tinygrad/tinygrad
2
0
Fork 3
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-08-29 10:56:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

nv_610 support (glm) (#16979)

Browse Source 
* nv_610 support

* unbump onnx

* fix autogen workflow
This commit is contained in:
George Hotz
2026-07-10 20:37:30 -07:00
committed by GitHub
parent 75a4bfddc9
commit 43ad225d36
4 changed files with 26576 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/autogen.yml
+1 -1
View File
@@ -42,7 +42,7 @@ jobs:
run: |
find tinygrad/runtime/autogen -type f -name "*.py" -not -path "*/amd/*" -not -name "__init__.py" -not -name "comgr.py" -not -name "metal.py" -not -name "iokit.py" -not -name "corefoundation.py" -not -name "libclang.py" -delete
python3 -c "from tinygrad.runtime.autogen import opencl"
python3 -c "from tinygrad.runtime.autogen import cuda, nvrtc, nvjitlink, nv_570, nv_580, nv"
python3 -c "from tinygrad.runtime.autogen import cuda, nvrtc, nvjitlink, nv_570, nv_580, nv_610, nv"
python3 -c "from tinygrad.runtime.autogen import comgr_3, hsa, hip, amd_gpu, sqtt, rocprof, amdgpu_kd, amdgpu_drm"
python3 -c "from tinygrad.runtime.autogen.am import *"
python3 -c "from tinygrad.runtime.autogen.nv_regs import *"
tinygrad/runtime/autogen/__init__.py
+3 -2
View File
@@ -3,7 +3,8 @@ from tinygrad.helpers import fetch, flatten, system, getenv
root = (here:=pathlib.Path(__file__).parent).parents[2]
nv_src = {"nv_570": "https://github.com/NVIDIA/open-gpu-kernel-modules/archive/81fe4fb417c8ac3b9bdcc1d56827d116743892a5.tar.gz",
"nv_580": "https://github.com/NVIDIA/open-gpu-kernel-modules/archive/2af9f1f0f7de4988432d4ae875b5858ffdb09cc2.tar.gz"}
"nv_580": "https://github.com/NVIDIA/open-gpu-kernel-modules/archive/2af9f1f0f7de4988432d4ae875b5858ffdb09cc2.tar.gz",
"nv_610": "https://github.com/NVIDIA/open-gpu-kernel-modules/archive/refs/tags/610.43.03.tar.gz"}
ffmpeg_src = "https://ffmpeg.org/releases/ffmpeg-8.0.1.tar.gz"
rocr_src = "https://github.com/ROCm/rocm-systems/archive/refs/tags/rocm-7.1.1.tar.gz"
linux_headers_deb = "https://snapshot.debian.org/archive/debian/20260207T145350Z/pool/main/l/linux/linux-libc-dev_6.18.9-1_all.deb"
@@ -60,7 +61,7 @@ def __getattr__(nm):
case "nvrtc": return load("nvrtc", ["{}/include/nvrtc.h"], dll="'nvrtc'", paths=nv_lib_path, srcs=nvrtc_src, prolog=["import sysconfig"])
case "nvjitlink": load("nvjitlink", [root/"extra/nvJitLink.h"], dll="'nvJitLink'", paths=nv_lib_path, prolog=["import sysconfig"])
case "kfd": return load("kfd", [root/"extra/hip_gpu_driver/kfd_ioctl.h"])
case "nv_570" | "nv_580":
case "nv_570" | "nv_580" | "nv_610":
return load(nm, [
*[root/"extra/nv_gpu_driver"/s for s in ["clc9b0.h", "clc6c0qmd.h","clcec0qmd.h", "nvdec_drv.h"]], "{}/kernel-open/common/inc/nvmisc.h",
*[f"{{}}/src/common/sdk/nvidia/inc/class/cl{s}.h" for s in ["0000", "0070", "0080", "2080", "2080_notification", "c56f", "c86f", "c96f", "c761",
tinygrad/runtime/autogen/nv_610.py
+26569
View File
File diff suppressed because one or more lines are too long
tinygrad/runtime/ops_nv.py
+3 -2
View File
@@ -10,7 +10,7 @@ from tinygrad.device import Compiled, BufferSpec
from tinygrad.helpers import getenv, mv_address, round_up, data64, data64_le, prod, OSX, hi32, lo32, PROFILE, ContextVar, VIZ, ProfileEvent
from tinygrad.renderer.ptx import PTXRenderer
from tinygrad.renderer.cstyle import CUDARenderer, NVCCRenderer
from tinygrad.runtime.autogen import nv_570, nv_580, mesa
from tinygrad.runtime.autogen import nv_570, nv_580, nv_610, mesa
from tinygrad.runtime.support.elf import elf_loader
from tinygrad.runtime.support.nv.nvdev import NVDev, NVMemoryManager
from tinygrad.runtime.support.system import System, PCIIfaceBase, MAP_FIXED
@@ -389,7 +389,8 @@ class NVKIface:
NVKIface.root = self.rm_alloc(0, nv_gpu.NV01_ROOT_CLIENT, None, root=0)
drvver = self.rm_control(self.root, nv_gpu.NV0000_CTRL_CMD_SYSTEM_GET_BUILD_VERSION_V2, nv_gpu.NV0000_CTRL_SYSTEM_GET_BUILD_VERSION_V2_PARAMS())
if int(drvver.driverVersionBuffer.decode().split('.')[0], 10) >= 580: nv_gpu = nv_580
if int(drvver.driverVersionBuffer.decode().split('.')[0], 10) >= 610: nv_gpu = nv_610
elif int(drvver.driverVersionBuffer.decode().split('.')[0], 10) >= 580: nv_gpu = nv_580
self.uvm(nv_gpu.UVM_INITIALIZE, nv_gpu.UVM_INITIALIZE_PARAMS())