Kernel

CUDA

SGEMM

✅ Naive

✅ Thread Tiling

✅ Thread Tiling Bank Free

SGEMV

✅ Naive

✅ Warp Reduce

Reduce

Transpose

Sort

✅ MergeSort

Softmax

✅ Naive

✅ WarpReduce

Triton

HGEMM

✅ Block Tiling

GEMV

Reduce

Transpose

Sort

Softmax

Build

python3 ./script.py {kernelName}

Dependence

• NVIDIA GPU
• OpenAI Triton >= 2.0