/**
* This method frees specific chunk of memory, described by AllocationPoint passed in
*
* @param point
*/
@Override
public void free(AllocationPoint point) {
switch (point.getAllocationStatus()) {
case HOST:
{
// cudaFreeHost call here
// FIXME: it would be nice to get rid of typecasting here
long reqMem = AllocationUtils.getRequiredMemory(point.getShape());
// log.info("Deallocating {} bytes on [HOST]", reqMem);
NativeOps nativeOps = NativeOpsHolder.getInstance().getDeviceNativeOps();
long result = nativeOps.freeHost(point.getPointers().getHostPointer().address());
// JCuda.cudaFreeHost(new Pointer(point.getPointers().getHostPointer().address()));
if (result == 0) throw new RuntimeException("Can't deallocate [HOST] memory...");
}
break;
case DEVICE:
{
// cudaFree call
// JCuda.cudaFree(new Pointer(point.getPointers().getDevicePointer().address()));
long reqMem = AllocationUtils.getRequiredMemory(point.getShape());
// log.info("Deallocating {} bytes on [DEVICE]", reqMem);
NativeOps nativeOps = NativeOpsHolder.getInstance().getDeviceNativeOps();
long result = nativeOps.freeDevice(point.getPointers().getDevicePointer().address(), 0);
if (result == 0) throw new RuntimeException("Can't deallocate [DEVICE] memory...");
}
break;
default:
throw new IllegalStateException(
"Can't free memory on target [" + point.getAllocationStatus() + "]");
}
}
/** @author [email protected] */ public class CudaDirectProvider implements MemoryProvider { protected static final long DEVICE_RESERVED_SPACE = 1024 * 1024 * 50L; private static Logger log = LoggerFactory.getLogger(CudaDirectProvider.class); protected NativeOps nativeOps = NativeOpsHolder.getInstance().getDeviceNativeOps(); private AtomicLong emergencyCounter = new AtomicLong(0); /** * This method provides PointersPair to memory chunk specified by AllocationShape * * @param shape shape of desired memory chunk * @param point target AllocationPoint structure * @param location either HOST or DEVICE * @return */ @Override public PointersPair malloc( AllocationShape shape, AllocationPoint point, AllocationStatus location) { switch (location) { case HOST: { Pointer devicePointer = new Pointer(); long reqMem = AllocationUtils.getRequiredMemory(shape); // log.info("Allocating {} bytes on [HOST]", reqMem); // FIXME: this is WRONG, and directly leads to memleak if (reqMem < 1) reqMem = 1; long pointer = nativeOps.mallocHost(reqMem, 0); if (pointer == 0) throw new RuntimeException("Can't allocate [HOST] memory: " + reqMem); Pointer hostPointer = new CudaPointer(pointer); /* JCuda.cudaHostGetDevicePointer( devicePointer, hostPointer, 0); */ PointersPair devicePointerInfo = new PointersPair(); devicePointerInfo.setDevicePointer(new CudaPointer(hostPointer, reqMem)); devicePointerInfo.setHostPointer(new CudaPointer(hostPointer, reqMem)); point.setPointers(devicePointerInfo); point.setAllocationStatus(AllocationStatus.HOST); return devicePointerInfo; } case DEVICE: { // cudaMalloc call long reqMem = AllocationUtils.getRequiredMemory(shape); // log.info("Allocating {} bytes on [DEVICE]", reqMem); // FIXME: this is WRONG, and directly leads to memleak if (reqMem < 1) reqMem = 1; /* if (reqMem == 65536 || reqMem == 1048576 || reqMem == 262144) emergencyCounter.incrementAndGet(); if (emergencyCounter.get() > 2000) throw new RuntimeException("PEW"); */ // FIXME: it would be nice to get rid of typecasting here long pointer = nativeOps.mallocDevice(reqMem, 0, 0); if (pointer == 0) return null; // throw new RuntimeException("Can't allocate [DEVICE] memory!"); Pointer devicePointer = new CudaPointer(pointer); PointersPair devicePointerInfo = point.getPointers(); if (devicePointerInfo == null) devicePointerInfo = new PointersPair(); devicePointerInfo.setDevicePointer(new CudaPointer(devicePointer, reqMem)); point.setAllocationStatus(AllocationStatus.DEVICE); return devicePointerInfo; } default: throw new IllegalStateException("Unsupported location for malloc: [" + location + "]"); } } /** * This method frees specific chunk of memory, described by AllocationPoint passed in * * @param point */ @Override public void free(AllocationPoint point) { switch (point.getAllocationStatus()) { case HOST: { // cudaFreeHost call here // FIXME: it would be nice to get rid of typecasting here long reqMem = AllocationUtils.getRequiredMemory(point.getShape()); // log.info("Deallocating {} bytes on [HOST]", reqMem); NativeOps nativeOps = NativeOpsHolder.getInstance().getDeviceNativeOps(); long result = nativeOps.freeHost(point.getPointers().getHostPointer().address()); // JCuda.cudaFreeHost(new Pointer(point.getPointers().getHostPointer().address())); if (result == 0) throw new RuntimeException("Can't deallocate [HOST] memory..."); } break; case DEVICE: { // cudaFree call // JCuda.cudaFree(new Pointer(point.getPointers().getDevicePointer().address())); long reqMem = AllocationUtils.getRequiredMemory(point.getShape()); // log.info("Deallocating {} bytes on [DEVICE]", reqMem); NativeOps nativeOps = NativeOpsHolder.getInstance().getDeviceNativeOps(); long result = nativeOps.freeDevice(point.getPointers().getDevicePointer().address(), 0); if (result == 0) throw new RuntimeException("Can't deallocate [DEVICE] memory..."); } break; default: throw new IllegalStateException( "Can't free memory on target [" + point.getAllocationStatus() + "]"); } } /** * This method checks specified device for specified amount of memory * * @param deviceId * @param requiredMemory * @return */ public boolean pingDeviceForFreeMemory(Integer deviceId, long requiredMemory) { /* long[] totalMem = new long[1]; long[] freeMem = new long[1]; JCuda.cudaMemGetInfo(freeMem, totalMem); long free = freeMem[0]; long total = totalMem[0]; long used = total - free; /* We don't want to allocate memory if it's too close to the end of available ram. */ // if (configuration != null && used > total * configuration.getMaxDeviceMemoryUsed()) return // false; /* if (free + requiredMemory < total * 0.85) return true; else return false; */ long freeMem = nativeOps.getDeviceFreeMemory(-1); if (freeMem - requiredMemory < DEVICE_RESERVED_SPACE) return false; else return true; } }