cory-master/BabelStream
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Check result of the final reduction

Browse Source
This commit is contained in:
Tom Deakin 2016-10-14 14:45:28 +01:00
parent 04ca357159
commit 275bfb2066
1 changed files with 16 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
main.cpp
View File

@ -44,7 +44,7 @@ bool use_float = false;
template <typename T> template <typename T>
void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>& b, std::vector<T>& c); void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>& b, std::vector<T>& c, T& sum);
template <typename T> template <typename T>
void run(); void run();
@ -94,6 +94,9 @@ void run()
<< " (=" << 3.0*ARRAY_SIZE*sizeof(T)*1.0E-9 << " GB)" << std::endl; << " (=" << 3.0*ARRAY_SIZE*sizeof(T)*1.0E-9 << " GB)" << std::endl;
std::cout.precision(ss); std::cout.precision(ss);
// Result of the Dot kernel
T sum;
Stream<T> *stream; Stream<T> *stream;
#if defined(CUDA) #if defined(CUDA)
@ -167,7 +170,7 @@ void run()
// Execute Dot // Execute Dot
t1 = std::chrono::high_resolution_clock::now(); t1 = std::chrono::high_resolution_clock::now();
stream->dot(); sum = stream->dot();
t2 = std::chrono::high_resolution_clock::now(); t2 = std::chrono::high_resolution_clock::now();
timings[4].push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count()); timings[4].push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count());
@ -175,7 +178,7 @@ void run()
// Check solutions // Check solutions
stream->read_arrays(a, b, c); stream->read_arrays(a, b, c);
check_solution<T>(num_times, a, b, c); check_solution<T>(num_times, a, b, c, sum);
// Display timing results // Display timing results
std::cout std::cout
@ -220,12 +223,13 @@ void run()
} }
template <typename T> template <typename T>
void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>& b, std::vector<T>& c) void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>& b, std::vector<T>& c, T& sum)
{ {
// Generate correct solution // Generate correct solution
T goldA = 0.1; T goldA = 0.1;
T goldB = 0.2; T goldB = 0.2;
T goldC = 0.0; T goldC = 0.0;
T golSum = 0.0;
const T scalar = 0.3; const T scalar = 0.3;
@ -238,6 +242,9 @@ void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>
goldA = goldB + scalar * goldC; goldA = goldB + scalar * goldC;
} }
// Do the reduction
goldSum = goldA * goldB * ntimes;
// Calculate the average error // Calculate the average error
double errA = std::accumulate(a.begin(), a.end(), 0.0, [&](double sum, const T val){ return sum + fabs(val - goldA); }); double errA = std::accumulate(a.begin(), a.end(), 0.0, [&](double sum, const T val){ return sum + fabs(val - goldA); });
errA /= a.size(); errA /= a.size();
@ -245,6 +252,7 @@ void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>
errB /= b.size(); errB /= b.size();
double errC = std::accumulate(c.begin(), c.end(), 0.0, [&](double sum, const T val){ return sum + fabs(val - goldC); }); double errC = std::accumulate(c.begin(), c.end(), 0.0, [&](double sum, const T val){ return sum + fabs(val - goldC); });
errC /= c.size(); errC /= c.size();
double errSum = fabs(sum - goldSum);
double epsi = std::numeric_limits<T>::epsilon() * 100.0; double epsi = std::numeric_limits<T>::epsilon() * 100.0;
@ -260,6 +268,10 @@ void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>
std::cerr std::cerr
<< "Validation failed on c[]. Average error " << errC << "Validation failed on c[]. Average error " << errC
<< std::endl; << std::endl;
if (errSum > epsi)
std::cerr
<< "Validation failed on sum. Error " << errSum
<< std::endl;
} }