Kernels
Ashiedu commited on
Commit
df5d07c
·
verified ·
1 Parent(s): 88c47e4

Initial kernel bundle upload: universal

Browse files
.gitattributes CHANGED
@@ -1,35 +1,3 @@
1
- *.7z filter=lfs diff=lfs merge=lfs -text
2
- *.arrow filter=lfs diff=lfs merge=lfs -text
3
- *.bin filter=lfs diff=lfs merge=lfs -text
4
- *.bz2 filter=lfs diff=lfs merge=lfs -text
5
- *.ckpt filter=lfs diff=lfs merge=lfs -text
6
- *.ftz filter=lfs diff=lfs merge=lfs -text
7
- *.gz filter=lfs diff=lfs merge=lfs -text
8
- *.h5 filter=lfs diff=lfs merge=lfs -text
9
- *.joblib filter=lfs diff=lfs merge=lfs -text
10
- *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
- *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
- *.model filter=lfs diff=lfs merge=lfs -text
13
- *.msgpack filter=lfs diff=lfs merge=lfs -text
14
- *.npy filter=lfs diff=lfs merge=lfs -text
15
- *.npz filter=lfs diff=lfs merge=lfs -text
16
- *.onnx filter=lfs diff=lfs merge=lfs -text
17
- *.ot filter=lfs diff=lfs merge=lfs -text
18
- *.parquet filter=lfs diff=lfs merge=lfs -text
19
- *.pb filter=lfs diff=lfs merge=lfs -text
20
- *.pickle filter=lfs diff=lfs merge=lfs -text
21
- *.pkl filter=lfs diff=lfs merge=lfs -text
22
- *.pt filter=lfs diff=lfs merge=lfs -text
23
  *.pth filter=lfs diff=lfs merge=lfs -text
24
- *.rar filter=lfs diff=lfs merge=lfs -text
25
- *.safetensors filter=lfs diff=lfs merge=lfs -text
26
- saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
- *.tar.* filter=lfs diff=lfs merge=lfs -text
28
- *.tar filter=lfs diff=lfs merge=lfs -text
29
- *.tflite filter=lfs diff=lfs merge=lfs -text
30
- *.tgz filter=lfs diff=lfs merge=lfs -text
31
- *.wasm filter=lfs diff=lfs merge=lfs -text
32
- *.xz filter=lfs diff=lfs merge=lfs -text
33
- *.zip filter=lfs diff=lfs merge=lfs -text
34
- *.zst filter=lfs diff=lfs merge=lfs -text
35
- *tfevents* filter=lfs diff=lfs merge=lfs -text
 
1
+ *.so filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  *.pth filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
.gitignore ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ build
2
+ *pyc
CARD.md ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: kernels
3
+ {% if license %}license: {{ license }}
4
+ {% endif %}---
5
+
6
+ This is the repository card of {{ repo_id }} that has been pushed on the Hub. It was built to be used with the [`kernels` library](https://github.com/huggingface/kernels). This card was automatically generated.
7
+
8
+ ## How to use
9
+ {% if functions %}
10
+
11
+ ```python
12
+ # make sure `kernels` is installed: `pip install -U kernels`
13
+ from kernels import get_kernel
14
+
15
+ kernel_module = get_kernel("{{ repo_id }}", version={{ version }})
16
+ {{ functions[0] }} = kernel_module.{{ functions[0] }}
17
+
18
+ {{ functions[0] }}(...)
19
+ ```
20
+ {% else %}
21
+
22
+ Usage example not available.
23
+ {% endif %}
24
+
25
+ ## Available functions
26
+ {% if functions %}
27
+ {% for func in functions %}
28
+ - `{{ func }}`
29
+ {% endfor %}
30
+ {% else %}
31
+
32
+ Function list not available.
33
+ {% endif %}
34
+ {% if layers %}
35
+
36
+ ## Available layers
37
+ {% for layer in layers %}
38
+ - `{{ layer }}`
39
+ {% endfor %}
40
+ {% endif %}
41
+
42
+ ## Benchmarks
43
+ {% if has_benchmark %}
44
+
45
+ Benchmarking script is available for this kernel. Run `kernels benchmark {{ repo_id }} --version {{ version }}`.
46
+ {% else %}
47
+
48
+ No benchmark available yet.
49
+ {% endif %}
50
+ {% if upstream %}
51
+
52
+ ## Upstream
53
+
54
+ The original source code for this kernel comes from {{ upstream }}.
55
+ {% endif %}
56
+ {% if source %}
57
+
58
+ ## Source
59
+
60
+ The kernel-builder formatted source for this kernel is available at {{ source }}.
61
+ {% endif %}
benchmarks/benchmark.py ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+
3
+ from kernels.benchmark import Benchmark
4
+
5
+
6
+ class UniversalBenchmark(Benchmark):
7
+ def setup(self):
8
+ self.size = 1024
9
+ self.input = torch.randn(self.size, self.size, device=self.device)
10
+ self.out = torch.empty_like(self.input)
11
+
12
+ def benchmark_base(self):
13
+ self.kernel.universal(self.out, self.input)
14
+
15
+ def verify_base(self) -> torch.Tensor:
16
+ # Reference implementation
17
+ return self.input.clone()
build.toml ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [general]
2
+ name = "universal"
3
+ license = "Apache-2.0"
4
+ backends = [
5
+ "rocm",
6
+ ]
7
+ version = 1
8
+ edition = 5
9
+
10
+ [general.hub]
11
+ repo-id = "Ashiedu/universal"
12
+
13
+ [torch]
14
+ src = [
15
+ "torch-ext/torch_binding.cpp",
16
+ "torch-ext/torch_binding.h",
17
+ ]
18
+
19
+ [kernel.universal_rocm]
20
+ backend = "rocm"
21
+ depends = ["torch"]
22
+ rocm-archs = [
23
+ "gfx1030",
24
+ ]
25
+ src = ["universal_cuda/universal.cu"]
example.py ADDED
@@ -0,0 +1,42 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # /// script
2
+ # requires-python = ">=3.13"
3
+ # dependencies = [
4
+ # "kernels",
5
+ # "numpy",
6
+ # "torch",
7
+ # ]
8
+ # ///
9
+
10
+ import platform
11
+ from pathlib import Path
12
+
13
+ import kernels
14
+ import torch
15
+
16
+ # Load the locally built kernel
17
+ kernel = kernels.get_local_kernel(Path("build"), "universal")
18
+
19
+ # Select device
20
+ if platform.system() == "Darwin":
21
+ device = torch.device("mps")
22
+ elif hasattr(torch, "xpu") and torch.xpu.is_available():
23
+ device = torch.device("xpu")
24
+ elif torch.version.cuda is not None and torch.cuda.is_available():
25
+ device = torch.device("cuda")
26
+ else:
27
+ device = torch.device("cpu")
28
+
29
+ print(f"Using device: {device}")
30
+
31
+ # Create input tensor
32
+ x = torch.tensor([1.0, 2.0, 3.0], device=device)
33
+ print(f"Input: {x}")
34
+
35
+ # Run kernel (adds 1 to each element)
36
+ result = kernel.universal(x)
37
+ print(f"Output: {result}")
38
+
39
+ # Verify result
40
+ expected = x + 1.0
41
+ assert torch.allclose(result, expected), "Kernel output doesn't match expected!"
42
+ print("Success!")
flake.lock ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "nodes": {
3
+ "flake-compat": {
4
+ "locked": {
5
+ "lastModified": 1767039857,
6
+ "narHash": "sha256-vNpUSpF5Nuw8xvDLj2KCwwksIbjua2LZCqhV1LNRDns=",
7
+ "owner": "edolstra",
8
+ "repo": "flake-compat",
9
+ "rev": "5edf11c44bc78a0d334f6334cdaf7d60d732daab",
10
+ "type": "github"
11
+ },
12
+ "original": {
13
+ "owner": "edolstra",
14
+ "repo": "flake-compat",
15
+ "type": "github"
16
+ }
17
+ },
18
+ "flake-utils": {
19
+ "inputs": {
20
+ "systems": "systems"
21
+ },
22
+ "locked": {
23
+ "lastModified": 1731533236,
24
+ "narHash": "sha256-l0KFg5HjrsfsO/JpG+r7fRrqm12kzFHyUHqHCVpMMbI=",
25
+ "owner": "numtide",
26
+ "repo": "flake-utils",
27
+ "rev": "11707dc2f618dd54ca8739b309ec4fc024de578b",
28
+ "type": "github"
29
+ },
30
+ "original": {
31
+ "owner": "numtide",
32
+ "repo": "flake-utils",
33
+ "type": "github"
34
+ }
35
+ },
36
+ "kernel-builder": {
37
+ "inputs": {
38
+ "flake-compat": "flake-compat",
39
+ "flake-utils": "flake-utils",
40
+ "nixpkgs": "nixpkgs",
41
+ "rust-overlay": "rust-overlay"
42
+ },
43
+ "locked": {
44
+ "lastModified": 1783738448,
45
+ "narHash": "sha256-vNb0C3LOAPlDPJH2OHbrVND5KrfvNXcl66nQaVp8rRU=",
46
+ "owner": "huggingface",
47
+ "repo": "kernels",
48
+ "rev": "570dcf4a810e6bd9d2433716c2bc6962c6ae46bb",
49
+ "type": "github"
50
+ },
51
+ "original": {
52
+ "owner": "huggingface",
53
+ "repo": "kernels",
54
+ "type": "github"
55
+ }
56
+ },
57
+ "nixpkgs": {
58
+ "locked": {
59
+ "lastModified": 1783284758,
60
+ "narHash": "sha256-tiQ8/qi8I45OOaBBYlVbXoAVkeQzvvTQOv5I45rMw5o=",
61
+ "owner": "NixOS",
62
+ "repo": "nixpkgs",
63
+ "rev": "ec1a11210589d294f0ac99d3290a27e6c73dfa1d",
64
+ "type": "github"
65
+ },
66
+ "original": {
67
+ "owner": "NixOS",
68
+ "repo": "nixpkgs",
69
+ "rev": "ec1a11210589d294f0ac99d3290a27e6c73dfa1d",
70
+ "type": "github"
71
+ }
72
+ },
73
+ "root": {
74
+ "inputs": {
75
+ "kernel-builder": "kernel-builder"
76
+ }
77
+ },
78
+ "rust-overlay": {
79
+ "inputs": {
80
+ "nixpkgs": [
81
+ "kernel-builder",
82
+ "nixpkgs"
83
+ ]
84
+ },
85
+ "locked": {
86
+ "lastModified": 1783320166,
87
+ "narHash": "sha256-l7C/OsjcnWDOk2K3ssj+SBduwL67LashjBqis9+t468=",
88
+ "owner": "oxalica",
89
+ "repo": "rust-overlay",
90
+ "rev": "20ee15370c9256669d66968b89ee20a4b0a4e673",
91
+ "type": "github"
92
+ },
93
+ "original": {
94
+ "owner": "oxalica",
95
+ "repo": "rust-overlay",
96
+ "type": "github"
97
+ }
98
+ },
99
+ "systems": {
100
+ "locked": {
101
+ "lastModified": 1681028828,
102
+ "narHash": "sha256-Vy1rq5AaRuLzOxct8nz4T6wlgyUR7zLU309k9mBC768=",
103
+ "owner": "nix-systems",
104
+ "repo": "default",
105
+ "rev": "da67096a3b9bf56a91d16901293e51ba5b49a27e",
106
+ "type": "github"
107
+ },
108
+ "original": {
109
+ "owner": "nix-systems",
110
+ "repo": "default",
111
+ "type": "github"
112
+ }
113
+ }
114
+ },
115
+ "root": "root",
116
+ "version": 7
117
+ }
flake.nix ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ inputs = {
3
+ kernel-builder.url = "github:huggingface/kernels";
4
+ };
5
+ outputs =
6
+ { self, kernel-builder, ... }:
7
+ kernel-builder.lib.genKernelFlakeOutputs {
8
+ inherit self;
9
+ path = ./.;
10
+ };
11
+ }
tests/__init__.py ADDED
File without changes
tests/test_universal.py ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import platform
2
+
3
+ import torch
4
+
5
+ import universal
6
+
7
+
8
+ def test_universal():
9
+ if platform.system() == "Darwin":
10
+ device = torch.device("mps")
11
+ elif hasattr(torch, "xpu") and torch.xpu.is_available():
12
+ device = torch.device("xpu")
13
+ elif torch.version.cuda is not None and torch.cuda.is_available():
14
+ device = torch.device("cuda")
15
+ else:
16
+ device = torch.device("cpu")
17
+
18
+ x = torch.randn(1024, 1024, dtype=torch.float32, device=device)
19
+ expected = x + 1.0
20
+ result = universal.universal(x)
21
+ torch.testing.assert_close(result, expected)
torch-ext/torch_binding.cpp ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #include <torch/library.h>
2
+
3
+ #include "registration.h"
4
+ #include "torch_binding.h"
5
+
6
+ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
7
+ ops.def("universal(Tensor! out, Tensor input) -> ()");
8
+ #if defined(CPU_KERNEL)
9
+ ops.impl("universal", torch::kCPU, &universal);
10
+ #elif defined(CUDA_KERNEL) || defined(ROCM_KERNEL)
11
+ ops.impl("universal", torch::kCUDA, &universal);
12
+ #elif defined(METAL_KERNEL)
13
+ ops.impl("universal", torch::kMPS, universal);
14
+ #elif defined(XPU_KERNEL)
15
+ ops.impl("universal", torch::kXPU, &universal);
16
+ #endif
17
+ }
18
+
19
+ REGISTER_EXTENSION(TORCH_EXTENSION_NAME)
torch-ext/torch_binding.h ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #include <torch/torch.h>
4
+
5
+ void universal(torch::Tensor &out, torch::Tensor const &input);
torch-ext/universal/__init__.py ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from typing import Optional
2
+
3
+ import torch
4
+
5
+ from ._ops import ops
6
+
7
+
8
+ def universal(x: torch.Tensor, out: Optional[torch.Tensor] = None) -> torch.Tensor:
9
+ if out is None:
10
+ out = torch.empty_like(x)
11
+ ops.universal(out, x)
12
+ return out
universal_cuda/universal.cu ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #include <ATen/cuda/CUDAContext.h>
2
+ #include <c10/cuda/CUDAGuard.h>
3
+ #include <torch/all.h>
4
+
5
+ __global__ void universal_kernel(float *__restrict__ out,
6
+ float const *__restrict__ input, const int n) {
7
+ int idx = blockIdx.x * blockDim.x + threadIdx.x;
8
+ if (idx < n) {
9
+ out[idx] = input[idx] + 1.0f;
10
+ }
11
+ }
12
+
13
+ void universal(torch::Tensor &out, torch::Tensor const &input) {
14
+ TORCH_CHECK(input.device().is_cuda(), "input must be a CUDA tensor");
15
+ TORCH_CHECK(input.is_contiguous(), "input must be contiguous");
16
+ TORCH_CHECK(input.scalar_type() == at::ScalarType::Float,
17
+ "universal only supports float32");
18
+ TORCH_CHECK(input.sizes() == out.sizes(),
19
+ "Tensors must have the same shape");
20
+ TORCH_CHECK(input.scalar_type() == out.scalar_type(),
21
+ "Tensors must have the same dtype");
22
+ TORCH_CHECK(input.device() == out.device(),
23
+ "Tensors must be on the same device");
24
+
25
+ int n = input.numel();
26
+ int threads = 256;
27
+ int blocks = (n + threads - 1) / threads;
28
+
29
+ const at::cuda::OptionalCUDAGuard device_guard(device_of(input));
30
+ const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
31
+ universal_kernel<<<blocks, threads, 0, stream>>>(
32
+ out.data_ptr<float>(), input.data_ptr<float>(), n);
33
+ }