Merge pull request #155 from wsttiger/master
Modify hip headers such that they are more compatible with strict C compiler
[ROCm/clr commit: 01163d327b]
Cette révision appartient à :
Maneesh Gupta
GitHub
@@ -247,6 +247,7 @@ while (@ARGV) {
|
||||
$ft{'err'} += s/\bcudaErrorInvalidValue\b/hipErrorInvalidValue/g;
|
||||
$ft{'err'} += s/\bcudaErrorInvalidResourceHandle\b/hipErrorInvalidResourceHandle/g;
|
||||
$ft{'err'} += s/\bcudaErrorInvalidDevice\b/hipErrorInvalidDevice/g;
|
||||
$ft{'err'} += s/\bcudaErrorInvalidDevicePointer\b/hipErrorInvalidDevicePointer/g;
|
||||
$ft{'err'} += s/\bcudaErrorNoDevice\b/hipErrorNoDevice/g;
|
||||
$ft{'err'} += s/\bcudaErrorNotReady\b/hipErrorNotReady/g;
|
||||
$ft{'err'} += s/\bcudaErrorUnknown\b/hipErrorUnknown/g;
|
||||
@@ -280,6 +281,7 @@ while (@ARGV) {
|
||||
$ft{'mem'} += s/\bcudaMemcpyKind\b/hipMemcpyKind/g;
|
||||
|
||||
$ft{'mem'} += s/\bcudaPointerAttributes\b/hipPointerAttribute_t/g;
|
||||
$ft{'mem'} += s/\bcudaPointerGetAttributes\b/hipPointerGetAttribute_t/g;
|
||||
|
||||
$ft{'mem'} += s/\bcudaMemcpy2D\b/hipMemcpy2D/g;
|
||||
$ft{'mem'} += s/\bcudaMemcpy2DToArray\b/hipMemcpy2DToArray/g;
|
||||
@@ -339,6 +341,7 @@ while (@ARGV) {
|
||||
$ft{'event'} += s/\bcudaEventElapsedTime\b/hipEventElapsedTime/g;
|
||||
$ft{'event'} += s/\bcudaEventSynchronize\b/hipEventSynchronize/g;
|
||||
$ft{'event'} += s/\bcudaEventDisableTiming\b/hipEventDisableTiming/g;
|
||||
$ft{'event'} += s/\bcudaEventQuery\b/hipEventQuery/g;
|
||||
|
||||
#--------
|
||||
# Streams
|
||||
@@ -489,15 +492,15 @@ while (@ARGV) {
|
||||
my $kernelName;
|
||||
|
||||
# Handle the <<numBlocks, blockDim, sharedSize, stream>>> syntax:
|
||||
$k += s/(\w+)\s*(<.*>)?\s*<<<\s*(.+)\s*,\s*(.+)\s*,\s*(.+)\s*,\s*(.+)\s*>>>([\s*\\]*)\(/hipLaunchKernel(HIP_KERNEL_NAME($1$2), dim3($3), dim3($4), $5, $6, /g;
|
||||
$k += s/(\w+)\s*(<.*>)?\s*<<<\s*(.+)\s*,\s*(.+)\s*,\s*(.+)\s*,\s*(.+)\s*>>>([\s*\\]*)\(/hipLaunchKernelGGL(($1$2), dim3($3), dim3($4), $5, $6, /g;
|
||||
$kernelName = $1 if $k;
|
||||
|
||||
# Handle the <<numBlocks, blockDim, sharedSize>>> syntax:
|
||||
$k += s/(\w+)\s*(<.*>)?\s*<<<\s*(.+)\s*,\s*(.+)\s*,\s*(.+)\s*>>>([\s*\\]*)\(/hipLaunchKernel(HIP_KERNEL_NAME($1$2), dim3($3), dim3($4), $5, 0, /g;
|
||||
$k += s/(\w+)\s*(<.*>)?\s*<<<\s*(.+)\s*,\s*(.+)\s*,\s*(.+)\s*>>>([\s*\\]*)\(/hipLaunchKernelGGL(($1$2), dim3($3), dim3($4), $5, 0, /g;
|
||||
$kernelName = $1 if $k;
|
||||
|
||||
# Handle the <<numBlocks, blockDim>>> syntax:
|
||||
$k += s/(\w+)\s*(<.*>)?\s*<<<\s*(.+)\s*,\s*(.+)\s*>>>([\s\\]*)\(/hipLaunchKernel(HIP_KERNEL_NAME($1$2), dim3($3), dim3($4), 0, 0, /g;
|
||||
$k += s/(\w+)\s*(<.*>)?\s*<<<\s*(.+)\s*,\s*(.+)\s*>>>([\s\\]*)\(/hipLaunchKernelGGL(($1$2), dim3($3), dim3($4), 0, 0, /g;
|
||||
$kernelName = $1 if $k;
|
||||
|
||||
$ft{'kern'} += $k;
|
||||
|
||||
@@ -161,6 +161,7 @@ typedef enum hipError_t {
|
||||
hipErrorProfilerNotInitialized = 6,
|
||||
hipErrorProfilerAlreadyStarted = 7,
|
||||
hipErrorProfilerAlreadyStopped = 8,
|
||||
hipErrorInsufficientDriver = 35,
|
||||
hipErrorInvalidImage = 200,
|
||||
hipErrorInvalidContext = 201, ///< Produced when input context is invalid.
|
||||
hipErrorContextAlreadyCurrent = 202,
|
||||
|
||||
@@ -31,6 +31,13 @@ THE SOFTWARE.
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
#ifdef __cplusplus
|
||||
#define __dparm(x) \
|
||||
= x
|
||||
#else
|
||||
#define __dparm(x)
|
||||
#endif
|
||||
|
||||
//TODO -move to include/hip_runtime_api.h as a common implementation.
|
||||
/**
|
||||
* Memory copy types
|
||||
@@ -102,8 +109,8 @@ typedef cudaEvent_t hipEvent_t;
|
||||
typedef cudaStream_t hipStream_t;
|
||||
typedef cudaIpcEventHandle_t hipIpcEventHandle_t;
|
||||
typedef cudaIpcMemHandle_t hipIpcMemHandle_t;
|
||||
typedef cudaLimit hipLimit_t;
|
||||
typedef cudaFuncCache hipFuncCache_t;
|
||||
typedef enum cudaLimit hipLimit_t;
|
||||
typedef enum cudaFuncCache hipFuncCache_t;
|
||||
typedef CUcontext hipCtx_t;
|
||||
typedef cudaSharedMemConfig hipSharedMemConfig;
|
||||
typedef CUfunc_cache hipFuncCache;
|
||||
@@ -112,10 +119,10 @@ typedef CUdevice hipDevice_t;
|
||||
typedef CUmodule hipModule_t;
|
||||
typedef CUfunction hipFunction_t;
|
||||
typedef CUdeviceptr hipDeviceptr_t;
|
||||
typedef cudaChannelFormatKind hipChannelFormatKind;
|
||||
typedef cudaChannelFormatDesc hipChannelFormatDesc;
|
||||
typedef cudaTextureReadMode hipTextureReadMode;
|
||||
typedef cudaArray hipArray;
|
||||
typedef enum cudaChannelFormatKind hipChannelFormatKind;
|
||||
typedef struct cudaChannelFormatDesc hipChannelFormatDesc;
|
||||
typedef enum cudaTextureReadMode hipTextureReadMode;
|
||||
typedef struct cudaArray hipArray;
|
||||
|
||||
// Flags that can be used with hipStreamCreateWithFlags
|
||||
#define hipStreamDefault cudaStreamDefault
|
||||
@@ -192,7 +199,7 @@ switch(hError) {
|
||||
}
|
||||
}
|
||||
|
||||
inline static cudaMemcpyKind hipMemcpyKindToCudaMemcpyKind(hipMemcpyKind kind) {
|
||||
inline static enum cudaMemcpyKind hipMemcpyKindToCudaMemcpyKind(hipMemcpyKind kind) {
|
||||
switch(kind) {
|
||||
case hipMemcpyHostToHost:
|
||||
return cudaMemcpyHostToHost;
|
||||
@@ -255,7 +262,7 @@ inline static hipError_t hipHostMalloc(void** ptr, size_t size, unsigned int fla
|
||||
return hipCUDAErrorTohipError(cudaHostAlloc(ptr, size, flags));
|
||||
}
|
||||
|
||||
inline static hipError_t hipMallocArray(hipArray** array, const hipChannelFormatDesc* desc, size_t width, size_t height, unsigned int flags) {
|
||||
inline static hipError_t hipMallocArray(hipArray** array, const struct hipChannelFormatDesc* desc, size_t width, size_t height, unsigned int flags) {
|
||||
return hipCUDAErrorTohipError(cudaMallocArray(array, desc, width, height, flags));
|
||||
}
|
||||
|
||||
@@ -294,8 +301,8 @@ inline static hipError_t hipSetDevice(int device) {
|
||||
|
||||
inline static hipError_t hipChooseDevice( int* device, const hipDeviceProp_t* prop )
|
||||
{
|
||||
cudaDeviceProp cdprop;
|
||||
memset(&cdprop,0x0,sizeof(cudaDeviceProp));
|
||||
struct cudaDeviceProp cdprop;
|
||||
memset(&cdprop,0x0,sizeof(struct cudaDeviceProp));
|
||||
cdprop.major= prop->major;
|
||||
cdprop.minor = prop->minor;
|
||||
cdprop.totalGlobalMem = prop->totalGlobalMem ;
|
||||
@@ -356,25 +363,24 @@ inline static hipError_t hipMemcpy(void* dst, const void* src, size_t sizeBytes,
|
||||
}
|
||||
|
||||
|
||||
inline static hipError_t hipMemcpyAsync(void* dst, const void* src, size_t sizeBytes, hipMemcpyKind copyKind, hipStream_t stream=0) {
|
||||
inline static hipError_t hipMemcpyAsync(void* dst, const void* src, size_t sizeBytes, hipMemcpyKind copyKind, hipStream_t stream __dparm(0)) {
|
||||
return hipCUDAErrorTohipError(cudaMemcpyAsync(dst, src, sizeBytes, hipMemcpyKindToCudaMemcpyKind(copyKind), stream));
|
||||
}
|
||||
|
||||
|
||||
inline static hipError_t hipMemcpyToSymbol(const void* symbol, const void* src, size_t sizeBytes, size_t offset = 0, hipMemcpyKind copyType = hipMemcpyHostToDevice) {
|
||||
inline static hipError_t hipMemcpyToSymbol(const void* symbol, const void* src, size_t sizeBytes, size_t offset __dparm(0), hipMemcpyKind copyType __dparm(hipMemcpyHostToDevice)) {
|
||||
return hipCUDAErrorTohipError(cudaMemcpyToSymbol(symbol, src, sizeBytes, offset, hipMemcpyKindToCudaMemcpyKind(copyType)));
|
||||
}
|
||||
|
||||
inline static hipError_t hipMemcpyToSymbolAsync(const void* symbol, const void* src, size_t sizeBytes, size_t offset, hipMemcpyKind copyType, hipStream_t stream = 0) {
|
||||
inline static hipError_t hipMemcpyToSymbolAsync(const void* symbol, const void* src, size_t sizeBytes, size_t offset, hipMemcpyKind copyType, hipStream_t stream __dparm(0)) {
|
||||
return hipCUDAErrorTohipError(cudaMemcpyToSymbolAsync(symbol, src, sizeBytes, offset, hipMemcpyKindToCudaMemcpyKind(copyType), stream));
|
||||
}
|
||||
|
||||
inline static hipError_t hipMemcpyFromSymbol(void *dst, const void* symbolName, size_t sizeBytes, size_t offset = 0, hipMemcpyKind kind = hipMemcpyDeviceToHost)
|
||||
inline static hipError_t hipMemcpyFromSymbol(void *dst, const void* symbolName, size_t sizeBytes, size_t offset __dparm(0), hipMemcpyKind kind __dparm(hipMemcpyDeviceToHost))
|
||||
{
|
||||
return hipCUDAErrorTohipError(cudaMemcpyFromSymbol(dst, symbolName, sizeBytes, offset, hipMemcpyKindToCudaMemcpyKind(kind)));
|
||||
}
|
||||
|
||||
inline static hipError_t hipMemcpyFromSymbolAsync(void *dst, const void* symbolName, size_t sizeBytes, size_t offset, hipMemcpyKind kind, hipStream_t stream = 0)
|
||||
inline static hipError_t hipMemcpyFromSymbolAsync(void *dst, const void* symbolName, size_t sizeBytes, size_t offset, hipMemcpyKind kind, hipStream_t stream __dparm(0))
|
||||
{
|
||||
return hipCUDAErrorTohipError(cudaMemcpyFromSymbolAsync(dst, symbolName, sizeBytes, offset, hipMemcpyKindToCudaMemcpyKind(kind), stream));
|
||||
}
|
||||
@@ -443,7 +449,7 @@ inline static hipError_t hipMemset(void* devPtr,int value, size_t count) {
|
||||
return hipCUDAErrorTohipError(cudaMemset(devPtr, value, count));
|
||||
}
|
||||
|
||||
inline static hipError_t hipMemsetAsync(void* devPtr,int value, size_t count, hipStream_t stream = 0) {
|
||||
inline static hipError_t hipMemsetAsync(void* devPtr,int value, size_t count, hipStream_t stream __dparm(0)) {
|
||||
return hipCUDAErrorTohipError(cudaMemsetAsync(devPtr, value, count, stream));
|
||||
}
|
||||
|
||||
@@ -454,7 +460,7 @@ inline static hipError_t hipMemsetD8(hipDeviceptr_t dest, unsigned char value,
|
||||
|
||||
inline static hipError_t hipGetDeviceProperties(hipDeviceProp_t *p_prop, int device)
|
||||
{
|
||||
cudaDeviceProp cdprop;
|
||||
struct cudaDeviceProp cdprop;
|
||||
cudaError_t cerror;
|
||||
cerror = cudaGetDeviceProperties(&cdprop,device);
|
||||
strncpy(p_prop->name,cdprop.name, 256);
|
||||
@@ -515,7 +521,7 @@ inline static hipError_t hipGetDeviceProperties(hipDeviceProp_t *p_prop, int dev
|
||||
|
||||
inline static hipError_t hipDeviceGetAttribute(int* pi, hipDeviceAttribute_t attr, int device)
|
||||
{
|
||||
cudaDeviceAttr cdattr;
|
||||
enum cudaDeviceAttr cdattr;
|
||||
cudaError_t cerror;
|
||||
|
||||
switch (attr) {
|
||||
@@ -591,7 +597,7 @@ inline static hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
|
||||
}
|
||||
|
||||
inline static hipError_t hipPointerGetAttributes(hipPointerAttribute_t *attributes, void* ptr){
|
||||
cudaPointerAttributes cPA;
|
||||
struct cudaPointerAttributes cPA;
|
||||
hipError_t err = hipCUDAErrorTohipError(cudaPointerGetAttributes(&cPA, ptr));
|
||||
if(err == hipSuccess){
|
||||
switch (cPA.memoryType){
|
||||
@@ -622,7 +628,7 @@ inline static hipError_t hipEventCreate( hipEvent_t* event)
|
||||
return hipCUDAErrorTohipError(cudaEventCreate(event));
|
||||
}
|
||||
|
||||
inline static hipError_t hipEventRecord( hipEvent_t event, hipStream_t stream = NULL)
|
||||
inline static hipError_t hipEventRecord( hipEvent_t event, hipStream_t stream __dparm(NULL))
|
||||
{
|
||||
return hipCUDAErrorTohipError(cudaEventRecord(event,stream));
|
||||
}
|
||||
@@ -755,18 +761,18 @@ inline static hipError_t hipMemcpyPeer ( void* dst, int dstDevice, const void*
|
||||
return hipCUDAErrorTohipError(cudaMemcpyPeer(dst, dstDevice, src, srcDevice, count));
|
||||
}
|
||||
|
||||
inline static hipError_t hipMemcpyPeerAsync ( void* dst, int dstDevice, const void* src, int srcDevice, size_t count, hipStream_t stream=0 )
|
||||
inline static hipError_t hipMemcpyPeerAsync ( void* dst, int dstDevice, const void* src, int srcDevice, size_t count, hipStream_t stream __dparm(0))
|
||||
{
|
||||
return hipCUDAErrorTohipError(cudaMemcpyPeerAsync(dst, dstDevice, src, srcDevice, count, stream));
|
||||
}
|
||||
|
||||
// Profile APIs:
|
||||
inline hipError_t hipProfilerStart()
|
||||
inline static hipError_t hipProfilerStart()
|
||||
{
|
||||
return hipCUDAErrorTohipError(cudaProfilerStart());
|
||||
}
|
||||
|
||||
inline hipError_t hipProfilerStop()
|
||||
inline static hipError_t hipProfilerStop()
|
||||
{
|
||||
return hipCUDAErrorTohipError(cudaProfilerStop());
|
||||
}
|
||||
|
||||
Référencer dans un nouveau ticket
Bloquer un utilisateur