| 1 | //===--------- Misc.cpp - OpenMP device misc interfaces ----------- C++ -*-===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // |
| 10 | //===----------------------------------------------------------------------===// |
| 11 | |
| 12 | #include "Allocator.h" |
| 13 | #include "Configuration.h" |
| 14 | #include "DeviceTypes.h" |
| 15 | #include "Shared/RPCOpcodes.h" |
| 16 | #include "shared/rpc.h" |
| 17 | |
| 18 | #include "Debug.h" |
| 19 | |
| 20 | namespace ompx { |
| 21 | namespace impl { |
| 22 | |
| 23 | /// Lookup a device-side function using a host pointer /p HstPtr using the table |
| 24 | /// provided by the device plugin. The table is an ordered pair of host and |
| 25 | /// device pointers sorted on the value of the host pointer. |
| 26 | void *indirectCallLookup(void *HstPtr) { |
| 27 | if (!HstPtr) |
| 28 | return nullptr; |
| 29 | |
| 30 | struct IndirectCallTable { |
| 31 | void *HstPtr; |
| 32 | void *DevPtr; |
| 33 | }; |
| 34 | IndirectCallTable *Table = |
| 35 | reinterpret_cast<IndirectCallTable *>(config::getIndirectCallTablePtr()); |
| 36 | uint64_t TableSize = config::getIndirectCallTableSize(); |
| 37 | |
| 38 | // If the table is empty we assume this is device pointer. |
| 39 | if (!Table || !TableSize) |
| 40 | return HstPtr; |
| 41 | |
| 42 | uint32_t Left = 0; |
| 43 | uint32_t Right = TableSize; |
| 44 | |
| 45 | // If the pointer is definitely not contained in the table we exit early. |
| 46 | if (HstPtr < Table[Left].HstPtr || HstPtr > Table[Right - 1].HstPtr) |
| 47 | return HstPtr; |
| 48 | |
| 49 | while (Left != Right) { |
| 50 | uint32_t Current = Left + (Right - Left) / 2; |
| 51 | if (Table[Current].HstPtr == HstPtr) |
| 52 | return Table[Current].DevPtr; |
| 53 | |
| 54 | if (HstPtr < Table[Current].HstPtr) |
| 55 | Right = Current; |
| 56 | else |
| 57 | Left = Current; |
| 58 | } |
| 59 | |
| 60 | // If we searched the whole table and found nothing this is a device pointer. |
| 61 | return HstPtr; |
| 62 | } |
| 63 | |
| 64 | /// The openmp client instance used to communicate with the server. |
| 65 | [[gnu::visibility("protected" ), |
| 66 | gnu::weak]] rpc::Client Client asm("__llvm_rpc_client" ); |
| 67 | |
| 68 | } // namespace impl |
| 69 | } // namespace ompx |
| 70 | |
| 71 | /// Interfaces |
| 72 | /// |
| 73 | ///{ |
| 74 | |
| 75 | extern "C" { |
| 76 | int32_t __kmpc_cancellationpoint(IdentTy *, int32_t, int32_t) { return 0; } |
| 77 | |
| 78 | int32_t __kmpc_cancel(IdentTy *, int32_t, int32_t) { return 0; } |
| 79 | |
| 80 | double omp_get_wtick(void) { |
| 81 | // The number of ticks per second for the AMDGPU clock varies by card and can |
| 82 | // only be retrieved by querying the driver. We rely on the device environment |
| 83 | // to inform us what the proper frequency is. NVPTX uses a nanosecond |
| 84 | // resolution, we could omit the global read but this makes it consistent. |
| 85 | return 1.0 / ompx::config::getClockFrequency(); |
| 86 | } |
| 87 | |
| 88 | double omp_get_wtime(void) { |
| 89 | return static_cast<double>(__builtin_readsteadycounter()) * omp_get_wtick(); |
| 90 | } |
| 91 | |
| 92 | void *__llvm_omp_indirect_call_lookup(void *HstPtr) { |
| 93 | return ompx::impl::indirectCallLookup(HstPtr); |
| 94 | } |
| 95 | |
| 96 | void *omp_alloc(size_t size, omp_allocator_handle_t allocator) { |
| 97 | switch (allocator) { |
| 98 | case omp_default_mem_alloc: |
| 99 | case omp_large_cap_mem_alloc: |
| 100 | case omp_const_mem_alloc: |
| 101 | case omp_high_bw_mem_alloc: |
| 102 | case omp_low_lat_mem_alloc: |
| 103 | return malloc(size: size); |
| 104 | default: |
| 105 | return nullptr; |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | void omp_free(void *ptr, omp_allocator_handle_t allocator) { |
| 110 | switch (allocator) { |
| 111 | case omp_default_mem_alloc: |
| 112 | case omp_large_cap_mem_alloc: |
| 113 | case omp_const_mem_alloc: |
| 114 | case omp_high_bw_mem_alloc: |
| 115 | case omp_low_lat_mem_alloc: |
| 116 | free(ptr: ptr); |
| 117 | case omp_null_allocator: |
| 118 | default: |
| 119 | return; |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | unsigned long long __llvm_omp_host_call(void *fn, void *data, size_t size) { |
| 124 | rpc::Client::Port Port = ompx::impl::Client.open<OFFLOAD_HOST_CALL>(); |
| 125 | Port.send_n(data, size); |
| 126 | Port.send([=](rpc::Buffer *buffer, uint32_t) { |
| 127 | buffer->data[0] = reinterpret_cast<uintptr_t>(fn); |
| 128 | }); |
| 129 | unsigned long long Ret; |
| 130 | Port.recv([&](rpc::Buffer *Buffer, uint32_t) { |
| 131 | Ret = static_cast<unsigned long long>(Buffer->data[0]); |
| 132 | }); |
| 133 | Port.close(); |
| 134 | return Ret; |
| 135 | } |
| 136 | } |
| 137 | |
| 138 | ///} |
| 139 | |