Skip to content

CUDA 12 GPUOcelot sm86 support #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
li0nr merged 10 commits into from
Aug 30, 2026
Merged

CUDA 12 GPUOcelot sm86 support #2

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
a245bc0
use CUDA 12 GPUOcelot build
li0nr Aug 25, 2026
c86a7a0
ci: update NVRTC to 12.9.86
li0nr Aug 25, 2026
0f16d58
min gpu arch is now sm_50.
li0nr Aug 28, 2026
0c07970
min gpu arch is now sm_86.
li0nr Aug 28, 2026
ec0681b
sm_86 CUDA half float casts
li0nr Aug 28, 2026
72c3147
handle CUDA half casts to ulong and bfloat16
li0nr Aug 28, 2026
b6aaa14
half floats in cuda 12 bulk fixes.
li0nr Aug 28, 2026
9a57f5d
ci: use GPUOcelot CUDA 12 v2 release
li0nr Aug 30, 2026
43294c8
ci: use CUDA 12 and GPUOcelot sm86
li0nr Aug 30, 2026
7b1883c
ci: keep existing CUDA library setup
li0nr Aug 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 11 additions & 5 deletions .github/actions/setup-tinygrad/action.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -250,10 +250,16 @@ runs:
if: inputs.cuda == 'true'
shell: bash
run: |
sudo mkdir -p /usr/local/cuda/targets/x86_64-linux
curl -fL https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvrtc/linux-x86_64/cuda_nvrtc-linux-x86_64-11.5.119-archive.tar.xz \
| sudo tar -xJ -C /usr/local/cuda/targets/x86_64-linux --strip-components=1
echo /usr/local/cuda/targets/x86_64-linux/lib | sudo tee /etc/ld.so.conf.d/cuda-nvrtc.conf
cuda_root=/usr/local/cuda/targets/x86_64-linux
sudo mkdir -p "$cuda_root"
for archive in \
cuda_nvrtc-linux-x86_64-12.9.86-archive.tar.xz \
cuda_cudart-linux-x86_64-12.9.37-archive.tar.xz; do
curl -fL "https://developer.download.nvidia.com/compute/cuda/redist/${archive%%-linux-x86_64-*}/linux-x86_64/$archive" \
| sudo tar -xJ -C "$cuda_root" --strip-components=1
done
echo "CUDA_PATH=$cuda_root" >> "$GITHUB_ENV"
echo "$cuda_root/lib" | sudo tee /etc/ld.so.conf.d/cuda-nvrtc.conf
sudo ldconfig

# **** gpuocelot ****
Expand All @@ -262,7 +268,7 @@ runs:
shell: bash
run: |
sudo mkdir -p /usr/local/lib
sudo curl --output-dir /usr/local/lib -fLO https://github.com/tinygrad/gpuocelot/releases/download/v0.1.0/libgpuocelot.${{ runner.os == 'Linux' && 'so' || 'dylib' }}
sudo curl --output-dir /usr/local/lib -fLO https://github.com/li0nr/gpuocelot/releases/download/cuda-12-support-sm86/libgpuocelot.${{ runner.os == 'Linux' && 'so' || 'dylib' }}

# **** WebGPU ****

Expand Down
2 changes: 1 addition & 1 deletion test/mockgpu/nv/nvdriver.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -168,7 +168,7 @@ def rm_control(self, argp):
else:
for i,c in enumerate(classes): params.classList[i] = c
elif struct.cmd == nv_gpu.NV2080_CTRL_CMD_GR_GET_INFO:
info = {nv_gpu.NV2080_CTRL_GR_INFO_INDEX_SM_VERSION: nv_gpu.NV2080_CTRL_GR_INFO_SM_VERSION_3_5,
info = {nv_gpu.NV2080_CTRL_GR_INFO_INDEX_SM_VERSION: nv_gpu.NV2080_CTRL_GR_INFO_SM_VERSION_8_6,
nv_gpu.NV2080_CTRL_GR_INFO_INDEX_LITTER_NUM_GPCS: 1,
nv_gpu.NV2080_CTRL_GR_INFO_INDEX_LITTER_NUM_TPC_PER_GPC: 1,
nv_gpu.NV2080_CTRL_GR_INFO_INDEX_LITTER_NUM_SM_PER_TPC: 1,
Expand Down
10 changes: 10 additions & 0 deletions tinygrad/renderer/cstyle.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -426,6 +426,16 @@ def __init__(self, target:Target, use_nvcc=False):
type_map = {dtypes.uint32: "uint", dtypes.bfloat16: "nv_bfloat16", dtypes.fp8e4m3: "__nv_fp8_e4m3", dtypes.fp8e5m2: "__nv_fp8_e5m2"}
extra_matcher = create_non_native_float_pats(dtypes.fp8s, casting=False) + PatternMatcher([
(UPat(Ops.CAST, dtypes.fp8s, UPat.var("x", dtypes.fp8s), name='y'), lambda x,y: x.cast(dtypes.float).cast(y.dtype) if x.dtype!=y.dtype else None),
(UPat(Ops.CAST, (dtypes.char, dtypes.uchar, dtypes.long, dtypes.ulong, dtypes.bfloat16), (UPat.var("x", dtypes.half),), name="y"),
lambda x,y: x.cast(dtypes.float).cast(y.dtype)),
(UPat(Ops.CAST, dtypes.half, (UPat.var("x", (dtypes.long, dtypes.ulong, dtypes.bfloat16)),)),
lambda x: x.cast(dtypes.float).cast(dtypes.half)),
(UPat(Ops.CAST, (dtypes.char, dtypes.uchar), (UPat.var("x", dtypes.bfloat16),), name="y"),
lambda x,y: x.cast(dtypes.float).cast(y.dtype)),
(UPat(Ops.CAST, (dtypes.long, dtypes.ulong), (UPat.var("x", dtypes.bfloat16),), name="y"),
lambda x,y: x.cast(dtypes.float).cast(y.dtype)),
(UPat(Ops.CAST, dtypes.bfloat16, (UPat.var("x", (dtypes.long, dtypes.ulong)),)),
lambda x: x.cast(dtypes.float).cast(dtypes.bfloat16)),
])
string_rewrite = PatternMatcher([
(UPat(Ops.BITCAST, name="x"), lambda ctx,x: f"tg_bitcast<{ctx.render_dtype(x.dtype)}>(({ctx.render_dtype(x.src[0].dtype)})({ctx[x.src[0]]}))"
Expand Down