move gpuctypes in tree (#3253)

* move gpuctypes in tree

* fix mypy

* regex exclude

* autogen sh

* mypy exclude

* does that fix it

* fix mypy

* add hip confirm

* verify all autogens

* build clang2py

* opencl headers

* gpu on 22.04

.github/workflows/test.yml

@@ -328,6 +328,13 @@ jobs:
         sudo apt install --no-install-recommends --allow-unauthenticated -y rocm-hip-libraries hip-dev
     - name: Install Python Dependencies
       run: pip install -e '.[testing]' --extra-index-url https://download.pytorch.org/whl/cpu
+    - name: Verify HIP autogen
+      run: |
+        cp tinygrad/autogen/hip.py /tmp/hip.py.bak
+        cp tinygrad/autogen/comgr.py /tmp/comgr.py.bak
+        ./autogen_stubs.sh hip
+        diff /tmp/hip.py.bak tinygrad/autogen/hip.py
+        diff /tmp/comgr.py.bak tinygrad/autogen/comgr.py
     - name: Test HIP compilation on RDNA3 [gfx1100]
       # test/test_symbolic_ops.py can't run here, it was comparing empty memory
       run: |
@@ -374,7 +381,7 @@ jobs:
         backend: [llvm, clang, gpu, cuda] #, triton] #, ptx]
 
     name: Tests on (${{ matrix.backend }})
-    runs-on: ${{ matrix.backend == 'gpu'  && 'ubuntu-20.04' || 'ubuntu-latest' }}
+    runs-on: ubuntu-latest
     timeout-minutes: 20
 
     steps:
@@ -402,7 +409,7 @@ jobs:
           echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
           echo "deb [ allow-insecure=yes ] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list
           sudo apt update
-          sudo apt install --allow-unauthenticated -y --no-install-recommends \
+          sudo apt install --allow-unauthenticated -y --no-install-recommends opencl-headers \
             intel-oneapi-runtime-openmp=2023.2.1-16 intel-oneapi-runtime-compilers-common=2023.2.1-16 intel-oneapi-runtime-compilers=2023.2.1-16 \
             intel-oneapi-runtime-dpcpp-sycl-opencl-cpu=2023.2.1-16 intel-oneapi-runtime-tbb-common=2021.10.0-49541 \
             intel-oneapi-runtime-tbb=2021.10.0-49541 intel-oneapi-runtime-opencl=2023.2.1-16
@@ -443,6 +450,18 @@ jobs:
         run: |
           python -c "from tinygrad import Device; assert Device.DEFAULT in ['LLVM','CLANG','CUDA','GPU'], Device.DEFAULT"
           DEBUG=5 PYTHONPATH=${{ github.workspace }} FORWARD_ONLY=1 python3 test/test_ops.py TestOps.test_add
+      - name: Verify OpenCL autogen
+        if: matrix.backend == 'gpu'
+        run: |
+          cp tinygrad/autogen/opencl.py /tmp/opencl.py.bak
+          ./autogen_stubs.sh opencl
+          diff /tmp/opencl.py.bak tinygrad/autogen/opencl.py
+      - name: Verify CUDA autogen
+        if: matrix.backend == 'cuda'
+        run: |
+          cp tinygrad/autogen/cuda.py /tmp/cuda.py.bak
+          ./autogen_stubs.sh cuda
+          diff /tmp/cuda.py.bak tinygrad/autogen/cuda.py
       - name: Run pytest (not cuda)
         if: matrix.backend!='cuda' && matrix.backend!='ptx' && matrix.backend!='triton'
         run: python -m pytest -n=auto test/ --durations=20

autogen_stubs.sh

@@ -0,0 +1,62 @@
+#!/bin/bash -e
+
+# setup instructions for clang2py
+if [[ ! $(clang2py -V) ]]; then
+  pushd .
+  cd /tmp
+  sudo apt-get install -y --no-install-recommends clang
+  pip install clang==14.0.6
+  git clone https://github.com/geohot/ctypeslib.git
+  cd ctypeslib
+  pip install --user .
+  clang2py -V
+  popd
+fi
+
+generate_opencl() {
+  clang2py /usr/include/CL/cl.h -o autogen/opencl.py -l /usr/lib/x86_64-linux-gnu/libOpenCL.so.1 -k cdefstum
+  sed -i '1s/^/# mypy: ignore-errors\n/' autogen/opencl.py
+  grep FIXME_STUB autogen/opencl.py || true
+  # hot patches
+  sed -i "s\import ctypes\import ctypes, ctypes.util\g" autogen/opencl.py
+  sed -i "s\ctypes.CDLL('/usr/lib/x86_64-linux-gnu/libOpenCL.so.1')\ctypes.CDLL(ctypes.util.find_library('OpenCL'))\g" autogen/opencl.py
+  python3 -c "import autogen.opencl"
+}
+
+generate_hip() {
+  clang2py /opt/rocm/include/hip/hip_ext.h /opt/rocm/include/hip/hiprtc.h \
+  /opt/rocm/include/hip/hip_runtime_api.h /opt/rocm/include/hip/driver_types.h \
+  --clang-args="-D__HIP_PLATFORM_AMD__ -I/opt/rocm/include -x c++" -o autogen/hip.py -l /opt/rocm/lib/libamdhip64.so
+  echo "hipDeviceProp_t = hipDeviceProp_tR0600" >> autogen/hip.py
+  echo "hipGetDeviceProperties = hipGetDevicePropertiesR0600" >> autogen/hip.py
+  sed -i '1s/^/# mypy: ignore-errors\n/' autogen/hip.py
+  grep FIXME_STUB autogen/hip.py || true
+  # we can trust HIP is always at /opt/rocm/lib
+  #sed -i "s\import ctypes\import ctypes, ctypes.util\g" autogen/hip.py
+  #sed -i "s\ctypes.CDLL('/opt/rocm/lib/libhiprtc.so')\ctypes.CDLL(ctypes.util.find_library('hiprtc'))\g" autogen/hip.py
+  #sed -i "s\ctypes.CDLL('/opt/rocm/lib/libamdhip64.so')\ctypes.CDLL(ctypes.util.find_library('amdhip64'))\g" autogen/hip.py
+  python3 -c "import autogen.hip"
+
+  clang2py /opt/rocm/include/amd_comgr/amd_comgr.h \
+  --clang-args="-D__HIP_PLATFORM_AMD__ -I/opt/rocm/include -x c++" -o autogen/comgr.py -l /opt/rocm/lib/libamd_comgr.so
+  sed -i '1s/^/# mypy: ignore-errors\n/' autogen/comgr.py
+  grep FIXME_STUB autogen/comgr.py || true
+  python3 -c "import autogen.comgr"
+}
+
+generate_cuda() {
+  clang2py /usr/include/cuda.h /usr/include/nvrtc.h -o autogen/cuda.py -l /usr/lib/x86_64-linux-gnu/libcuda.so -l /usr/lib/x86_64-linux-gnu/libnvrtc.so
+  sed -i "s\import ctypes\import ctypes, ctypes.util\g" autogen/cuda.py
+  sed -i "s\ctypes.CDLL('/usr/lib/x86_64-linux-gnu/libcuda.so')\ctypes.CDLL(ctypes.util.find_library('cuda'))\g" autogen/cuda.py
+  sed -i "s\ctypes.CDLL('/usr/lib/x86_64-linux-gnu/libnvrtc.so')\ctypes.CDLL(ctypes.util.find_library('nvrtc'))\g" autogen/cuda.py
+  sed -i '1s/^/# mypy: ignore-errors\n/' autogen/cuda.py
+  grep FIXME_STUB autogen/cuda.py || true
+}
+
+cd tinygrad
+if [ "$1" == "opencl" ]; then generate_opencl
+elif [ "$1" == "hip" ]; then generate_hip
+elif [ "$1" == "cuda" ]; then generate_cuda
+elif [ "$1" == "all" ]; then generate_opencl; generate_hip; generate_cuda
+else echo "usage: $0 <type>"
+fi

extra/dist/world.py

@@ -5,7 +5,7 @@ from tinygrad.helpers import DEBUG, colored, getenv
 from tinygrad.lazy import LazyBuffer
 from tinygrad.runtime.lib import RawBuffer, RawBufferCopyInOut
 try:
-  import gpuctypes.hip as hip
+  import tinygrad.autogen.hip as hip
   from tinygrad.runtime.ops_hip import RawHIPBuffer, check
 except: RawHIPBuffer = None
 from tinygrad.runtime.ops_disk import RawDiskBuffer

extra/hip_events.py

@@ -1,5 +1,5 @@
 import ctypes
-import gpuctypes.hip as hip
+import tinygrad.autogen.hip as hip
 from tinygrad.runtime.ops_hip import check
 from tinygrad.helpers import init_c_var
 

mypy.ini

@@ -7,4 +7,4 @@ explicit_package_bases = True
 warn_unreachable = True
 warn_redundant_casts = True
 # NOTE: had to comment this out to make mypy pass on both CI and OSX
-#warn_unused_ignores = True
+#warn_unused_ignores = True

ruff.toml

@@ -31,5 +31,6 @@ exclude = [
   "examples/",
   "extra/",
   "openpilot/",
+  "tinygrad/autogen",
 ]
 

setup.py

@@ -14,13 +14,13 @@ setup(name='tinygrad',
       license='MIT',
       long_description=long_description,
       long_description_content_type='text/markdown',
-      packages = ['tinygrad', 'tinygrad.codegen', 'tinygrad.nn', 'tinygrad.renderer',
+      packages = ['tinygrad', 'tinygrad.autogen', 'tinygrad.codegen', 'tinygrad.nn', 'tinygrad.renderer',
                   'tinygrad.runtime', 'tinygrad.runtime.graph', 'tinygrad.shape', 'tinygrad.features'],
       classifiers=[
         "Programming Language :: Python :: 3",
         "License :: OSI Approved :: MIT License"
       ],
-      install_requires=["numpy", "tqdm", "gpuctypes",
+      install_requires=["numpy", "tqdm",
                         "pyobjc-framework-Metal; platform_system=='Darwin'",
                         "pyobjc-framework-libdispatch; platform_system=='Darwin'"],
       python_requires='>=3.8',

sz.py

@@ -12,6 +12,7 @@ def gen_stats(base_path="."):
   for path, _, files in os.walk(os.path.join(base_path, "tinygrad")):
     for name in files:
       if not name.endswith(".py"): continue
+      if 'tinygrad/autogen' in path: continue
       filepath = os.path.join(path, name)
       relfilepath = os.path.relpath(filepath, base_path)
       with tokenize.open(filepath) as file_:

tinygrad/autogen/comgr.py

@@ -0,0 +1,865 @@
+# mypy: ignore-errors
+# -*- coding: utf-8 -*-
+#
+# TARGET arch is: ['-D__HIP_PLATFORM_AMD__', '-I/opt/rocm/include', '-x', 'c++']
+# WORD_SIZE is: 8
+# POINTER_SIZE is: 8
+# LONGDOUBLE_SIZE is: 16
+#
+import ctypes
+
+
+def string_cast(char_pointer, encoding='utf-8', errors='strict'):
+    value = ctypes.cast(char_pointer, ctypes.c_char_p).value
+    if value is not None and encoding is not None:
+        value = value.decode(encoding, errors=errors)
+    return value
+
+
+def char_pointer_cast(string, encoding='utf-8'):
+    if encoding is not None:
+        try:
+            string = string.encode(encoding)
+        except AttributeError:
+            # In Python3, bytes has no encode attribute
+            pass
+    string = ctypes.c_char_p(string)
+    return ctypes.cast(string, ctypes.POINTER(ctypes.c_char))
+
+
+
+_libraries = {}
+_libraries['libamd_comgr.so'] = ctypes.CDLL('/opt/rocm/lib/libamd_comgr.so')
+c_int128 = ctypes.c_ubyte*16
+c_uint128 = c_int128
+void = None
+if ctypes.sizeof(ctypes.c_longdouble) == 16:
+    c_long_double_t = ctypes.c_longdouble
+else:
+    c_long_double_t = ctypes.c_ubyte*16
+
+class AsDictMixin:
+    @classmethod
+    def as_dict(cls, self):
+        result = {}
+        if not isinstance(self, AsDictMixin):
+            # not a structure, assume it's already a python object
+            return self
+        if not hasattr(cls, "_fields_"):
+            return result
+        # sys.version_info >= (3, 5)
+        # for (field, *_) in cls._fields_:  # noqa
+        for field_tuple in cls._fields_:  # noqa
+            field = field_tuple[0]
+            if field.startswith('PADDING_'):
+                continue
+            value = getattr(self, field)
+            type_ = type(value)
+            if hasattr(value, "_length_") and hasattr(value, "_type_"):
+                # array
+                if not hasattr(type_, "as_dict"):
+                    value = [v for v in value]
+                else:
+                    type_ = type_._type_
+                    value = [type_.as_dict(v) for v in value]
+            elif hasattr(value, "contents") and hasattr(value, "_type_"):
+                # pointer
+                try:
+                    if not hasattr(type_, "as_dict"):
+                        value = value.contents
+                    else:
+                        type_ = type_._type_
+                        value = type_.as_dict(value.contents)
+                except ValueError:
+                    # nullptr
+                    value = None
+            elif isinstance(value, AsDictMixin):
+                # other structure
+                value = type_.as_dict(value)
+            result[field] = value
+        return result
+
+
+class Structure(ctypes.Structure, AsDictMixin):
+
+    def __init__(self, *args, **kwds):
+        # We don't want to use positional arguments fill PADDING_* fields
+
+        args = dict(zip(self.__class__._field_names_(), args))
+        args.update(kwds)
+        super(Structure, self).__init__(**args)
+
+    @classmethod
+    def _field_names_(cls):
+        if hasattr(cls, '_fields_'):
+            return (f[0] for f in cls._fields_ if not f[0].startswith('PADDING'))
+        else:
+            return ()
+
+    @classmethod
+    def get_type(cls, field):
+        for f in cls._fields_:
+            if f[0] == field:
+                return f[1]
+        return None
+
+    @classmethod
+    def bind(cls, bound_fields):
+        fields = {}
+        for name, type_ in cls._fields_:
+            if hasattr(type_, "restype"):
+                if name in bound_fields:
+                    if bound_fields[name] is None:
+                        fields[name] = type_()
+                    else:
+                        # use a closure to capture the callback from the loop scope
+                        fields[name] = (
+                            type_((lambda callback: lambda *args: callback(*args))(
+                                bound_fields[name]))
+                        )
+                    del bound_fields[name]
+                else:
+                    # default callback implementation (does nothing)
+                    try:
+                        default_ = type_(0).restype().value
+                    except TypeError:
+                        default_ = None
+                    fields[name] = type_((
+                        lambda default_: lambda *args: default_)(default_))
+            else:
+                # not a callback function, use default initialization
+                if name in bound_fields:
+                    fields[name] = bound_fields[name]
+                    del bound_fields[name]
+                else:
+                    fields[name] = type_()
+        if len(bound_fields) != 0:
+            raise ValueError(
+                "Cannot bind the following unknown callback(s) {}.{}".format(
+                    cls.__name__, bound_fields.keys()
+            ))
+        return cls(**fields)
+
+
+class Union(ctypes.Union, AsDictMixin):
+    pass
+
+
+
+
+
+
+# values for enumeration 'amd_comgr_status_s'
+amd_comgr_status_s__enumvalues = {
+    0: 'AMD_COMGR_STATUS_SUCCESS',
+    1: 'AMD_COMGR_STATUS_ERROR',
+    2: 'AMD_COMGR_STATUS_ERROR_INVALID_ARGUMENT',
+    3: 'AMD_COMGR_STATUS_ERROR_OUT_OF_RESOURCES',
+}
+AMD_COMGR_STATUS_SUCCESS = 0
+AMD_COMGR_STATUS_ERROR = 1
+AMD_COMGR_STATUS_ERROR_INVALID_ARGUMENT = 2
+AMD_COMGR_STATUS_ERROR_OUT_OF_RESOURCES = 3
+amd_comgr_status_s = ctypes.c_uint32 # enum
+amd_comgr_status_t = amd_comgr_status_s
+amd_comgr_status_t__enumvalues = amd_comgr_status_s__enumvalues
+
+# values for enumeration 'amd_comgr_language_s'
+amd_comgr_language_s__enumvalues = {
+    0: 'AMD_COMGR_LANGUAGE_NONE',
+    1: 'AMD_COMGR_LANGUAGE_OPENCL_1_2',
+    2: 'AMD_COMGR_LANGUAGE_OPENCL_2_0',
+    3: 'AMD_COMGR_LANGUAGE_HC',
+    4: 'AMD_COMGR_LANGUAGE_HIP',
+    4: 'AMD_COMGR_LANGUAGE_LAST',
+}
+AMD_COMGR_LANGUAGE_NONE = 0
+AMD_COMGR_LANGUAGE_OPENCL_1_2 = 1
+AMD_COMGR_LANGUAGE_OPENCL_2_0 = 2
+AMD_COMGR_LANGUAGE_HC = 3
+AMD_COMGR_LANGUAGE_HIP = 4
+AMD_COMGR_LANGUAGE_LAST = 4
+amd_comgr_language_s = ctypes.c_uint32 # enum
+amd_comgr_language_t = amd_comgr_language_s
+amd_comgr_language_t__enumvalues = amd_comgr_language_s__enumvalues
+try:
+    amd_comgr_status_string = _libraries['libamd_comgr.so'].amd_comgr_status_string
+    amd_comgr_status_string.restype = amd_comgr_status_t
+    amd_comgr_status_string.argtypes = [amd_comgr_status_t, ctypes.POINTER(ctypes.POINTER(ctypes.c_char))]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_version = _libraries['libamd_comgr.so'].amd_comgr_get_version
+    amd_comgr_get_version.restype = None
+    amd_comgr_get_version.argtypes = [ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+
+# values for enumeration 'amd_comgr_data_kind_s'
+amd_comgr_data_kind_s__enumvalues = {
+    0: 'AMD_COMGR_DATA_KIND_UNDEF',
+    1: 'AMD_COMGR_DATA_KIND_SOURCE',
+    2: 'AMD_COMGR_DATA_KIND_INCLUDE',
+    3: 'AMD_COMGR_DATA_KIND_PRECOMPILED_HEADER',
+    4: 'AMD_COMGR_DATA_KIND_DIAGNOSTIC',
+    5: 'AMD_COMGR_DATA_KIND_LOG',
+    6: 'AMD_COMGR_DATA_KIND_BC',
+    7: 'AMD_COMGR_DATA_KIND_RELOCATABLE',
+    8: 'AMD_COMGR_DATA_KIND_EXECUTABLE',
+    9: 'AMD_COMGR_DATA_KIND_BYTES',
+    16: 'AMD_COMGR_DATA_KIND_FATBIN',
+    17: 'AMD_COMGR_DATA_KIND_AR',
+    18: 'AMD_COMGR_DATA_KIND_BC_BUNDLE',
+    19: 'AMD_COMGR_DATA_KIND_AR_BUNDLE',
+    19: 'AMD_COMGR_DATA_KIND_LAST',
+}
+AMD_COMGR_DATA_KIND_UNDEF = 0
+AMD_COMGR_DATA_KIND_SOURCE = 1
+AMD_COMGR_DATA_KIND_INCLUDE = 2
+AMD_COMGR_DATA_KIND_PRECOMPILED_HEADER = 3
+AMD_COMGR_DATA_KIND_DIAGNOSTIC = 4
+AMD_COMGR_DATA_KIND_LOG = 5
+AMD_COMGR_DATA_KIND_BC = 6
+AMD_COMGR_DATA_KIND_RELOCATABLE = 7
+AMD_COMGR_DATA_KIND_EXECUTABLE = 8
+AMD_COMGR_DATA_KIND_BYTES = 9
+AMD_COMGR_DATA_KIND_FATBIN = 16
+AMD_COMGR_DATA_KIND_AR = 17
+AMD_COMGR_DATA_KIND_BC_BUNDLE = 18
+AMD_COMGR_DATA_KIND_AR_BUNDLE = 19
+AMD_COMGR_DATA_KIND_LAST = 19
+amd_comgr_data_kind_s = ctypes.c_uint32 # enum
+amd_comgr_data_kind_t = amd_comgr_data_kind_s
+amd_comgr_data_kind_t__enumvalues = amd_comgr_data_kind_s__enumvalues
+class struct_amd_comgr_data_s(Structure):
+    pass
+
+struct_amd_comgr_data_s._pack_ = 1 # source:False
+struct_amd_comgr_data_s._fields_ = [
+    ('handle', ctypes.c_uint64),
+]
+
+amd_comgr_data_t = struct_amd_comgr_data_s
+class struct_amd_comgr_data_set_s(Structure):
+    pass
+
+struct_amd_comgr_data_set_s._pack_ = 1 # source:False
+struct_amd_comgr_data_set_s._fields_ = [
+    ('handle', ctypes.c_uint64),
+]
+
+amd_comgr_data_set_t = struct_amd_comgr_data_set_s
+class struct_amd_comgr_action_info_s(Structure):
+    pass
+
+struct_amd_comgr_action_info_s._pack_ = 1 # source:False
+struct_amd_comgr_action_info_s._fields_ = [
+    ('handle', ctypes.c_uint64),
+]
+
+amd_comgr_action_info_t = struct_amd_comgr_action_info_s
+class struct_amd_comgr_metadata_node_s(Structure):
+    pass
+
+struct_amd_comgr_metadata_node_s._pack_ = 1 # source:False
+struct_amd_comgr_metadata_node_s._fields_ = [
+    ('handle', ctypes.c_uint64),
+]
+
+amd_comgr_metadata_node_t = struct_amd_comgr_metadata_node_s
+class struct_amd_comgr_symbol_s(Structure):
+    pass
+
+struct_amd_comgr_symbol_s._pack_ = 1 # source:False
+struct_amd_comgr_symbol_s._fields_ = [
+    ('handle', ctypes.c_uint64),
+]
+
+amd_comgr_symbol_t = struct_amd_comgr_symbol_s
+class struct_amd_comgr_disassembly_info_s(Structure):
+    pass
+
+struct_amd_comgr_disassembly_info_s._pack_ = 1 # source:False
+struct_amd_comgr_disassembly_info_s._fields_ = [
+    ('handle', ctypes.c_uint64),
+]
+
+amd_comgr_disassembly_info_t = struct_amd_comgr_disassembly_info_s
+class struct_amd_comgr_symbolizer_info_s(Structure):
+    pass
+
+struct_amd_comgr_symbolizer_info_s._pack_ = 1 # source:False
+struct_amd_comgr_symbolizer_info_s._fields_ = [
+    ('handle', ctypes.c_uint64),
+]
+
+amd_comgr_symbolizer_info_t = struct_amd_comgr_symbolizer_info_s
+try:
+    amd_comgr_get_isa_count = _libraries['libamd_comgr.so'].amd_comgr_get_isa_count
+    amd_comgr_get_isa_count.restype = amd_comgr_status_t
+    amd_comgr_get_isa_count.argtypes = [ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+size_t = ctypes.c_uint64
+try:
+    amd_comgr_get_isa_name = _libraries['libamd_comgr.so'].amd_comgr_get_isa_name
+    amd_comgr_get_isa_name.restype = amd_comgr_status_t
+    amd_comgr_get_isa_name.argtypes = [size_t, ctypes.POINTER(ctypes.POINTER(ctypes.c_char))]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_isa_metadata = _libraries['libamd_comgr.so'].amd_comgr_get_isa_metadata
+    amd_comgr_get_isa_metadata.restype = amd_comgr_status_t
+    amd_comgr_get_isa_metadata.argtypes = [ctypes.POINTER(ctypes.c_char), ctypes.POINTER(struct_amd_comgr_metadata_node_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_create_data = _libraries['libamd_comgr.so'].amd_comgr_create_data
+    amd_comgr_create_data.restype = amd_comgr_status_t
+    amd_comgr_create_data.argtypes = [amd_comgr_data_kind_t, ctypes.POINTER(struct_amd_comgr_data_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_release_data = _libraries['libamd_comgr.so'].amd_comgr_release_data
+    amd_comgr_release_data.restype = amd_comgr_status_t
+    amd_comgr_release_data.argtypes = [amd_comgr_data_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_data_kind = _libraries['libamd_comgr.so'].amd_comgr_get_data_kind
+    amd_comgr_get_data_kind.restype = amd_comgr_status_t
+    amd_comgr_get_data_kind.argtypes = [amd_comgr_data_t, ctypes.POINTER(amd_comgr_data_kind_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_set_data = _libraries['libamd_comgr.so'].amd_comgr_set_data
+    amd_comgr_set_data.restype = amd_comgr_status_t
+    amd_comgr_set_data.argtypes = [amd_comgr_data_t, size_t, ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+uint64_t = ctypes.c_uint64
+try:
+    amd_comgr_set_data_from_file_slice = _libraries['libamd_comgr.so'].amd_comgr_set_data_from_file_slice
+    amd_comgr_set_data_from_file_slice.restype = amd_comgr_status_t
+    amd_comgr_set_data_from_file_slice.argtypes = [amd_comgr_data_t, ctypes.c_int32, uint64_t, uint64_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_set_data_name = _libraries['libamd_comgr.so'].amd_comgr_set_data_name
+    amd_comgr_set_data_name.restype = amd_comgr_status_t
+    amd_comgr_set_data_name.argtypes = [amd_comgr_data_t, ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_data = _libraries['libamd_comgr.so'].amd_comgr_get_data
+    amd_comgr_get_data.restype = amd_comgr_status_t
+    amd_comgr_get_data.argtypes = [amd_comgr_data_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_data_name = _libraries['libamd_comgr.so'].amd_comgr_get_data_name
+    amd_comgr_get_data_name.restype = amd_comgr_status_t
+    amd_comgr_get_data_name.argtypes = [amd_comgr_data_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_data_isa_name = _libraries['libamd_comgr.so'].amd_comgr_get_data_isa_name
+    amd_comgr_get_data_isa_name.restype = amd_comgr_status_t
+    amd_comgr_get_data_isa_name.argtypes = [amd_comgr_data_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_create_symbolizer_info = _libraries['libamd_comgr.so'].amd_comgr_create_symbolizer_info
+    amd_comgr_create_symbolizer_info.restype = amd_comgr_status_t
+    amd_comgr_create_symbolizer_info.argtypes = [amd_comgr_data_t, ctypes.CFUNCTYPE(None, ctypes.POINTER(ctypes.c_char), ctypes.POINTER(None)), ctypes.POINTER(struct_amd_comgr_symbolizer_info_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_destroy_symbolizer_info = _libraries['libamd_comgr.so'].amd_comgr_destroy_symbolizer_info
+    amd_comgr_destroy_symbolizer_info.restype = amd_comgr_status_t
+    amd_comgr_destroy_symbolizer_info.argtypes = [amd_comgr_symbolizer_info_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_symbolize = _libraries['libamd_comgr.so'].amd_comgr_symbolize
+    amd_comgr_symbolize.restype = amd_comgr_status_t
+    amd_comgr_symbolize.argtypes = [amd_comgr_symbolizer_info_t, uint64_t, ctypes.c_bool, ctypes.POINTER(None)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_data_metadata = _libraries['libamd_comgr.so'].amd_comgr_get_data_metadata
+    amd_comgr_get_data_metadata.restype = amd_comgr_status_t
+    amd_comgr_get_data_metadata.argtypes = [amd_comgr_data_t, ctypes.POINTER(struct_amd_comgr_metadata_node_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_destroy_metadata = _libraries['libamd_comgr.so'].amd_comgr_destroy_metadata
+    amd_comgr_destroy_metadata.restype = amd_comgr_status_t
+    amd_comgr_destroy_metadata.argtypes = [amd_comgr_metadata_node_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_create_data_set = _libraries['libamd_comgr.so'].amd_comgr_create_data_set
+    amd_comgr_create_data_set.restype = amd_comgr_status_t
+    amd_comgr_create_data_set.argtypes = [ctypes.POINTER(struct_amd_comgr_data_set_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_destroy_data_set = _libraries['libamd_comgr.so'].amd_comgr_destroy_data_set
+    amd_comgr_destroy_data_set.restype = amd_comgr_status_t
+    amd_comgr_destroy_data_set.argtypes = [amd_comgr_data_set_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_data_set_add = _libraries['libamd_comgr.so'].amd_comgr_data_set_add
+    amd_comgr_data_set_add.restype = amd_comgr_status_t
+    amd_comgr_data_set_add.argtypes = [amd_comgr_data_set_t, amd_comgr_data_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_data_set_remove = _libraries['libamd_comgr.so'].amd_comgr_data_set_remove
+    amd_comgr_data_set_remove.restype = amd_comgr_status_t
+    amd_comgr_data_set_remove.argtypes = [amd_comgr_data_set_t, amd_comgr_data_kind_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_data_count = _libraries['libamd_comgr.so'].amd_comgr_action_data_count
+    amd_comgr_action_data_count.restype = amd_comgr_status_t
+    amd_comgr_action_data_count.argtypes = [amd_comgr_data_set_t, amd_comgr_data_kind_t, ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_data_get_data = _libraries['libamd_comgr.so'].amd_comgr_action_data_get_data
+    amd_comgr_action_data_get_data.restype = amd_comgr_status_t
+    amd_comgr_action_data_get_data.argtypes = [amd_comgr_data_set_t, amd_comgr_data_kind_t, size_t, ctypes.POINTER(struct_amd_comgr_data_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_create_action_info = _libraries['libamd_comgr.so'].amd_comgr_create_action_info
+    amd_comgr_create_action_info.restype = amd_comgr_status_t
+    amd_comgr_create_action_info.argtypes = [ctypes.POINTER(struct_amd_comgr_action_info_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_destroy_action_info = _libraries['libamd_comgr.so'].amd_comgr_destroy_action_info
+    amd_comgr_destroy_action_info.restype = amd_comgr_status_t
+    amd_comgr_destroy_action_info.argtypes = [amd_comgr_action_info_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_set_isa_name = _libraries['libamd_comgr.so'].amd_comgr_action_info_set_isa_name
+    amd_comgr_action_info_set_isa_name.restype = amd_comgr_status_t
+    amd_comgr_action_info_set_isa_name.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_get_isa_name = _libraries['libamd_comgr.so'].amd_comgr_action_info_get_isa_name
+    amd_comgr_action_info_get_isa_name.restype = amd_comgr_status_t
+    amd_comgr_action_info_get_isa_name.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_set_language = _libraries['libamd_comgr.so'].amd_comgr_action_info_set_language
+    amd_comgr_action_info_set_language.restype = amd_comgr_status_t
+    amd_comgr_action_info_set_language.argtypes = [amd_comgr_action_info_t, amd_comgr_language_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_get_language = _libraries['libamd_comgr.so'].amd_comgr_action_info_get_language
+    amd_comgr_action_info_get_language.restype = amd_comgr_status_t
+    amd_comgr_action_info_get_language.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(amd_comgr_language_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_set_options = _libraries['libamd_comgr.so'].amd_comgr_action_info_set_options
+    amd_comgr_action_info_set_options.restype = amd_comgr_status_t
+    amd_comgr_action_info_set_options.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_get_options = _libraries['libamd_comgr.so'].amd_comgr_action_info_get_options
+    amd_comgr_action_info_get_options.restype = amd_comgr_status_t
+    amd_comgr_action_info_get_options.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_set_option_list = _libraries['libamd_comgr.so'].amd_comgr_action_info_set_option_list
+    amd_comgr_action_info_set_option_list.restype = amd_comgr_status_t
+    amd_comgr_action_info_set_option_list.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_char) * 0, size_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_get_option_list_count = _libraries['libamd_comgr.so'].amd_comgr_action_info_get_option_list_count
+    amd_comgr_action_info_get_option_list_count.restype = amd_comgr_status_t
+    amd_comgr_action_info_get_option_list_count.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_get_option_list_item = _libraries['libamd_comgr.so'].amd_comgr_action_info_get_option_list_item
+    amd_comgr_action_info_get_option_list_item.restype = amd_comgr_status_t
+    amd_comgr_action_info_get_option_list_item.argtypes = [amd_comgr_action_info_t, size_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_set_working_directory_path = _libraries['libamd_comgr.so'].amd_comgr_action_info_set_working_directory_path
+    amd_comgr_action_info_set_working_directory_path.restype = amd_comgr_status_t
+    amd_comgr_action_info_set_working_directory_path.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_get_working_directory_path = _libraries['libamd_comgr.so'].amd_comgr_action_info_get_working_directory_path
+    amd_comgr_action_info_get_working_directory_path.restype = amd_comgr_status_t
+    amd_comgr_action_info_get_working_directory_path.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_set_logging = _libraries['libamd_comgr.so'].amd_comgr_action_info_set_logging
+    amd_comgr_action_info_set_logging.restype = amd_comgr_status_t
+    amd_comgr_action_info_set_logging.argtypes = [amd_comgr_action_info_t, ctypes.c_bool]
+except AttributeError:
+    pass
+try:
+    amd_comgr_action_info_get_logging = _libraries['libamd_comgr.so'].amd_comgr_action_info_get_logging
+    amd_comgr_action_info_get_logging.restype = amd_comgr_status_t
+    amd_comgr_action_info_get_logging.argtypes = [amd_comgr_action_info_t, ctypes.POINTER(ctypes.c_bool)]
+except AttributeError:
+    pass
+
+# values for enumeration 'amd_comgr_action_kind_s'
+amd_comgr_action_kind_s__enumvalues = {
+    0: 'AMD_COMGR_ACTION_SOURCE_TO_PREPROCESSOR',
+    1: 'AMD_COMGR_ACTION_ADD_PRECOMPILED_HEADERS',
+    2: 'AMD_COMGR_ACTION_COMPILE_SOURCE_TO_BC',
+    3: 'AMD_COMGR_ACTION_ADD_DEVICE_LIBRARIES',
+    4: 'AMD_COMGR_ACTION_LINK_BC_TO_BC',
+    5: 'AMD_COMGR_ACTION_OPTIMIZE_BC_TO_BC',
+    6: 'AMD_COMGR_ACTION_CODEGEN_BC_TO_RELOCATABLE',
+    7: 'AMD_COMGR_ACTION_CODEGEN_BC_TO_ASSEMBLY',
+    8: 'AMD_COMGR_ACTION_LINK_RELOCATABLE_TO_RELOCATABLE',
+    9: 'AMD_COMGR_ACTION_LINK_RELOCATABLE_TO_EXECUTABLE',
+    10: 'AMD_COMGR_ACTION_ASSEMBLE_SOURCE_TO_RELOCATABLE',
+    11: 'AMD_COMGR_ACTION_DISASSEMBLE_RELOCATABLE_TO_SOURCE',
+    12: 'AMD_COMGR_ACTION_DISASSEMBLE_EXECUTABLE_TO_SOURCE',
+    13: 'AMD_COMGR_ACTION_DISASSEMBLE_BYTES_TO_SOURCE',
+    14: 'AMD_COMGR_ACTION_COMPILE_SOURCE_TO_FATBIN',
+    15: 'AMD_COMGR_ACTION_COMPILE_SOURCE_WITH_DEVICE_LIBS_TO_BC',
+    15: 'AMD_COMGR_ACTION_LAST',
+}
+AMD_COMGR_ACTION_SOURCE_TO_PREPROCESSOR = 0
+AMD_COMGR_ACTION_ADD_PRECOMPILED_HEADERS = 1
+AMD_COMGR_ACTION_COMPILE_SOURCE_TO_BC = 2
+AMD_COMGR_ACTION_ADD_DEVICE_LIBRARIES = 3
+AMD_COMGR_ACTION_LINK_BC_TO_BC = 4
+AMD_COMGR_ACTION_OPTIMIZE_BC_TO_BC = 5
+AMD_COMGR_ACTION_CODEGEN_BC_TO_RELOCATABLE = 6
+AMD_COMGR_ACTION_CODEGEN_BC_TO_ASSEMBLY = 7
+AMD_COMGR_ACTION_LINK_RELOCATABLE_TO_RELOCATABLE = 8
+AMD_COMGR_ACTION_LINK_RELOCATABLE_TO_EXECUTABLE = 9
+AMD_COMGR_ACTION_ASSEMBLE_SOURCE_TO_RELOCATABLE = 10
+AMD_COMGR_ACTION_DISASSEMBLE_RELOCATABLE_TO_SOURCE = 11
+AMD_COMGR_ACTION_DISASSEMBLE_EXECUTABLE_TO_SOURCE = 12
+AMD_COMGR_ACTION_DISASSEMBLE_BYTES_TO_SOURCE = 13
+AMD_COMGR_ACTION_COMPILE_SOURCE_TO_FATBIN = 14
+AMD_COMGR_ACTION_COMPILE_SOURCE_WITH_DEVICE_LIBS_TO_BC = 15
+AMD_COMGR_ACTION_LAST = 15
+amd_comgr_action_kind_s = ctypes.c_uint32 # enum
+amd_comgr_action_kind_t = amd_comgr_action_kind_s
+amd_comgr_action_kind_t__enumvalues = amd_comgr_action_kind_s__enumvalues
+try:
+    amd_comgr_do_action = _libraries['libamd_comgr.so'].amd_comgr_do_action
+    amd_comgr_do_action.restype = amd_comgr_status_t
+    amd_comgr_do_action.argtypes = [amd_comgr_action_kind_t, amd_comgr_action_info_t, amd_comgr_data_set_t, amd_comgr_data_set_t]
+except AttributeError:
+    pass
+
+# values for enumeration 'amd_comgr_metadata_kind_s'
+amd_comgr_metadata_kind_s__enumvalues = {
+    0: 'AMD_COMGR_METADATA_KIND_NULL',
+    1: 'AMD_COMGR_METADATA_KIND_STRING',
+    2: 'AMD_COMGR_METADATA_KIND_MAP',
+    3: 'AMD_COMGR_METADATA_KIND_LIST',
+    3: 'AMD_COMGR_METADATA_KIND_LAST',
+}
+AMD_COMGR_METADATA_KIND_NULL = 0
+AMD_COMGR_METADATA_KIND_STRING = 1
+AMD_COMGR_METADATA_KIND_MAP = 2
+AMD_COMGR_METADATA_KIND_LIST = 3
+AMD_COMGR_METADATA_KIND_LAST = 3
+amd_comgr_metadata_kind_s = ctypes.c_uint32 # enum
+amd_comgr_metadata_kind_t = amd_comgr_metadata_kind_s
+amd_comgr_metadata_kind_t__enumvalues = amd_comgr_metadata_kind_s__enumvalues
+try:
+    amd_comgr_get_metadata_kind = _libraries['libamd_comgr.so'].amd_comgr_get_metadata_kind
+    amd_comgr_get_metadata_kind.restype = amd_comgr_status_t
+    amd_comgr_get_metadata_kind.argtypes = [amd_comgr_metadata_node_t, ctypes.POINTER(amd_comgr_metadata_kind_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_metadata_string = _libraries['libamd_comgr.so'].amd_comgr_get_metadata_string
+    amd_comgr_get_metadata_string.restype = amd_comgr_status_t
+    amd_comgr_get_metadata_string.argtypes = [amd_comgr_metadata_node_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_metadata_map_size = _libraries['libamd_comgr.so'].amd_comgr_get_metadata_map_size
+    amd_comgr_get_metadata_map_size.restype = amd_comgr_status_t
+    amd_comgr_get_metadata_map_size.argtypes = [amd_comgr_metadata_node_t, ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_iterate_map_metadata = _libraries['libamd_comgr.so'].amd_comgr_iterate_map_metadata
+    amd_comgr_iterate_map_metadata.restype = amd_comgr_status_t
+    amd_comgr_iterate_map_metadata.argtypes = [amd_comgr_metadata_node_t, ctypes.CFUNCTYPE(amd_comgr_status_s, struct_amd_comgr_metadata_node_s, struct_amd_comgr_metadata_node_s, ctypes.POINTER(None)), ctypes.POINTER(None)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_metadata_lookup = _libraries['libamd_comgr.so'].amd_comgr_metadata_lookup
+    amd_comgr_metadata_lookup.restype = amd_comgr_status_t
+    amd_comgr_metadata_lookup.argtypes = [amd_comgr_metadata_node_t, ctypes.POINTER(ctypes.c_char), ctypes.POINTER(struct_amd_comgr_metadata_node_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_metadata_list_size = _libraries['libamd_comgr.so'].amd_comgr_get_metadata_list_size
+    amd_comgr_get_metadata_list_size.restype = amd_comgr_status_t
+    amd_comgr_get_metadata_list_size.argtypes = [amd_comgr_metadata_node_t, ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_index_list_metadata = _libraries['libamd_comgr.so'].amd_comgr_index_list_metadata
+    amd_comgr_index_list_metadata.restype = amd_comgr_status_t
+    amd_comgr_index_list_metadata.argtypes = [amd_comgr_metadata_node_t, size_t, ctypes.POINTER(struct_amd_comgr_metadata_node_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_iterate_symbols = _libraries['libamd_comgr.so'].amd_comgr_iterate_symbols
+    amd_comgr_iterate_symbols.restype = amd_comgr_status_t
+    amd_comgr_iterate_symbols.argtypes = [amd_comgr_data_t, ctypes.CFUNCTYPE(amd_comgr_status_s, struct_amd_comgr_symbol_s, ctypes.POINTER(None)), ctypes.POINTER(None)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_symbol_lookup = _libraries['libamd_comgr.so'].amd_comgr_symbol_lookup
+    amd_comgr_symbol_lookup.restype = amd_comgr_status_t
+    amd_comgr_symbol_lookup.argtypes = [amd_comgr_data_t, ctypes.POINTER(ctypes.c_char), ctypes.POINTER(struct_amd_comgr_symbol_s)]
+except AttributeError:
+    pass
+
+# values for enumeration 'amd_comgr_symbol_type_s'
+amd_comgr_symbol_type_s__enumvalues = {
+    -1: 'AMD_COMGR_SYMBOL_TYPE_UNKNOWN',
+    0: 'AMD_COMGR_SYMBOL_TYPE_NOTYPE',
+    1: 'AMD_COMGR_SYMBOL_TYPE_OBJECT',
+    2: 'AMD_COMGR_SYMBOL_TYPE_FUNC',
+    3: 'AMD_COMGR_SYMBOL_TYPE_SECTION',
+    4: 'AMD_COMGR_SYMBOL_TYPE_FILE',
+    5: 'AMD_COMGR_SYMBOL_TYPE_COMMON',
+    10: 'AMD_COMGR_SYMBOL_TYPE_AMDGPU_HSA_KERNEL',
+}
+AMD_COMGR_SYMBOL_TYPE_UNKNOWN = -1
+AMD_COMGR_SYMBOL_TYPE_NOTYPE = 0
+AMD_COMGR_SYMBOL_TYPE_OBJECT = 1
+AMD_COMGR_SYMBOL_TYPE_FUNC = 2
+AMD_COMGR_SYMBOL_TYPE_SECTION = 3
+AMD_COMGR_SYMBOL_TYPE_FILE = 4
+AMD_COMGR_SYMBOL_TYPE_COMMON = 5
+AMD_COMGR_SYMBOL_TYPE_AMDGPU_HSA_KERNEL = 10
+amd_comgr_symbol_type_s = ctypes.c_int32 # enum
+amd_comgr_symbol_type_t = amd_comgr_symbol_type_s
+amd_comgr_symbol_type_t__enumvalues = amd_comgr_symbol_type_s__enumvalues
+
+# values for enumeration 'amd_comgr_symbol_info_s'
+amd_comgr_symbol_info_s__enumvalues = {
+    0: 'AMD_COMGR_SYMBOL_INFO_NAME_LENGTH',
+    1: 'AMD_COMGR_SYMBOL_INFO_NAME',
+    2: 'AMD_COMGR_SYMBOL_INFO_TYPE',
+    3: 'AMD_COMGR_SYMBOL_INFO_SIZE',
+    4: 'AMD_COMGR_SYMBOL_INFO_IS_UNDEFINED',
+    5: 'AMD_COMGR_SYMBOL_INFO_VALUE',
+    5: 'AMD_COMGR_SYMBOL_INFO_LAST',
+}
+AMD_COMGR_SYMBOL_INFO_NAME_LENGTH = 0
+AMD_COMGR_SYMBOL_INFO_NAME = 1
+AMD_COMGR_SYMBOL_INFO_TYPE = 2
+AMD_COMGR_SYMBOL_INFO_SIZE = 3
+AMD_COMGR_SYMBOL_INFO_IS_UNDEFINED = 4
+AMD_COMGR_SYMBOL_INFO_VALUE = 5
+AMD_COMGR_SYMBOL_INFO_LAST = 5
+amd_comgr_symbol_info_s = ctypes.c_uint32 # enum
+amd_comgr_symbol_info_t = amd_comgr_symbol_info_s
+amd_comgr_symbol_info_t__enumvalues = amd_comgr_symbol_info_s__enumvalues
+try:
+    amd_comgr_symbol_get_info = _libraries['libamd_comgr.so'].amd_comgr_symbol_get_info
+    amd_comgr_symbol_get_info.restype = amd_comgr_status_t
+    amd_comgr_symbol_get_info.argtypes = [amd_comgr_symbol_t, amd_comgr_symbol_info_t, ctypes.POINTER(None)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_create_disassembly_info = _libraries['libamd_comgr.so'].amd_comgr_create_disassembly_info
+    amd_comgr_create_disassembly_info.restype = amd_comgr_status_t
+    amd_comgr_create_disassembly_info.argtypes = [ctypes.POINTER(ctypes.c_char), ctypes.CFUNCTYPE(ctypes.c_uint64, ctypes.c_uint64, ctypes.POINTER(ctypes.c_char), ctypes.c_uint64, ctypes.POINTER(None)), ctypes.CFUNCTYPE(None, ctypes.POINTER(ctypes.c_char), ctypes.POINTER(None)), ctypes.CFUNCTYPE(None, ctypes.c_uint64, ctypes.POINTER(None)), ctypes.POINTER(struct_amd_comgr_disassembly_info_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_destroy_disassembly_info = _libraries['libamd_comgr.so'].amd_comgr_destroy_disassembly_info
+    amd_comgr_destroy_disassembly_info.restype = amd_comgr_status_t
+    amd_comgr_destroy_disassembly_info.argtypes = [amd_comgr_disassembly_info_t]
+except AttributeError:
+    pass
+try:
+    amd_comgr_disassemble_instruction = _libraries['libamd_comgr.so'].amd_comgr_disassemble_instruction
+    amd_comgr_disassemble_instruction.restype = amd_comgr_status_t
+    amd_comgr_disassemble_instruction.argtypes = [amd_comgr_disassembly_info_t, uint64_t, ctypes.POINTER(None), ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_demangle_symbol_name = _libraries['libamd_comgr.so'].amd_comgr_demangle_symbol_name
+    amd_comgr_demangle_symbol_name.restype = amd_comgr_status_t
+    amd_comgr_demangle_symbol_name.argtypes = [amd_comgr_data_t, ctypes.POINTER(struct_amd_comgr_data_s)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_populate_mangled_names = _libraries['libamd_comgr.so'].amd_comgr_populate_mangled_names
+    amd_comgr_populate_mangled_names.restype = amd_comgr_status_t
+    amd_comgr_populate_mangled_names.argtypes = [amd_comgr_data_t, ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_get_mangled_name = _libraries['libamd_comgr.so'].amd_comgr_get_mangled_name
+    amd_comgr_get_mangled_name.restype = amd_comgr_status_t
+    amd_comgr_get_mangled_name.argtypes = [amd_comgr_data_t, size_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_populate_name_expression_map = _libraries['libamd_comgr.so'].amd_comgr_populate_name_expression_map
+    amd_comgr_populate_name_expression_map.restype = amd_comgr_status_t
+    amd_comgr_populate_name_expression_map.argtypes = [amd_comgr_data_t, ctypes.POINTER(ctypes.c_uint64)]
+except AttributeError:
+    pass
+try:
+    amd_comgr_map_name_expression_to_symbol_name = _libraries['libamd_comgr.so'].amd_comgr_map_name_expression_to_symbol_name
+    amd_comgr_map_name_expression_to_symbol_name.restype = amd_comgr_status_t
+    amd_comgr_map_name_expression_to_symbol_name.argtypes = [amd_comgr_data_t, ctypes.POINTER(ctypes.c_uint64), ctypes.POINTER(ctypes.c_char), ctypes.POINTER(ctypes.c_char)]
+except AttributeError:
+    pass
+class struct_code_object_info_s(Structure):
+    pass
+
+struct_code_object_info_s._pack_ = 1 # source:False
+struct_code_object_info_s._fields_ = [
+    ('isa', ctypes.POINTER(ctypes.c_char)),
+    ('size', ctypes.c_uint64),
+    ('offset', ctypes.c_uint64),
+]
+
+amd_comgr_code_object_info_t = struct_code_object_info_s
+try:
+    amd_comgr_lookup_code_object = _libraries['libamd_comgr.so'].amd_comgr_lookup_code_object
+    amd_comgr_lookup_code_object.restype = amd_comgr_status_t
+    amd_comgr_lookup_code_object.argtypes = [amd_comgr_data_t, ctypes.POINTER(struct_code_object_info_s), size_t]
+except AttributeError:
+    pass
+__all__ = \
+    ['AMD_COMGR_ACTION_ADD_DEVICE_LIBRARIES',
+    'AMD_COMGR_ACTION_ADD_PRECOMPILED_HEADERS',
+    'AMD_COMGR_ACTION_ASSEMBLE_SOURCE_TO_RELOCATABLE',
+    'AMD_COMGR_ACTION_CODEGEN_BC_TO_ASSEMBLY',
+    'AMD_COMGR_ACTION_CODEGEN_BC_TO_RELOCATABLE',
+    'AMD_COMGR_ACTION_COMPILE_SOURCE_TO_BC',
+    'AMD_COMGR_ACTION_COMPILE_SOURCE_TO_FATBIN',
+    'AMD_COMGR_ACTION_COMPILE_SOURCE_WITH_DEVICE_LIBS_TO_BC',
+    'AMD_COMGR_ACTION_DISASSEMBLE_BYTES_TO_SOURCE',
+    'AMD_COMGR_ACTION_DISASSEMBLE_EXECUTABLE_TO_SOURCE',
+    'AMD_COMGR_ACTION_DISASSEMBLE_RELOCATABLE_TO_SOURCE',
+    'AMD_COMGR_ACTION_LAST', 'AMD_COMGR_ACTION_LINK_BC_TO_BC',
+    'AMD_COMGR_ACTION_LINK_RELOCATABLE_TO_EXECUTABLE',
+    'AMD_COMGR_ACTION_LINK_RELOCATABLE_TO_RELOCATABLE',
+    'AMD_COMGR_ACTION_OPTIMIZE_BC_TO_BC',
+    'AMD_COMGR_ACTION_SOURCE_TO_PREPROCESSOR',
+    'AMD_COMGR_DATA_KIND_AR', 'AMD_COMGR_DATA_KIND_AR_BUNDLE',
+    'AMD_COMGR_DATA_KIND_BC', 'AMD_COMGR_DATA_KIND_BC_BUNDLE',
+    'AMD_COMGR_DATA_KIND_BYTES', 'AMD_COMGR_DATA_KIND_DIAGNOSTIC',
+    'AMD_COMGR_DATA_KIND_EXECUTABLE', 'AMD_COMGR_DATA_KIND_FATBIN',
+    'AMD_COMGR_DATA_KIND_INCLUDE', 'AMD_COMGR_DATA_KIND_LAST',
+    'AMD_COMGR_DATA_KIND_LOG',
+    'AMD_COMGR_DATA_KIND_PRECOMPILED_HEADER',
+    'AMD_COMGR_DATA_KIND_RELOCATABLE', 'AMD_COMGR_DATA_KIND_SOURCE',
+    'AMD_COMGR_DATA_KIND_UNDEF', 'AMD_COMGR_LANGUAGE_HC',
+    'AMD_COMGR_LANGUAGE_HIP', 'AMD_COMGR_LANGUAGE_LAST',
+    'AMD_COMGR_LANGUAGE_NONE', 'AMD_COMGR_LANGUAGE_OPENCL_1_2',
+    'AMD_COMGR_LANGUAGE_OPENCL_2_0', 'AMD_COMGR_METADATA_KIND_LAST',
+    'AMD_COMGR_METADATA_KIND_LIST', 'AMD_COMGR_METADATA_KIND_MAP',
+    'AMD_COMGR_METADATA_KIND_NULL', 'AMD_COMGR_METADATA_KIND_STRING',
+    'AMD_COMGR_STATUS_ERROR',
+    'AMD_COMGR_STATUS_ERROR_INVALID_ARGUMENT',
+    'AMD_COMGR_STATUS_ERROR_OUT_OF_RESOURCES',
+    'AMD_COMGR_STATUS_SUCCESS', 'AMD_COMGR_SYMBOL_INFO_IS_UNDEFINED',
+    'AMD_COMGR_SYMBOL_INFO_LAST', 'AMD_COMGR_SYMBOL_INFO_NAME',
+    'AMD_COMGR_SYMBOL_INFO_NAME_LENGTH', 'AMD_COMGR_SYMBOL_INFO_SIZE',
+    'AMD_COMGR_SYMBOL_INFO_TYPE', 'AMD_COMGR_SYMBOL_INFO_VALUE',
+    'AMD_COMGR_SYMBOL_TYPE_AMDGPU_HSA_KERNEL',
+    'AMD_COMGR_SYMBOL_TYPE_COMMON', 'AMD_COMGR_SYMBOL_TYPE_FILE',
+    'AMD_COMGR_SYMBOL_TYPE_FUNC', 'AMD_COMGR_SYMBOL_TYPE_NOTYPE',
+    'AMD_COMGR_SYMBOL_TYPE_OBJECT', 'AMD_COMGR_SYMBOL_TYPE_SECTION',
+    'AMD_COMGR_SYMBOL_TYPE_UNKNOWN', 'amd_comgr_action_data_count',
+    'amd_comgr_action_data_get_data',
+    'amd_comgr_action_info_get_isa_name',
+    'amd_comgr_action_info_get_language',
+    'amd_comgr_action_info_get_logging',
+    'amd_comgr_action_info_get_option_list_count',
+    'amd_comgr_action_info_get_option_list_item',
+    'amd_comgr_action_info_get_options',
+    'amd_comgr_action_info_get_working_directory_path',
+    'amd_comgr_action_info_set_isa_name',
+    'amd_comgr_action_info_set_language',
+    'amd_comgr_action_info_set_logging',
+    'amd_comgr_action_info_set_option_list',
+    'amd_comgr_action_info_set_options',
+    'amd_comgr_action_info_set_working_directory_path',
+    'amd_comgr_action_info_t', 'amd_comgr_action_kind_s',
+    'amd_comgr_action_kind_t', 'amd_comgr_action_kind_t__enumvalues',
+    'amd_comgr_code_object_info_t', 'amd_comgr_create_action_info',
+    'amd_comgr_create_data', 'amd_comgr_create_data_set',
+    'amd_comgr_create_disassembly_info',
+    'amd_comgr_create_symbolizer_info', 'amd_comgr_data_kind_s',
+    'amd_comgr_data_kind_t', 'amd_comgr_data_kind_t__enumvalues',
+    'amd_comgr_data_set_add', 'amd_comgr_data_set_remove',
+    'amd_comgr_data_set_t', 'amd_comgr_data_t',
+    'amd_comgr_demangle_symbol_name', 'amd_comgr_destroy_action_info',
+    'amd_comgr_destroy_data_set',
+    'amd_comgr_destroy_disassembly_info',
+    'amd_comgr_destroy_metadata', 'amd_comgr_destroy_symbolizer_info',
+    'amd_comgr_disassemble_instruction',
+    'amd_comgr_disassembly_info_t', 'amd_comgr_do_action',
+    'amd_comgr_get_data', 'amd_comgr_get_data_isa_name',
+    'amd_comgr_get_data_kind', 'amd_comgr_get_data_metadata',
+    'amd_comgr_get_data_name', 'amd_comgr_get_isa_count',
+    'amd_comgr_get_isa_metadata', 'amd_comgr_get_isa_name',
+    'amd_comgr_get_mangled_name', 'amd_comgr_get_metadata_kind',
+    'amd_comgr_get_metadata_list_size',
+    'amd_comgr_get_metadata_map_size',
+    'amd_comgr_get_metadata_string', 'amd_comgr_get_version',
+    'amd_comgr_index_list_metadata', 'amd_comgr_iterate_map_metadata',
+    'amd_comgr_iterate_symbols', 'amd_comgr_language_s',
+    'amd_comgr_language_t', 'amd_comgr_language_t__enumvalues',
+    'amd_comgr_lookup_code_object',
+    'amd_comgr_map_name_expression_to_symbol_name',
+    'amd_comgr_metadata_kind_s', 'amd_comgr_metadata_kind_t',
+    'amd_comgr_metadata_kind_t__enumvalues',
+    'amd_comgr_metadata_lookup', 'amd_comgr_metadata_node_t',
+    'amd_comgr_populate_mangled_names',
+    'amd_comgr_populate_name_expression_map',
+    'amd_comgr_release_data', 'amd_comgr_set_data',
+    'amd_comgr_set_data_from_file_slice', 'amd_comgr_set_data_name',
+    'amd_comgr_status_s', 'amd_comgr_status_string',
+    'amd_comgr_status_t', 'amd_comgr_status_t__enumvalues',
+    'amd_comgr_symbol_get_info', 'amd_comgr_symbol_info_s',
+    'amd_comgr_symbol_info_t', 'amd_comgr_symbol_info_t__enumvalues',
+    'amd_comgr_symbol_lookup', 'amd_comgr_symbol_t',
+    'amd_comgr_symbol_type_s', 'amd_comgr_symbol_type_t',
+    'amd_comgr_symbol_type_t__enumvalues', 'amd_comgr_symbolize',
+    'amd_comgr_symbolizer_info_t', 'size_t',
+    'struct_amd_comgr_action_info_s', 'struct_amd_comgr_data_s',
+    'struct_amd_comgr_data_set_s',
+    'struct_amd_comgr_disassembly_info_s',
+    'struct_amd_comgr_metadata_node_s', 'struct_amd_comgr_symbol_s',
+    'struct_amd_comgr_symbolizer_info_s', 'struct_code_object_info_s',
+    'uint64_t']

tinygrad/runtime/graph/cuda.py

@@ -1,6 +1,6 @@
 import ctypes
 from typing import Any, Optional, Tuple, Dict, List, cast
-import gpuctypes.cuda as cuda
+import tinygrad.autogen.cuda as cuda
 from tinygrad.helpers import init_c_var, encode_args_cuda_style, all_same, GraphException
 from tinygrad.device import CompiledASTRunner, update_stats, Buffer
 from tinygrad.runtime.ops_cuda import check, cu_time_execution

tinygrad/runtime/graph/hip.py

@@ -1,6 +1,6 @@
 import ctypes
 from typing import Tuple
-import gpuctypes.hip as hip
+import tinygrad.autogen.hip as hip
 from tinygrad.helpers import init_c_var, time_execution_cuda_style
 from tinygrad.runtime.ops_hip import check
 from tinygrad.runtime.graph.cuda import CUDAGraph

tinygrad/runtime/ops_cuda.py

@@ -2,7 +2,7 @@ from __future__ import annotations
 import subprocess, hashlib, tempfile, ctypes, ctypes.util, functools, re
 from pathlib import Path
 from typing import Tuple, Optional
-import gpuctypes.cuda as cuda
+import tinygrad.autogen.cuda as cuda
 from tinygrad.helpers import DEBUG, getenv, from_mv, init_c_var, colored, cpu_time_execution, compile_cuda_style, encode_args_cuda_style, time_execution_cuda_style  # noqa: E501
 from tinygrad.device import Compiled, LRUAllocator, MallocAllocator
 from tinygrad.codegen.kernel import LinearizerOptions
@@ -22,7 +22,7 @@ CUDACPU = getenv("CUDACPU") == 1
 if CUDACPU:
   gpuocelot_lib = ctypes.CDLL(ctypes.util.find_library("gpuocelot"))
   gpuocelot_lib.ptx_run.argtypes = [ctypes.c_char_p, ctypes.c_int, ctypes.POINTER(ctypes.c_void_p), ctypes.c_int, ctypes.c_int, ctypes.c_int, ctypes.c_int, ctypes.c_int, ctypes.c_int, ctypes.c_int]  # noqa: E501
-  cuda.cuLaunchKernel = lambda src, gx, gy, gz, lx, ly, lz, shared, stream, unused_extra, args: gpuocelot_lib.ptx_run(src, len(args), (ctypes.c_void_p * len(args))(*[ctypes.cast(x, ctypes.c_void_p) for x in args]), lx, ly, lz, gx, gy, gz, shared)  # noqa: E501
+  cuda.cuLaunchKernel = lambda src, gx, gy, gz, lx, ly, lz, shared, stream, unused_extra, args: gpuocelot_lib.ptx_run(src, len(args), (ctypes.c_void_p * len(args))(*[ctypes.cast(x, ctypes.c_void_p) for x in args]), lx, ly, lz, gx, gy, gz, shared)  # type: ignore  # noqa: E501
 
 def check(status):
   if status != 0: raise RuntimeError(f"CUDA Error {status}, {ctypes.string_at(init_c_var(ctypes.POINTER(ctypes.c_char)(), lambda x: cuda.cuGetErrorString(status, ctypes.byref(x)))).decode()}")  # noqa: E501

tinygrad/runtime/ops_gpu.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 from typing import Tuple, Optional, List, cast
 import ctypes, functools, hashlib
-import gpuctypes.opencl as cl
+import tinygrad.autogen.opencl as cl
 from tinygrad.helpers import init_c_var, to_char_p_p, from_mv, OSX, DEBUG
 from tinygrad.codegen.kernel import LinearizerOptions
 from tinygrad.renderer.cstyle import OpenCLRenderer
@@ -41,13 +41,14 @@ class CLProgram:
     if hasattr(self, 'kernel'): check(cl.clReleaseKernel(self.kernel))
     if hasattr(self, 'program'): check(cl.clReleaseProgram(self.program))
 
-  def __call__(self, *bufs:cl.cl_mem, global_size:Tuple[int,int,int]=(1,1,1), local_size:Optional[Tuple[int,int,int]]=None, vals:Tuple[int, ...]=(), wait=False) -> Optional[float]:  # noqa: E501
+  def __call__(self, *bufs:ctypes._CData, global_size:Tuple[int,int,int]=(1,1,1), local_size:Optional[Tuple[int,int,int]]=None, vals:Tuple[int, ...]=(), wait=False) -> Optional[float]:  # noqa: E501
     for i,b in enumerate(bufs): cl.clSetKernelArg(self.kernel, i, ctypes.sizeof(b), ctypes.byref(b))
-    for i,b in enumerate(vals,start=len(bufs)): cl.clSetKernelArg(self.kernel, i, 4, ctypes.byref(ctypes.c_int32(b)))
+    for i,v in enumerate(vals,start=len(bufs)): cl.clSetKernelArg(self.kernel, i, 4, ctypes.byref(ctypes.c_int32(v)))
     if local_size is not None: global_size = cast(Tuple[int,int,int], tuple(int(g*l) for g,l in zip(global_size, local_size)))
     event = cl.cl_event() if wait else None
     check(cl.clEnqueueNDRangeKernel(self.device.queue, self.kernel, len(global_size), None, (ctypes.c_size_t * len(global_size))(*global_size), (ctypes.c_size_t * len(local_size))(*local_size) if local_size else None, 0, None, event))  # noqa: E501
     if wait:
+      assert event is not None
       check(cl.clWaitForEvents(1, ctypes.byref(event)))
       start = init_c_var(ctypes.c_uint64(), lambda x: check(cl.clGetEventProfilingInfo(event, cl.CL_PROFILING_COMMAND_START, ctypes.sizeof(x), ctypes.byref(x), None)))  # noqa: E501
       end = init_c_var(ctypes.c_uint64(), lambda x: check(cl.clGetEventProfilingInfo(event, cl.CL_PROFILING_COMMAND_END, ctypes.sizeof(x), ctypes.byref(x), None)))  # noqa: E501
@@ -58,19 +59,19 @@ class CLAllocator(LRUAllocator):
   def __init__(self, device:CLDevice):
     self.device = device
     super().__init__()
-  def _alloc(self, size:int) -> cl.cl_mem:
+  def _alloc(self, size:int) -> ctypes._CData:
     return checked(cl.clCreateBuffer(self.device.context, cl.CL_MEM_READ_WRITE, size, None, ctypes.byref(status := ctypes.c_int32())), status)
-  def _alloc_with_options(self, size:int, options:BufferOptions) -> cl.cl_mem:
+  def _alloc_with_options(self, size:int, options:BufferOptions) -> ctypes._CData:
     if options.image is not None:
       return checked(cl.clCreateImage2D(self.device.context, cl.CL_MEM_READ_WRITE,
                                         cl.cl_image_format(cl.CL_RGBA, {2: cl.CL_HALF_FLOAT, 4: cl.CL_FLOAT}[options.image.itemsize]),
                                         options.image.shape[1], options.image.shape[0], 0, None, ctypes.byref(status := ctypes.c_int32())), status)
     else: return self._alloc(size)
-  def _free(self, buf:cl.cl_mem): check(cl.clReleaseMemObject(buf))
+  def _free(self, buf:ctypes._CData): check(cl.clReleaseMemObject(buf))
-  def copyin(self, dest:cl.cl_mem, src:memoryview):
+  def copyin(self, dest:ctypes._CData, src:memoryview):
     check(cl.clEnqueueWriteBuffer(self.device.queue, dest, False, 0, len(src)*src.itemsize, from_mv(src), 0, None, None))
     self.device.pending_copyin.append(src)    # NOTE: these can't be freed until the GPU actually executes this command
-  def copyout(self, dest:memoryview, src:cl.cl_mem):
+  def copyout(self, dest:memoryview, src:ctypes._CData):
     check(cl.clEnqueueReadBuffer(self.device.queue, src, False, 0, len(dest)*dest.itemsize, from_mv(dest), 0, None, None))
     self.device.synchronize()
 

tinygrad/runtime/ops_hip.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 import ctypes, functools, subprocess, io
 from typing import Tuple, TypeVar, List, Any, cast, Set
-import gpuctypes.hip as hip
+import tinygrad.autogen.hip as hip
 from tinygrad.helpers import DEBUG, getenv, init_c_var
 from tinygrad.helpers import from_mv, round_up, to_mv, colored, init_c_struct_t, to_char_p_p, get_bytes
 from tinygrad.device import Compiled, LRUAllocator, MallocAllocator, BufferOptions, JITRunner, Device, Buffer, update_stats
@@ -133,7 +133,7 @@ class HIPDevice(Compiled):
   def __init__(self, device:str=""):
     self.device = int(device.split(":")[1]) if ":" in device else 0
     self.arch = init_c_var(hip.hipDeviceProp_t(), lambda x: check(hip.hipGetDeviceProperties(x, self.device))).gcnArchName.decode() if not MOCKHIP else "gfx1100"  # noqa: E501
-    self.pending_copyin: List[hip.hipDeviceptr_t] = []
+    self.pending_copyin: List[ctypes.c_void_p] = []
     self.track_cross_buffer: List[Any] = []
     self.peers: Set[int] = set()