admin/omath
1
0
Fork 0
mirror of https://github.com/orange-cpp/omath.git synced 2026-02-12 22:53:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

removed brackets

Browse Source 
Initial plan

Initial exploration and analysis complete

Co-authored-by: orange-cpp <59374393+orange-cpp@users.noreply.github.com>

Optimize performance: A* pathfinding, Vector3 hash, pattern scanner, AVX2 code, and serialization

Co-authored-by: orange-cpp <59374393+orange-cpp@users.noreply.github.com>

Add bounds check for navigation mesh serialization

Co-authored-by: orange-cpp <59374393+orange-cpp@users.noreply.github.com>

Document serialization limitation for large neighbor counts

Co-authored-by: orange-cpp <59374393+orange-cpp@users.noreply.github.com>

Add _codeql_build_dir to gitignore

Co-authored-by: orange-cpp <59374393+orange-cpp@users.noreply.github.com>

Removes codeql detected source root

Eliminates the automatically generated file that was causing issues.

This file was added by codeql and no longer needed.

revert
cleaned up gitignore

moved to anon namespace

Improves navigation mesh serialization and clamping

Ensures correct serialization of navigation meshes by clamping neighbor counts to prevent data corruption when exceeding uint16_t limits.

Updates data types to `std::uint8_t` and `std::size_t` for consistency.
Uses `std::copy_n` instead of `std::memcpy` for deserialization.
This commit is contained in:
Orange
2025-10-30 05:38:11 +03:00
parent 765d5e7216
commit eec34c1efb
5 changed files with 133 additions and 99 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
source/pathfinding/a_star.cpp
View File

@@ -4,9 +4,25 @@
#include "omath/pathfinding/a_star.hpp"
#include <algorithm>
#include <optional>
#include <queue>
#include <unordered_map>
#include <unordered_set>
namespace
{
struct OpenListNode final
{
omath::Vector3<float> position;
float f_cost;
[[nodiscard]]
bool operator>(const OpenListNode& other) const noexcept
{
return f_cost > other.f_cost;
}
};
}
namespace omath::pathfinding
{
struct PathNode final
@@ -37,23 +53,13 @@ namespace omath::pathfinding
std::ranges::reverse(path);
return path;
}
auto Astar::get_perfect_node(const std::unordered_map<Vector3<float>, PathNode>& open_list,
const Vector3<float>& end_vertex) noexcept
{
return std::ranges::min_element(open_list,
[&end_vertex](const auto& a, const auto& b)
{
const float fa = a.second.g_cost + a.first.distance_to(end_vertex);
const float fb = b.second.g_cost + b.first.distance_to(end_vertex);
return fa < fb;
});
}
std::vector<Vector3<float>> Astar::find_path(const Vector3<float>& start, const Vector3<float>& end,
const NavigationMesh& nav_mesh) noexcept
{
std::unordered_map<Vector3<float>, PathNode> closed_list;
std::unordered_map<Vector3<float>, PathNode> open_list;
std::unordered_map<Vector3<float>, PathNode> node_data;
std::priority_queue<OpenListNode, std::vector<OpenListNode>, std::greater<>> open_list;
auto maybe_start_vertex = nav_mesh.get_closest_vertex(start);
auto maybe_end_vertex = nav_mesh.get_closest_vertex(end);
@@ -64,20 +70,27 @@ namespace omath::pathfinding
const auto start_vertex = maybe_start_vertex.value();
const auto end_vertex = maybe_end_vertex.value();
open_list.emplace(start_vertex, PathNode{std::nullopt, 0.f});
node_data.emplace(start_vertex, PathNode{std::nullopt, 0.f});
open_list.push({start_vertex, start_vertex.distance_to(end_vertex)});
while (!open_list.empty())
{
auto current_it = get_perfect_node(open_list, end_vertex);
auto current = open_list.top().position;
open_list.pop();
const auto current = current_it->first;
const auto current_node = current_it->second;
if (closed_list.contains(current))
continue;
auto current_node_it = node_data.find(current);
if (current_node_it == node_data.end())
continue;
const auto current_node = current_node_it->second;
if (current == end_vertex)
return reconstruct_final_path(closed_list, current);
closed_list.emplace(current, current_node);
open_list.erase(current_it);
for (const auto& neighbor: nav_mesh.get_neighbors(current))
{
@@ -86,11 +99,14 @@ namespace omath::pathfinding
const float tentative_g_cost = current_node.g_cost + neighbor.distance_to(current);
// ReSharper disable once CppTooWideScopeInitStatement
const auto open_it = open_list.find(neighbor);
auto node_it = node_data.find(neighbor);
if (open_it == open_list.end() || tentative_g_cost < open_it->second.g_cost)
open_list[neighbor] = PathNode{current, tentative_g_cost};
if (node_it == node_data.end() || tentative_g_cost < node_it->second.g_cost)
{
node_data[neighbor] = PathNode{current, tentative_g_cost};
const float f_cost = tentative_g_cost + neighbor.distance_to(end_vertex);
open_list.push({neighbor, f_cost});
}
}
}

49
source/pathfinding/navigation_mesh.cpp
View File

@@ -3,6 +3,8 @@
//
#include "omath/pathfinding/navigation_mesh.hpp"
#include <algorithm>
#include <cstring>
#include <limits>
#include <stdexcept>
namespace omath::pathfinding
{
@@ -30,55 +32,68 @@ namespace omath::pathfinding
std::vector<uint8_t> NavigationMesh::serialize() const noexcept
{
auto dump_to_vector = []<typename T>(const T& t, std::vector<uint8_t>& vec)
std::vector<std::uint8_t> raw;
// Pre-calculate total size for better performance
std::size_t total_size = 0;
for (const auto& [vertex, neighbors] : m_vertex_map)
{
for (size_t i = 0; i < sizeof(t); i++)
vec.push_back(*(reinterpret_cast<const uint8_t*>(&t) + i));
total_size += sizeof(vertex) + sizeof(std::uint16_t) + sizeof(Vector3<float>) * neighbors.size();
}
raw.reserve(total_size);
auto dump_to_vector = [&raw]<typename T>(const T& t)
{
const auto* byte_ptr = reinterpret_cast<const std::uint8_t*>(&t);
raw.insert(raw.end(), byte_ptr, byte_ptr + sizeof(T));
};
std::vector<uint8_t> raw;
for (const auto& [vertex, neighbors]: m_vertex_map)
for (const auto& [vertex, neighbors] : m_vertex_map)
{
const auto neighbors_count = neighbors.size();
// Clamp neighbors count to fit in uint16_t (prevents silent data corruption)
// NOTE: If neighbors.size() > 65535, only the first 65535 neighbors will be serialized.
// This is a limitation of the current serialization format using uint16_t for count.
const auto clamped_count =
std::min<std::size_t>(neighbors.size(), std::numeric_limits<std::uint16_t>::max());
const auto neighbors_count = static_cast<std::uint16_t>(clamped_count);
dump_to_vector(vertex, raw);
dump_to_vector(neighbors_count, raw);
dump_to_vector(vertex);
dump_to_vector(neighbors_count);
for (const auto& neighbor: neighbors)
dump_to_vector(neighbor, raw);
// Only serialize up to the clamped count
for (std::size_t i = 0; i < clamped_count; ++i)
dump_to_vector(neighbors[i]);
}
return raw;
}
void NavigationMesh::deserialize(const std::vector<uint8_t>& raw) noexcept
{
auto load_from_vector = [](const std::vector<uint8_t>& vec, size_t& offset, auto& value)
auto load_from_vector = [](const std::vector<uint8_t>& vec, std::size_t& offset, auto& value)
{
if (offset + sizeof(value) > vec.size())
{
throw std::runtime_error("Deserialize: Invalid input data size.");
}
std::copy_n(vec.data() + offset, sizeof(value), reinterpret_cast<uint8_t*>(&value));
offset += sizeof(value);
};
m_vertex_map.clear();
size_t offset = 0;
std::size_t offset = 0;
while (offset < raw.size())
{
Vector3<float> vertex;
load_from_vector(raw, offset, vertex);
uint16_t neighbors_count;
std::uint16_t neighbors_count;
load_from_vector(raw, offset, neighbors_count);
std::vector<Vector3<float>> neighbors;
neighbors.reserve(neighbors_count);
for (size_t i = 0; i < neighbors_count; ++i)
for (std::size_t i = 0; i < neighbors_count; ++i)
{
Vector3<float> neighbor;
load_from_vector(raw, offset, neighbor);

119
source/projectile_prediction/proj_pred_engine_avx2.cpp
View File

@@ -18,97 +18,104 @@ namespace omath::projectile_prediction
[[maybe_unused]] const Target& target) const
{
#if defined(OMATH_USE_AVX2) && defined(__i386__) && defined(__x86_64__)
const float bulletGravity = m_gravityConstant * projectile.m_gravityScale;
const float v0 = projectile.m_launchSpeed;
const float v0Sqr = v0 * v0;
const Vector3 projOrigin = projectile.m_origin;
const float bullet_gravity = m_gravity_constant * projectile.m_gravity_scale;
const float v0 = projectile.m_launch_speed;
const float v0_sqr = v0 * v0;
const Vector3 proj_origin = projectile.m_origin;
constexpr int SIMD_FACTOR = 8;
float currentTime = m_simulationTimeStep;
float current_time = m_simulation_time_step;
for (; currentTime <= m_maximumSimulationTime; currentTime += m_simulationTimeStep * SIMD_FACTOR)
for (; current_time <= m_maximum_simulation_time; current_time += m_simulation_time_step * SIMD_FACTOR)
{
const __m256 times
= _mm256_setr_ps(currentTime, currentTime + m_simulationTimeStep,
currentTime + m_simulationTimeStep * 2, currentTime + m_simulationTimeStep * 3,
currentTime + m_simulationTimeStep * 4, currentTime + m_simulationTimeStep * 5,
currentTime + m_simulationTimeStep * 6, currentTime + m_simulationTimeStep * 7);
= _mm256_setr_ps(current_time, current_time + m_simulation_time_step,
current_time + m_simulation_time_step * 2, current_time + m_simulation_time_step * 3,
current_time + m_simulation_time_step * 4, current_time + m_simulation_time_step * 5,
current_time + m_simulation_time_step * 6, current_time + m_simulation_time_step * 7);
const __m256 targetX
const __m256 target_x
= _mm256_fmadd_ps(_mm256_set1_ps(target.m_velocity.x), times, _mm256_set1_ps(target.m_origin.x));
const __m256 targetY
const __m256 target_y
= _mm256_fmadd_ps(_mm256_set1_ps(target.m_velocity.y), times, _mm256_set1_ps(target.m_origin.y));
const __m256 timesSq = _mm256_mul_ps(times, times);
const __m256 targetZ = _mm256_fmadd_ps(_mm256_set1_ps(target.m_velocity.z), times,
_mm256_fnmadd_ps(_mm256_set1_ps(0.5f * m_gravityConstant), timesSq,
const __m256 times_sq = _mm256_mul_ps(times, times);
const __m256 target_z = _mm256_fmadd_ps(_mm256_set1_ps(target.m_velocity.z), times,
_mm256_fnmadd_ps(_mm256_set1_ps(0.5f * m_gravity_constant), times_sq,
_mm256_set1_ps(target.m_origin.z)));
const __m256 deltaX = _mm256_sub_ps(targetX, _mm256_set1_ps(projOrigin.x));
const __m256 deltaY = _mm256_sub_ps(targetY, _mm256_set1_ps(projOrigin.y));
const __m256 deltaZ = _mm256_sub_ps(targetZ, _mm256_set1_ps(projOrigin.z));
const __m256 delta_x = _mm256_sub_ps(target_x, _mm256_set1_ps(proj_origin.x));
const __m256 delta_y = _mm256_sub_ps(target_y, _mm256_set1_ps(proj_origin.y));
const __m256 delta_z = _mm256_sub_ps(target_z, _mm256_set1_ps(proj_origin.z));
const __m256 dSqr = _mm256_add_ps(_mm256_mul_ps(deltaX, deltaX), _mm256_mul_ps(deltaY, deltaY));
const __m256 d_sqr = _mm256_add_ps(_mm256_mul_ps(delta_x, delta_x), _mm256_mul_ps(delta_y, delta_y));
const __m256 bgTimesSq = _mm256_mul_ps(_mm256_set1_ps(bulletGravity), timesSq);
const __m256 term = _mm256_add_ps(deltaZ, _mm256_mul_ps(_mm256_set1_ps(0.5f), bgTimesSq));
const __m256 termSq = _mm256_mul_ps(term, term);
const __m256 numerator = _mm256_add_ps(dSqr, termSq);
const __m256 denominator = _mm256_add_ps(timesSq, _mm256_set1_ps(1e-8f)); // Avoid division by zero
const __m256 requiredV0Sqr = _mm256_div_ps(numerator, denominator);
const __m256 bg_times_sq = _mm256_mul_ps(_mm256_set1_ps(bullet_gravity), times_sq);
const __m256 term = _mm256_add_ps(delta_z, _mm256_mul_ps(_mm256_set1_ps(0.5f), bg_times_sq));
const __m256 term_sq = _mm256_mul_ps(term, term);
const __m256 numerator = _mm256_add_ps(d_sqr, term_sq);
const __m256 denominator = _mm256_add_ps(times_sq, _mm256_set1_ps(1e-8f)); // Avoid division by zero
const __m256 required_v0_sqr = _mm256_div_ps(numerator, denominator);
const __m256 v0SqrVec = _mm256_set1_ps(v0Sqr + 1e-3f);
const __m256 mask = _mm256_cmp_ps(requiredV0Sqr, v0SqrVec, _CMP_LE_OQ);
const __m256 v0_sqr_vec = _mm256_set1_ps(v0_sqr + 1e-3f);
const __m256 mask = _mm256_cmp_ps(required_v0_sqr, v0_sqr_vec, _CMP_LE_OQ);
const unsigned validMask = _mm256_movemask_ps(mask);
const unsigned valid_mask = _mm256_movemask_ps(mask);
if (!validMask)
if (!valid_mask)
continue;
alignas(32) float validTimes[SIMD_FACTOR];
_mm256_store_ps(validTimes, times);
alignas(32) float valid_times[SIMD_FACTOR];
_mm256_store_ps(valid_times, times);
for (int i = 0; i < SIMD_FACTOR; ++i)
{
if (!(validMask & (1 << i)))
if (!(valid_mask & (1 << i)))
continue;
const float candidateTime = validTimes[i];
const float candidate_time = valid_times[i];
if (candidateTime > m_maximumSimulationTime)
if (candidate_time > m_maximum_simulation_time)
continue;
// Fine search around candidate time
for (float fineTime = candidateTime - m_simulationTimeStep * 2;
fineTime <= candidateTime + m_simulationTimeStep * 2; fineTime += m_simulationTimeStep)
for (float fine_time = candidate_time - m_simulation_time_step * 2;
fine_time <= candidate_time + m_simulation_time_step * 2; fine_time += m_simulation_time_step)
{
if (fineTime < 0)
if (fine_time < 0)
continue;
const Vector3 targetPos = target.PredictPosition(fineTime, m_gravityConstant);
const auto pitch = CalculatePitch(projOrigin, targetPos, bulletGravity, v0, fineTime);
// Manually compute predicted target position to avoid adding method to Target
Vector3 target_pos = target.m_origin + target.m_velocity * fine_time;
if (target.m_is_airborne)
target_pos.z -= 0.5f * m_gravity_constant * fine_time * fine_time;
const auto pitch = calculate_pitch(proj_origin, target_pos, bullet_gravity, v0, fine_time);
if (!pitch)
continue;
const Vector3 delta = targetPos - projOrigin;
const Vector3 delta = target_pos - proj_origin;
const float d = std::sqrt(delta.x * delta.x + delta.y * delta.y);
const float height = d * std::tan(angles::DegreesToRadians(*pitch));
return Vector3(targetPos.x, targetPos.y, projOrigin.z + height);
const float height = d * std::tan(angles::degrees_to_radians(*pitch));
return Vector3(target_pos.x, target_pos.y, proj_origin.z + height);
}
}
}
// Fallback scalar processing for remaining times
for (; currentTime <= m_maximumSimulationTime; currentTime += m_simulationTimeStep)
for (; current_time <= m_maximum_simulation_time; current_time += m_simulation_time_step)
{
const Vector3 targetPos = target.PredictPosition(currentTime, m_gravityConstant);
const auto pitch = CalculatePitch(projOrigin, targetPos, bulletGravity, v0, currentTime);
Vector3 target_pos = target.m_origin + target.m_velocity * current_time;
if (target.m_is_airborne)
target_pos.z -= 0.5f * m_gravity_constant * current_time * current_time;
const auto pitch = calculate_pitch(proj_origin, target_pos, bullet_gravity, v0, current_time);
if (!pitch)
continue;
const Vector3 delta = targetPos - projOrigin;
const Vector3 delta = target_pos - proj_origin;
const float d = std::sqrt(delta.x * delta.x + delta.y * delta.y);
const float height = d * std::tan(angles::DegreesToRadians(*pitch));
return Vector3(targetPos.x, targetPos.y, projOrigin.z + height);
const float height = d * std::tan(angles::degrees_to_radians(*pitch));
return Vector3(target_pos.x, target_pos.y, proj_origin.z + height);
}
return std::nullopt;
@@ -134,22 +141,22 @@ namespace omath::projectile_prediction
return std::nullopt;
const Vector3 delta = target_pos - proj_origin;
const float dSqr = delta.x * delta.x + delta.y * delta.y;
const float d_sqr = delta.x * delta.x + delta.y * delta.y;
const float h = delta.z;
const float term = h + 0.5f * bullet_gravity * time * time;
const float requiredV0Sqr = (dSqr + term * term) / (time * time);
const float v0Sqr = v0 * v0;
const float required_v0_sqr = (d_sqr + term * term) / (time * time);
const float v0_sqr = v0 * v0;
if (requiredV0Sqr > v0Sqr + 1e-3f)
if (required_v0_sqr > v0_sqr + 1e-3f)
return std::nullopt;
if (dSqr == 0.0f)
if (d_sqr == 0.0f)
return term >= 0.0f ? 90.0f : -90.0f;
const float d = std::sqrt(dSqr);
const float tanTheta = term / d;
return angles::RadiansToDegrees(std::atan(tanTheta));
const float d = std::sqrt(d_sqr);
const float tan_theta = term / d;
return angles::radians_to_degrees(std::atan(tan_theta));
#else
throw std::runtime_error(
std::format("{} AVX2 feature is not enabled!", std::source_location::current().function_name()));