Fix undefined THREADS_PER_WARP in ln_fwd_kernels.cuh

[Copilot]

The forward layer norm kernel was using THREADS_PER_WARP without declaring it as a local enum, causing build failures after clang-format was applied.

Changes

• Added missing enum { THREADS_PER_WARP = Ktraits::THREADS_PER_WARP }; declaration in ln_fwd_kernel to match the pattern used in ln_bwd_kernel

The constant was referenced at lines 36, 37, and 42 but never brought into local scope from the Ktraits template parameter.

template <typename Ktraits>
__global__ __launch_bounds__(Ktraits::THREADS_PER_CTA) void ln_fwd_kernel(FwdParams params) {
  enum { ROWS_PER_CTA = Ktraits::ROWS_PER_CTA };
  // ... other enums ...
  enum { THREADS_PER_WARP = Ktraits::THREADS_PER_WARP };  // Added
  enum { CTAS_PER_ROW = Ktraits::CTAS_PER_ROW };

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.