Enable swizzling SMEM for transposed dot operand

ROCm / triton

Development repository for the Triton language and compiler

MIT License

89 stars 27 forks source link

Enable swizzling SMEM for transposed dot operand #474

Closed htyu closed 8 months ago

htyu commented 8 months ago

Transposed operand will be accessed in an opposite order from the original operand. Enabling swizzling seems to help performance. I'm seeing 10% performance improvement for our internal model.