[GDA] Alltoall optimization - single warp (#319)

* Remove testing of data types
As the collective is templated, we are just testing if sizeof(T) works

* Added single threaded varients

* Applied thread puts optimization to barrier

* Apply single threaded optimization to alltoall

* This optimization only works on bnxt, so place a switch to protect it

* Handle the edge case where the thread count is smaller than the number of PEs
Šī revīzija ir iekļauta:
Yiltan
2025-11-19 14:25:29 -05:00
revīziju iesūtīja GitHub
vecāks 618bdef082
revīzija 1347d5d628
10 mainīti faili ar 236 papildinājumiem un 16 dzēšanām
+2 -2
Parādīt failu
@@ -156,7 +156,7 @@ void GDABackend::setup_host_ctx() {
void GDABackend::setup_default_ctx() {
TeamInfo *tinfo = team_tracker.get_team_world()->tinfo_wrt_world;
default_context_proxy_ = GDADefaultContextProxyT(this, tinfo);
default_context_proxy_ = GDADefaultContextProxyT(this, tinfo, gda_provider);
}
void GDABackend::setup_ctxs() {
@@ -166,7 +166,7 @@ void GDABackend::setup_ctxs() {
CHECK_HIP(hipMalloc(&ctx_array, sizeof(GDAContext) * envvar::max_num_contexts));
// 0th context is default context
for (size_t i = 0; i < envvar::max_num_contexts; i++) {
new (&ctx_array[i]) GDAContext(this, i + 1);
new (&ctx_array[i]) GDAContext(this, i + 1, gda_provider);
ctx_free_list.get()->push_back(ctx_array + i);
}
}
+88
Parādīt failu
@@ -213,6 +213,10 @@ __device__ void QueuePair::bnxt_quiet() {
}
}
__device__ void QueuePair::bnxt_quiet_single() {
poll_cq_until(sq.depth);
}
__device__ void QueuePair::bnxt_post_wqe_rma(int pe, int32_t length, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode) {
uint64_t active_lane_mask;
uint8_t active_lane_count;
@@ -301,6 +305,90 @@ __device__ void QueuePair::bnxt_post_wqe_rma(int pe, int32_t length, uintptr_t *
}
}
__device__ void QueuePair::bnxt_post_wqe_rma_single(int pe, int32_t length, uintptr_t *laddr,
uintptr_t *raddr, uint8_t opcode) {
uint64_t active_lane_mask;
uint8_t active_lane_count;
uint8_t active_lane_id;
struct bnxt_re_bsqe hdr;
struct bnxt_re_rdma rdma;
struct bnxt_re_sge sge;
struct bnxt_re_bsqe *hdr_ptr;
struct bnxt_re_rdma *rdma_ptr;
struct bnxt_re_sge *sge_ptr;
uint32_t wqe_size;
uint32_t wqe_type;
uint32_t hdr_flags;
uint32_t inline_msg;
aquire_lock(&sq.lock);
inline_msg = length <= inline_threshold &&
opcode == gda_op_rdma_write;
poll_cq_until(GDA_BNXT_WQE_SLOT_COUNT);
hdr_ptr = (struct bnxt_re_bsqe*) bnxt_re_get_hwqe(&sq, 0);
rdma_ptr = (struct bnxt_re_rdma*) bnxt_re_get_hwqe(&sq, 1);
sge_ptr = (struct bnxt_re_sge*) bnxt_re_get_hwqe(&sq, 2);
/* Populate Header Segment */
wqe_type = BNXT_RE_HDR_WT_MASK & opcode;
wqe_size = BNXT_RE_HDR_WS_MASK & GDA_BNXT_WQE_SLOT_COUNT;
hdr_flags = ((uint32_t) BNXT_RE_HDR_FLAGS_MASK)
& ((uint32_t) BNXT_RE_WR_FLAGS_SIGNALED);
if (inline_msg) {
hdr_flags |= ((uint32_t) BNXT_RE_WR_FLAGS_INLINE);
}
hdr.rsv_ws_fl_wt = (wqe_size << BNXT_RE_HDR_WS_SHIFT)
| (hdr_flags << BNXT_RE_HDR_FLAGS_SHIFT)
| wqe_type;
hdr.key_immd = 0;
hdr.lhdr.qkey_len = length;
/* Populate RDMA Segment */
rdma.rva = (uint64_t) raddr;
rdma.rkey = rkey;
if (!inline_msg) {
/* Populate SG Segment */
sge.pa = (uint64_t) laddr;
sge.lkey = lkey;
sge.length = length;
}
/* Write WQE to SQ */
memcpy(hdr_ptr, &hdr, sizeof(struct bnxt_re_bsqe));
memcpy(rdma_ptr, &rdma, sizeof(struct bnxt_re_rdma));
if (inline_msg) {
memcpy(sge_ptr, laddr, length);
} else {
memcpy(sge_ptr, &sge, sizeof(struct bnxt_re_sge));
}
/* Populate MSN Table */
bnxt_re_fill_psns_for_msntbl(&sq, length);
/* Update SQ Pointer */
bnxt_re_incr_tail(&sq, GDA_BNXT_WQE_SLOT_COUNT);
/* Ring Doorbell
* Doorbell ring must be serialized as we cannot have all threads write to the same address */
active_lane_mask = get_active_lane_mask();
active_lane_count = get_active_lane_count(active_lane_mask);
active_lane_id = get_active_lane_num(active_lane_mask);
for (int i = 0; i < active_lane_count; i++) {
if (i == active_lane_id) {
bnxt_ring_doorbell(sq.tail);
release_lock(&sq.lock);
}
}
}
__device__ uint64_t QueuePair::bnxt_post_wqe_amo(int pe, int32_t length, uintptr_t *raddr, uint8_t opcode,
int64_t atomic_data, int64_t atomic_cmp, bool fetching) {
uint64_t active_lane_mask;
+6 -1
Parādīt failu
@@ -34,7 +34,7 @@
namespace rocshmem {
__host__ GDAContext::GDAContext(Backend *b, unsigned int ctx_id)
__host__ GDAContext::GDAContext(Backend *b, unsigned int ctx_id, int gda_provider)
: Context(b, false) {
GDABackend *backend{static_cast<GDABackend *>(b)};
base_heap = backend->heap.get_heap_bases().data();
@@ -56,6 +56,7 @@ __host__ GDAContext::GDAContext(Backend *b, unsigned int ctx_id)
ipcImpl_.pes_with_ipc_avail = backend->ipcImpl.pes_with_ipc_avail;
ctx_id_ = ctx_id;
gda_provider_ = gda_provider;
}
__host__ GDAContext::~GDAContext() {
@@ -147,6 +148,10 @@ __device__ void GDAContext::pe_quiet(size_t pe) {
qps[pe].quiet();
}
__device__ void GDAContext::pe_quiet_single(size_t pe) {
qps[pe].quiet_single();
}
__device__ void *GDAContext::shmem_ptr(const void *dest, int pe) {
void *ret = nullptr;
int local_pe{-1};
+12 -1
Parādīt failu
@@ -34,7 +34,7 @@ class QueuePair;
class GDAContext : public Context {
public:
__host__ GDAContext(Backend *b, unsigned int ctx_id);
__host__ GDAContext(Backend *b, unsigned int ctx_id, int gda_provider);
__host__ ~GDAContext();
@@ -63,6 +63,7 @@ class GDAContext : public Context {
__device__ void quiet_wave();
__device__ void pe_quiet(size_t pe);
__device__ void pe_quiet_single(size_t pe);
__device__ void *shmem_ptr(const void *dest, int pe);
@@ -257,6 +258,10 @@ class GDAContext : public Context {
__device__ void alltoall_linear(rocshmem_team_t team, T *dest,
const T *source, int nelems);
template <typename T>
__device__ void alltoall_linear_thread_puts(rocshmem_team_t team, T *dest,
const T *source, int nelems);
__device__ void internal_sync(int pe, int PE_start, int stride, int PE_size,
int64_t *pSync);
@@ -272,6 +277,10 @@ class GDAContext : public Context {
__device__ void internal_direct_barrier_wg(int pe, int PE_start, int stride,
int n_pes, int64_t *pSync);
__device__ void internal_direct_barrier_wg_thread_puts(int pe, int PE_start,
int stride, int n_pes,
int64_t *pSync);
__device__ void internal_atomic_barrier(int pe, int PE_start, int stride,
int n_pes, int64_t *pSync);
@@ -298,6 +307,8 @@ class GDAContext : public Context {
*/
unsigned int ctx_id_{};
int gda_provider_{0};
public:
QueuePair *qps{nullptr};
+49
Parādīt failu
@@ -121,6 +121,55 @@ __device__ void GDAContext::internal_direct_barrier_wg(int pe, int PE_start,
}
}
__device__ void GDAContext::internal_direct_barrier_wg_thread_puts(int pe, int PE_start,
int stride, int n_pes,
int64_t *pSync) {
int64_t flag_val{1};
if (pe == PE_start) {
int tid = get_flat_block_id();
int step_size = min(get_flat_block_size(), WF_SIZE);
// Go through all PE offsets (except current offset = 0)
// and wait until they all reach
for (int j = tid + 1; j < n_pes; j+= step_size) {
wait_until(&pSync[j], ROCSHMEM_CMP_EQ, flag_val);
pSync[j] = ROCSHMEM_SYNC_VALUE;
}
__syncthreads();
// Announce to other PEs that all have reached
for (int i = tid + 1, j = PE_start + stride + tid;
i < n_pes;
i+= step_size, j += (step_size * stride)) {
uint64_t L_offset = reinterpret_cast<char*>(&pSync[0]) - base_heap[my_pe];
qps[j].put_nbi_single(base_heap[j] + L_offset, &flag_val, sizeof(long), j);
}
for (int i = tid + 1, j = PE_start + stride + tid;
i < n_pes;
i+= step_size, j += (step_size * stride)) {
pe_quiet_single(j);
}
__syncthreads();
if (is_thread_zero_in_block()) {
pSync[0] = ROCSHMEM_SYNC_VALUE;
}
} else {
if (is_thread_zero_in_block()) {
// Mark current PE offset as reached
size_t pe_offset = (pe - PE_start) / stride;
putmem(&pSync[pe_offset], &flag_val, sizeof(long), PE_start);
wait_until(&pSync[0], ROCSHMEM_CMP_EQ, flag_val);
pSync[0] = ROCSHMEM_SYNC_VALUE;
__threadfence_system();
}
}
}
__device__ void GDAContext::internal_atomic_barrier(int pe, int PE_start,
int stride, int n_pes,
int64_t *pSync) {
+36 -2
Parādīt failu
@@ -32,6 +32,7 @@
#include "gda_team.hpp"
#include "queue_pair.hpp"
#include "rocshmem_calc.hpp"
#include "backend_gda.hpp"
#include <hip/hip_runtime.h>
@@ -604,7 +605,11 @@ __device__ void GDAContext::internal_broadcast(T *dst, const T *src, int nelems,
template <typename T>
__device__ void GDAContext::alltoall(rocshmem_team_t team, T *dst,
const T *src, int nelems) {
alltoall_linear(team, dst, src, nelems);
if (gda_provider_ == GDAProvider::BNXT) {
alltoall_linear_thread_puts(team, dst, src, nelems);
} else {
alltoall_linear(team, dst, src, nelems);
}
}
template <typename T>
@@ -620,7 +625,6 @@ __device__ void GDAContext::alltoall_linear(rocshmem_team_t team, T *dst,
int wf_id = get_flat_block_id() / WF_SIZE;
int wf_count = (int) ceil((double)get_flat_block_size() / (double)WF_SIZE);
bool wf_leader = 0 == get_active_lane_num();
// Have each PE put their designated data to the other PEs
for (int j = wf_id; j < pe_size; j+= wf_count) {
@@ -637,6 +641,36 @@ __device__ void GDAContext::alltoall_linear(rocshmem_team_t team, T *dst,
internal_sync_wg(my_pe, pe_start, stride, pe_size, pSync);
}
template <typename T>
__device__ void GDAContext::alltoall_linear_thread_puts(rocshmem_team_t team, T *dst,
const T *src, int nelems) {
GDATeam *team_obj = reinterpret_cast<GDATeam *>(team);
int pe_start = team_obj->tinfo_wrt_world->pe_start;
int pe_size = team_obj->num_pes;
int stride = team_obj->tinfo_wrt_world->stride;
long *pSync = team_obj->alltoall_pSync;
int my_pe_in_team = team_obj->my_pe;
int tid = get_flat_block_id();
int step_size = min(get_flat_block_size(), WF_SIZE);
// Have each PE put their designated data to the other PEs
for (int j = tid; j < pe_size; j+= step_size) {
int dest_pe = team_obj->get_pe_in_world(j);
uint64_t L_offset = reinterpret_cast<char*>(&dst[my_pe_in_team * nelems]) - base_heap[my_pe];
qps[dest_pe].put_nbi_single(base_heap[dest_pe] + L_offset, &src[j * nelems], nelems * sizeof(T), dest_pe);
}
for (int j = tid; j < pe_size; j+= step_size) {
int dest_pe = team_obj->get_pe_in_world(j);
pe_quiet_single(dest_pe);
}
// wait until everyone has obtained their designated data
internal_direct_barrier_wg_thread_puts(my_pe, pe_start, stride, pe_size, pSync);
}
template <typename T>
__device__ void GDAContext::fcollect(rocshmem_team_t team, T *dst,
const T *src, int nelems) {
+2 -1
Parādīt failu
@@ -44,10 +44,11 @@ class GDADefaultContextProxy {
* Placement new the memory which is allocated by proxy_
*/
explicit GDADefaultContextProxy(GDABackend* backend, TeamInfo *tinfo,
int gda_provider,
size_t num_elems = 1)
: constructed_{true}, proxy_{num_elems} {
auto ctx{proxy_.get()};
new (ctx) GDAContext(reinterpret_cast<Backend*>(backend), 0);
new (ctx) GDAContext(reinterpret_cast<Backend*>(backend), 0, gda_provider);
ctx->tinfo = tinfo;
rocshmem_ctx_t local{ctx, tinfo};
set_internal_ctx(&local);
+36 -3
Parādīt failu
@@ -138,19 +138,19 @@ __device__ void QueuePair::post_wqe_rma_turn(int pe, int32_t size, uintptr_t *la
uint8_t lane = __ffsll((unsigned long long)turns) - 1;
int pe_turn = __shfl(pe, lane);
if (pe_turn == pe) {
post_wqe_rma_single(pe, size, laddr, raddr, opcode);
post_wqe_rma_mt(pe, size, laddr, raddr, opcode);
need_turn = false;
}
turns = __ballot(need_turn);
}
} else {
if (is_thread_zero_in_wave()) {
post_wqe_rma_single(pe, size, laddr, raddr, opcode);
post_wqe_rma_mt(pe, size, laddr, raddr, opcode);
}
}
}
__device__ void QueuePair::post_wqe_rma_single(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode) {
__device__ void QueuePair::post_wqe_rma_mt(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode) {
switch (gda_provider_) {
#if defined(GDA_MLX5)
case GDAProvider::MLX5:
@@ -167,6 +167,19 @@ __device__ void QueuePair::post_wqe_rma_single(int pe, int32_t size, uintptr_t *
}
}
__device__ void QueuePair::post_wqe_rma_single(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode) {
switch (gda_provider_) {
#if defined(GDA_BNXT)
case GDAProvider::BNXT:
return bnxt_post_wqe_rma_single(pe, size, laddr, raddr, opcode);
#endif
case GDAProvider::IONIC:
case GDAProvider::MLX5:
default:
assert(false /* invalid nic provider */);
}
}
__device__ uint64_t QueuePair::post_wqe_amo(int pe, int32_t size, uintptr_t *raddr, uint8_t opcode,
int64_t atomic_data, int64_t atomic_cmp, bool fetching) {
switch (gda_provider_) {
@@ -214,6 +227,20 @@ __device__ void QueuePair::quiet(Collectivity cy) {
}
}
__device__ void QueuePair::quiet_single() {
switch (gda_provider_) {
#if defined(GDA_BNXT)
case GDAProvider::BNXT:
bnxt_quiet_single();
return;
#endif
case GDAProvider::MLX5:
case GDAProvider::IONIC:
default:
assert(false /* invalid nic provider */);
}
}
/******************************************************************************
****************************** SHMEM INTERFACE *******************************
*****************************************************************************/
@@ -223,6 +250,12 @@ __device__ void QueuePair::put_nbi(void *dest, const void *source, size_t nelems
post_wqe_rma(pe, nelems, src, dst, gda_op_rdma_write, cy);
}
__device__ void QueuePair::put_nbi_single(void *dest, const void *source, size_t nelems, int pe) {
uintptr_t *src = reinterpret_cast<uintptr_t*>(const_cast<void*>(source));
uintptr_t *dst = reinterpret_cast<uintptr_t*>(dest);
post_wqe_rma_single(pe, nelems, src, dst, gda_op_rdma_write);
}
__device__ void QueuePair::get_nbi(void *dest, const void *source, size_t nelems, int pe, Collectivity cy) {
uintptr_t *src = reinterpret_cast<uintptr_t*>(const_cast<void*>(source));
uintptr_t *dst = reinterpret_cast<uintptr_t*>(dest);
+5
Parādīt failu
@@ -77,6 +77,7 @@ class QueuePair {
* @param[in] pe Destination processing element of data transmission.
*/
__device__ void put_nbi(void *dest, const void *source, size_t nelems, int pe, Collectivity cy = THREAD);
__device__ void put_nbi_single(void *dest, const void *source, size_t nelems, int pe);
/**
* @brief Create and enqueue a non-blocking get work queue entry (wqe).
@@ -92,6 +93,7 @@ class QueuePair {
* @brief Empty all completions from the completion queue.
*/
__device__ void quiet(Collectivity cy = THREAD);
__device__ void quiet_single();
/**
* @brief Create and enqueue an atomic fetch work queue entry (wqe).
@@ -165,6 +167,7 @@ class QueuePair {
__device__ __attribute__((noinline)) void post_wqe_rma(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode, Collectivity cy);
__device__ __attribute__((noinline)) void post_wqe_rma_turn(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode, Collectivity cy);
__device__ __attribute__((noinline)) void post_wqe_rma_single(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode);
__device__ __attribute__((noinline)) void post_wqe_rma_mt(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode);
#if defined(GDA_MLX5)
__device__ uint64_t mlx5_post_wqe_amo(int pe, int32_t size, uintptr_t *raddr, uint8_t opcode, int64_t atomic_data, int64_t atomic_cmp, bool fetch);
@@ -174,7 +177,9 @@ class QueuePair {
#if defined(GDA_BNXT)
__device__ uint64_t bnxt_post_wqe_amo(int pe, int32_t size, uintptr_t *raddr, uint8_t opcode, int64_t atomic_data, int64_t atomic_cmp, bool fetch);
__device__ void bnxt_post_wqe_rma(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode);
__device__ void bnxt_post_wqe_rma_single(int pe, int32_t size, uintptr_t *laddr, uintptr_t *raddr, uint8_t opcode);
__device__ void bnxt_quiet();
__device__ void bnxt_quiet_single();
#endif
#if defined(GDA_IONIC)
__device__ uint64_t ionic_post_wqe_amo(int pe, int32_t size, uintptr_t *raddr, uint8_t opcode, int64_t atomic_data, int64_t atomic_cmp, bool fetch);
-6
Parādīt failu
@@ -225,13 +225,7 @@ std::vector<Tester*> Tester::create(TesterArguments args) {
if (rank == 0) {
std::cout << "Alltoall Test ###" << std::endl;
}
testers.push_back(new TeamAlltoallTester<int64_t>(args));
testers.push_back(new TeamAlltoallTester<int>(args));
testers.push_back(new TeamAlltoallTester<long long>(args));
testers.push_back(new TeamAlltoallTester<float>(args));
testers.push_back(new TeamAlltoallTester<double>(args));
testers.push_back(new TeamAlltoallTester<char>(args));
testers.push_back(new TeamAlltoallTester<unsigned char>(args));
return testers;
case TeamFCollectTestType:
if (rank == 0) {