GPU: allow standalone benchmark to work on macOS with Metal

Author: ktf

GPU/GPUTracking/CMakeLists.txt

@@ -513,9 +513,7 @@ if(CPUDEVICE_ENABLED)
 endif()
 
 if(ALIGPU_BUILD_TYPE STREQUAL "O2" OR ALIGPU_BUILD_TYPE STREQUAL "Standalone")
-  if(NOT ${CMAKE_SYSTEM} MATCHES Darwin)
-    add_subdirectory(Standalone/Benchmark)
-  endif()
+  add_subdirectory(Standalone/Benchmark)
 endif()
 
 if(GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_GPU})

GPU/GPUTracking/Definitions/GPUSettingsList.h

@@ -348,8 +348,8 @@ AddOption(keepAllMemory, bool, false, "", 0, "Allocate all memory on both device
 AddOption(keepDisplayMemory, bool, false, "", 0, "Like keepAllMemory, but only for memory required for event display")
 AddOption(disableMemoryReuse, bool, false, "", 0, "Disable memory reusage (for debugging only)")
 AddOption(memoryAllocationStrategy, int8_t, 0, "", 0, "Memory Allocation Strategy (0 = auto, 1 = individual allocations, 2 = single global allocation)")
-AddOption(forceMemoryPoolSize, uint64_t, 1, "memSize", 0, "Force size of allocated GPU / page locked host memory (0 = disable, 1 / 2 = auto, other values set the size of the memory pool)", min(0ul))
-AddOption(forceHostMemoryPoolSize, uint64_t, 0, "hostMemSize", 0, "Force size of allocated host page locked host memory (overriding memSize)", min(0ul))
+AddOption(forceMemoryPoolSize, uint64_t, 1, "memSize", 0, "Force size of allocated GPU / page locked host memory (0 = disable, 1 / 2 = auto, other values set the size of the memory pool)", min(uint64_t(0)))
+AddOption(forceHostMemoryPoolSize, uint64_t, 0, "hostMemSize", 0, "Force size of allocated host page locked host memory (overriding memSize)", min(uint64_t(0)))
 AddOption(memoryScalingFactor, float, 1.f, "", 0, "Factor to apply to all memory scalers")
 AddOption(memoryScalingFuzz, uint64_t, 0, "", 0, "Fuzz the memoryScalingFactor (0 disable, 1 enable, >1 set seed", def(1))
 AddOption(tpcInputWithClusterRejection, uint8_t, 0, "", 0, "Indicate whether the TPC input is CTF data with cluster rejection, to tune buffer estimations")

GPU/GPUTracking/Definitions/GPUSettingsListStandalone.h

@@ -47,7 +47,7 @@ EndConfig()
 BeginConfig(GPUSettingsStandalone, configStandalone)
 AddOption(runGPU, uint8_t, 1, "", 'g', "Use GPU for processing", message("GPU processing enabled"), set(2))
 AddOptionSet(runGPU, uint8_t, 0, "", 'c', "Use CPU for processing", message("CPU enabled"))
-AddOption(gpuType, std::string, "AUTO", "", 0, "GPU type (CUDA / HIP / OCL / CPUDEVICE) or CPU or AUTO")
+AddOption(gpuType, std::string, "AUTO", "", 0, "GPU type (CUDA / HIP / OCL / METAL / CPUDEVICE) or CPU or AUTO")
 AddOption(runGPUforce, bool, true, "", 0, "Force usage of the specified GPU device type, no CPU fallback")
 AddOption(promptOnExit, bool, false, "", 0, "Prompt for keypress before exiting")
 AddOption(continueOnError, bool, false, "", 0, "Continue processing after an error")
@@ -64,8 +64,8 @@ AddOption(eventDisplay, int32_t, 0, "display", 'd', "Show standalone event displ
 AddOption(eventGenerator, bool, false, "", 0, "Run event generator")
 AddOption(cont, bool, false, "", 0, "Process continuous timeframe data, even if input is triggered")
 AddOption(setMaxTimeBin, int32_t, -2, "", 0, "maximum time bin of continuous data, 0 for triggered events, -1 for automatic continuous mode, -2 for automatic continuous / triggered")
-AddOption(outputcontrolmem, uint64_t, 0, "outputMemory", 0, "Use predefined output buffer of this size", min(0ul), message("Using %s bytes as output memory"))
-AddOption(inputcontrolmem, uint64_t, 0, "inputMemory", 0, "Use predefined input buffer of this size", min(0ul), message("Using %s bytes as input memory"))
+AddOption(outputcontrolmem, uint64_t, 0, "outputMemory", 0, "Use predefined output buffer of this size", min(uint64_t(0)), message("Using %s bytes as output memory"))
+AddOption(inputcontrolmem, uint64_t, 0, "inputMemory", 0, "Use predefined input buffer of this size", min(uint64_t(0)), message("Using %s bytes as input memory"))
 AddOption(cpuAffinity, int32_t, -1, "", 0, "Pin CPU affinity to this CPU core", min(-1))
 AddOption(fifoScheduler, bool, false, "", 0, "Use FIFO realtime scheduler", message("Setting FIFO scheduler: %s"))
 AddOption(fpe, int8_t, -1, "", 0, "Trap on floating point exceptions (-1 = if no ffast-math)")

GPU/GPUTracking/Standalone/Benchmark/standalone.cxx

@@ -124,6 +124,7 @@ int32_t ReadConfiguration(int argc, char** argv)
   setlocale(LC_ALL, "en_US.utf-8");
   setlocale(LC_NUMERIC, "en_US.utf-8");
   if (configStandalone.cpuAffinity != -1) {
+#ifdef __linux__
     cpu_set_t mask;
     CPU_ZERO(&mask);
     CPU_SET(configStandalone.cpuAffinity, &mask);
@@ -133,8 +134,13 @@ int32_t ReadConfiguration(int argc, char** argv)
       printf("Error setting CPU affinity\n");
       return 1;
     }
+#else
+    printf("CPU affinity not supported on this platform\n");
+    return 1;
+#endif
   }
   if (configStandalone.fifoScheduler) {
+#ifdef __linux__
     printf("Setting FIFO scheduler\n");
     sched_param param;
     sched_getparam(0, &param);
@@ -143,13 +149,21 @@ int32_t ReadConfiguration(int argc, char** argv)
       printf("Error setting scheduler\n");
       return 1;
     }
+#else
+    printf("FIFO scheduler not supported on this platform\n");
+    return 1;
+#endif
   }
 #ifdef __FAST_MATH__
   if (configStandalone.fpe == 1) {
 #else
   if (configStandalone.fpe) {
 #endif
+#ifdef __linux__
     feenableexcept(FE_INVALID | FE_DIVBYZERO | FE_OVERFLOW);
+#else
+    printf("Trapping floating point exceptions not supported on this platform\n");
+#endif
   }
   bool detMode = false, noFTZMode = false;
 #ifdef GPUCA_DETERMINISTIC_MODE
@@ -1021,7 +1035,9 @@ int32_t main(int argc, char** argv)
 
 #ifndef _WIN32
   if (configStandalone.proc.runQA && configStandalone.fpe) {
+#ifdef __linux__
     fedisableexcept(FE_INVALID | FE_DIVBYZERO | FE_OVERFLOW);
+#endif
   }
 #endif