source: CIVL/examples/translation/cuda/civlc-cuda.cvl@ 054e457

#include "civlc.h"
#include <string.h>

$scope _global_scope = $here;
$scope _host_scope;

// Cuda data types

typedef struct {
        unsigned int x, y, z;
} dim3;

typedef struct {
        unsigned int x, y, z;
} uint3;

enum cudaMemcpyKind {
        cudaMemcpyHostToHost,
        cudaMemcpyHostToDevice,
        cudaMemcpyDeviceToHost,
        cudaMemcpyDeviceToDevice,
        cudaMemcpyDefault
};

enum cudaError {
        cudaSuccess,
        cudaErrorMissingConfiguration,
        cudaErrorMemoryAllocation,
        cudaErrorInitializationError,
        cudaErrorLaunchFailure,
        cudaErrorPriorLaunchFailure,
        cudaErrorLaunchTimeout,
        cudaErrorLaunchOutOfResources,
        cudaErrorInvalidDeviceFunction,
        cudaErrorInvalidConfiguration,
        cudaErrorInvalidDevice,
        cudaErrorInvalidValue,
        cudaErrorInvalidPitchValue,
        cudaErrorInvalidSymbol,
        cudaErrorMapBufferObjectFailed,
        cudaErrorUnmapBufferObjectFailed,
        cudaErrorInvalidHostPointer,
        cudaErrorInvalidDevicePointer,
        cudaErrorInvalidTexture,
        cudaErrorInvalidTextureBinding,
        cudaErrorInvalidChannelDescriptor,
        cudaErrorInvalidMemcpyDirection,
        cudaErrorAddressOfConstant,
        cudaErrorTextureFetchFailed,
        cudaErrorTextureNotBound,
        cudaErrorSynchronizationError,
        cudaErrorInvalidFilterSetting,
        cudaErrorInvalidNormSetting,
        cudaErrorMixedDeviceExecution,
        cudaErrorCudartUnloading,
        cudaErrorUnknown,
        cudaErrorNotYetImplemented,
        cudaErrorMemoryValueTooLarge,
        cudaErrorInvalidResourceHandle,
        cudaErrorNotReady,
        cudaErrorInsufficientDriver,
        cudaErrorSetOnActiveProcess,
        cudaErrorInvalidSurface,
        cudaErrorNoDevice,
        cudaErrorECCUncorrectable,
        cudaErrorSharedObjectSymbolNotFound,
        cudaErrorSharedObjectInitFailed,
        cudaErrorUnsupportedLimit,
        cudaErrorDuplicateVariableName,
        cudaErrorDuplicateTextureName,
        cudaErrorDuplicateSurfaceName,
        cudaErrorDevicesUnavailable,
        cudaErrorInvalidKernelImage,
        cudaErrorNoKernelImageForDevice,
        cudaErrorIncompatibleDriverContext,
        cudaErrorPeerAccessAlreadyEnabled,
        cudaErrorPeerAccessNotEnabled,
        cudaErrorDeviceAlreadyInUse,
        cudaErrorProfilerDisabled,
        cudaErrorProfilerNotInitialized,
        cudaErrorProfilerAlreadyStarted,
        cudaErrorProfilerAlreadyStopped,
        cudaErrorAssert,
        cudaErrorTooManyPeers,
        cudaErrorHostMemoryAlreadyRegistered,
        cudaErrorHostMemoryNotRegistered,
        cudaErrorOperatingSystem,
        cudaErrorStartupFailure,
        cudaErrorApiFailureBase
};
typedef enum cudaError cudaError_t;

// civl versions of cuda functions
uint3 make_uint3(unsigned int a, unsigned int b, unsigned int c) {
        uint3 t;
        t.x = a; t.y = b; t.z = c;
        return t;
}

cudaError_t cudaMemcpy ( void *dst, const void *src, size_t count, enum cudaMemcpyKind kind ) {

        /*
        switch (kind) {
                case cudaMemcpyHostToHost:
                        $assert($scopeof(*src) <= _host_scope);
                        $assert($scopeof(*dst) <= _host_scope);
                        break;
                case cudaMemcpyHostToDevice:
                        $assert($scopeof(*src) <= _host_scope);
                        $assert($scopeof(*dst) > _host_scope);
                        break;
                case cudaMemcpyDeviceToHost:
                        $assert($scopeof(*src) > _host_scope);
                        $assert($scopeof(*dst) <= _host_scope);
                        break;
                case cudaMemcpyDeviceToDevice:
                        $assert($scopeof(*src) > _host_scope);
                        $assert($scopeof(*dst) > _host_scope);
                        break;
                default:
                        break;
        }
        */
        memcpy(dst, src, count);
        return cudaSuccess;
}

cudaError_t cudaFree(void *devPtr) {
        $assert($scopeof(*devPtr) > _host_scope);
        $free(devPtr);
        return cudaSuccess;
}

cudaError_t cudaSetDevice(int device_id) {
        return cudaSuccess;
}

cudaError_t cudaGetDeviceCount(int *count) {
        // possibly this should return an value specified as $input?
        *count = 1;
        return cudaSuccess;
}

// helper function, only called by transformed program
int _index (dim3 size, uint3 indexes) {

        return indexes.x + size.x * (indexes.y + size.y * indexes.z);

}

dim3 _toDim3(int x) {
        dim3 d;
        d.x = x; d.y = 1; d.z = 1;
        return d;
}

void _createProcs(dim3 dim, void procFn(uint3)) {
        $proc procs[dim.x][dim.y][dim.z];
        for (int x = 0; x < dim.x; x++) {
                for (int y = 0; y < dim.y; y++) {
                        for (int z = 0; z < dim.z; z++) {
                                uint3 bid = make_uint3(x, y, z);
                                procs[x][y][z] = $spawn procFn(bid);
                        }

                }
        }
        for (int x = 0; x < dim.x; x++) {
                for (int y = 0; y < dim.y; y++) {
                        for (int z = 0; z < dim.z; z++) {
                                $wait(procs[x][y][z]);
                        }
                }
        }
}

typedef struct {
        $proc finalProcess;
} _CUstream;
typedef _CUstream* cudaStream_t;

struct {
        cudaStream_t stream;
        struct _cudaStreamNode_ *next;
} _cudaStreamNode_;
typedef struct _cudaStreamNode_ _cudaStreamNode;

typedef struct {
        _cudaStreamNode *headNode;
        $proc lastProc;
} _cudaContext;

_cudaContext _context = { .headNode = NULL, .lastProc = $proc_null };

cudaError_t cudaStreamCreate(cudaStream_t *pStream) {
        *pStream = (cudaStream_t)$malloc($root, sizeof(_CUstream));
        (*pStream)->finalProcess = $proc_null;
        _cudaStreamNode *newNode = (_cudaStreamNode*)$malloc($root, sizeof(_cudaStreamNode));
        _cudaStreamNode *oldHead = _context.headNode;
        newNode->stream = *pStream;
        newNode->next = oldHead;
        _context.headNode = newNode;

        return cudaSuccess;
}

cudaError_t cudaStreamDestroy(cudaStream_t pStream) {
        _cudaStreamNode *prevNode = NULL;
        _cudaStreamNode *curNode = _context.headNode;
        while (curNode != NULL) {
                if (pStream == curNode->stream) {
                        if (prevNode == NULL) {
                                _context.headNode = NULL;
                        } else {
                                prevNode->next = curNode->next;
                                $free(pStream);
                                $free(curNode);
                        }
                        return cudaSuccess;
                }
        }
        $assert($false);
        return cudaErrorInvalidResourceHandle;
}

void _tryWait($proc p) {
        if (p != $proc_null) $wait(p);
}

void _initKernel(cudaStream_t stream, void (*k)($proc)) {
        $proc prevProc;
        if (stream == NULL) {
                cudaDeviceSynchronize();
                prevProc = _context.lastProc;
                _context.lastProc = $spawn k(prevProc);
        } else {
                prevProc = stream->lastProc;
                stream->lastProc = $spawn k(prevProc);
        }
}