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

Revert "Add nstream kernel from PRK"

Browse Source 
This reverts commit 1e94a41f3c.
This commit is contained in:
Tom Deakin 2021-02-02 11:25:27 +00:00
parent 877f820282
commit f99f8d35d9
2 changed files with 5 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Stream.h
View File

@ -29,7 +29,6 @@ class Stream
virtual void mul() = 0; virtual void mul() = 0;
virtual void add() = 0; virtual void add() = 0;
virtual void triad() = 0; virtual void triad() = 0;
virtual void nstream() = 0;
virtual T dot() = 0; virtual T dot() = 0;
// Copy memory between host and device // Copy memory between host and device

21
main.cpp
View File

@ -186,7 +186,7 @@ void run()
T sum; T sum;
// List of times // List of times
std::vector<std::vector<double>> timings(6); std::vector<std::vector<double>> timings(5);
// Declare timers // Declare timers
std::chrono::high_resolution_clock::time_point t1, t2; std::chrono::high_resolution_clock::time_point t1, t2;
@ -218,17 +218,11 @@ void run()
t2 = std::chrono::high_resolution_clock::now(); t2 = std::chrono::high_resolution_clock::now();
timings[3].push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count()); timings[3].push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count());
// Execute nstream
t1 = std::chrono::high_resolution_clock::now();
stream->nstream();
t2 = std::chrono::high_resolution_clock::now();
timings[4].push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count());
// Execute Dot // Execute Dot
t1 = std::chrono::high_resolution_clock::now(); t1 = std::chrono::high_resolution_clock::now();
sum = stream->dot(); sum = stream->dot();
t2 = std::chrono::high_resolution_clock::now(); t2 = std::chrono::high_resolution_clock::now();
timings[5].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());
} }
@ -268,17 +262,16 @@ void run()
std::string labels[6] = {"Copy", "Mul", "Add", "Triad", "nstream", "Dot"}; std::string labels[5] = {"Copy", "Mul", "Add", "Triad", "Dot"};
size_t sizes[6] = { size_t sizes[5] = {
2 * sizeof(T) * ARRAY_SIZE, 2 * sizeof(T) * ARRAY_SIZE,
2 * sizeof(T) * ARRAY_SIZE, 2 * sizeof(T) * ARRAY_SIZE,
3 * sizeof(T) * ARRAY_SIZE, 3 * sizeof(T) * ARRAY_SIZE,
3 * sizeof(T) * ARRAY_SIZE, 3 * sizeof(T) * ARRAY_SIZE,
4 * sizeof(T) * ARRAY_SIZE,
2 * sizeof(T) * ARRAY_SIZE 2 * sizeof(T) * ARRAY_SIZE
}; };
for (int i = 0; i < 6; i++) for (int i = 0; i < 5; i++)
{ {
// Get min/max; ignore the first result // Get min/max; ignore the first result
auto minmax = std::minmax_element(timings[i].begin()+1, timings[i].end()); auto minmax = std::minmax_element(timings[i].begin()+1, timings[i].end());
@ -480,10 +473,6 @@ void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>
goldC = goldA + goldB; goldC = goldA + goldB;
} }
goldA = goldB + scalar * goldC; goldA = goldB + scalar * goldC;
if (!triad_only)
{
goldA += goldB + scalar * goldC;
}
} }
// Do the reduction // Do the reduction