_DATA_LAYOUT
 
 
 registry = Registry()

numba_dppy/ocl/stubs.py

@@ -2,7 +2,7 @@
 from numba.core import types, ir, typing
 from numba.core.rewrites.macros import Macro
 
-from numba.dppl.target import SPIR_LOCAL_ADDRSPACE
+from numba_dppy.target import SPIR_LOCAL_ADDRSPACE
 
 _stub_error = NotImplementedError("This is a stub.")
 

numba_dppy/spirv_generator.py

@@ -7,7 +7,7 @@
 import tempfile
 
 from numba import config
-from numba.dppl.target import LINK_ATOMIC
+from numba_dppy.target import LINK_ATOMIC
 
 
 def _raise_bad_env_path(msg, path, extra=None):

numba_dppy/target.py

@@ -74,7 +74,7 @@ def _init_data_model_manager():
 
 def _replace_numpy_ufunc_with_opencl_supported_functions():
     from numba.np.ufunc_db import _ufunc_db as ufunc_db
-    from numba.dppl.ocl.mathimpl import lower_ocl_impl, sig_mapper
+    from numba_dppy.ocl.mathimpl import lower_ocl_impl, sig_mapper
 
     ufuncs = [("fabs", np.fabs), ("exp", np.exp), ("log", np.log),
               ("log10", np.log10), ("expm1", np.expm1), ("log1p", np.log1p),
@@ -96,7 +96,7 @@ class DPPLTargetContext(BaseContext):
     generic_addrspace = SPIR_GENERIC_ADDRSPACE
 
     def init(self):
-        self._internal_codegen = codegen.JITSPIRVCodegen("numba.dppl.jit")
+        self._internal_codegen = codegen.JITSPIRVCodegen("numba_dppy.jit")
         self._target_data = (ll.create_target_data(codegen
                                 .SPIR_DATA_LAYOUT[utils.MACHINE_BITS]))
         # Override data model manager to SPIR model
@@ -117,7 +117,7 @@ def init(self):
 
 
     def replace_numpy_ufunc_with_opencl_supported_functions(self):
-        from numba.dppl.ocl.mathimpl import lower_ocl_impl, sig_mapper
+        from numba_dppy.ocl.mathimpl import lower_ocl_impl, sig_mapper
 
         ufuncs = [("fabs", np.fabs), ("exp", np.exp), ("log", np.log),
                   ("log10", np.log10), ("expm1", np.expm1), ("log1p", np.log1p),