tvm_tirx.backend.cuda.op.cta_max

tvm_tirx.backend.cuda.op.cta_max(value, num_warps, scratch)

Convenience wrapper: cta_reduce(value, "max", num_warps, scratch).