diff --git a/mlx/backend/metal/kernels/fp_quantized_nax.h b/mlx/backend/metal/kernels/fp_quantized_nax.h index 7da4dd02..28c4ec82 100644 --- a/mlx/backend/metal/kernels/fp_quantized_nax.h +++ b/mlx/backend/metal/kernels/fp_quantized_nax.h @@ -375,11 +375,11 @@ METAL_FUNC void fp_qmm_n_impl( static_assert(BK % SIMD_SIZE == 0, "BK should be divisible by SIMD_SIZE"); (void)lid; + (void)M; constexpr int pack_factor = get_pack_factor<8>(); constexpr int bytes_per_pack = get_bytes_per_pack(); - constexpr int BK_padded = (BK + 16 / sizeof(T)); constexpr int BN_padded = (BN + 16 / sizeof(T)); using loader_w_t = QuantizedBlockLoader<