portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

HIP improvements

Browse Source
This commit is contained in:
Peter Boyle 2020-09-16 00:32:27 +01:00
parent 288c615782
commit 4677c40195
1 changed files with 8 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
Grid/threads/Accelerator.cc
View File

@ -55,6 +55,7 @@ void acceleratorInit(void)
printf("AcceleratorCudaInit[%d]: ========================\n",rank); printf("AcceleratorCudaInit[%d]: ========================\n",rank);
printf("AcceleratorCudaInit[%d]: Device identifier: %s\n",rank, prop.name); printf("AcceleratorCudaInit[%d]: Device identifier: %s\n",rank, prop.name);
GPU_PROP_FMT(totalGlobalMem,"%lld"); GPU_PROP_FMT(totalGlobalMem,"%lld");
GPU_PROP(managedMemory); GPU_PROP(managedMemory);
GPU_PROP(isMultiGpuBoard); GPU_PROP(isMultiGpuBoard);
@ -109,20 +110,24 @@ void acceleratorInit(void)
if ((localRankStr = getenv(ENV_RANK_OMPI )) != NULL) { world_rank = atoi(localRankStr);} if ((localRankStr = getenv(ENV_RANK_OMPI )) != NULL) { world_rank = atoi(localRankStr);}
if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);} if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
printf("world_rank %d has %d devices\n",world_rank,nDevices);
size_t totalDeviceMem=0;
for (int i = 0; i < nDevices; i++) { for (int i = 0; i < nDevices; i++) {
#define GPU_PROP_FMT(canMapHostMemory,FMT) printf("AcceleratorHipInit: " #canMapHostMemory ": " FMT" \n",prop.canMapHostMemory); #define GPU_PROP_FMT(canMapHostMemory,FMT) printf("AcceleratorHipInit: " #canMapHostMemory ": " FMT" \n",prop.canMapHostMemory);
#define GPU_PROP(canMapHostMemory) GPU_PROP_FMT(canMapHostMemory,"%d"); #define GPU_PROP(canMapHostMemory) GPU_PROP_FMT(canMapHostMemory,"%d");
hipGetDeviceProperties(&gpu_props[i], i); hipGetDeviceProperties(&gpu_props[i], i);
hipDeviceProp_t prop;
prop = gpu_props[i];
totalDeviceMem = prop.totalGlobalMem;
if ( world_rank == 0) { if ( world_rank == 0) {
hipDeviceProp_t prop;
prop = gpu_props[i];
printf("AcceleratorHipInit: ========================\n"); printf("AcceleratorHipInit: ========================\n");
printf("AcceleratorHipInit: Device Number : %d\n", i); printf("AcceleratorHipInit: Device Number : %d\n", i);
printf("AcceleratorHipInit: ========================\n"); printf("AcceleratorHipInit: ========================\n");
printf("AcceleratorHipInit: Device identifier: %s\n", prop.name); printf("AcceleratorHipInit: Device identifier: %s\n", prop.name);
GPU_PROP_FMT(totalGlobalMem,"%lld");
// GPU_PROP(managedMemory); // GPU_PROP(managedMemory);
GPU_PROP(isMultiGpuBoard); GPU_PROP(isMultiGpuBoard);
GPU_PROP(warpSize); GPU_PROP(warpSize);
@ -131,6 +136,7 @@ void acceleratorInit(void)
// GPU_PROP(singleToDoublePrecisionPerfRatio); // GPU_PROP(singleToDoublePrecisionPerfRatio);
} }
} }
MemoryManager::DeviceMaxBytes = (8*totalDeviceMem)/10; // Assume 80% ours
#undef GPU_PROP_FMT #undef GPU_PROP_FMT
#undef GPU_PROP #undef GPU_PROP
#ifdef GRID_IBM_SUMMIT #ifdef GRID_IBM_SUMMIT