Fix wrong kernel picking

mconcas · mconcas · commit 2aedb24b18b7 · 2021-10-07T15:39:24.000+02:00
diff --git a/GPU/GPUbenchmark/benchmark.cxx b/GPU/GPUbenchmark/benchmark.cxx
@@ -13,6 +13,7 @@
 /// \author mconcas@cern.ch
 ///
 #include "Shared/Kernels.h"
+#define VERSION "version 0.1-latest-#6773"
 
 bool parseArgs(o2::benchmark::benchmarkOpts& conf, int argc, const char* argv[])
 {
@@ -21,11 +22,12 @@ bool parseArgs(o2::benchmark::benchmarkOpts& conf, int argc, const char* argv[])
   bpo::options_description options("Benchmark options");
   options.add_options()(
     "help,h", "Print help message.")(
+    "version,v", "print version")(
     "device,d", bpo::value<int>()->default_value(0), "Id of the device to run test on, EPN targeted.")(
-    "test,t", bpo::value<std::vector<std::string>>()->multitoken()->default_value(std::vector<std::string>{"read", "write", "copy"}, "read, write, copy"), "Tests to be performed.")(
-    "kind,k", bpo::value<std::vector<std::string>>()->multitoken()->default_value(std::vector<std::string>{"char", "int", "ulong"}, "char, int, ulong"), "Test data type to be used.")(
-    "mode,m", bpo::value<std::vector<std::string>>()->multitoken()->default_value(std::vector<std::string>{"seq", "con"}, "seq, con"), "Mode: sequential or concurrent.")(
-    "pool,p", bpo::value<std::vector<std::string>>()->multitoken()->default_value(std::vector<std::string>{"sb", "mb"}, "sb, mb"), "Pool strategy: single or multi blocks.")(
+    "test,t", bpo::value<std::vector<std::string>>()->multitoken()->default_value(std::vector<std::string>{"read", "write", "copy"}, "read write copy"), "Tests to be performed.")(
+    "kind,k", bpo::value<std::vector<std::string>>()->multitoken()->default_value(std::vector<std::string>{"char", "int", "ulong"}, "char int ulong"), "Test data type to be used.")(
+    "mode,m", bpo::value<std::vector<std::string>>()->multitoken()->default_value(std::vector<std::string>{"seq", "con"}, "seq con"), "Mode: sequential or concurrent.")(
+    "pool,p", bpo::value<std::vector<std::string>>()->multitoken()->default_value(std::vector<std::string>{"sb", "mb"}, "sb mb"), "Pool strategy: single or multi blocks.")(
     "chunkSize,c", bpo::value<float>()->default_value(1.f), "Size of scratch partitions (GB).")(
     "freeMemFraction,f", bpo::value<float>()->default_value(0.95f), "Fraction of free memory to be allocated (min: 0.f, max: 1.f).")(
     "launches,l", bpo::value<int>()->default_value(10), "Number of iterations in reading kernels.")(
@@ -39,6 +41,11 @@ bool parseArgs(o2::benchmark::benchmarkOpts& conf, int argc, const char* argv[])
       return false;
     }
 
+    if (vm.count("version")) {
+      std::cout << VERSION << std::endl;
+      return false;
+    }
+
     bpo::notify(vm);
   } catch (const bpo::error& e) {
     std::cerr << e.what() << "\n\n";
diff --git a/GPU/GPUbenchmark/cuda/Kernels.cu b/GPU/GPUbenchmark/cuda/Kernels.cu
@@ -391,12 +391,15 @@ void GPUbenchmark<chunk_t>::runTest(Test test, Mode mode, KernelConfig config)
   switch (test) {
     case Test::Read: {
       kernel = (config == KernelConfig::Single) ? &gpu::readChunkSBKernel<chunk_t> : &gpu::readChunkMBKernel<chunk_t>;
+      break;
     }
     case Test::Write: {
       kernel = (config == KernelConfig::Single) ? &gpu::writeChunkSBKernel<chunk_t> : &gpu::writeChunkMBKernel<chunk_t>;
+      break;
     }
     case Test::Copy: {
       kernel = (config == KernelConfig::Single) ? &gpu::copyChunkSBKernel<chunk_t> : &gpu::copyChunkMBKernel<chunk_t>;
+      break;
     }
   }
 
@@ -472,5 +475,4 @@ template class GPUbenchmark<int>;
 // template class GPUbenchmark<int4>;
 
 } // namespace benchmark
-} // namespace o2
-void readConcurrent(KernelConfig config);
+} // namespace o2

Original file line number	Diff line number	Diff line change
`@@ -391,12 +391,15 @@ void GPUbenchmark<chunk_t>::runTest(Test test, Mode mode, KernelConfig config)`
`391`	`391`	`switch (test) {`
`392`	`392`	`case Test::Read: {`
`393`	`393`	`kernel = (config == KernelConfig::Single) ? &gpu::readChunkSBKernel<chunk_t> : &gpu::readChunkMBKernel<chunk_t>;`
	`394`	`+ break;`
`394`	`395`	`}`
`395`	`396`	`case Test::Write: {`
`396`	`397`	`kernel = (config == KernelConfig::Single) ? &gpu::writeChunkSBKernel<chunk_t> : &gpu::writeChunkMBKernel<chunk_t>;`
	`398`	`+ break;`
`397`	`399`	`}`
`398`	`400`	`case Test::Copy: {`
`399`	`401`	`kernel = (config == KernelConfig::Single) ? &gpu::copyChunkSBKernel<chunk_t> : &gpu::copyChunkMBKernel<chunk_t>;`
	`402`	`+ break;`
`400`	`403`	`}`
`401`	`404`	`}`
`402`	`405`
`@@ -472,5 +475,4 @@ template class GPUbenchmark<int>;`
`472`	`475`	`// template class GPUbenchmark<int4>;`
`473`	`476`
`474`	`477`	`} // namespace benchmark`
`475`		`-} // namespace o2`
`476`		`-void readConcurrent(KernelConfig config);`
	`478`	`+} // namespace o2`