/* Copyright (c) 2015 - 2021 Advanced Micro Devices, Inc. All rights reserved. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ #include "test_common.h" #include #ifdef __linux__ #include #elif defined(_WIN32) #include #endif // standard global variables that can be set on command line size_t N = 4 * 1024 * 1024; char memsetval = 0x42; int memsetD32val = 0xDEADBEEF; short memsetD16val = 0xDEAD; char memsetD8val = 0xDE; int iterations = 1; unsigned blocksPerCU = 6; // to hide latency unsigned threadsPerBlock = 256; int textureFilterMode = 0; // 0: hipFilterModePoint; 1: hipFilterModeLinear int p_gpuDevice = 0; unsigned p_verbose = 0; int p_tests = -1; /*which tests to run. Interpretation is left to each test. default:all*/ int debug_test = 0; #ifdef _WIN64 const char* HIP_VISIBLE_DEVICES_STR = "HIP_VISIBLE_DEVICES="; const char* CUDA_VISIBLE_DEVICES_STR = "CUDA_VISIBLE_DEVICES="; const char* PATH_SEPERATOR_STR = "\\"; const char* NULL_DEVICE = "NUL:"; #else const char* HIP_VISIBLE_DEVICES_STR = "HIP_VISIBLE_DEVICES"; const char* CUDA_VISIBLE_DEVICES_STR = "CUDA_VISIBLE_DEVICES"; const char* PATH_SEPERATOR_STR = "/"; const char* NULL_DEVICE = "/dev/null"; #endif #ifdef _WIN64 // Windows does not have rand_r, use srand and rand instead. int rand_r(unsigned int* s) { srand(*s); return rand(); } #endif // Get Free Memory from the system static size_t getMemoryAmount() { #if __linux__ struct sysinfo info; int _ = sysinfo(&info); return info.freeram / (1024 * 1024); // MB #elif defined(_WIN32) MEMORYSTATUSEX statex; statex.dwLength = sizeof(statex); GlobalMemoryStatusEx(&statex); return (statex.ullAvailPhys / (1024 * 1024)); // MB #endif } size_t getHostThreadCount(const size_t memPerThread, const size_t maxThreads) { if (memPerThread == 0) return 0; auto memAmount = getMemoryAmount(); const auto processor_count = std::thread::hardware_concurrency(); if (processor_count == 0 || memAmount == 0) return 0; size_t thread_count = 0; if ((processor_count * memPerThread) < memAmount) thread_count = processor_count; else thread_count = reinterpret_cast(memAmount / memPerThread); if (maxThreads > 0) { return (thread_count > maxThreads) ? maxThreads : thread_count; } return thread_count; } // Function to determine if the device is of gfx11 architecture bool IsGfx11() { #if defined(__HIP_PLATFORM_NVIDIA__) return false; #elif defined(__HIP_PLATFORM_AMD__) int device = -1; hipDeviceProp_t props{}; HIPCHECK(hipGetDevice(&device)); HIPCHECK(hipGetDeviceProperties(&props, device)); // Get GCN Arch Name and compare to check if it is gfx11 std::string arch = std::string(props.gcnArchName); auto pos = arch.find(":"); if (pos != std::string::npos) arch = arch.substr(0, pos); if(arch.size() >= 5) arch = arch.substr(0,5); return (arch == std::string("gfx11")) ? true : false; #else std::cout<<"Have to be either Nvidia or AMD platform, asserting"<= argc || !HipTest::parseSize(argv[i], &N)) { failed("Bad N size argument"); } } else if (!strcmp(arg, "--threadsPerBlock")) { if (++i >= argc || !HipTest::parseUInt(argv[i], &threadsPerBlock)) { failed("Bad threadsPerBlock argument"); } } else if (!strcmp(arg, "--blocksPerCU")) { if (++i >= argc || !HipTest::parseUInt(argv[i], &blocksPerCU)) { failed("Bad blocksPerCU argument"); } } else if (!strcmp(arg, "--memsetval")) { int ex; if (++i >= argc || !HipTest::parseInt(argv[i], &ex)) { failed("Bad memsetval argument"); } memsetval = ex; } else if (!strcmp(arg, "--memsetD32val")) { int ex; if (++i >= argc || !HipTest::parseInt(argv[i], &ex)) { failed("Bad memsetD32val argument"); } memsetD32val = ex; } else if (!strcmp(arg, "--memsetD16val")) { int ex; if (++i >= argc || !HipTest::parseInt(argv[i], &ex)) { failed("Bad memsetD16val argument"); } memsetD16val = ex; } else if (!strcmp(arg, "--memsetD8val")) { int ex; if (++i >= argc || !HipTest::parseInt(argv[i], &ex)) { failed("Bad memsetD8val argument"); } memsetD8val = ex; } else if (!strcmp(arg, "--textureFilterMode")) { int mode; if (++i >= argc || !HipTest::parseInt(argv[i], &mode)) { failed("Bad textureFilterMode argument"); } textureFilterMode = mode; } else if (!strcmp(arg, "--iterations") || (!strcmp(arg, "-i"))) { if (++i >= argc || !HipTest::parseInt(argv[i], &iterations)) { failed("Bad iterations argument"); } } else if (!strcmp(arg, "--gpu") || (!strcmp(arg, "-gpuDevice")) || (!strcmp(arg, "-g"))) { if (++i >= argc || !HipTest::parseInt(argv[i], &p_gpuDevice)) { failed("Bad gpuDevice argument"); } } else if (!strcmp(arg, "--verbose") || (!strcmp(arg, "-v"))) { if (++i >= argc || !HipTest::parseUInt(argv[i], &p_verbose)) { failed("Bad verbose argument"); } } else if (!strcmp(arg, "--tests") || (!strcmp(arg, "-t"))) { if (++i >= argc || !HipTest::parseInt(argv[i], &p_tests)) { failed("Bad tests argument"); } } else if (!strcmp(arg, "--debug") || (!strcmp(arg, "-d"))) { if (++i >= argc || !HipTest::parseInt(argv[i], &debug_test)) { failed("Bad tests argument"); } } else { if (failOnUndefinedArg) { failed("Bad argument '%s'", arg); } else { argv[extraArgs++] = argv[i]; } } }; return extraArgs; } unsigned setNumBlocks(unsigned blocksPerCU, unsigned threadsPerBlock, size_t N) { int device; HIPCHECK(hipGetDevice(&device)); hipDeviceProp_t props; HIPCHECK(hipGetDeviceProperties(&props, device)); unsigned blocks = props.multiProcessorCount * blocksPerCU; if (blocks * threadsPerBlock > N) { blocks = (N + threadsPerBlock - 1) / threadsPerBlock; } return blocks; } } // namespace HipTest