StarPilot/tinygrad_repo/tinygrad/runtime/support/compiler_amd.py

import ctypes, hashlib, tempfile, subprocess, pathlib
from tinygrad.helpers import amdgpu_disassemble, getenv
from tinygrad.runtime.autogen import comgr
try:
  comgr.amd_comgr_get_version(ctypes.byref(major:=ctypes.c_uint64()), ctypes.byref(minor:=ctypes.c_uint64()))
  if major.value >= 3:
    # in comgr 3 the values of enums in headers were changed: https://github.com/ROCm/llvm-project/issues/272
    import tinygrad.runtime.autogen.comgr_3 as comgr # type: ignore[no-redef]
    assert comgr.AMD_COMGR_LANGUAGE_HIP == 3
except AttributeError: pass  # ignore if ROCm isn't installed
from tinygrad.device import Compiler, CompileError
from tinygrad.runtime.support import c
from tinygrad.helpers import to_char_p_p

def check(status):
  if status != 0:
    comgr.amd_comgr_status_string(status, ctypes.byref(status_str := ctypes.POINTER(ctypes.c_char)()))
    raise RuntimeError(f"comgr fail {status}, {ctypes.string_at(status_str).decode()}")

def _get_comgr_data(data_set, data_type):
  check(comgr.amd_comgr_action_data_get_data(data_set, data_type, 0, ctypes.byref(data_exec := comgr.amd_comgr_data_t())))
  check(comgr.amd_comgr_get_data(data_exec, ctypes.byref(sz := ctypes.c_uint64()), None))
  check(comgr.amd_comgr_get_data(data_exec, ctypes.byref(sz), (dat := ctypes.create_string_buffer(sz.value))))
  check(comgr.amd_comgr_release_data(data_exec))
  return bytes(dat)

# amd_comgr_action_info_set_options was deprecated
def set_options(action_info, options:bytes):
  # TODO: this type should be correct in the autogen stub
  @comgr.dll.bind(comgr.amd_comgr_status_t, comgr.amd_comgr_action_info_t, c.POINTER[c.POINTER[ctypes.c_char]], comgr.size_t)
  def amd_comgr_action_info_set_option_list(ai, o, c) -> comgr.amd_comgr_status_t: pass # type: ignore[empty-body]
  return amd_comgr_action_info_set_option_list(action_info, to_char_p_p(options_list:=options.split(b' ')), len(options_list))

# AMD_COMGR_SAVE_TEMPS=1 AMD_COMGR_REDIRECT_LOGS=stdout AMD_COMGR_EMIT_VERBOSE_LOGS=1
def compile_hip(prg:str, arch="gfx1100", asm=False) -> bytes:
  check(comgr.amd_comgr_create_action_info(ctypes.byref(action_info := comgr.amd_comgr_action_info_t())))
  check(comgr.amd_comgr_action_info_set_language(action_info, comgr.AMD_COMGR_LANGUAGE_HIP))
  check(comgr.amd_comgr_action_info_set_isa_name(action_info, b"amdgcn-amd-amdhsa--" + arch.encode()))
  check(comgr.amd_comgr_action_info_set_logging(action_info, True))

  check(comgr.amd_comgr_create_data_set(ctypes.byref(data_set_src := comgr.amd_comgr_data_set_t())))
  check(comgr.amd_comgr_create_data_set(ctypes.byref(data_set_bc := comgr.amd_comgr_data_set_t())))
  check(comgr.amd_comgr_create_data_set(ctypes.byref(data_set_reloc := comgr.amd_comgr_data_set_t())))
  check(comgr.amd_comgr_create_data_set(ctypes.byref(data_set_exec := comgr.amd_comgr_data_set_t())))

  check(comgr.amd_comgr_create_data(comgr.AMD_COMGR_DATA_KIND_SOURCE, ctypes.byref(data_src := comgr.amd_comgr_data_t())))
  check(comgr.amd_comgr_set_data(data_src, len(rprg := prg.encode()), rprg))

  if asm:
    check(comgr.amd_comgr_set_data_name(data_src, b"<null>.s"))
    check(comgr.amd_comgr_data_set_add(data_set_src, data_src))
    status = comgr.amd_comgr_do_action(comgr.AMD_COMGR_ACTION_ASSEMBLE_SOURCE_TO_RELOCATABLE, action_info, data_set_src, data_set_reloc)
    if status != 0:
      print(_get_comgr_data(data_set_reloc, comgr.AMD_COMGR_DATA_KIND_LOG).decode())
      raise RuntimeError("assemble failed")
  else:
    check(comgr.amd_comgr_set_data_name(data_src, b"<null>"))
    check(comgr.amd_comgr_data_set_add(data_set_src, data_src))
    # -include hiprtc_runtime.h was removed
    options = [
      "-O3", "-mcumode", "--hip-version=6.0.32830", "-DHIP_VERSION_MAJOR=6", "-DHIP_VERSION_MINOR=0", "-DHIP_VERSION_PATCH=32830",
      "-D__HIPCC_RTC__", "-std=c++14", "-nogpuinc", "-Wno-gnu-line-marker", "-Wno-missing-prototypes", f"--offload-arch={arch}",
      "-I/opt/rocm/include", "-Xclang -disable-llvm-passes", "-Xclang -aux-triple", "-Xclang x86_64-unknown-linux-gnu"]
    check(set_options(action_info, ' '.join(options).encode()))
    status = comgr.amd_comgr_do_action(comgr.AMD_COMGR_ACTION_COMPILE_SOURCE_WITH_DEVICE_LIBS_TO_BC, action_info, data_set_src, data_set_bc)
    if status != 0:
      print(_get_comgr_data(data_set_bc, comgr.AMD_COMGR_DATA_KIND_LOG).decode())
      raise RuntimeError("compile failed")
    check(set_options(action_info, b"-O3 -mllvm -amdgpu-internalize-symbols"))
    check(comgr.amd_comgr_do_action(comgr.AMD_COMGR_ACTION_CODEGEN_BC_TO_RELOCATABLE, action_info, data_set_bc, data_set_reloc))

  check(set_options(action_info, b""))
  check(comgr.amd_comgr_do_action(comgr.AMD_COMGR_ACTION_LINK_RELOCATABLE_TO_EXECUTABLE, action_info, data_set_reloc, data_set_exec))
  ret = _get_comgr_data(data_set_exec, comgr.AMD_COMGR_DATA_KIND_EXECUTABLE)
  check(comgr.amd_comgr_release_data(data_src))
  for x in [data_set_src, data_set_bc, data_set_reloc, data_set_exec]: check(comgr.amd_comgr_destroy_data_set(x))
  check(comgr.amd_comgr_destroy_action_info(action_info))
  return ret

class HIPCompiler(Compiler):
  def __init__(self, arch:str):
    assert comgr.dll.nm in c.DLL._loaded_, f"comgr not available: {comgr.dll.emsg}"
    self.arch = arch
    super().__init__(f"compile_hip_{self.arch}")
  def compile(self, src:str) -> bytes:
    try: return compile_hip(src, self.arch, src.split('\n', 1)[0].strip() == '.text')
    except RuntimeError as e: raise CompileError(e) from e
  def disassemble(self, lib:bytes): amdgpu_disassemble(lib)

class HIPCCCompiler(Compiler):
  def __init__(self, arch:str, extra_options:list[str]=[]):
    self.arch, self.extra_options, self.no_hipcc = arch, extra_options, getenv("NO_HIPCC")
    super().__init__(f"compile_hipcc_{self.arch}_{hashlib.sha256(' '.join(extra_options).encode()).hexdigest()[:8]}"+
                     ("_nohipcc" if self.no_hipcc else ""))
  def compile(self, src:str) -> bytes:
    if self.no_hipcc: return b""
    with tempfile.NamedTemporaryFile(suffix=".cpp") as srcf, tempfile.NamedTemporaryFile(suffix=".bc") as bcf:
      with tempfile.NamedTemporaryFile(suffix=".hsaco") as libf:
        srcf.write(src.encode())
        srcf.flush()

        rocm_path = getenv("ROCM_PATH", "/opt/rocm")
        subprocess.run(["hipcc", "-c", "-emit-llvm", "--cuda-device-only", "-O3", "-mcumode",
                        f"--offload-arch={self.arch}", f"-I{rocm_path}/include/hip", "-o", bcf.name, srcf.name] + self.extra_options, check=True)
        subprocess.run(["hipcc", "-target", "amdgcn-amd-amdhsa", f"-mcpu={self.arch}",
                        "-O3", "-mllvm", "-amdgpu-internalize-symbols", "-c", "-o", libf.name, bcf.name] + self.extra_options, check=True)

        return pathlib.Path(libf.name).read_bytes()
  def disassemble(self, lib:bytes): amdgpu_disassemble(lib)