// TODO: Can we do additional checks?
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuArrayGetDescriptor)(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, CUarray hArray) {
OrigFn fn;
CUresult result;
CUcontext ctx = NULL;
cgCtxListType *ctxNode;
cgArrListType *node;
VALGRIND_GET_ORIG_FN(fn);
cgLock();
CALL_FN_W_WW(result, fn, pArrayDescriptor, hArray);
// Determine context of current thread ..
cgGetCtx(&ctx);
// .. locate the respective ctx node ..
ctxNode = cgFindCtx(ctx);
// .. and finally locate the array in the context's list of arrays.
node = cgFindArr(ctxNode, hArray);
if (result == CUDA_SUCCESS && !node) {
VALGRIND_PRINTF("cuArrayGetDescriptor returned successfully, but array not found\n");
VALGRIND_PRINTF_BACKTRACE(" in cudagrind's internal list. Reason: Unknown\n");
} else if (result != CUDA_SUCCESS && node) {
VALGRIND_PRINTF("cuArrayGetDescriptor returned with error code: %d,\n", result);
VALGRIND_PRINTF_BACKTRACE(" but array is found in cudagrind's internal list.\n");
} else if (result != CUDA_SUCCESS) {
VALGRIND_PRINTF("cuArrayGetDescriptor returned with error code: %d,\n", result);
VALGRIND_PRINTF_BACKTRACE(" possible reason: Wrong context or array not previously created.\n");
}
cgUnlock();
return result;
}
/*
* Adds the array residing in context ctx and referenced by dptr to the allocated array list.
*
* Input:
* CUcontext ctx - CUDA context of array
* CUarray dptr - device pointer of array
*/
void cgAddArr(CUcontext ctx, CUarray dptr) {
cgCtxListType *ctxNode;
cgArrListType *arrNode;
// Initialize desc with dummy values
CUDA_ARRAY3D_DESCRIPTOR desc = {0, 0, CU_AD_FORMAT_FLOAT, 0, 0, 0};
CUresult error;
// Locate or create entry for context
ctxNode = cgFindCtx(ctx);
// Nasty hack:
// If we call cuArray3DGetDescriptor first, its wrapper will
// produce a spurious (albeit, at this point, legit) error message,
// because the array is not contained in our internal list. But if
// we call cgCtxAddArr first, we are missing the array descriptor!
// -> We add the array with a dummy descriptor which in turn allows
// us to fetch the original descriptor through the driver without
// triggering the checks in the cuArray3DGetDescriptor wrapper.
//
// 1. Add array to internal list, including dummy descriptor
cgCtxAddArr(ctxNode, dptr, desc);
// 2. Get real descriptor, the wrapper wont complain now
error = cuArray3DGetDescriptor(&desc, dptr);
if (error != CUDA_SUCCESS) {
VALGRIND_PRINTF("Error: cuArray3DGetDescriptor returned with %d in cgAddArr.\n", error);
}
// 3. Fetch node of array and set the real descriptor
arrNode = cgFindArr(ctxNode, dptr);
arrNode->desc = desc;
#ifdef CUDAGRIND_DEBUG
VALGRIND_PRINTF("Context list after 'cgAddArr' operation:\n");
printCtxList();
#endif
}
int
main (int argc, char **argv)
{
int x = 0;
x += VALGRIND_PRINTF("Yo ");
x += VALGRIND_PRINTF("Yo ");
x += VALGRIND_PRINTF("Ma\n");
fprintf(stderr, "%d\n", x);
x = VALGRIND_PRINTF_BACKTRACE("Backtrace line one\nLine two:\n");
fprintf(stderr, "%d\n", x);
return 0;
}
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuMemcpyDtoDAsync)(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream) {
int error = 0;
long vgErrorAddress;
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&hStream, sizeof(CUstream));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'hStream' in call to cuMemcpyDtoDAsync not defined.\n");
}
cgLock();
CUcontext ctx = NULL;
cgCtxListType *nodeCtx;
cgMemListType *nodeMemDst, *nodeMemSrc;
// Get current context ..
cgGetCtx(&ctx);
nodeCtx = cgFindCtx(ctx);
// .. and locate memory if we are handling device memory
nodeMemDst = cgFindMem(nodeCtx, dstDevice);
nodeMemSrc = cgFindMem(nodeCtx, srcDevice);
if (nodeMemDst && nodeMemDst->locked & 2 && nodeMemDst->stream != hStream) {
error++;
VALGRIND_PRINTF("Error: Concurrent write and read access by different streams.\n");
}
if (nodeMemSrc && nodeMemSrc->locked && nodeMemSrc->stream != hStream) {
error++;
VALGRIND_PRINTF("Error: Concurrent write and read access by different streams.\n");
}
if (nodeMemDst) {
nodeMemDst->locked = nodeMemDst->locked | 2;
nodeMemDst->stream = hStream;
}
if (nodeMemSrc) {
nodeMemSrc->locked = nodeMemSrc->locked | 1;
nodeMemSrc->stream = hStream;
}
cgUnlock();
if (error) {
VALGRIND_PRINTF_BACKTRACE("");
}
return cuMemcpyDtoD(dstDevice, srcDevice, ByteCount);
}
/*
* Removes the entry of array referenced by dptr from ctxNode's array list.
*
* Input:
* cgCtxListType *ctxNode - The node from which dptr is to be removed
* CUarray dptr - The device pointer of the to be removed array entry
*/
void cgCtxDelArr(cgCtxListType *ctxNode, CUarray dptr) {
cgArrListType *toFree, *node = ctxNode->array;
int deleted = 0;
// Run through list of memory segments and remove it if it's found
if (node) {
if (node->dptr == dptr) {
ctxNode->array = node->next;
toFree = node;
deleted = 1;
} else {
while (node->next && node->next->dptr != dptr) {
node = node->next;
}
// If node->next is not NULL it has to contain dptr now
if (node->next) {
toFree = node->next;
node->next = node->next->next;
deleted = 1;
}
}
}
// Print error if the to be deletec device pointer can not be found
if (!deleted) {
VALGRIND_PRINTF("Error: Tried to remove non-existant device array reference in cgCtxDelArr.\n");
VALGRIND_PRINTF_BACKTRACE("Possible reason: Wrong CUDA context or double free on device array pointer.\n");
} else { // Else free the memory used by the node ..
free(toFree);
// TODO: Also remove the context entry if it's empty? And where?
}
}
/*
* Adds an entry for the array referenced by dptr with given descriptor desc to the given context node ctxNode.
*
* cgCtxLisType *ctxNode - Node containing memory information for context
* CUarray dptr - Device array reference (1D,2D or 3D)
* CUDA_ARRAY3D_DESCRIPTOR desc- Descripter array referenced by dptr
*/
void cgCtxAddArr(cgCtxListType *ctxNode, CUarray dptr, CUDA_ARRAY3D_DESCRIPTOR desc) {
cgArrListType *node;
// Create new entry if list is still empty
if (!(ctxNode->array)) {
ctxNode->array = (cgArrListType*)malloc(sizeof(cgArrListType));
node = ctxNode->array;
node->dptr = dptr;
node->desc = desc;
node->locked = 0;
// Do not have to set node->stream here
node->next = NULL;
} else {
node = ctxNode->array;
while (node->next && node->dptr != dptr) {
node = node->next;
}
if (node->dptr != dptr) {
node->next = (cgArrListType*)malloc(sizeof(cgArrListType));
node = node->next;
node->dptr = dptr;
node->desc = desc;
node->locked = 0;
// Do not have to set node->stream here
node->next = NULL;
} else {
VALGRIND_PRINTF("Error: Tried to add already existing array reference in cgCtxAddArr.\n");
VALGRIND_PRINTF_BACKTRACE("Possible reason: Unknown. This should not have happened!");
}
}
}
/*
* Adds an entry for the device memory referenced by dptr of given size to the given context node ctxNode.
*
* cgCtxLisType *ctxNode - Node containing memory information for context
* CUdeviceptr dptr - Device pointer to memory
* size_t size - Size of memory referenced by dptr
*/
void cgCtxAddMem(cgCtxListType *ctxNode, CUdeviceptr dptr, size_t size) {
cgMemListType *node;
// Create new entry if list is still empty
if (!(ctxNode->memory)) {
ctxNode->memory = (cgMemListType*)malloc(sizeof(cgMemListType));
node = ctxNode->memory;
node->dptr = dptr;
node->isSymbol = 0;
node->size = size;
node->locked = 0;
// Do not have to set node->stream here
node->next = NULL;
} else {
node = ctxNode->memory;
while (node->next && node->dptr != dptr) {
node = node->next;
}
if (node->dptr != dptr) {
node->next = (cgMemListType*)malloc(sizeof(cgMemListType));
node = node->next;
node->dptr = dptr;
node->isSymbol = 0;
node->size = size;
node->locked = 0;
// Do not have to set node->stream here
node->next = NULL;
} else {
VALGRIND_PRINTF("Error: Tried to add already existing device pointer in cgCtxAddMem.\n");
VALGRIND_PRINTF_BACKTRACE("Possible reason: Unknown. This should not have happened!");
}
}
}
int
main (int argc, char **argv)
{
int x = VALGRIND_PRINTF("Yo");
printf ("%d\n", x);
return 0;
}
// The gcc constructor facility is used to initialize the POSIX Mutex that makes
// Cudagrind threadsafe and output a welcome Message when it's being run.
void __attribute__ ((constructor)) cgConstructor() {
#ifndef CUDAGRIND_NOPTHREAD
// If CUDAGRIND_NOTHREADS is
int error;
// Set to PTHREAD_MUTEX_RECURSIVE to enable recursive calls inside Cudagrind wrappers
error = pthread_mutexattr_settype(&cgMutexAttr, PTHREAD_MUTEX_RECURSIVE);
if (error) {
VALGRIND_PRINTF("Error: Constructor failed to set mutex attribute with error %d.\n", error);
}
error = pthread_mutex_init(&cgMutex, &cgMutexAttr);
if (error) {
VALGRIND_PRINTF("Error: Constructor failed to set mutex attribute with error %d.\n", error);
}
#endif
VALGRIND_PRINTF("\nWelcome to Cudagrind version %d.%d.%d\n\n", CG_VERSION_MAJOR, CG_VERSION_MINOR, CG_VERSION_REVISION);
}
// Helper function that prints the current ctx/gpu-mem list on the screen
void printCtxList() {
cgCtxListType *ctxList = cgCtxList;
cgMemListType *memList;
cgArrListType *arrList;
if (cgCtxList) {
while (ctxList) {
VALGRIND_PRINTF("CTX %lu: ", ctxList->ctx);
memList = ctxList->memory;
while (memList) {
VALGRIND_PRINTF("%lu (%lu) -> ", memList->dptr, memList->size);
memList = memList->next;
}
VALGRIND_PRINTF("\n");
if (ctxList->array) {
VALGRIND_PRINTF(" Arr: ");
arrList = ctxList->array;
// TODO: More debug output (size) for arrays?
while (arrList) {
VALGRIND_PRINTF("%lu -> ", arrList->dptr);
arrList = arrList->next;
}
VALGRIND_PRINTF("\n");
}
ctxList = ctxList->next;
}
} else {
VALGRIND_PRINTF("CTX list is currently empty!\n");
}
}
ucs_log_func_rc_t
ucs_log_default_handler(const char *file, unsigned line, const char *function,
ucs_log_level_t level, const char *prefix,
const char *message, va_list ap)
{
size_t buffer_size = ucs_config_memunits_get(ucs_global_opts.log_buffer_size,
256, 2048);
const char *short_file;
struct timeval tv;
size_t length;
char *buf;
char *valg_buf;
if (!ucs_log_enabled(level) && (level != UCS_LOG_LEVEL_PRINT)) {
return UCS_LOG_FUNC_RC_CONTINUE;
}
buf = ucs_alloca(buffer_size + 1);
buf[buffer_size] = 0;
strncpy(buf, prefix, buffer_size);
length = strlen(buf);
vsnprintf(buf + length, buffer_size - length, message, ap);
short_file = strrchr(file, '/');
short_file = (short_file == NULL) ? file : short_file + 1;
gettimeofday(&tv, NULL);
if (level <= ucs_global_opts.log_level_trigger) {
ucs_handle_error(ucs_log_level_names[level], "%13s:%-4u %s: %s",
short_file, line, ucs_log_level_names[level], buf);
} else if (RUNNING_ON_VALGRIND) {
valg_buf = ucs_alloca(buffer_size + 1);
snprintf(valg_buf, buffer_size,
"[%lu.%06lu] %16s:%-4u %-4s %-5s %s\n", tv.tv_sec, tv.tv_usec,
short_file, line, "UCX", ucs_log_level_names[level], buf);
VALGRIND_PRINTF("%s", valg_buf);
} else if (ucs_log_initialized) {
fprintf(ucs_log_file,
"[%lu.%06lu] [%s:%-5d:%d] %16s:%-4u %-4s %-5s %s\n",
tv.tv_sec, tv.tv_usec, ucs_log_hostname, ucs_log_pid,
ucs_log_get_thread_num(), short_file, line, "UCX",
ucs_log_level_names[level], buf);
} else {
fprintf(stdout,
"[%lu.%06lu] %16s:%-4u %-4s %-5s %s\n",
tv.tv_sec, tv.tv_usec, short_file, line,
"UCX", ucs_log_level_names[level], buf);
}
/* flush the log file if the log_level of this message is fatal or error */
if (level <= UCS_LOG_LEVEL_ERROR) {
ucs_log_flush();
}
return UCS_LOG_FUNC_RC_CONTINUE;
}
// The gcc destructor facility is used to clear the POSIX Mutex
void __attribute__ ((destructor)) cgDestructor() {
#ifndef CUDAGRIND_NOPTHREAD
int error;
error = pthread_mutex_destroy(&cgMutex);
if (error) {
VALGRIND_PRINTF("Error: Deconstructor failed to free mutex with error %d.\n", error);
}
#endif
// Search through whole context and memory/array lists for unfreed memory.
// Memory allocated to handle symboles will be ignored.
cgCtxListType *nodeCtx = cgCtxList;
cgMemListType *nodeMem;
cgArrListType *nodeArr;
int countMem = 0, countArr = 0;
size_t sizeMem = 0, sizeArr = 0;
while (nodeCtx) {
nodeMem = nodeCtx->memory;
while (nodeMem) {
if (!(nodeMem->isSymbol)) {
countMem++;
sizeMem += nodeMem->size;
}
nodeMem = nodeMem->next;
}
nodeArr = nodeCtx->array;
while (nodeArr) {
countArr++;
// TODO: Calculate amount of lost memory for arrays
nodeArr = nodeArr->next;
}
nodeCtx = nodeCtx->next;
}
if (countMem || countArr) {
VALGRIND_PRINTF("Cudagrind leak report:\n");
if (countMem) {
VALGRIND_PRINTF(" %d device memory region (%lu bytes) not freed.\n", countMem, (unsigned long)sizeMem);
}
if (countArr) {
VALGRIND_PRINTF(" %d device arrays not freed.\n", countArr);
}
}
}
/*
* Adds the memory residing in context ctx and referenced by dptr to the allocated memory list.
*
* Input:
* CUcontext ctx - CUDA context of memory
* CUdeviceptr dptr - device pointer of memory
* size_t bytesize - size of memory in bytes
*/
void cgAddMem(CUcontext ctx, CUdeviceptr dptr, size_t bytesize) {
cgCtxListType *ctxNode;
// Locate or create entry for context
ctxNode = cgFindCtx(ctx);
cgCtxAddMem(ctxNode, dptr, bytesize);
#ifdef CUDAGRIND_DEBUG
VALGRIND_PRINTF("Context list after 'cgAddMem' operation:\n");
printCtxList();
#endif
}
bool valgrindCheckObjectInPool(MM_GCExtensionsBase *extensions, uintptr_t baseAddress)
{
#if defined(VALGRIND_REQUEST_LOGS)
VALGRIND_PRINTF("Checking for an object at 0x%lx\n", baseAddress);
#endif /* defined(VALGRIND_REQUEST_LOGS) */
MUTEX_ENTER(extensions->memcheckHashTableMutex);
bool exists = hashTableFind(extensions->memcheckHashTable, &baseAddress) != NULL ? true : false;
MUTEX_EXIT(extensions->memcheckHashTableMutex);
return exists;
}
/*
* Removes the entry of the array residing in context ctx and referenced by dptr from allocated array list.
*
* Input:
* CUcontext ctx - context
* CUarray dptr - device pointer of to be removed 2D array
*/
void cgDelArr(CUcontext ctx, CUarray dptr) {
cgCtxListType *ctxNode;
// Locate the entry for the context ..
// TODO: What happens if context does not exist anymore (we create it ..
// but is that what we really want?!)
ctxNode = cgFindCtx(ctx);
cgCtxDelArr(ctxNode, dptr);
#ifdef CUDAGRIND_DEBUG
VALGRIND_PRINTF("Context list after 'cgDelArr' operation:\n");
printCtxList();
#endif
}
void
memory_status_calibrate (const char *marker)
{
int l, d, r, s;
VALGRIND_PRINTF ("Calibrating: %s\n", marker);
VALGRIND_DO_LEAK_CHECK;
VALGRIND_COUNT_LEAKS(l, d, r, s);
_memory_baseline_leaks = l;
_memory_baseline_errors = VALGRIND_COUNT_ERRORS;
}
int
memory_status_verify (const char *marker)
{
int l, d, r, s, status;
VALGRIND_PRINTF ("Verifying: %s\n", marker);
status = MEMORY_OK;
if (VALGRIND_COUNT_ERRORS > _memory_baseline_errors)
status |= MEMORY_ERROR;
VALGRIND_DO_LEAK_CHECK;
VALGRIND_COUNT_LEAKS(l, d, r, s);
if (l > _memory_baseline_leaks)
status |= MEMORY_LEAK;
return status;
}
void* child_fn_1 ( void* arg )
{
const char* threadname = "try1";
int r;
# if !defined(VGO_darwin)
pthread_setname_np(pthread_self(), threadname);
# else
pthread_setname_np(threadname);
# endif
bad_things(3);
VALGRIND_PRINTF("%s", "I am in child_fn_1\n");
r = pthread_create(&children[2], NULL, child_fn_2, NULL);
assert(!r);
r = pthread_join(children[2], NULL);
assert(!r);
return NULL;
}
/*
* pmemobj_vg_check_no_undef -- (internal) check whether there are any undefined
* regions
*/
static void
pmemobj_vg_check_no_undef(struct pmemobjpool *pop)
{
LOG(4, "pop %p", pop);
struct {
void *start, *end;
} undefs[MAX_UNDEFS];
int num_undefs = 0;
VALGRIND_DO_DISABLE_ERROR_REPORTING;
char *addr_start = pop->addr;
char *addr_end = addr_start + pop->size;
while (addr_start < addr_end) {
char *noaccess = (char *)VALGRIND_CHECK_MEM_IS_ADDRESSABLE(
addr_start, addr_end - addr_start);
if (noaccess == NULL)
noaccess = addr_end;
while (addr_start < noaccess) {
char *undefined =
(char *)VALGRIND_CHECK_MEM_IS_DEFINED(
addr_start, noaccess - addr_start);
if (undefined) {
addr_start = undefined;
#ifdef VALGRIND_CHECK_MEM_IS_UNDEFINED
addr_start = (char *)
VALGRIND_CHECK_MEM_IS_UNDEFINED(
addr_start, noaccess - addr_start);
if (addr_start == NULL)
addr_start = noaccess;
#else
while (addr_start < noaccess &&
VALGRIND_CHECK_MEM_IS_DEFINED(
addr_start, 1))
addr_start++;
#endif
if (num_undefs < MAX_UNDEFS) {
undefs[num_undefs].start = undefined;
undefs[num_undefs].end = addr_start - 1;
num_undefs++;
}
} else
addr_start = noaccess;
}
#ifdef VALGRIND_CHECK_MEM_IS_UNADDRESSABLE
addr_start = (char *)VALGRIND_CHECK_MEM_IS_UNADDRESSABLE(
addr_start, addr_end - addr_start);
if (addr_start == NULL)
addr_start = addr_end;
#else
while (addr_start < addr_end &&
(char *)VALGRIND_CHECK_MEM_IS_ADDRESSABLE(
addr_start, 1) == addr_start)
addr_start++;
#endif
}
VALGRIND_DO_ENABLE_ERROR_REPORTING;
if (num_undefs) {
/*
* How to resolve this error:
* If it's part of the free space Valgrind should be told about
* it by VALGRIND_DO_MAKE_MEM_NOACCESS request. If it's
* allocated - initialize it or use VALGRIND_DO_MAKE_MEM_DEFINED
* request.
*/
VALGRIND_PRINTF("Part of the pool is left in undefined state on"
" boot. This is pmemobj's bug.\nUndefined"
" regions:\n");
for (int i = 0; i < num_undefs; ++i)
VALGRIND_PRINTF(" [%p, %p]\n", undefs[i].start,
undefs[i].end);
if (num_undefs == MAX_UNDEFS)
VALGRIND_PRINTF(" ...\n");
/* Trigger error. */
VALGRIND_CHECK_MEM_IS_DEFINED(undefs[0].start, 1);
}
}
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuMemsetD2D32)(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height) {
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_5W(result, fn, dstDevice, dstPitch, ui, Width, Height);
// Check if function parameters are defined.
// TODO: Warning or error in case of a partially undefined ui?
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&dstDevice, sizeof(CUdeviceptr));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'dstDevice' in call to cuMemsetD2D32 not defined.\n");
}
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&dstPitch, sizeof(size_t));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'dstPitch' in call to cuMemsetD2D32 not defined.\n");
}
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&ui, sizeof(ui));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Warning: 'ui' in call to cuMemsetD2D32 is not fully defined.\n");
}
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&Width, sizeof(size_t));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'Width' in call to cuMemsetD2D32 not defined.\n");
}
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&Height, sizeof(size_t));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'Height' in call to cuMemsetD2D32 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 cuMemsetD2D32.\n");
} else {
// If memory is allocated, check size of available memory
dstSize = nodeMemDst->size - (dstDevice - nodeMemDst->dptr);
// The whole memory block of dstPitch*Height must fit into memory
if (dstSize < sizeof(ui) * dstPitch * Height) {
error++;
VALGRIND_PRINTF("Error: Allocated device memory too small in call to cuMemsetD2D32.\n"
" Expected %lu allocated bytes but only found %lu.\n",
sizeof(ui) * dstPitch * Height, dstSize);
}
// Check if dstDevice and dstPitch are both properly aligned
// TODO: Is this a valid check? (see also cuMemsetD32)
if (dstDevice % 4) {
error++;
VALGRIND_PRINTF("Error: Pointer dstDevice in call to cuMemsetD2D32 not four byte aligned.\n");
}
if (dstPitch % 4) {
error++;
VALGRIND_PRINTF("Error: Destination pitch in call to cuMemsetD2D32 not four byte aligned.\n");
}
}
// Make sure pitch is big enough to accommodate asked for Width
if (dstPitch < Width) {
error++;
VALGRIND_PRINTF("Error: dstPitch smaller than Width in call to cuMemsetD2D32.\n");
}
if (error) {
VALGRIND_PRINTF_BACKTRACE("");
}
cgUnlock();
return result;
}
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuMemcpy3D)(const CUDA_MEMCPY3D *pCopy) {
OrigFn fn;
CUresult result;
CUcontext ctx = NULL;
int error = 0, error_addressable, error_defined;
long vgErrorAddress = 0, vgErrorAddressDefined = 0;
VALGRIND_GET_ORIG_FN(fn);
cgLock();
CALL_FN_W_W(result, fn, pCopy);
// Check if pCopy is null, not allocated or undefined.
// For obvious reasons we skip the following checks if either condition is true.
if (!pCopy) {
error++;
VALGRIND_PRINTF_BACKTRACE("Error: pCopy in call to cuMemcpy3D is NULL.\n");
cgUnlock();
return result;
} else if ( vgErrorAddress = VALGRIND_CHECK_MEM_IS_ADDRESSABLE(pCopy, sizeof(CUDA_MEMCPY3D)) ) {
error++;
VALGRIND_PRINTF_BACKTRACE("Error: pCopy in call to cuMemcpy3D points to unallocated memory.\n");
cgUnlock();
return result;
} // It makes no sense to check _IS_DEFINED on the whole structure, since only part of it is used!
// General checks of constaints imposed by reference manual
if (pCopy->srcMemoryType != CU_MEMORYTYPE_ARRAY) {
if (pCopy->srcPitch && pCopy->srcPitch < pCopy->WidthInBytes + pCopy->srcXInBytes) {
error++;
VALGRIND_PRINTF("Error: srcPitch < WidthInBytes+srcXInBytes in cuMemcpy3D.\n");
}
if (pCopy->srcHeight && pCopy->srcHeight < pCopy->Height + pCopy->srcY) {
error++;
VALGRIND_PRINTF("Error: srcHeight < Height+srcY in cuMemcpy3D.\n");
}
}
if (pCopy->dstMemoryType != CU_MEMORYTYPE_ARRAY) {
if (pCopy->dstPitch && pCopy->dstPitch < pCopy->WidthInBytes + pCopy->dstXInBytes) {
error++;
VALGRIND_PRINTF("Error: dstPitch < WidthInBytes+dstXInBytes in cuMemcpy3D.\n");
}
if (pCopy->dstHeight && pCopy->dstHeight < pCopy->Height + pCopy->dstY) {
error++;
VALGRIND_PRINTF("Error: dstHeight < Height+dstY in cuMemcpy3D.\n");
}
}
switch (pCopy->srcMemoryType) {
case CU_MEMORYTYPE_UNIFIED:
// TODO: How do we handle unified memory?
break;
case CU_MEMORYTYPE_HOST: {
void *line;
error_addressable = 0;
error_defined = 0;
// TODO: Is Height, Depth > 1, even for 1D/2D copy operations?
for (int i = 0 ; i < pCopy->Height ; i++) {
for (int j = 0 ; j < pCopy->Depth ; j++) {
line = (void*)(
(char*)pCopy->srcHost
+ ((pCopy->srcZ + j) * pCopy->srcHeight + (pCopy->srcY + i))*pCopy->srcPitch
+ pCopy->srcXInBytes
);
vgErrorAddress = VALGRIND_CHECK_MEM_IS_ADDRESSABLE(line, (size_t)pCopy->WidthInBytes);
if (vgErrorAddress) {
error_addressable++;
} else {
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(line, (size_t)pCopy->WidthInBytes);
if (vgErrorAddress) {
error_defined++;
}
}
}
}
// TODO: Can we give precise information about location of error?
if (error_addressable) {
error++;
VALGRIND_PRINTF("Error: (Part of) source host memory not allocated\n"
" in call to cuMemcpy3D.\n");
}
if (error_defined) {
error++;
VALGRIND_PRINTF("Error: (Part of) source host memory not defined\n"
" in call to cuMemcpy3D.\n");
}
break;
}
case CU_MEMORYTYPE_DEVICE: {
// ptrEnd points to the end of the memory area which pCopy->srcDevice points into
CUdeviceptr line, ptrEnd;
cgMemListType *nodeMem;
// TODO: Check if pCopy->srcDevice is defined?
cgGetCtx(&ctx);
nodeMem = cgFindMem(cgFindCtx(ctx), pCopy->srcDevice);
// We only track addressable status (whether memory is allocated) for device memory regions
error_addressable = 0;
if (nodeMem) {
ptrEnd = nodeMem->dptr + nodeMem->size;
/*
for (int i = 0 ; i < pCopy->Height ; i++) {
for (int j = 0 ; j < pCopy->Depth ; j++) {
line = (CUdeviceptr)(
pCopy->srcDevice
+ ((pCopy->srcZ + j) * pCopy->srcHeight + (pCopy->srcY + i)) * pCopy->srcPitch
+ pCopy->srcXInBytes
);
// Is there enough allocated memory left to statisfy the current line?
if (ptrEnd - line < pCopy->WidthInBytes) {
error_addressable++;
}
}
}
*/
// Device memory should not be fragmented, so we only check the very last slice of memory
line = (CUdeviceptr)(
pCopy->srcDevice
+ (
(pCopy->srcZ + pCopy->Depth - 1) * pCopy->srcHeight
+ (pCopy->srcY + pCopy->Height - 1)
) * pCopy->srcPitch
+ pCopy->srcXInBytes);
if (ptrEnd - line < pCopy->WidthInBytes) {
error_addressable++;
}
} else {
error_addressable++;
}
if (error_addressable) {
error++;
VALGRIND_PRINTF("Error: (Part of) source device memory not allocated\n"
" in call to cuMemcpy3D.\n");
}
break;
}
case CU_MEMORYTYPE_ARRAY: {
CUDA_ARRAY3D_DESCRIPTOR descriptor;
int bytesPerElement;
int widthInBytes;
// Fetch array descriptor ..
cuArray3DGetDescriptor(&descriptor, pCopy->srcArray);
bytesPerElement = cgArrDescBytesPerElement(&descriptor);
if (!bytesPerElement) {
error++;
VALGRIND_PRINTF("Error: Unknown Format value in src array descriptor in cuMemcpy3D.\n");
}
widthInBytes = bytesPerElement * descriptor.Width;
// .. and check if dimensions are conform to the ones requested in pCopy
if (widthInBytes - pCopy->srcXInBytes < pCopy->WidthInBytes) {
error++;
VALGRIND_PRINTF("Error: Available width of %u bytes in source array is smaller than\n"
" requested Width of %u bytes in pCopy of cuMemcpy3D.\n",
widthInBytes - pCopy->srcXInBytes, pCopy->WidthInBytes);
}
if (pCopy->Height > 1 && descriptor.Height - pCopy->srcY < pCopy->Height) {
error++;
VALGRIND_PRINTF("Error: Available Height of %u in source array is smaller than\n"
" requested Height of %u in pCopy of cuMemcpy3D.\n",
descriptor.Height - pCopy->srcY, pCopy->Height);
}
if (pCopy->Depth > 1 && descriptor.Depth - pCopy->srcZ < pCopy->Depth) {
error++;
VALGRIND_PRINTF("Error: Available Depth of %u in source array is smaller than\n"
" requested Depth of %u in pCopy of cuMemcpy3D.\n",
descriptor.Depth - pCopy->srcY, pCopy->Height);
}
break;
}
default:
error++;
VALGRIND_PRINTF("Error: Unknown source memory type %d in cuMemcpy3D\n");
break;
}
switch (pCopy->dstMemoryType) {
case CU_MEMORYTYPE_UNIFIED:
// TODO: How do we handle unified memory?
break;
case CU_MEMORYTYPE_HOST: {
void *line;
error_addressable = 0;
error_defined = 0;
// TODO: Is Height, Depth > 1, even for 1D/2D copy operations?
for (int i = 0 ; i < pCopy->Height ; i++) {
for (int j = 0 ; j < pCopy->Depth ; j++) {
line = (void*)(
(char*)pCopy->dstHost
+ ((pCopy->dstZ + j) * pCopy->dstHeight + (pCopy->dstY + i))*pCopy->dstPitch
+ pCopy->dstXInBytes
);
// Unlike for the source operand we only need to check allocation status here
vgErrorAddress = VALGRIND_CHECK_MEM_IS_ADDRESSABLE(line, (size_t)pCopy->WidthInBytes);
if (vgErrorAddress) {
error_addressable++;
}
}
}
// TODO: Can we give precise information about location of error?
if (error_addressable) {
error++;
VALGRIND_PRINTF("Error: (Part of) destination host memory not allocated\n"
" in call to cuMemcpy3D.\n");
}
break;
}
case CU_MEMORYTYPE_DEVICE: {
// ptrEnd points to the end of the memory area which pCopy->dstDevice points into
CUdeviceptr line, ptrEnd;
cgMemListType *nodeMem;
// TODO: Check if pCopy->dstDevice is defined?
cgGetCtx(&ctx);
nodeMem = cgFindMem(cgFindCtx(ctx), pCopy->dstDevice);
// We only track addressable status (whether memory is allocated) for device memory regions
error_addressable = 0;
if (nodeMem) {
ptrEnd = nodeMem->dptr + nodeMem->size;
/*
for (int i = 0 ; i < pCopy->Height ; i++) {
for (int j = 0 ; j < pCopy->Depth ; j++) {
line = (CUdeviceptr)(
pCopy->dstDevice
+ ((pCopy->dstZ + j) * pCopy->dstHeight + (pCopy->dstY + i)) * pCopy->dstPitch
+ pCopy->dstXInBytes
);
// Is there enough allocated memory left to statisfy the current line?
if (ptrEnd - line < pCopy->WidthInBytes) {
error_addressable++;
}
}
}
*/
// Device memory should not be fragmented, so we only check the very last slice of memory
line = (CUdeviceptr)(
pCopy->dstDevice
+ (
(pCopy->dstZ + pCopy->Depth - 1) * pCopy->dstHeight
+ (pCopy->dstY + pCopy->Height - 1)
) * pCopy->dstPitch
+ pCopy->dstXInBytes);
if (ptrEnd - line < pCopy->WidthInBytes) {
error_addressable++;
}
} else {
error_addressable++;
}
if (error_addressable) {
error++;
VALGRIND_PRINTF("Error: (Part of) destination device memory not allocated\n"
" in call to cuMemcpy3D.\n");
}
break;
}
case CU_MEMORYTYPE_ARRAY: {
CUDA_ARRAY3D_DESCRIPTOR descriptor;
int bytesPerElement;
int widthInBytes;
// Fetch array descriptor ..
cuArray3DGetDescriptor(&descriptor, pCopy->dstArray);
bytesPerElement = cgArrDescBytesPerElement(&descriptor);
if (!bytesPerElement) {
error++;
VALGRIND_PRINTF("Error: Unknown Format value in dst array descriptor in cuMemcpy3D.\n");
}
widthInBytes = bytesPerElement * descriptor.Width;
// .. and check if dimensions are conform to the ones requested in pCopy
if (widthInBytes - pCopy->dstXInBytes < pCopy->WidthInBytes) {
error++;
VALGRIND_PRINTF("Error: Available width of %u bytes in destination array is smaller than\n"
" requested Width of %u bytes in pCopy of cuMemcpy3D.\n",
widthInBytes - pCopy->dstXInBytes, pCopy->WidthInBytes);
}
if (pCopy->Height > 1 && descriptor.Height - pCopy->dstY < pCopy->Height) {
error++;
VALGRIND_PRINTF("Error: Available Height of %u in destination array is smaller than\n"
" requested Height of %u in pCopy of cuMemcpy3D.\n",
descriptor.Height - pCopy->dstY, pCopy->Height);
}
if (pCopy->Depth > 1 && descriptor.Depth - pCopy->dstZ < pCopy->Depth) {
error++;
VALGRIND_PRINTF("Error: Available Depth of %u in destination array is smaller than\n"
" requested Depth of %u in pCopy of cuMemcpy3D.\n",
descriptor.Depth - pCopy->dstZ, pCopy->Depth);
}
break;
}
default:
error++;
VALGRIND_PRINTF("Error: Unknown destination memory type %d in cuMemcpy3D\n");
break;
}
if (error) {
VALGRIND_PRINTF_BACKTRACE(" %d errors detected in call to cuMemcpy3D.", error);
}
cgUnlock();
return result;
}
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuMemcpyHtoA)(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount) {
OrigFn fn;
CUresult result;
CUcontext ctx = NULL;
cgCtxListType *nodeCtx;
cgArrListType *nodeArrDst;
cgMemListType *nodeMemSrc;
long vgErrorAddress;
int error = 0;
VALGRIND_GET_ORIG_FN(fn);
cgLock();
CALL_FN_W_WWWW(result, fn, dstArray, dstOffset, srcHost, ByteCount);
// Check if actual function parameters are defined
if (VALGRIND_CHECK_MEM_IS_DEFINED(&dstArray, sizeof(CUarray))) {
error++;
VALGRIND_PRINTF("Error: dstArray in call to cuMemcpyAtoD is not defined.\n");
} else if (!dstArray) {
error++;
VALGRIND_PRINTF("Error: dstArray in call to cuMemcpyAtoD is NULL.\n");
}
if (VALGRIND_CHECK_MEM_IS_DEFINED(&dstOffset, sizeof(size_t))) {
error++;
VALGRIND_PRINTF("Error: dstOffset in call to cuMemcpyAtoD is not defined.\n");
}
if (VALGRIND_CHECK_MEM_IS_DEFINED(&srcHost, sizeof(CUdeviceptr))) {
error++;
VALGRIND_PRINTF("Error: srcHost in call to cuMemcpyAtoD is not defined.\n");
} else if (!srcHost) {
error++;
VALGRIND_PRINTF("Error: srcDevice in call to cuMemcpyAtoD is NULL");
}
if (VALGRIND_CHECK_MEM_IS_DEFINED(&ByteCount, sizeof(size_t))) {
error++;
VALGRIND_PRINTF("Error: ByteCount in call to cuMemcpyAtoD is not defined.\n");
}
cgGetCtx(&ctx);
nodeCtx = cgFindCtx(ctx);
nodeArrDst = cgFindArr(nodeCtx, dstArray);
if (srcHost) {
vgErrorAddress = VALGRIND_CHECK_MEM_IS_ADDRESSABLE(srcHost, ByteCount);
// Check if memory referenced by srcHost has been allocated.
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: Source host memory in cuMemcpyHtoA is not not allocated.\n"
" Expected %l bytes but only found %l.\n",
ByteCount, vgErrorAddress - (long)srcHost);
} else {
// If allocated, now check if host memory is defined.
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(srcHost, ByteCount);
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: Source host memory in cuMemcpyHtoA is not defined.\n"
" Expected %l bytes but only found %l.\n",
ByteCount, vgErrorAddress - (long)srcHost);
}
}
}
if (nodeArrDst) {
// Check if array is 1-dimensional or big enough in first dimension
if (nodeArrDst->desc.Height > 1 || nodeArrDst->desc.Depth > 1) {
if (nodeArrDst->desc.Width - dstOffset < ByteCount) {
error++;
VALGRIND_PRINTF("Error: Destination array in cuMemcpyAtoD is 2-dimensional\n"
" and ByteCount bigger than available width in first dimension.\n");
} else {
VALGRIND_PRINTF("Warning: Destination array in cuMemcpyAtoD is 2-dimensional.\n");
}
} else if (nodeArrDst->desc.Width - dstOffset < ByteCount) {
// If array is 1D, check size.
VALGRIND_PRINTF("Error: Destination array in cuMemcpyAtoD is too small.\n"
" Expected %l bytes but only found %l.\n",
ByteCount, nodeArrDst->desc.Width - dstOffset);
error++;
}
} else {
error++;
VALGRIND_PRINTF("Error: Destination array not allocated in call to cuMemcpyAtoD.\n");
}
cgUnlock();
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;
}
static cairo_test_status_t
cairo_test_for_target (cairo_test_context_t *ctx,
const cairo_boilerplate_target_t *target,
int dev_offset,
cairo_bool_t similar)
{
cairo_test_status_t status;
cairo_surface_t *surface = NULL;
cairo_t *cr;
const char *empty_str = "";
char *offset_str;
char *base_name, *base_path;
char *out_png_path;
char *ref_path = NULL, *ref_png_path, *cmp_png_path = NULL;
char *new_path = NULL, *new_png_path;
char *xfail_path = NULL, *xfail_png_path;
char *base_ref_png_path;
char *base_new_png_path;
char *base_xfail_png_path;
char *diff_png_path;
char *test_filename = NULL, *pass_filename = NULL, *fail_filename = NULL;
cairo_test_status_t ret;
cairo_content_t expected_content;
cairo_font_options_t *font_options;
const char *format;
cairo_bool_t have_output = FALSE;
cairo_bool_t have_result = FALSE;
void *closure;
double width, height;
cairo_bool_t have_output_dir;
#if HAVE_MEMFAULT
int malloc_failure_iterations = ctx->malloc_failure;
int last_fault_count = 0;
#endif
/* Get the strings ready that we'll need. */
format = cairo_boilerplate_content_name (target->content);
if (dev_offset)
xasprintf (&offset_str, ".%d", dev_offset);
else
offset_str = (char *) empty_str;
xasprintf (&base_name, "%s.%s.%s%s%s",
ctx->test_name,
target->name,
format,
similar ? ".similar" : "",
offset_str);
if (offset_str != empty_str)
free (offset_str);
ref_png_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
target->name,
target->basename,
format,
CAIRO_TEST_REF_SUFFIX,
CAIRO_TEST_PNG_EXTENSION);
new_png_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
target->name,
target->basename,
format,
CAIRO_TEST_NEW_SUFFIX,
CAIRO_TEST_PNG_EXTENSION);
xfail_png_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
target->name,
target->basename,
format,
CAIRO_TEST_XFAIL_SUFFIX,
CAIRO_TEST_PNG_EXTENSION);
base_ref_png_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
NULL, NULL,
format,
CAIRO_TEST_REF_SUFFIX,
CAIRO_TEST_PNG_EXTENSION);
base_new_png_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
NULL, NULL,
format,
CAIRO_TEST_NEW_SUFFIX,
CAIRO_TEST_PNG_EXTENSION);
base_xfail_png_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
NULL, NULL,
format,
CAIRO_TEST_XFAIL_SUFFIX,
CAIRO_TEST_PNG_EXTENSION);
if (target->file_extension != NULL) {
ref_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
target->name,
target->basename,
format,
CAIRO_TEST_REF_SUFFIX,
target->file_extension);
new_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
target->name,
target->basename,
format,
CAIRO_TEST_NEW_SUFFIX,
target->file_extension);
xfail_path = cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
target->name,
target->basename,
format,
CAIRO_TEST_XFAIL_SUFFIX,
target->file_extension);
}
have_output_dir = _cairo_test_mkdir (ctx->output);
xasprintf (&base_path, "%s/%s",
have_output_dir ? ctx->output : ".",
base_name);
xasprintf (&out_png_path, "%s" CAIRO_TEST_OUT_PNG, base_path);
xasprintf (&diff_png_path, "%s" CAIRO_TEST_DIFF_PNG, base_path);
if (ctx->test->requirements != NULL) {
const char *required;
required = target->is_vector ? "target=raster" : "target=vector";
if (strstr (ctx->test->requirements, required) != NULL) {
cairo_test_log (ctx, "Error: Skipping for %s target %s\n",
target->is_vector ? "vector" : "raster",
target->name);
ret = CAIRO_TEST_UNTESTED;
goto UNWIND_STRINGS;
}
required = target->is_recording ? "target=!recording" : "target=recording";
if (strstr (ctx->test->requirements, required) != NULL) {
cairo_test_log (ctx, "Error: Skipping for %s target %s\n",
target->is_recording ? "recording" : "non-recording",
target->name);
ret = CAIRO_TEST_UNTESTED;
goto UNWIND_STRINGS;
}
}
width = ctx->test->width;
height = ctx->test->height;
if (width && height) {
width += dev_offset;
height += dev_offset;
}
#if HAVE_MEMFAULT
REPEAT:
MEMFAULT_CLEAR_FAULTS ();
MEMFAULT_RESET_LEAKS ();
ctx->last_fault_count = 0;
last_fault_count = MEMFAULT_COUNT_FAULTS ();
/* Pre-initialise fontconfig so that the configuration is loaded without
* malloc failures (our primary goal is to test cairo fault tolerance).
*/
#if HAVE_FCINIT
FcInit ();
#endif
MEMFAULT_ENABLE_FAULTS ();
#endif
have_output = FALSE;
have_result = FALSE;
/* Run the actual drawing code. */
ret = CAIRO_TEST_SUCCESS;
surface = (target->create_surface) (base_path,
target->content,
width, height,
ctx->test->width + 25 * NUM_DEVICE_OFFSETS,
ctx->test->height + 25 * NUM_DEVICE_OFFSETS,
CAIRO_BOILERPLATE_MODE_TEST,
&closure);
if (surface == NULL) {
cairo_test_log (ctx, "Error: Failed to set %s target\n", target->name);
ret = CAIRO_TEST_UNTESTED;
goto UNWIND_STRINGS;
}
#if HAVE_MEMFAULT
if (ctx->malloc_failure &&
MEMFAULT_COUNT_FAULTS () - last_fault_count > 0 &&
cairo_surface_status (surface) == CAIRO_STATUS_NO_MEMORY)
{
goto REPEAT;
}
#endif
if (cairo_surface_status (surface)) {
MF (MEMFAULT_PRINT_FAULTS ());
cairo_test_log (ctx, "Error: Created an error surface: %s\n",
cairo_status_to_string (cairo_surface_status (surface)));
ret = CAIRO_TEST_FAILURE;
goto UNWIND_STRINGS;
}
/* Check that we created a surface of the expected type. */
if (cairo_surface_get_type (surface) != target->expected_type) {
MF (MEMFAULT_PRINT_FAULTS ());
cairo_test_log (ctx, "Error: Created surface is of type %d (expected %d)\n",
cairo_surface_get_type (surface), target->expected_type);
ret = CAIRO_TEST_UNTESTED;
goto UNWIND_SURFACE;
}
/* Check that we created a surface of the expected content,
* (ignore the artificial CAIRO_TEST_CONTENT_COLOR_ALPHA_FLATTENED value).
*/
expected_content = cairo_boilerplate_content (target->content);
if (cairo_surface_get_content (surface) != expected_content) {
MF (MEMFAULT_PRINT_FAULTS ());
cairo_test_log (ctx, "Error: Created surface has content %d (expected %d)\n",
cairo_surface_get_content (surface), expected_content);
ret = CAIRO_TEST_FAILURE;
goto UNWIND_SURFACE;
}
if (cairo_surface_set_user_data (surface,
&cairo_boilerplate_output_basename_key,
base_path,
NULL))
{
#if HAVE_MEMFAULT
cairo_surface_destroy (surface);
if (target->cleanup)
target->cleanup (closure);
goto REPEAT;
#else
ret = CAIRO_TEST_FAILURE;
goto UNWIND_SURFACE;
#endif
}
cairo_surface_set_device_offset (surface, dev_offset, dev_offset);
cr = cairo_create (surface);
if (cairo_set_user_data (cr, &_cairo_test_context_key, (void*) ctx, NULL)) {
#if HAVE_MEMFAULT
cairo_destroy (cr);
cairo_surface_destroy (surface);
if (target->cleanup)
target->cleanup (closure);
goto REPEAT;
#else
ret = CAIRO_TEST_FAILURE;
goto UNWIND_CAIRO;
#endif
}
if (similar)
cairo_push_group_with_content (cr, expected_content);
/* Clear to transparent (or black) depending on whether the target
* surface supports alpha. */
cairo_save (cr);
cairo_set_operator (cr, CAIRO_OPERATOR_CLEAR);
cairo_paint (cr);
cairo_restore (cr);
/* Set all components of font_options to avoid backend differences
* and reduce number of needed reference images. */
font_options = cairo_font_options_create ();
cairo_font_options_set_hint_style (font_options, CAIRO_HINT_STYLE_NONE);
cairo_font_options_set_hint_metrics (font_options, CAIRO_HINT_METRICS_ON);
cairo_font_options_set_antialias (font_options, CAIRO_ANTIALIAS_GRAY);
cairo_set_font_options (cr, font_options);
cairo_font_options_destroy (font_options);
cairo_save (cr);
alarm (ctx->timeout);
status = (ctx->test->draw) (cr, ctx->test->width, ctx->test->height);
alarm (0);
cairo_restore (cr);
if (similar) {
cairo_pop_group_to_source (cr);
cairo_set_operator (cr, CAIRO_OPERATOR_SOURCE);
cairo_paint (cr);
}
#if HAVE_MEMFAULT
MEMFAULT_DISABLE_FAULTS ();
/* repeat test after malloc failure injection */
if (ctx->malloc_failure &&
MEMFAULT_COUNT_FAULTS () - last_fault_count > 0 &&
(status == CAIRO_TEST_NO_MEMORY ||
cairo_status (cr) == CAIRO_STATUS_NO_MEMORY ||
cairo_surface_status (surface) == CAIRO_STATUS_NO_MEMORY))
{
cairo_destroy (cr);
cairo_surface_destroy (surface);
if (target->cleanup)
target->cleanup (closure);
cairo_debug_reset_static_data ();
#if HAVE_FCFINI
FcFini ();
#endif
if (MEMFAULT_COUNT_LEAKS () > 0) {
MEMFAULT_PRINT_FAULTS ();
MEMFAULT_PRINT_LEAKS ();
}
goto REPEAT;
}
#endif
/* Then, check all the different ways it could fail. */
if (status) {
cairo_test_log (ctx, "Error: Function under test failed\n");
ret = status;
goto UNWIND_CAIRO;
}
#if HAVE_MEMFAULT
if (MEMFAULT_COUNT_FAULTS () - last_fault_count > 0 &&
MEMFAULT_HAS_FAULTS ())
{
VALGRIND_PRINTF ("Unreported memfaults...");
MEMFAULT_PRINT_FAULTS ();
}
#endif
if (target->finish_surface != NULL) {
#if HAVE_MEMFAULT
/* We need to re-enable faults as most recording-surface processing
* is done during cairo_surface_finish().
*/
MEMFAULT_CLEAR_FAULTS ();
last_fault_count = MEMFAULT_COUNT_FAULTS ();
MEMFAULT_ENABLE_FAULTS ();
#endif
/* also check for infinite loops whilst replaying */
alarm (ctx->timeout);
status = target->finish_surface (surface);
alarm (0);
#if HAVE_MEMFAULT
MEMFAULT_DISABLE_FAULTS ();
if (ctx->malloc_failure &&
MEMFAULT_COUNT_FAULTS () - last_fault_count > 0 &&
status == CAIRO_STATUS_NO_MEMORY)
{
cairo_destroy (cr);
cairo_surface_destroy (surface);
if (target->cleanup)
target->cleanup (closure);
cairo_debug_reset_static_data ();
#if HAVE_FCFINI
FcFini ();
#endif
if (MEMFAULT_COUNT_LEAKS () > 0) {
MEMFAULT_PRINT_FAULTS ();
MEMFAULT_PRINT_LEAKS ();
}
goto REPEAT;
}
#endif
if (status) {
cairo_test_log (ctx, "Error: Failed to finish surface: %s\n",
cairo_status_to_string (status));
ret = CAIRO_TEST_FAILURE;
goto UNWIND_CAIRO;
}
}
/* Skip image check for tests with no image (width,height == 0,0) */
if (ctx->test->width != 0 && ctx->test->height != 0) {
cairo_surface_t *ref_image;
cairo_surface_t *test_image;
cairo_surface_t *diff_image;
buffer_diff_result_t result;
cairo_status_t diff_status;
if (ref_png_path == NULL) {
cairo_test_log (ctx, "Error: Cannot find reference image for %s\n",
base_name);
/* we may be running this test to generate reference images */
_xunlink (ctx, out_png_path);
/* be more generous as we may need to use external renderers */
alarm (4 * ctx->timeout);
test_image = target->get_image_surface (surface, 0,
ctx->test->width,
ctx->test->height);
alarm (0);
diff_status = cairo_surface_write_to_png (test_image, out_png_path);
cairo_surface_destroy (test_image);
if (diff_status) {
if (cairo_surface_status (test_image) == CAIRO_STATUS_INVALID_STATUS)
ret = CAIRO_TEST_CRASHED;
else
ret = CAIRO_TEST_FAILURE;
cairo_test_log (ctx,
"Error: Failed to write output image: %s\n",
cairo_status_to_string (diff_status));
}
have_output = TRUE;
ret = CAIRO_TEST_XFAILURE;
goto UNWIND_CAIRO;
}
if (target->file_extension != NULL) { /* compare vector surfaces */
char *filenames[] = {
ref_png_path,
ref_path,
new_png_path,
new_path,
xfail_png_path,
xfail_path,
base_ref_png_path,
base_new_png_path,
base_xfail_png_path,
};
xasprintf (&test_filename, "%s.out%s",
base_path, target->file_extension);
xasprintf (&pass_filename, "%s.pass%s",
base_path, target->file_extension);
xasprintf (&fail_filename, "%s.fail%s",
base_path, target->file_extension);
if (cairo_test_file_is_older (pass_filename,
filenames,
ARRAY_SIZE (filenames)))
{
_xunlink (ctx, pass_filename);
}
if (cairo_test_file_is_older (fail_filename,
filenames,
ARRAY_SIZE (filenames)))
{
_xunlink (ctx, fail_filename);
}
if (cairo_test_files_equal (out_png_path, ref_path)) {
cairo_test_log (ctx, "Vector surface matches reference.\n");
have_output = FALSE;
ret = CAIRO_TEST_SUCCESS;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, new_path)) {
cairo_test_log (ctx, "Vector surface matches current failure.\n");
have_output = FALSE;
ret = CAIRO_TEST_NEW;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, xfail_path)) {
cairo_test_log (ctx, "Vector surface matches known failure.\n");
have_output = FALSE;
ret = CAIRO_TEST_XFAILURE;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (test_filename, pass_filename)) {
/* identical output as last known PASS */
cairo_test_log (ctx, "Vector surface matches last pass.\n");
have_output = TRUE;
ret = CAIRO_TEST_SUCCESS;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (test_filename, fail_filename)) {
/* identical output as last known FAIL, fail */
cairo_test_log (ctx, "Vector surface matches last fail.\n");
have_result = TRUE; /* presume these were kept around as well */
have_output = TRUE;
ret = CAIRO_TEST_FAILURE;
goto UNWIND_CAIRO;
}
}
/* be more generous as we may need to use external renderers */
alarm (4 * ctx->timeout);
test_image = target->get_image_surface (surface, 0,
ctx->test->width,
ctx->test->height);
alarm (0);
if (cairo_surface_status (test_image)) {
cairo_test_log (ctx, "Error: Failed to extract image: %s\n",
cairo_status_to_string (cairo_surface_status (test_image)));
if (cairo_surface_status (test_image) == CAIRO_STATUS_INVALID_STATUS)
ret = CAIRO_TEST_CRASHED;
else
ret = CAIRO_TEST_FAILURE;
cairo_surface_destroy (test_image);
goto UNWIND_CAIRO;
}
_xunlink (ctx, out_png_path);
diff_status = cairo_surface_write_to_png (test_image, out_png_path);
if (diff_status) {
cairo_test_log (ctx, "Error: Failed to write output image: %s\n",
cairo_status_to_string (diff_status));
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_FAILURE;
goto UNWIND_CAIRO;
}
have_output = TRUE;
/* binary compare png files (no decompression) */
if (target->file_extension == NULL) {
char *filenames[] = {
ref_png_path,
new_png_path,
xfail_png_path,
base_ref_png_path,
base_new_png_path,
base_xfail_png_path,
};
xasprintf (&test_filename, "%s", out_png_path);
xasprintf (&pass_filename, "%s.pass.png", base_path);
xasprintf (&fail_filename, "%s.fail.png", base_path);
if (cairo_test_file_is_older (pass_filename,
filenames,
ARRAY_SIZE (filenames)))
{
_xunlink (ctx, pass_filename);
}
if (cairo_test_file_is_older (fail_filename,
filenames,
ARRAY_SIZE (filenames)))
{
_xunlink (ctx, fail_filename);
}
if (cairo_test_files_equal (test_filename, pass_filename)) {
cairo_test_log (ctx, "PNG file exactly matches last pass.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_SUCCESS;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, ref_png_path)) {
cairo_test_log (ctx, "PNG file exactly matches reference image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_SUCCESS;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, new_png_path)) {
cairo_test_log (ctx, "PNG file exactly matches current failure image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_NEW;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, xfail_png_path)) {
cairo_test_log (ctx, "PNG file exactly matches known failure image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_XFAILURE;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (test_filename, fail_filename)) {
cairo_test_log (ctx, "PNG file exactly matches last fail.\n");
have_result = TRUE; /* presume these were kept around as well */
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_FAILURE;
goto UNWIND_CAIRO;
}
} else {
if (cairo_test_files_equal (out_png_path, ref_png_path)) {
cairo_test_log (ctx, "PNG file exactly matches reference image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_SUCCESS;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, new_png_path)) {
cairo_test_log (ctx, "PNG file exactly matches current failure image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_NEW;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, xfail_png_path)) {
cairo_test_log (ctx, "PNG file exactly matches known failure image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_XFAILURE;
goto UNWIND_CAIRO;
}
}
if (cairo_test_files_equal (out_png_path, base_ref_png_path)) {
cairo_test_log (ctx, "PNG file exactly reference image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_SUCCESS;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, base_new_png_path)) {
cairo_test_log (ctx, "PNG file exactly current failure image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_NEW;
goto UNWIND_CAIRO;
}
if (cairo_test_files_equal (out_png_path, base_xfail_png_path)) {
cairo_test_log (ctx, "PNG file exactly known failure image.\n");
have_result = TRUE;
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_XFAILURE;
goto UNWIND_CAIRO;
}
/* first compare against the ideal reference */
ref_image = cairo_test_get_reference_image (ctx, base_ref_png_path,
target->content == CAIRO_TEST_CONTENT_COLOR_ALPHA_FLATTENED);
if (cairo_surface_status (ref_image)) {
cairo_test_log (ctx, "Error: Cannot open reference image for %s: %s\n",
base_ref_png_path,
cairo_status_to_string (cairo_surface_status (ref_image)));
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_FAILURE;
goto UNWIND_CAIRO;
}
diff_image = cairo_image_surface_create (CAIRO_FORMAT_ARGB32,
ctx->test->width,
ctx->test->height);
cmp_png_path = base_ref_png_path;
diff_status = image_diff (ctx,
test_image, ref_image, diff_image,
&result);
_xunlink (ctx, diff_png_path);
if (diff_status ||
image_diff_is_failure (&result, target->error_tolerance))
{
/* that failed, so check against the specific backend */
ref_image = cairo_test_get_reference_image (ctx, ref_png_path,
target->content == CAIRO_TEST_CONTENT_COLOR_ALPHA_FLATTENED);
if (cairo_surface_status (ref_image)) {
cairo_test_log (ctx, "Error: Cannot open reference image for %s: %s\n",
ref_png_path,
cairo_status_to_string (cairo_surface_status (ref_image)));
cairo_surface_destroy (test_image);
ret = CAIRO_TEST_FAILURE;
goto UNWIND_CAIRO;
}
cmp_png_path = ref_png_path;
diff_status = image_diff (ctx,
test_image, ref_image,
diff_image,
&result);
if (diff_status)
{
cairo_test_log (ctx, "Error: Failed to compare images: %s\n",
cairo_status_to_string (diff_status));
ret = CAIRO_TEST_FAILURE;
}
else if (image_diff_is_failure (&result, target->error_tolerance))
{
ret = CAIRO_TEST_FAILURE;
diff_status = cairo_surface_write_to_png (diff_image,
diff_png_path);
if (diff_status) {
cairo_test_log (ctx, "Error: Failed to write differences image: %s\n",
cairo_status_to_string (diff_status));
} else {
have_result = TRUE;
}
cairo_test_copy_file (test_filename, fail_filename);
}
else
{ /* success */
cairo_test_copy_file (test_filename, pass_filename);
}
}
else
{ /* success */
cairo_test_copy_file (test_filename, pass_filename);
}
/* If failed, compare against the current image output,
* and attempt to detect systematic failures.
*/
if (ret == CAIRO_TEST_FAILURE) {
char *image_out_path;
image_out_path =
cairo_test_reference_filename (ctx,
base_name,
ctx->test_name,
"image",
"image",
format,
CAIRO_TEST_OUT_SUFFIX,
CAIRO_TEST_PNG_EXTENSION);
if (image_out_path != NULL) {
if (cairo_test_files_equal (out_png_path,
image_out_path))
{
ret = CAIRO_TEST_XFAILURE;
}
else
{
ref_image =
cairo_image_surface_create_from_png (image_out_path);
if (cairo_surface_status (ref_image) == CAIRO_STATUS_SUCCESS)
{
diff_status = image_diff (ctx,
test_image, ref_image,
diff_image,
&result);
if (diff_status == CAIRO_STATUS_SUCCESS &&
!image_diff_is_failure (&result, target->error_tolerance))
{
ret = CAIRO_TEST_XFAILURE;
}
cairo_surface_destroy (ref_image);
}
}
free (image_out_path);
}
}
cairo_surface_destroy (test_image);
cairo_surface_destroy (diff_image);
}
if (cairo_status (cr) != CAIRO_STATUS_SUCCESS) {
cairo_test_log (ctx, "Error: Function under test left cairo status in an error state: %s\n",
cairo_status_to_string (cairo_status (cr)));
ret = CAIRO_TEST_ERROR;
goto UNWIND_CAIRO;
}
UNWIND_CAIRO:
free (test_filename);
free (fail_filename);
free (pass_filename);
test_filename = fail_filename = pass_filename = NULL;
#if HAVE_MEMFAULT
if (ret == CAIRO_TEST_FAILURE)
MEMFAULT_PRINT_FAULTS ();
#endif
cairo_destroy (cr);
UNWIND_SURFACE:
cairo_surface_destroy (surface);
if (target->cleanup)
target->cleanup (closure);
#if HAVE_MEMFAULT
cairo_debug_reset_static_data ();
#if HAVE_FCFINI
FcFini ();
#endif
if (MEMFAULT_COUNT_LEAKS () > 0) {
if (ret != CAIRO_TEST_FAILURE)
MEMFAULT_PRINT_FAULTS ();
MEMFAULT_PRINT_LEAKS ();
}
if (ret == CAIRO_TEST_SUCCESS && --malloc_failure_iterations > 0)
goto REPEAT;
#endif
if (have_output)
cairo_test_log (ctx, "OUTPUT: %s\n", out_png_path);
if (have_result) {
if (cmp_png_path == NULL) {
/* XXX presume we matched the normal ref last time */
cmp_png_path = ref_png_path;
}
cairo_test_log (ctx,
"REFERENCE: %s\nDIFFERENCE: %s\n",
cmp_png_path, diff_png_path);
}
UNWIND_STRINGS:
free (out_png_path);
free (ref_png_path);
free (base_ref_png_path);
free (ref_path);
free (new_png_path);
free (base_new_png_path);
free (new_path);
free (xfail_png_path);
free (base_xfail_png_path);
free (xfail_path);
free (diff_png_path);
free (base_path);
free (base_name);
return ret;
}
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuMemcpy2DAsync)(const CUDA_MEMCPY2D *pCopy, CUstream hStream) {
int error = 0;
long vgErrorAddress;
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&hStream, sizeof(CUstream));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'hStream' in call to cuMemcpy2DAsync not defined.\n");
}
cgLock();
CUcontext ctx = NULL;
cgGetCtx(&ctx);
// Check if destination (device) memory/array is already being written to.
switch (pCopy->dstMemoryType) {
case CU_MEMORYTYPE_DEVICE: {
cgMemListType *nodeMem;
nodeMem = cgFindMem(cgFindCtx(ctx), pCopy->dstDevice);
if (nodeMem) {
// Are we trying to read a memory region that's being written by diffrent stream?
if (nodeMem->locked & 2 && nodeMem->stream != hStream) {
error++;
VALGRIND_PRINTF("Error: Concurrent write and read access by different streams.\n");
}
nodeMem->locked = nodeMem->locked | 1;
nodeMem->stream = hStream;
}
break;
}
case CU_MEMORYTYPE_ARRAY: {
cgArrListType *nodeArr;
nodeArr = cgFindArr(cgFindCtx(ctx), pCopy->dstArray);
if (nodeArr) {
// Are we trying to read an array that's being written by different stream?
if (nodeArr->locked & 2 && nodeArr->stream != hStream) {
error++;
VALGRIND_PRINTF("Error: Concurrent write and read access to array by different streams.\n");
}
nodeArr->locked = nodeArr->locked | 1;
nodeArr->stream = hStream;
}
break;
}
}
// Check if source (device) memory/array is already being written to/read from.
switch (pCopy->srcMemoryType) {
case CU_MEMORYTYPE_DEVICE: {
cgMemListType *nodeMem;
nodeMem = cgFindMem(cgFindCtx(ctx), pCopy->srcDevice);
if (nodeMem) {
// Are we trying to read a memory region that's being written by diffrent stream?
if (nodeMem->locked && nodeMem->stream != hStream) {
error++;
VALGRIND_PRINTF("Error: Concurrent write and read access by different streams.\n");
}
nodeMem->locked = nodeMem->locked | 2;
nodeMem->stream = hStream;
}
break;
}
case CU_MEMORYTYPE_ARRAY: {
cgArrListType *nodeArr;
nodeArr = cgFindArr(cgFindCtx(ctx), pCopy->srcArray);
if (nodeArr) {
// Are we trying to read an array that's being written by different stream?
if (nodeArr->locked && nodeArr->stream != hStream) {
error++;
VALGRIND_PRINTF("Error: Concurrent write and read access to array by different streams.\n");
}
nodeArr->locked = nodeArr->locked | 2;
nodeArr->stream = hStream;
}
break;
}
}
cgUnlock();
if (error) {
VALGRIND_PRINTF_BACKTRACE("");
}
return cuMemcpy2D(pCopy);
}
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;
}
