| 88 | using TmaDescriptor = CUtensorMap; |
| 89 | |
| 90 | HOST_DEVICE |
| 91 | void |
| 92 | prefetch_tma_descriptor(TmaDescriptor const* desc_ptr) |
| 93 | { |
| 94 | uint64_t gmem_int_desc = reinterpret_cast<uint64_t>(desc_ptr); |
| 95 | // Prefetch TMA Descriptor using generic addressing (i.e. no specific state space: const or param) |
| 96 | asm volatile ( |
| 97 | "prefetch.tensormap [%0];" |
| 98 | : |
| 99 | : "l"(gmem_int_desc) |
| 100 | : "memory"); |
| 101 | } |
nothing calls this directly
no outgoing calls
no test coverage detected