#TileLang-Sunrise

TileLang-Sunrise is the Sunrise S2/TANG backend distribution of Tile Language (tile-lang), a concise domain-specific language designed to streamline the development of high-performance GPU/CPU/accelerator kernels (e.g., GEMM, Dequant GEMM, FlashAttention, LinearAttention). By employing a Pythonic syntax with an underlying compiler infrastructure on top of TVM, TileLang-Sunrise allows developers to focus on productivity without sacrificing the low-level optimizations necessary for state-of-the-art performance.

#Upstream TileLang Latest News

02/02/2026 🧩: Check out TileLang Puzzles, a fun and interactive way to learn TileLang programming with 10 progressively harder puzzles!
12/18/2025 🚀: Added CuTeDSL backend support, enabling compilation to NVIDIA CUTLASS CuTe DSL! Join us in building and optimizing this exciting new backend: Issue #1454.
12/17/2025 🔬: Integrated Z3 theorem prover into TVM Arith Analyzer, bringing SMT-based symbolic reasoning for enhanced optimizations and automatic correctness verification!
10/31/2025 🔧: Migrated to apache-tvm-ffi, significantly reducing CPU overhead!
10/30/2025 📦: We have released v0.1.6.post2, which is the last version compatible with Python 3.8.
10/07/2025 🍎: Added Apple Metal Device support, check out Pull Request #799 for details.
09/29/2025 🎉: Thrilled to announce that AscendC and AscendNPU IR backends targeting Huawei Ascend chips are now supported! Check out the preview here: 🔗 link. This includes implementations across two branches: ascendc_pto and npuir. Feel free to explore and share your feedback!
07/04/2025 🚀: Introduced T.gemm_sp for 2:4 sparse tensor core support, check out Pull Request #526 for details.
06/05/2025 ✨: Added NVRTC Backend to significantly reduce compilation time for cute templates!
04/14/2025 🚀: Added high-performance FlashMLA implementation for AMD MI300X, achieving performance parity with hand-optimized assembly kernels of Aiter! See example_mla_amd for details.
03/03/2025 🚀: Added high-performance MLA Decoding support using only 80 lines of Python code, achieving performance on par with FlashMLA on H100 (see example_mla_decode.py)! We also provide documentation explaining how TileLang achieves this.
02/15/2025 ✨: Added WebGPU Codegen support, see Pull Request #86!
02/12/2025 ✨: Excited to announce the release of v0.1.0!
02/10/2025 🚀: Added debug tools for TileLang—T.print for printing variables/buffers (docs) and a memory layout plotter (examples/plot_layout).
01/20/2025 ✨: We are excited to announce that tile-lang, a dsl for high performance AI workloads, is now open source and available to the public!

#Tested Devices

TileLang-Sunrise supports Sunrise S2 accelerators through the TANG backend.

#OP Implementation Examples

TileLang-Sunrise provides the building blocks to implement a wide variety of operators. Some examples include:

Within the examples directory, you will also find additional complex kernels—such as convolutions, forward/backward passes for FlashAttention, more operators will continuously be added.

#Installation

#Environment Dependencies

The Sunrise backend requires a compatible TANG Runtime, torch_ptpu 0.2.3+torch 2.10, and matching Triton packages. Configure the following paths for your local environment before installation:

TANGRT_LIB_PATH="/usr/local/tangrt/lib/linux-x86_64:/usr/lib64"
export LD_LIBRARY_PATH=${TANGRT_LIB_PATH}:$LD_LIBRARY_PATH
export STPU_TANGRT_PATH=/usr/local/tangrt

# Required by torch_ptpu; replace these paths for your local environment.
export PYTHON_INCLUDE_DIR=/path/to/python/include
export VENDOR_INCLUDE_DIRS=/usr/local/tangrt/include
export PTPU_PATH=/path/to/python/site-packages/torch_ptpu

#Build from Source

Install the required system dependencies, then install TileLang-Sunrise locally:

# install required system dependencies
sudo apt-get update
sudo apt-get install -y python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev

pip install -e . -v # remove -e option if you don't want to install in editable mode, -v for verbose output

We currently provide three ways to install TileLang-Sunrise from source:

#Quick Start

In this section, you'll learn how to write and verify a straightforward GEMM (matrix multiplication) kernel using TileLang-Sunrise on Sunrise S2.

#GEMM Example

The following example defines a GEMM kernel and validates its output against PyTorch.

import tilelang
import tilelang.language as T
import torch


@tilelang.jit(out_idx=[-1], target="tang")
def matmul(M, N, K):
    block_M = block_N = 64
    block_K = 32

    @T.prim_func
    def gemm(
        A: T.Tensor((M, K), "float16"),
        B: T.Tensor((K, N), "float16"),
        C: T.Tensor((M, N), "float16"),
    ):
        with T.Kernel(
            T.ceildiv(N, block_N),
            T.ceildiv(M, block_M),
            threads=128,
        ) as (bx, by):
            A_shared = T.alloc_shared((block_M, block_K), "float16")
            B_shared = T.alloc_shared((block_K, block_N), "float16")
            C_local = T.alloc_fragment((block_M, block_N), "float")
            T.clear(C_local)

            for ko in T.Pipelined(T.ceildiv(K, block_K), num_stages=2):
                T.copy(A[by * block_M, ko * block_K], A_shared)
                T.copy(B[ko * block_K, bx * block_N], B_shared)
                T.gemm(A_shared, B_shared, C_local)

            T.copy(C_local, C[by * block_M, bx * block_N])

    return gemm


M = N = K = 1024
kernel = matmul(M, N, K)
A = torch.randn((M, K), device="ptpu", dtype=torch.float16)
B = torch.randn((K, N), device="ptpu", dtype=torch.float16)
C = kernel(A, B)
expected = torch.matmul(A.cpu(), B.cpu())
torch.testing.assert_close(C.cpu(), expected, atol=1e-2, rtol=1e-2)
print("GEMM correctness check passed.")

#Dive Deep into TileLang-Sunrise Beyond GEMM

In addition to GEMM, we provide a variety of examples to showcase the versatility and power of TileLang-Sunrise, including:

Dequantize GEMM: Achieve high-performance dequantization by fine-grained control over per-thread operations, with many features now adopted as default behaviors in BitBLAS, which utilizing magic layout transformation and intrins to accelerate dequantize gemm.
FlashAttention: Enable cross-operator fusion with simple and intuitive syntax, and we also provide an example of auto tuning.
LinearAttention: Examples include RetNet and Mamba implementations.
Convolution: Implementations of Convolution with IM2Col.

#Upcoming Features

Check the upstream TileLang v0.2.0 release plan for upcoming features.

Upstream TileLang has now been used in project BitBLAS and AttentionEngine.

#Join the Discussion

Welcome to join the upstream TileLang Discord community for discussions, support, and collaboration!

#Acknowledgments

We would like to express our gratitude to the TVM community for their invaluable contributions. The initial version of this project was mainly developed by LeiWang1999, chengyupku and nox-410 with supervision from Prof. Zhi Yang at Peking University. Part of this work was carried out during an internship at Microsoft Research, where Dr. Lingxiao Ma, Dr. Yuqing Xia, Dr. Jilong Xue, and Dr. Fan Yang offered valuable advice and support. We deeply appreciate their mentorship and contributions.

Instalar no iPhone/iPad

Tilelang Sunrise