Added Signalling Operations
This commit is contained in:
1 orang tua
308816bc5e
melakukan
d8f44e4436
9 files changed
+425
No files matched your search
@@ -222,4 +222,109 @@ __device__ void IPCContext::internal_getmem_wave(void *dest,
|
||||
ipcImpl_.ipcFence();
|
||||
}
|
||||
|
||||
__device__ void IPCContext::putmem_signal(void *dest, const void *source, size_t nelems,
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op,
|
||||
int pe) {
|
||||
putmem(dest, source, nelems, pe);
|
||||
fence();
|
||||
|
||||
switch (sig_op) {
|
||||
case ROC_SHMEM_SIGNAL_SET:
|
||||
amo_set<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
|
||||
break;
|
||||
case ROC_SHMEM_SIGNAL_ADD:
|
||||
amo_add<uint64_t>(static_cast<void*>(dest), signal, pe);
|
||||
break;
|
||||
default:
|
||||
DPRINTF("[%s] Invalid sig_op value (%d)\n", __func__, sig_op);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
__device__ void IPCContext::putmem_signal_wg(void *dest, const void *source, size_t nelems,
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op,
|
||||
int pe) {
|
||||
putmem_wg(dest, source, nelems, pe);
|
||||
fence();
|
||||
|
||||
if (is_thread_zero_in_block()) {
|
||||
switch (sig_op) {
|
||||
case ROC_SHMEM_SIGNAL_SET:
|
||||
amo_set<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
|
||||
break;
|
||||
case ROC_SHMEM_SIGNAL_ADD:
|
||||
amo_add<uint64_t>(static_cast<void*>(dest), signal, pe);
|
||||
break;
|
||||
default:
|
||||
DPRINTF("[%s] Invalid sig_op value (%d)\n", __func__, sig_op);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
__device__ void IPCContext::putmem_signal_wave(void *dest, const void *source, size_t nelems,
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op,
|
||||
int pe) {
|
||||
putmem_wave(dest, source, nelems, pe);
|
||||
fence();
|
||||
|
||||
if (is_thread_zero_in_wave()) {
|
||||
switch (sig_op) {
|
||||
case ROC_SHMEM_SIGNAL_SET:
|
||||
amo_set<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
|
||||
break;
|
||||
case ROC_SHMEM_SIGNAL_ADD:
|
||||
amo_add<uint64_t>(static_cast<void*>(dest), signal, pe);
|
||||
break;
|
||||
default:
|
||||
DPRINTF("[%s] Invalid sig_op value (%d)\n", __func__, sig_op);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
__device__ void IPCContext::putmem_signal_nbi(void *dest, const void *source, size_t nelems,
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op,
|
||||
int pe) {
|
||||
putmem_signal(dest, source, nelems, sig_addr, signal, sig_op, pe);
|
||||
}
|
||||
|
||||
__device__ void IPCContext::putmem_signal_nbi_wg(void *dest, const void *source, size_t nelems,
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op,
|
||||
int pe) {
|
||||
putmem_signal_wg(dest, source, nelems, sig_addr, signal, sig_op, pe);
|
||||
}
|
||||
|
||||
__device__ void IPCContext::putmem_signal_nbi_wave(void *dest, const void *source, size_t nelems,
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op,
|
||||
int pe) {
|
||||
putmem_signal_wave(dest, source, nelems, sig_addr, signal, sig_op, pe);
|
||||
}
|
||||
|
||||
__device__ uint64_t IPCContext::signal_fetch(const uint64_t *sig_addr) {
|
||||
uint64_t *dst = const_cast<uint64_t*>(sig_addr);
|
||||
return amo_fetch_add<uint64_t>(static_cast<void*>(dst), 0, my_pe);
|
||||
}
|
||||
|
||||
__device__ uint64_t IPCContext::signal_fetch_wg(const uint64_t *sig_addr) {
|
||||
__shared__ uint64_t value;
|
||||
if (is_thread_zero_in_block()) {
|
||||
uint64_t *dst = const_cast<uint64_t*>(sig_addr);
|
||||
value = amo_fetch_add<uint64_t>(static_cast<void*>(dst), 0, my_pe);
|
||||
}
|
||||
__threadfence_block();
|
||||
return value;
|
||||
}
|
||||
|
||||
__device__ uint64_t IPCContext::signal_fetch_wave(const uint64_t *sig_addr) {
|
||||
uint64_t value;
|
||||
if (is_thread_zero_in_wave()) {
|
||||
uint64_t *dst = const_cast<uint64_t*>(sig_addr);
|
||||
value = amo_fetch_add<uint64_t>(static_cast<void*>(dst), 0, my_pe);
|
||||
}
|
||||
__threadfence_block();
|
||||
value = __shfl(value, 0);
|
||||
return value;
|
||||
}
|
||||
|
||||
} // namespace rocshmem
|
||||
@@ -186,6 +186,27 @@ class IPCContext : public Context {
|
||||
template <typename T>
|
||||
__device__ void get_nbi_wave(T *dest, const T *source, size_t nelems, int pe);
|
||||
|
||||
#define IPC_CONTEXT_PUT_SIGNAL_DEC(SUFFIX) \
|
||||
template <typename T> \
|
||||
__device__ void put_signal##SUFFIX(T *dest, const T *source, size_t nelems, \
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op, \
|
||||
int pe); \
|
||||
\
|
||||
__device__ void putmem_signal##SUFFIX(void *dest, const void *source, size_t nelems, \
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op, \
|
||||
int pe);
|
||||
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEC()
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEC(_wg)
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEC(_wave)
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEC(_nbi)
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEC(_nbi_wg)
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEC(_nbi_wave)
|
||||
|
||||
__device__ uint64_t signal_fetch(const uint64_t *sig_addr);
|
||||
__device__ uint64_t signal_fetch_wg(const uint64_t *sig_addr);
|
||||
__device__ uint64_t signal_fetch_wave(const uint64_t *sig_addr);
|
||||
|
||||
private:
|
||||
|
||||
//context class has IpcImpl object (ipcImpl_)
|
||||
|
||||
@@ -578,6 +578,25 @@ __device__ void IPCContext::get_nbi_wave(T *dest, const T *source,
|
||||
getmem_nbi_wave(dest, source, nelems * sizeof(T), pe);
|
||||
}
|
||||
|
||||
#define IPC_CONTEXT_PUT_SIGNAL_DEF(SUFFIX) \
|
||||
template <typename T> \
|
||||
__device__ void IPCContext::put_signal##SUFFIX(T *dest, const T *source, size_t nelems, \
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op, \
|
||||
int pe) { \
|
||||
putmem_signal##SUFFIX(dest, source, nelems * sizeof(T), sig_addr, signal, sig_op, pe); \
|
||||
} \
|
||||
\
|
||||
template <typename T> \
|
||||
__device__ void IPCContext::put_signal_nbi##SUFFIX(T *dest, const T *source, size_t nelems, \
|
||||
uint64_t *sig_addr, uint64_t signal, int sig_op, \
|
||||
int pe) { \
|
||||
putmem_signal##SUFFIX(dest, source, nelems * sizeof(T), sig_addr, signal, sig_op, pe); \
|
||||
}
|
||||
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEF()
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEF(_wg)
|
||||
IPC_CONTEXT_PUT_SIGNAL_DEF(_wave)
|
||||
|
||||
} // namespace rocshmem
|
||||
|
||||
#endif // LIBRARY_SRC_IPC_CONTEXT_TMPL_DEVICE_HPP_
|
||||
Reference in new issue
Block a user