perf(engine_multiscale): performance enhancments due to improved hashing, locality, and data structure optimization

This particular commit speeds up QSE solving for systems where reverse reactions and engine caching is disabled by about 24%
This commit is contained in:
2025-10-22 15:08:49 -04:00
parent b20d3467fa
commit 0581f69c48
6 changed files with 41 additions and 47 deletions

View File

@@ -1312,11 +1312,9 @@ namespace gridfire {
if ( p != 0) { return false; }
const double T9 = tx[0];
// This is an interesting problem because the reverse rate should only ever be computed for strong reactions
// Which do not depend on rho or Y. However, the signature requires them...
// For now, we just pass dummy values for rho and Y
// We can pass a dummy comp and rho because reverse rates should only be calculated for strong reactions whose
// rates of progression do not depend on composition or density.
const double reverseRate = m_engine.calculateReverseRate(m_reaction, T9, 0.0, {});
// std::cout << m_reaction.peName() << " reverseRate: " << reverseRate << " at T9: " << T9 << "\n";
ty[0] = reverseRate; // Store the reverse rate in the output vector
if (vx.size() > 0) {
@@ -1335,9 +1333,6 @@ namespace gridfire {
const double T9 = tx[0];
const double reverseRate = ty[0];
// This is an interesting problem because the reverse rate should only ever be computed for strong reactions
// Which do not depend on rho or Y. However, the signature requires them...
// For now, we just pass dummy values for rho and Y
const double derivative = m_engine.calculateReverseRateTwoBodyDerivative(m_reaction, T9, 0, {}, reverseRate);
px[0] = py[0] * derivative; // Return the derivative of the reverse rate with respect to T9

View File

@@ -5,6 +5,7 @@
#include "gridfire/partition/partition_ground.h"
#include "gridfire/partition/partition_rauscher_thielemann.h"
#include "gridfire/utils/hashing.h"
#include "quill/LogMacros.h"
namespace gridfire::partition {
@@ -25,13 +26,14 @@ namespace gridfire::partition {
}
double CompositePartitionFunction::evaluate(int z, int a, double T9) const {
LOG_TRACE_L3(m_logger, "Evaluating partition function for Z={} A={} T9={}", z, a, T9);
const uint_fast32_t hash = utils::hash_atomic(a, z);
if (m_supportCache.contains(hash)) {
return m_supportCache.at(hash).evaluate(z, a, T9);
}
for (const auto& partitionFunction : m_partitionFunctions) {
if (partitionFunction->supports(z, a)) {
LOG_TRACE_L3(m_logger, "Partition function of type {} supports Z={} A={}", partitionFunction->type(), z, a);
m_supportCache.emplace(hash, *partitionFunction);
return partitionFunction->evaluate(z, a, T9);
} else {
LOG_TRACE_L3(m_logger, "Partition function of type {} does not support Z={} A={}", partitionFunction->type(), z, a);
}
}
LOG_ERROR(
@@ -46,9 +48,13 @@ namespace gridfire::partition {
}
double CompositePartitionFunction::evaluateDerivative(int z, int a, double T9) const {
const uint_fast32_t hash = utils::hash_atomic(a, z);
if (m_supportCache.contains(hash)) {
return m_supportCache.at(hash).evaluateDerivative(z, a, T9);
}
for (const auto& partitionFunction : m_partitionFunctions) {
if (partitionFunction->supports(z, a)) {
LOG_TRACE_L3(m_logger, "Evaluating derivative of partition function for Z={} A={} T9={}", z, a, T9);
m_supportCache.emplace(hash, *partitionFunction);
return partitionFunction->evaluateDerivative(z, a, T9);
}
}
@@ -64,9 +70,12 @@ namespace gridfire::partition {
}
bool CompositePartitionFunction::supports(int z, int a) const {
const uint_fast32_t hash = utils::hash_atomic(a, z);
if (m_supportCache.contains(hash)) {
return true;
}
for (const auto& partitionFunction : m_partitionFunctions) {
if (partitionFunction->supports(z, a)) {
LOG_TRACE_L2(m_logger, "Partition function supports Z={} A={}", z, a);
return true;
}
}

View File

@@ -1,6 +1,7 @@
#include "gridfire/reaction/weak/weak_interpolator.h"
#include "gridfire/reaction/reaction.h"
#include "gridfire/reaction/weak/weak.h"
#include "gridfire/utils/hashing.h"
#include <algorithm>
#include <map>
@@ -19,9 +20,9 @@ namespace gridfire::rates::weak {
WeakRateInterpolator::WeakRateInterpolator(const RowDataTable &raw_data) {
// Group all raw data rows by their isotope ID.
std::map<uint32_t, std::vector<const RateDataRow*>> grouped_rows;
std::unordered_map<uint32_t, std::vector<const RateDataRow*>> grouped_rows;
for (const auto& row : raw_data) {
grouped_rows[pack_isotope_id(row.A, row.Z)].push_back(&row);
grouped_rows[utils::hash_atomic(row.A, row.Z)].push_back(&row);
}
// Process each isotope's data to build a simple 2D grid.
@@ -48,22 +49,8 @@ namespace gridfire::rates::weak {
for (size_t i = 0; i < nt9; i++) { t9_map[grid.t9_axis[i]] = i; }
for (size_t j = 0; j < nrhoYe; j++) { rhoYe_map[grid.rhoYe_axis[j]] = j; }
// Use a set to detect duplicate (T9, rhoYe) pairs, which would be a data error.
std::set<std::pair<float, float>> seen_coords;
// Populate the 2D grid.
for (const auto* row: rows) {
if (auto [it, inserted] = seen_coords.insert({row->t9, row->log_rhoye}); !inserted) {
auto A = static_cast<uint16_t>(isotope_id >> 8);
auto Z = static_cast<uint8_t>(isotope_id & 0xFF);
std::string msg = std::format(
"Duplicate data point for isotope (A={}, Z={}) at (T9={}, log10(rho*Ye)={}) in weak rate table. This indicates corrupted or malformed input data and should be taken as an unrecoverable error.",
A, Z, row->t9, row->log_rhoye
);
LOG_ERROR(m_logger, "{}", msg);
throw std::runtime_error(msg);
}
size_t i_t9 = t9_map.at(row->t9);
size_t j_rhoYe = rhoYe_map.at(row->log_rhoye);
@@ -107,7 +94,7 @@ namespace gridfire::rates::weak {
const double t9,
const double log_rhoYe
) const {
const auto it = m_rate_table.find(pack_isotope_id(A, Z));
const auto it = m_rate_table.find(utils::hash_atomic(A, Z));
if (it == m_rate_table.end()) {
return std::unexpected(InterpolationError{InterpolationErrorType::UNKNOWN_SPECIES_ERROR});
}
@@ -222,9 +209,4 @@ namespace gridfire::rates::weak {
return result;
}
uint32_t WeakRateInterpolator::pack_isotope_id(const uint16_t A, const uint8_t Z) {
return (static_cast<uint32_t>(A) << 8) | static_cast<uint32_t>(Z);
}
}