diff --git a/zig/kernels/cuda/contract_controls.cu b/zig/kernels/cuda/contract_controls.cu index d047c7353d..3e977c4c1d 100644 --- a/zig/kernels/cuda/contract_controls.cu +++ b/zig/kernels/cuda/contract_controls.cu @@ -364,6 +364,8 @@ extern "C" __global__ void __launch_bounds__(128) tf_ctl_pocket_in_fmax(const __ const float z = fmaxf(s, 0.0f); act[static_cast(r) * 64 + j] = z * z; } +} + // Gate 3: moss_image.cu's steps, each with one of the publisher's bf16 points dropped; their range checks must go red. namespace tf_ctl_moss { typedef __nv_bfloat16 bf16;