kiyosi

外观

<kiyosi/pricing/settings/monte_carlo.hpp> ​

cpp
#include <kiyosi/pricing/settings/monte_carlo.hpp>

MonteCarloBackend ​

cpp
enum class MonteCarloBackend

Execution backends supported by Monte Carlo engines. CUDA operations use the calling thread's current device and leave it unchanged. Each call owns a nonblocking stream and waits only for its own work, including cleanup. The device and runtime must support CUDA stream-ordered memory allocation.

枚举值

  • cpu — Host CPU implementation.
  • cuda — CUDA implementation, when built and available.

kiyosi::MonteCarloSettings ​

cpp
struct kiyosi::MonteCarloSettings

CPU calls own their parallel work and complete synchronously. Concurrent calls on distinct engines or const calls on a shared engine are safe without concurrent mutation. Thread caps apply per call, not process-wide. Fixed seeds reproduce results across thread counts within a library version, build and standard-library implementation; CUDA is a separate backend. Allocation and internal runtime exceptions propagate; no cancellation interface is provided. Vanilla simulation settings, validated when creating an engine; an absent seed draws from the system entropy source. Vanilla engines accept 1..10,000,000 paths and 2..10,000 steps; American pricing before expiry requires at least 3 steps. European pricing samples the exact terminal distribution without intermediate steps.

成员 ​

path_count ​

cpp
int kiyosi::MonteCarloSettings::path_count

Number of simulated paths; 1..10,000,000.

step_count ​

cpp
int kiyosi::MonteCarloSettings::step_count

American grid points; 2..10,000 (American: >= 3); validated but unused for European pricing.

seed ​

cpp
std::optional<std::uint64_t> kiyosi::MonteCarloSettings::seed

Deterministic seed, or system entropy when absent.

backend ​

cpp
MonteCarloBackend kiyosi::MonteCarloSettings::backend

Execution backend.

thread_count ​

cpp
int kiyosi::MonteCarloSettings::thread_count

CPU concurrency cap: 0 automatic, 1 serial; CUDA requires 0.

kiyosi::TradingDayMonteCarloSettings ​

cpp
struct kiyosi::TradingDayMonteCarloSettings

Accumulator and autocallable products step the trading calendar directly, so no step count is needed; settings are validated when creating an engine. These engines accept 1..10,000,000 paths. CPU paths use separate PCG32 streams, so early knock-out cannot shift later paths' draws. Seeded prices are reproducible within a library version, standard-library implementation and backend.

成员 ​

path_count ​

cpp
int kiyosi::TradingDayMonteCarloSettings::path_count

Number of simulated paths; 1..10,000,000.

seed ​

cpp
std::optional<std::uint64_t> kiyosi::TradingDayMonteCarloSettings::seed

Deterministic seed, or system entropy when absent.

backend ​

cpp
MonteCarloBackend kiyosi::TradingDayMonteCarloSettings::backend

Execution backend.

thread_count ​

cpp
int kiyosi::TradingDayMonteCarloSettings::thread_count

CPU concurrency cap: 0 automatic, 1 serial; CUDA requires 0.