AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #38 from Yiltan/ro/implement-sigops

Browse Source 
Implements Signalling Operations for RO
This commit is contained in:
Yiltan
2025-02-10 15:10:07 -05:00
committed by GitHub
parent 94144f4460 f1c25f7e19
commit 495cd6970b
5 changed files with 150 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/reverse_offload/context_ro_device.cpp
+106
View File
@@ -21,6 +21,7 @@
*****************************************************************************/
#include "context_ro_device.hpp"
#include "context_ro_tmpl_device.hpp"
#include <hip/hip_runtime.h>
#include <hip/amd_detail/amd_device_functions.h>
@@ -348,6 +349,111 @@ __device__ void ROContext::getmem_nbi_wave(void *dest, const void *source,
}
}
__device__ void ROContext::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 ROCSHMEM_SIGNAL_SET:
amo_set<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
break;
case ROCSHMEM_SIGNAL_ADD:
amo_add<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
break;
default:
DPRINTF("[%s] Invalid sig_op value (%d)\n", __func__, sig_op);
break;
}
}
__device__ void ROContext::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 ROCSHMEM_SIGNAL_SET:
amo_set<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
break;
case ROCSHMEM_SIGNAL_ADD:
amo_add<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
break;
default:
DPRINTF("[%s] Invalid sig_op value (%d)\n", __func__, sig_op);
break;
}
}
}
__device__ void ROContext::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 ROCSHMEM_SIGNAL_SET:
amo_set<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
break;
case ROCSHMEM_SIGNAL_ADD:
amo_add<uint64_t>(static_cast<void*>(sig_addr), signal, pe);
break;
default:
DPRINTF("[%s] Invalid sig_op value (%d)\n", __func__, sig_op);
break;
}
}
}
__device__ void ROContext::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 ROContext::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 ROContext::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 ROContext::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 ROContext::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 ROContext::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;
}
__device__ uint64_t number_active_lanes() {
return __popcll(__ballot(1));
}