mawei/onepilot
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update251203 (#233)

Browse Source
This commit is contained in:
carrot
2025-12-03 10:28:27 +09:00
committed by GitHub
parent d6899edd97
commit c5ebcbcb97
347 changed files with 8678 additions and 13489 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tinygrad_repo/test/mockgpu/amd/amdgpu.py
+1 -13
View File
@@ -1,5 +1,6 @@
import ctypes, time
from test.mockgpu.gpu import VirtGPU
from test.mockgpu.helpers import _try_dlopen_remu
from tinygrad.helpers import getbits, to_mv, init_c_struct_t
import tinygrad.runtime.autogen.amd_gpu as amd_gpu, tinygrad.runtime.autogen.am.pm4_nv as pm4
@@ -24,19 +25,6 @@ WAIT_REG_MEM_FUNCTION_EQ = 3 # ==
WAIT_REG_MEM_FUNCTION_NEQ = 4 # !=
WAIT_REG_MEM_FUNCTION_GEQ = 5 # >=
REMU_PATHS = ["extra/remu/target/release/libremu.so", "libremu.so", "/usr/local/lib/libremu.so",
"extra/remu/target/release/libremu.dylib", "libremu.dylib", "/usr/local/lib/libremu.dylib", "/opt/homebrew/lib/libremu.dylib"]
def _try_dlopen_remu():
for path in REMU_PATHS:
try:
remu = ctypes.CDLL(path)
remu.run_asm.restype = ctypes.c_int32
remu.run_asm.argtypes = [ctypes.c_void_p, ctypes.c_uint32, ctypes.c_uint32, ctypes.c_uint32, ctypes.c_uint32,
ctypes.c_uint32, ctypes.c_uint32, ctypes.c_uint32, ctypes.c_void_p]
except OSError: pass
else: return remu
print("Could not find libremu.so")
return None
remu = _try_dlopen_remu()
def create_sdma_packets():
tinygrad_repo/test/mockgpu/cuda/cuda.py
+6 -5
View File
@@ -2,16 +2,14 @@ from __future__ import annotations
from typing import Any
import ctypes, time
from tinygrad.runtime.autogen import cuda as orig_cuda
from test.mockgpu.helpers import _try_dlopen_gpuocelot
from tinygrad.helpers import mv_address
for attr in dir(orig_cuda):
if not attr.startswith('__'):
globals()[attr] = getattr(orig_cuda, attr)
try:
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
except Exception: pass
gpuocelot_lib = _try_dlopen_gpuocelot()
# Global state
class CUDAState:
@@ -130,7 +128,10 @@ def cuModuleUnload(hmod) -> int:
def cuLaunchKernel(f, gx: int, gy: int, gz: int, lx: int, ly: int, lz: int, sharedMemBytes: int,
hStream: Any, kernelParams: Any, extra: Any) -> int:
cargs = [ctypes.cast(getattr(extra, field[0]), ctypes.c_void_p) for field in extra._fields_]
gpuocelot_lib.ptx_run(ctypes.cast(f.value, ctypes.c_char_p), len(cargs), (ctypes.c_void_p*len(cargs))(*cargs), lx, ly, lz, gx, gy, gz, 0)
try: gpuocelot_lib.ptx_run(ctypes.cast(f.value, ctypes.c_char_p), len(cargs), (ctypes.c_void_p*len(cargs))(*cargs), lx, ly, lz, gx, gy, gz, 0)
except Exception as e:
print("Error in cuLaunchKernel:", e)
return orig_cuda.CUDA_ERROR_LAUNCH_FAILED
return orig_cuda.CUDA_SUCCESS
def cuDeviceComputeCapability(major, minor, dev: int) -> int:
tinygrad_repo/test/mockgpu/helpers.py
+29
View File
@@ -0,0 +1,29 @@
import ctypes, ctypes.util
def _try_dlopen_gpuocelot():
GPUOCELOT_PATHS = [ctypes.util.find_library("gpuocelot")] if ctypes.util.find_library("gpuocelot") is not None else []
GPUOCELOT_PATHS += ["libgpuocelot.so", "/usr/local/lib/libgpuocelot.so",
"libgpuocelot.dylib", "/usr/local/lib/libgpuocelot.dylib", "/opt/homebrew/lib/libgpuocelot.dylib"]
for path in GPUOCELOT_PATHS:
try:
gpuocelot_lib = ctypes.CDLL(path)
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]
except OSError: pass
else: return gpuocelot_lib
print("Could not find libgpuocelot.so")
return None
def _try_dlopen_remu():
REMU_PATHS = ["extra/remu/target/release/libremu.so", "libremu.so", "/usr/local/lib/libremu.so",
"extra/remu/target/release/libremu.dylib", "libremu.dylib", "/usr/local/lib/libremu.dylib", "/opt/homebrew/lib/libremu.dylib"]
for path in REMU_PATHS:
try:
remu = ctypes.CDLL(path)
remu.run_asm.restype = ctypes.c_int32
remu.run_asm.argtypes = [ctypes.c_void_p, ctypes.c_uint32, ctypes.c_uint32, ctypes.c_uint32, ctypes.c_uint32,
ctypes.c_uint32, ctypes.c_uint32, ctypes.c_uint32, ctypes.c_void_p]
except OSError: pass
else: return remu
print("Could not find libremu.so")
return None
tinygrad_repo/test/mockgpu/nv/nvgpu.py
+6 -5
View File
@@ -2,6 +2,7 @@ import ctypes, ctypes.util, time
import tinygrad.runtime.autogen.nv_gpu as nv_gpu
from enum import Enum, auto
from test.mockgpu.gpu import VirtGPU
from test.mockgpu.helpers import _try_dlopen_gpuocelot
from tinygrad.helpers import to_mv, init_c_struct_t
def make_qmd_struct_type():
@@ -16,10 +17,7 @@ def make_qmd_struct_type():
qmd_struct_t = make_qmd_struct_type()
assert ctypes.sizeof(qmd_struct_t) == 0x40 * 4
try:
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
except Exception: pass
gpuocelot_lib = _try_dlopen_gpuocelot()
class SchedResult(Enum): CONT = auto(); YIELD = auto() # noqa: E702
@@ -99,7 +97,10 @@ class GPFIFO:
cargs = [ctypes.cast(args[i], ctypes.c_void_p) for i in range(args_cnt)] + [ctypes.cast(vals[i], ctypes.c_void_p) for i in range(vals_cnt)]
gx, gy, gz = qmd.cta_raster_width, qmd.cta_raster_height, qmd.cta_raster_depth
lx, ly, lz = qmd.cta_thread_dimension0, qmd.cta_thread_dimension1, qmd.cta_thread_dimension2
gpuocelot_lib.ptx_run(ctypes.cast(prg_addr, ctypes.c_char_p), args_cnt+vals_cnt, (ctypes.c_void_p*len(cargs))(*cargs), lx, ly, lz, gx, gy, gz, 0)
try:
gpuocelot_lib.ptx_run(ctypes.cast(prg_addr, ctypes.c_char_p), args_cnt+vals_cnt,
(ctypes.c_void_p*len(cargs))(*cargs), lx, ly, lz, gx, gy, gz, 0)
except Exception as e: print("failed to execute:", e)
if qmd.release0_enable:
rel0 = to_mv(qmd.release0_address_lower + (qmd.release0_address_upper << 32), 0x10).cast('Q')
rel0[0] = qmd.release0_payload_lower + (qmd.release0_payload_upper << 32)