Merge branch 'remu-dev-mac'

This commit is contained in:
qazal
2024-02-15 12:16:51 +02:00
2 changed files with 13 additions and 34 deletions
+11 -31
View File
@@ -1,12 +1,13 @@
import ctypes
import tinygrad.runtime.autogen.comgr as comgr
def check(status):
import tinygrad.runtime.autogen.comgr as comgr
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):
import tinygrad.runtime.autogen.comgr as comgr
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))))
@@ -15,33 +16,12 @@ def _get_comgr_data(data_set, data_type):
# AMD_COMGR_SAVE_TEMPS=1 AMD_COMGR_REDIRECT_LOGS=stdout AMD_COMGR_EMIT_VERBOSE_LOGS=1
def compile_hip(prg:str, arch="gfx1100") -> 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))
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
check(comgr.amd_comgr_action_info_set_options(action_info, b"-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 --offload-arch=gfx1100 -I/opt/rocm/include -Xclang -disable-llvm-passes")) # noqa: E501
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(comgr.amd_comgr_action_info_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(comgr.amd_comgr_action_info_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
import http.client, urllib.parse
params = urllib.parse.urlencode({'code': prg})
headers = {"Content-type": "application/x-www-form-urlencoded", "Accept": "text/plain"}
conn = http.client.HTTPConnection("temps-mbp.home", 80)
conn.request("POST", "/", params, headers)
response = conn.getresponse()
asm = response.read().decode()
conn.close()
return asm.encode("utf-8")
+2 -3
View File
@@ -1,7 +1,6 @@
from __future__ import annotations
import ctypes, functools, subprocess, io
from typing import Tuple, TypeVar, List, Any, cast, Set
import tinygrad.runtime.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
from tinygrad.device import Compiled, LRUAllocator, BufferOptions, JITRunner, Device, Buffer, MallocAllocator, update_stats, Compiler
@@ -9,7 +8,7 @@ from tinygrad.renderer.cstyle import HIPRenderer
from tinygrad.codegen.kernel import LinearizerOptions
from tinygrad.runtime.compiler.hip_comgr import compile_hip
hip = None
class HIPCompiler(Compiler):
linearizer_opts = LinearizerOptions("HIP", has_tensor_cores=True)
def __init__(self, arch:str):
@@ -150,7 +149,7 @@ class HIPWaitEvent(JITRunner):
update_stats(colored("wait", "RED"), 0, 0, {}, None, 1, jit, device=self.dname)
if getenv("HIPCPU"):
rhip = ctypes.CDLL("/usr/local/lib/libremu.so")
rhip = ctypes.CDLL("/Users/qazal/code/tinygrad/remu/target/release/libremu.dylib")
class RHIPProgram:
def __init__(self, name:str, lib:bytes):
self.name, self.lib = name, lib