Performance Ninja -- False Sharing

[@virajcz7143]

beautiful! watched so many videos but finally understood it after watching your video! great infographics!

[@easyperf3992]

Here is the complete code of the solution:
```
#include "solution.hpp"
#include
#include
#include
#include
size_t solution(const std::vector &data, int thread_count) {
// Using std::atomic counters to disallow compiler to promote `target`
// memory location into a register. This way we ensure that the store
// to `target` stays inside the loop.
#define CACHELINE_ALIGN alignas(64)
struct CACHELINE_ALIGN Accumulator {
std::atomic value = 0;
};
std::vector accumulators(thread_count);
#pragma omp parallel num_threads(thread_count) default(none) \
shared(accumulators, data)
{
int target_index = omp_get_thread_num();
auto &target = accumulators[target_index];
#pragma omp for
for (int i = 0; i < data.size(); i++) {
// Perform computation on each input
auto item = data[i];
item += 1000;
item ^= 0xADEDAE;
item |= (item >> 24);
// Write result to accumulator
target.value += item % 13;
}
}
size_t result = 0;
for (const auto &accumulator : accumulators) {
result += accumulator.value;
}
return result;
}
```

[@easyperf3992]

perf c2c:
joemario.github.io/blog/2016/09/01/c2c-blog/