tirx-kernels 0.1.1

High-performance GPU kernels (GEMM + FlashAttention4) written in TIRx

tirx-kernels

High-performance GPU kernels written in TIRx.

Kernels

Name	dtype	Arch
fp16_bf16_gemm	fp16 / bf16	sm_100a
fp8_blockwise_gemm	fp8 (blockwise)	sm_100a
nvfp4_gemm	nvfp4	sm_100a
flash_attention4	bf16	sm_100a

Installation

pip install tirx-kernels          # from a release
# or, from a checkout:
pip install -e .

External dependencies

These are not on PyPI and must be installed/available separately. They are imported lazily, so import tirx_kernels and kernel discovery work without them — they are only needed to actually compile/run a kernel:

Dependency	Needed by	Notes
tvm.tirx	all kernels (compile + run)	The TIRx compiler. Put it on PYTHONPATH, e.g. /path/to/tir/python.
torch	all kernels	CUDA build matching your GPU.
deep_gemm	fp8_blockwise_gemm data/baseline	Used for fp8 quantization helpers and the reference.
flashinfer	nvfp4_gemm data/baseline	Used for nvfp4 quantization and the baseline.

Usage

Command line

# List discovered kernels (with their config labels)
python -m tirx_kernels.registry --format json

# Run correctness tests (optionally filter by kernel / config label)
python -m tirx_kernels.test
python -m tirx_kernels.test --kernel fp16_bf16_gemm
python -m tirx_kernels.test --kernel fp16_bf16_gemm --config bf16_1024x1024x1024

# Benchmark
python -m tirx_kernels.bench --kernel nvfp4_gemm

Programmatic API

Every kernel module exposes a small, uniform interface (see tirx_kernels/_protocol.py):

from tirx_kernels.registry import discover_kernels

kernels = discover_kernels()          # {name: module}
mod = kernels["fp16_bf16_gemm"]

mod.run_test(M=1024, N=1024, K=1024)  # compile + run + correctness check
mod.run_bench(M=1024, N=1024, K=1024) # profile (needs a GPU)

func = mod.get_kernel(M=1024, N=1024, K=1024)  # the TIRx PrimFunc

Each module also provides KERNEL_META (name / category / compute_capability) and CONFIGS (the test/bench parameter sweeps) that the registry and CLI use.

License

Apache License 2.0. See LICENSE.