// pthread_cond_timedwait
PTH_FUNC(int, pthreadZucondZutimedwaitZAZa, // pthread_cond_timedwait@*
pthread_cond_t* cond, pthread_mutex_t* mutex,
struct timespec* abstime)
{
int ret;
OrigFn fn;
unsigned long mutex_is_valid;
VALGRIND_GET_ORIG_FN(fn);
if (TRACE_PTH_FNS) {
fprintf(stderr, "<< pthread_cond_timedwait %p %p %p",
cond, mutex, abstime);
fflush(stderr);
}
/* Tell the tool a cond-wait is about to happen, so it can check
for bogus argument values. In return it tells us whether it
thinks the mutex is valid or not. */
DO_CREQ_W_WW(mutex_is_valid,
_VG_USERREQ__HG_PTHREAD_COND_WAIT_PRE,
pthread_cond_t*,cond, pthread_mutex_t*,mutex);
assert(mutex_is_valid == 1 || mutex_is_valid == 0);
/* Tell the tool we're about to drop the mutex. This reflects the
fact that in a cond_wait, we show up holding the mutex, and the
call atomically drops the mutex and waits for the cv to be
signalled. */
if (mutex_is_valid) {
DO_CREQ_v_W(_VG_USERREQ__HG_PTHREAD_MUTEX_UNLOCK_PRE,
pthread_mutex_t*,mutex);
}
CALL_FN_W_WWW(ret, fn, cond,mutex,abstime);
if ((ret == 0 || ret == ETIMEDOUT) && mutex_is_valid) {
/* and now we have the mutex again */
DO_CREQ_v_W(_VG_USERREQ__HG_PTHREAD_MUTEX_LOCK_POST,
pthread_mutex_t*,mutex);
}
if (ret == 0 && mutex_is_valid) {
DO_CREQ_v_WW(_VG_USERREQ__HG_PTHREAD_COND_WAIT_POST,
pthread_cond_t*,cond, pthread_mutex_t*,mutex);
}
if (ret != 0 && ret != ETIMEDOUT) {
DO_PthAPIerror( "pthread_cond_timedwait", ret );
}
if (TRACE_PTH_FNS) {
fprintf(stderr, " cotimedwait -> %d >>\n", ret);
}
return ret;
}
static __always_inline
int sem_init_intercept(sem_t *sem, int pshared, unsigned int value)
{
int ret;
OrigFn fn;
VALGRIND_GET_ORIG_FN(fn);
VALGRIND_DO_CLIENT_REQUEST_STMT(VG_USERREQ__PRE_SEM_INIT,
sem, pshared, value, 0, 0);
CALL_FN_W_WWW(ret, fn, sem, pshared, value);
VALGRIND_DO_CLIENT_REQUEST_STMT(VG_USERREQ__POST_SEM_INIT,
sem, 0, 0, 0, 0);
return ret;
}
static __always_inline
int pthread_barrier_init_intercept(pthread_barrier_t* barrier,
const pthread_barrierattr_t* attr,
unsigned count)
{
int ret;
OrigFn fn;
VALGRIND_GET_ORIG_FN(fn);
VALGRIND_DO_CLIENT_REQUEST_STMT(VG_USERREQ__PRE_BARRIER_INIT,
barrier, pthread_barrier, count, 0, 0);
CALL_FN_W_WWW(ret, fn, barrier, attr, count);
VALGRIND_DO_CLIENT_REQUEST_STMT(VG_USERREQ__POST_BARRIER_INIT,
barrier, pthread_barrier, 0, 0, 0);
return ret;
}
static __always_inline
int pthread_cond_timedwait_intercept(pthread_cond_t *cond,
pthread_mutex_t *mutex,
const struct timespec* abstime)
{
int ret;
OrigFn fn;
VALGRIND_GET_ORIG_FN(fn);
VALGRIND_DO_CLIENT_REQUEST_STMT(VG_USERREQ__PRE_COND_WAIT,
cond, mutex, DRD_(mutex_type)(mutex), 0, 0);
CALL_FN_W_WWW(ret, fn, cond, mutex, abstime);
VALGRIND_DO_CLIENT_REQUEST_STMT(VG_USERREQ__POST_COND_WAIT,
cond, mutex, 1, 0, 0);
return ret;
}
// sem_init
PTH_FUNC(int, semZuinitZa, // sem_init*
sem_t *sem,
int pshared,
unsigned int value)
{
int ret;
int res;
OrigFn fn;
VALGRIND_GET_ORIG_FN(fn);
VALGRIND_DO_CLIENT_REQUEST(res, -1, VG_USERREQ__PRE_SEM_INIT,
sem, pshared, value, 0, 0);
CALL_FN_W_WWW(ret, fn, sem, pshared, value);
VALGRIND_DO_CLIENT_REQUEST(res, -1, VG_USERREQ__POST_SEM_INIT,
sem, 0, 0, 0, 0);
return ret;
}
// pthread_barrier_init
PTH_FUNC(int, pthreadZubarrierZuinit, // pthread_barrier_init
pthread_barrier_t* barrier,
const pthread_barrierattr_t* attr,
unsigned count)
{
int ret;
int res;
OrigFn fn;
VALGRIND_GET_ORIG_FN(fn);
VALGRIND_DO_CLIENT_REQUEST(res, -1, VG_USERREQ__PRE_BARRIER_INIT,
barrier, pthread_barrier, count, 0, 0);
CALL_FN_W_WWW(ret, fn, barrier, attr, count);
VALGRIND_DO_CLIENT_REQUEST(res, -1, VG_USERREQ__POST_BARRIER_INIT,
barrier, pthread_barrier, 0, 0, 0);
return ret;
}
// pthread_cond_timedwait
PTH_FUNC(int, pthreadZucondZutimedwaitZa, // pthread_cond_timedwait*
pthread_cond_t *cond,
pthread_mutex_t *mutex,
const struct timespec* abstime)
{
int ret;
int res;
OrigFn fn;
VALGRIND_GET_ORIG_FN(fn);
VALGRIND_DO_CLIENT_REQUEST(res, -1, VG_USERREQ__PRE_COND_WAIT,
cond, mutex, DRD_(mutex_type)(mutex), 0, 0);
CALL_FN_W_WWW(ret, fn, cond, mutex, abstime);
VALGRIND_DO_CLIENT_REQUEST(res, -1, VG_USERREQ__POST_COND_WAIT,
cond, mutex, 1, 0, 0);
return ret;
}
int VG_WRAP_FUNCTION_ZU(VG_Z_LIBC_SONAME, setenv)
(const char* name, const char* value, int overwrite)
{
OrigFn fn;
Word result;
const char* p;
VALGRIND_GET_ORIG_FN(fn);
if (name)
for (p = name; *p; p++)
__asm__ __volatile__("" ::: "memory");
if (value)
for (p = value; *p; p++)
__asm__ __volatile__("" ::: "memory");
VALGRIND_CHECK_VALUE_IS_DEFINED (overwrite);
CALL_FN_W_WWW(result, fn, name, value, overwrite);
return result;
}
int VG_WRAP_FUNCTION_ZU(VG_Z_LIBC_SONAME, setenv)
(const char* name, const char* value, int overwrite)
{
OrigFn fn;
Word result;
const char* p;
VALGRIND_GET_ORIG_FN(fn);
/* Now by walking over the string we magically produce
traces when hitting undefined memory. */
if (name)
for (p = name; *p; p++)
__asm__ __volatile__("" ::: "memory");
if (value)
for (p = value; *p; p++)
__asm__ __volatile__("" ::: "memory");
VALGRIND_CHECK_VALUE_IS_DEFINED (overwrite);
CALL_FN_W_WWW(result, fn, name, value, overwrite);
return result;
}
// Copy Host->Device
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuMemcpyHtoD)(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount) {
OrigFn fn;
CUresult result;
CUcontext ctx = NULL;
cgCtxListType *nodeCtx;
cgMemListType *nodeMem;
size_t dstSize;
long vgErrorAddress, vgErrorAddressDstDevice, vgErrorAddressSrcHost;
VALGRIND_GET_ORIG_FN(fn);
cgLock();
vgErrorAddressDstDevice = VALGRIND_CHECK_MEM_IS_DEFINED(&dstDevice, sizeof(void*));
vgErrorAddressSrcHost = VALGRIND_CHECK_MEM_IS_DEFINED(&srcHost, sizeof(CUdeviceptr));
// TODO: Currently errors are exclusive .. i.e. with undefined src and NULL
// dst pointer, only the undefined pointer is reported.
if (vgErrorAddressDstDevice || vgErrorAddressSrcHost) {
VALGRIND_PRINTF("Error:");
if (vgErrorAddressDstDevice) {
VALGRIND_PRINTF(" destination device");
if (vgErrorAddressSrcHost) {
VALGRIND_PRINTF(" and");
}
}
if (vgErrorAddressSrcHost) {
VALGRIND_PRINTF(" source host");
}
VALGRIND_PRINTF_BACKTRACE(" pointer in cuMemcpyHtoD not defined.\n");
} else if (dstDevice != 0 && srcHost != NULL) {
cgGetCtx(&ctx);
// Check allocation status and available size on device
nodeCtx = cgFindCtx(ctx);
nodeMem = cgFindMem(nodeCtx, dstDevice);
if (!nodeMem) {
VALGRIND_PRINTF("Error: Device memory during host->device memory copy is not allocated.");
} else {
dstSize = nodeMem->size - (dstDevice - nodeMem->dptr);
if (dstSize < ByteCount) {
VALGRIND_PRINTF("Error: Allocated device memory too small for host->device memory copy.\n");
VALGRIND_PRINTF(" Expected %lu allocated bytes but only found %lu.", ByteCount, dstSize);
}
}
if (!nodeMem || dstSize < ByteCount) {
VALGRIND_PRINTF_BACKTRACE("\n");
}
// Check allocation and definedness for host memory
vgErrorAddress = VALGRIND_CHECK_MEM_IS_ADDRESSABLE(srcHost, ByteCount);
if (vgErrorAddress) {
VALGRIND_PRINTF("Error: Host memory during host->device memory copy is not allocated.\n");
VALGRIND_PRINTF(" Expected %lu allocated bytes but only found %lu.", ByteCount, vgErrorAddress - (long)srcHost);
} else {
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(srcHost, ByteCount);
if (vgErrorAddress) {
VALGRIND_PRINTF("Error: Host memory during host->device memory copy is not defined.\n");
VALGRIND_PRINTF(" Expected %lu defined bytes but only found %lu.", ByteCount, vgErrorAddress - (long)srcHost);
}
}
if (vgErrorAddress) {
VALGRIND_PRINTF_BACKTRACE("\n");
}
} else {
VALGRIND_PRINTF("Error: cuMemcpyHtoD called with NULL");
if (dstDevice == 0) {
VALGRIND_PRINTF(" device");
if (srcHost == NULL) VALGRIND_PRINTF(" and");
}
if (srcHost == NULL) {
VALGRIND_PRINTF(" host");
}
VALGRIND_PRINTF_BACKTRACE(" pointer.\n");
}
CALL_FN_W_WWW(result, fn, dstDevice, srcHost, ByteCount);
cgUnlock();
return result;
}
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuMemsetD16)(CUdeviceptr dstDevice, unsigned short us, size_t N) {
OrigFn fn;
CUresult result;
CUcontext ctx = NULL;
cgMemListType *nodeMemDst;
int error = 0;
long vgErrorAddress;
size_t dstSize;
VALGRIND_GET_ORIG_FN(fn);
cgLock();
CALL_FN_W_WWW(result, fn, dstDevice, us, N);
// Check if function parameters are defined.
// TODO: Warning or error in case of a partially undefined us?
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&dstDevice, sizeof(CUdeviceptr));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'dstDevice' in call to cuMemsetD16 not defined.\n");
}
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&us, sizeof(us));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Warning: 'us' in call to cuMemsetD16 is not fully defined.\n");
}
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&N, sizeof(size_t));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'N' in call to cuMemsetD16 not defined.\n");
}
// Fetch current context
cgGetCtx(&ctx);
nodeMemDst = cgFindMem(cgFindCtx(ctx), dstDevice);
// Check if memory has been allocated
if (!nodeMemDst) {
error++;
VALGRIND_PRINTF("Error: Destination device memory not allocated in call to cuMemsetD16.\n");
} else {
// If memory is allocated, check size of available memory
dstSize = nodeMemDst->size - (dstDevice - nodeMemDst->dptr);
if (dstSize < sizeof(unsigned short) * N) {
error++;
VALGRIND_PRINTF("Error: Allocated device memory too small in call to cuMemsetD16.\n"
" Expected %lu allocated bytes but only found %lu.\n",
sizeof(unsigned short) * N, dstSize);
}
// Check if pointer is properly two byte aligned.
// TODO: Is this a valid check?
if (dstDevice % 2) {
error++;
VALGRIND_PRINTF("Error: Pointer dstDevice in call to cuMemsetD16 not two byte aligned.\n");
}
}
if (error) {
VALGRIND_PRINTF_BACKTRACE("");
}
cgUnlock();
return result;
}
