Atomic design pattern

All programs below support command line flag:

1. -c <cycles> cycles are measured in power of 2. For example -c 1 will be 2 cycles, -c 10 will be 1024 cycles.
2. -t <threads> number of threads

No contention tests

The first set of tests have no contention. It measures the raw latency of thread local variable, pthread mutex, atomic mutex, etc. User can also apply -t <threads> to see how performance degrades due to context switches.

• nc_thread_local
• nc_address
• nc_atomic_mutex1
• nc_atomic_mutex2
• nc_atomic_mutex3
• nc_atomic_mutex4
• nc_atomic_mutex5
• nc_pthread_mutex
• nc_pthread_rwlock_r
• nc_pthread_rwlock_w
• nc_atomic_rwlock_r
• nc_atomic_rwlock_w
• nc_atomic_punch_card_r
• nc_atomic_punch_card_w

Contention tests

• atomic_wait_free
• atomic_wait_free_8
• atomic_wait_free_16
• atomic_wait_free_32
• atomic_wait_free_64
• pthread_mutex
• atomic_mutex1
• atomic_mutex2
• atomic_mutex3
• atomic_mutex4
• atomic_mutex5
• pthread_mutex_8
• atomic_mutex1_8
• atomic_mutex2_8
• atomic_mutex3_8
• atomic_mutex4_8
• atomic_mutex5_8
• pthread_mutex_16
• atomic_mutex1_16
• atomic_mutex2_16
• atomic_mutex3_16
• atomic_mutex4_16
• atomic_mutex5_16
• pthread_mutex_32
• atomic_mutex1_32
• atomic_mutex2_32
• atomic_mutex3_32
• atomic_mutex4_32
• atomic_mutex5_32
• pthread_mutex_64
• atomic_mutex1_64
• atomic_mutex2_64
• atomic_mutex3_64
• atomic_mutex4_64
• atomic_mutex5_64
• pthread_rwlock
• atomic_rwlock
• atomic_punch_card
• pthread_rwlock_8
• atomic_rwlock_8
• atomic_punch_card_8
• pthread_rwlock_16
• atomic_rwlock_16
• atomic_punch_card_16
• pthread_rwlock_32
• atomic_rwlock_32
• atomic_punch_card_32
• pthread_rwlock_64
• atomic_rwlock_64
• atomic_punch_card_64

misc

• atomic_rwlock2
• atomic_cas
• atomic_cas_handtune

Compile:

autoreconf -vif
./configure
make

See comments for how it works.

Reference:

• [http://en.cppreference.com/w/c/atomic][]