Files
rocm-systems/runtime/hsa-runtime/core/runtime/interrupt_signal.cpp
T
David Yat Sin 8d3fee5095 Use HybridMutex for signal mutexes
Implement HybridMutex to improve latencies compared to KernelMutex when
there is contention between several threads calling hsa_signal_create
and hsa_amd_signal_async_handler.

Change-Id: If53377033e749b0050727964c9303f09b02527cc
2024-01-16 21:29:39 +00:00

422 regels
14 KiB
C++

////////////////////////////////////////////////////////////////////////////////
//
// The University of Illinois/NCSA
// Open Source License (NCSA)
//
// Copyright (c) 2014-2020, Advanced Micro Devices, Inc. All rights reserved.
//
// Developed by:
//
// AMD Research and AMD HSA Software Development
//
// Advanced Micro Devices, Inc.
//
// www.amd.com
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to
// deal with the Software without restriction, including without limitation
// the rights to use, copy, modify, merge, publish, distribute, sublicense,
// and/or sell copies of the Software, and to permit persons to whom the
// Software is furnished to do so, subject to the following conditions:
//
// - Redistributions of source code must retain the above copyright notice,
// this list of conditions and the following disclaimers.
// - Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimers in
// the documentation and/or other materials provided with the distribution.
// - Neither the names of Advanced Micro Devices, Inc,
// nor the names of its contributors may be used to endorse or promote
// products derived from this Software without specific prior written
// permission.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
// THE CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
// OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
// ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
// DEALINGS WITH THE SOFTWARE.
//
////////////////////////////////////////////////////////////////////////////////
#include "core/inc/interrupt_signal.h"
#include "core/inc/runtime.h"
#include "core/util/timer.h"
#include "core/util/locks.h"
#if defined(__i386__) || defined(__x86_64__)
#include <mwaitxintrin.h>
#define MWAITX_ECX_TIMER_ENABLE 0x2 // BIT(1)
#endif
namespace rocr {
namespace core {
HsaEvent* InterruptSignal::EventPool::alloc() {
ScopedAcquire<HybridMutex> lock(&lock_);
if (events_.empty()) {
if (!allEventsAllocated) {
HsaEvent* evt = InterruptSignal::CreateEvent(HSA_EVENTTYPE_SIGNAL, false);
if (evt == nullptr) allEventsAllocated = true;
return evt;
}
return nullptr;
}
HsaEvent* ret = events_.back().release();
events_.pop_back();
return ret;
}
void InterruptSignal::EventPool::free(HsaEvent* evt) {
if (evt == nullptr) return;
ScopedAcquire<HybridMutex> lock(&lock_);
events_.push_back(unique_event_ptr(evt));
}
int InterruptSignal::rtti_id_ = 0;
HsaEvent* InterruptSignal::CreateEvent(HSA_EVENTTYPE type, bool manual_reset) {
HsaEventDescriptor event_descriptor;
event_descriptor.EventType = type;
event_descriptor.SyncVar.SyncVar.UserData = NULL;
event_descriptor.SyncVar.SyncVarSize = sizeof(hsa_signal_value_t);
event_descriptor.NodeId = 0;
HsaEvent* ret = NULL;
if (HSAKMT_STATUS_SUCCESS ==
hsaKmtCreateEvent(&event_descriptor, manual_reset, false, &ret)) {
if (type == HSA_EVENTTYPE_MEMORY) {
memset(&ret->EventData.EventData.MemoryAccessFault.Failure, 0,
sizeof(HsaAccessAttributeFailure));
} else if (type == HSA_EVENTTYPE_HW_EXCEPTION) {
memset(&ret->EventData.EventData.HwException, 0, sizeof(HsaHwException));
}
}
return ret;
}
void InterruptSignal::DestroyEvent(HsaEvent* evt) { hsaKmtDestroyEvent(evt); }
InterruptSignal::InterruptSignal(hsa_signal_value_t initial_value, HsaEvent* use_event)
: LocalSignal(initial_value, false), Signal(signal()) {
if (use_event != nullptr) {
event_ = use_event;
free_event_ = false;
} else {
event_ = Runtime::runtime_singleton_->GetEventPool()->alloc();
free_event_ = true;
}
if (event_ != nullptr) {
signal_.event_id = event_->EventId;
signal_.event_mailbox_ptr = event_->EventData.HWData2;
} else {
signal_.event_id = 0;
signal_.event_mailbox_ptr = 0;
}
signal_.kind = AMD_SIGNAL_KIND_USER;
}
InterruptSignal::~InterruptSignal() {
if (free_event_) Runtime::runtime_singleton_->GetEventPool()->free(event_);
}
hsa_signal_value_t InterruptSignal::LoadRelaxed() {
return hsa_signal_value_t(
atomic::Load(&signal_.value, std::memory_order_relaxed));
}
hsa_signal_value_t InterruptSignal::LoadAcquire() {
return hsa_signal_value_t(
atomic::Load(&signal_.value, std::memory_order_acquire));
}
void InterruptSignal::StoreRelaxed(hsa_signal_value_t value) {
atomic::Store(&signal_.value, int64_t(value), std::memory_order_relaxed);
SetEvent();
}
void InterruptSignal::StoreRelease(hsa_signal_value_t value) {
atomic::Store(&signal_.value, int64_t(value), std::memory_order_release);
SetEvent();
}
hsa_signal_value_t InterruptSignal::WaitRelaxed(
hsa_signal_condition_t condition, hsa_signal_value_t compare_value,
uint64_t timeout, hsa_wait_state_t wait_hint) {
Retain();
MAKE_SCOPE_GUARD([&]() { Release(); });
uint32_t prior = waiting_++;
MAKE_SCOPE_GUARD([&]() { waiting_--; });
uint64_t event_age = 1;
if (!core::Runtime::runtime_singleton_->KfdVersion().supports_event_age) {
event_age = 0;
// Allow only the first waiter to sleep. Without event age tracking,
// race condition can cause some threads to sleep without wakeup since missing interrupt.
if (prior != 0) wait_hint = HSA_WAIT_STATE_ACTIVE;
}
int64_t value;
timer::fast_clock::time_point start_time = timer::fast_clock::now();
// Set a polling timeout value
// Should be a few times bigger than null kernel latency
const timer::fast_clock::duration kMaxElapsed = std::chrono::microseconds(200);
uint64_t hsa_freq;
HSA::hsa_system_get_info(HSA_SYSTEM_INFO_TIMESTAMP_FREQUENCY, &hsa_freq);
const timer::fast_clock::duration fast_timeout =
timer::duration_from_seconds<timer::fast_clock::duration>(
double(timeout) / double(hsa_freq));
bool condition_met = false;
#if defined(__i386__) || defined(__x86_64__)
if (g_use_mwaitx) _mm_monitorx(const_cast<int64_t*>(&signal_.value), 0, 0);
#endif
while (true) {
if (!IsValid()) return 0;
value = atomic::Load(&signal_.value, std::memory_order_relaxed);
switch (condition) {
case HSA_SIGNAL_CONDITION_EQ: {
condition_met = (value == compare_value);
break;
}
case HSA_SIGNAL_CONDITION_NE: {
condition_met = (value != compare_value);
break;
}
case HSA_SIGNAL_CONDITION_GTE: {
condition_met = (value >= compare_value);
break;
}
case HSA_SIGNAL_CONDITION_LT: {
condition_met = (value < compare_value);
break;
}
default:
return 0;
}
if (condition_met) return hsa_signal_value_t(value);
timer::fast_clock::time_point time = timer::fast_clock::now();
if (time - start_time > fast_timeout) {
value = atomic::Load(&signal_.value, std::memory_order_relaxed);
return hsa_signal_value_t(value);
}
if (wait_hint == HSA_WAIT_STATE_ACTIVE) {
#if defined(__i386__) || defined(__x86_64__)
if (g_use_mwaitx) {
_mm_mwaitx(0, 0, 0);
_mm_monitorx(const_cast<int64_t*>(&signal_.value), 0, 0);
}
#endif
continue;
}
if (time - start_time < kMaxElapsed) {
// os::uSleep(20);
#if defined(__i386__) || defined(__x86_64__)
if (g_use_mwaitx) {
_mm_mwaitx(0, 60000, MWAITX_ECX_TIMER_ENABLE);
_mm_monitorx(const_cast<int64_t*>(&signal_.value), 0, 0);
}
#endif
continue;
}
uint32_t wait_ms;
auto time_remaining = fast_timeout - (time - start_time);
uint64_t ct=timer::duration_cast<std::chrono::milliseconds>(
time_remaining).count();
wait_ms = (ct>0xFFFFFFFEu) ? 0xFFFFFFFEu : ct;
hsaKmtWaitOnEvent_Ext(event_, wait_ms, &event_age);
}
}
hsa_signal_value_t InterruptSignal::WaitAcquire(
hsa_signal_condition_t condition, hsa_signal_value_t compare_value,
uint64_t timeout, hsa_wait_state_t wait_hint) {
hsa_signal_value_t ret =
WaitRelaxed(condition, compare_value, timeout, wait_hint);
std::atomic_thread_fence(std::memory_order_acquire);
return ret;
}
void InterruptSignal::AndRelaxed(hsa_signal_value_t value) {
atomic::And(&signal_.value, int64_t(value), std::memory_order_relaxed);
SetEvent();
}
void InterruptSignal::AndAcquire(hsa_signal_value_t value) {
atomic::And(&signal_.value, int64_t(value), std::memory_order_acquire);
SetEvent();
}
void InterruptSignal::AndRelease(hsa_signal_value_t value) {
atomic::And(&signal_.value, int64_t(value), std::memory_order_release);
SetEvent();
}
void InterruptSignal::AndAcqRel(hsa_signal_value_t value) {
atomic::And(&signal_.value, int64_t(value), std::memory_order_acq_rel);
SetEvent();
}
void InterruptSignal::OrRelaxed(hsa_signal_value_t value) {
atomic::Or(&signal_.value, int64_t(value), std::memory_order_relaxed);
SetEvent();
}
void InterruptSignal::OrAcquire(hsa_signal_value_t value) {
atomic::Or(&signal_.value, int64_t(value), std::memory_order_acquire);
SetEvent();
}
void InterruptSignal::OrRelease(hsa_signal_value_t value) {
atomic::Or(&signal_.value, int64_t(value), std::memory_order_release);
SetEvent();
}
void InterruptSignal::OrAcqRel(hsa_signal_value_t value) {
atomic::Or(&signal_.value, int64_t(value), std::memory_order_acq_rel);
SetEvent();
}
void InterruptSignal::XorRelaxed(hsa_signal_value_t value) {
atomic::Xor(&signal_.value, int64_t(value), std::memory_order_relaxed);
SetEvent();
}
void InterruptSignal::XorAcquire(hsa_signal_value_t value) {
atomic::Xor(&signal_.value, int64_t(value), std::memory_order_acquire);
SetEvent();
}
void InterruptSignal::XorRelease(hsa_signal_value_t value) {
atomic::Xor(&signal_.value, int64_t(value), std::memory_order_release);
SetEvent();
}
void InterruptSignal::XorAcqRel(hsa_signal_value_t value) {
atomic::Xor(&signal_.value, int64_t(value), std::memory_order_acq_rel);
SetEvent();
}
void InterruptSignal::AddRelaxed(hsa_signal_value_t value) {
atomic::Add(&signal_.value, int64_t(value), std::memory_order_relaxed);
SetEvent();
}
void InterruptSignal::AddAcquire(hsa_signal_value_t value) {
atomic::Add(&signal_.value, int64_t(value), std::memory_order_acquire);
SetEvent();
}
void InterruptSignal::AddRelease(hsa_signal_value_t value) {
atomic::Add(&signal_.value, int64_t(value), std::memory_order_release);
SetEvent();
}
void InterruptSignal::AddAcqRel(hsa_signal_value_t value) {
atomic::Add(&signal_.value, int64_t(value), std::memory_order_acq_rel);
SetEvent();
}
void InterruptSignal::SubRelaxed(hsa_signal_value_t value) {
atomic::Sub(&signal_.value, int64_t(value), std::memory_order_relaxed);
SetEvent();
}
void InterruptSignal::SubAcquire(hsa_signal_value_t value) {
atomic::Sub(&signal_.value, int64_t(value), std::memory_order_acquire);
SetEvent();
}
void InterruptSignal::SubRelease(hsa_signal_value_t value) {
atomic::Sub(&signal_.value, int64_t(value), std::memory_order_release);
SetEvent();
}
void InterruptSignal::SubAcqRel(hsa_signal_value_t value) {
atomic::Sub(&signal_.value, int64_t(value), std::memory_order_acq_rel);
SetEvent();
}
hsa_signal_value_t InterruptSignal::ExchRelaxed(hsa_signal_value_t value) {
hsa_signal_value_t ret = hsa_signal_value_t(atomic::Exchange(
&signal_.value, int64_t(value), std::memory_order_relaxed));
SetEvent();
return ret;
}
hsa_signal_value_t InterruptSignal::ExchAcquire(hsa_signal_value_t value) {
hsa_signal_value_t ret = hsa_signal_value_t(atomic::Exchange(
&signal_.value, int64_t(value), std::memory_order_acquire));
SetEvent();
return ret;
}
hsa_signal_value_t InterruptSignal::ExchRelease(hsa_signal_value_t value) {
hsa_signal_value_t ret = hsa_signal_value_t(atomic::Exchange(
&signal_.value, int64_t(value), std::memory_order_release));
SetEvent();
return ret;
}
hsa_signal_value_t InterruptSignal::ExchAcqRel(hsa_signal_value_t value) {
hsa_signal_value_t ret = hsa_signal_value_t(atomic::Exchange(
&signal_.value, int64_t(value), std::memory_order_acq_rel));
SetEvent();
return ret;
}
hsa_signal_value_t InterruptSignal::CasRelaxed(hsa_signal_value_t expected,
hsa_signal_value_t value) {
hsa_signal_value_t ret = hsa_signal_value_t(
atomic::Cas(&signal_.value, int64_t(value), int64_t(expected),
std::memory_order_relaxed));
SetEvent();
return ret;
}
hsa_signal_value_t InterruptSignal::CasAcquire(hsa_signal_value_t expected,
hsa_signal_value_t value) {
hsa_signal_value_t ret = hsa_signal_value_t(
atomic::Cas(&signal_.value, int64_t(value), int64_t(expected),
std::memory_order_acquire));
SetEvent();
return ret;
}
hsa_signal_value_t InterruptSignal::CasRelease(hsa_signal_value_t expected,
hsa_signal_value_t value) {
hsa_signal_value_t ret = hsa_signal_value_t(
atomic::Cas(&signal_.value, int64_t(value), int64_t(expected),
std::memory_order_release));
SetEvent();
return ret;
}
hsa_signal_value_t InterruptSignal::CasAcqRel(hsa_signal_value_t expected,
hsa_signal_value_t value) {
hsa_signal_value_t ret = hsa_signal_value_t(
atomic::Cas(&signal_.value, int64_t(value), int64_t(expected),
std::memory_order_acq_rel));
SetEvent();
return ret;
}
} // namespace core
} // namespace rocr