Superficial changes

include/doGemm.hh

@@ -283,8 +283,55 @@ class doGemm {
 		gpuResult_once.gflops =
 						calcGflops(flops, iterations_, gpuResult_once.runtime);
 
+#if !defined GPU_DEFAULT
+    if (!((std::fabs(cpuResult.checksum - gpuResult_once.checksum) < 0.5) &&
+          (std::fabs(cpuResult.checksum - gpuResult_always.checksum) < 0.5) &&
+          (std::fabs(cpuResult.checksum - gpuResult_unified.checksum) < 0.5))) {
+      std::cerr << "ERROR - " << getKernelName()
+                << " kernel checksums do not match:\n\tInput dimensions: N="
+                << N << std::endl;
+      std::cerr << std::setprecision(10)
+                << "\tCPU Checksum = " << cpuResult.checksum << std::endl;
+      std::cerr << std::setprecision(10)
+                << "\tGPU (Once) Checksum = " << gpuResult_once.checksum
+                << std::endl;
+      std::cerr << std::setprecision(10)
+                << "\tGPU (Always) Checksum = " << gpuResult_always.checksum
+                << std::endl;
+      std::cerr << std::setprecision(10)
+                << "\tGPU (Unified) Checksum = " << gpuResult_unified.checksum
+                << std::endl;
+      exit(1);
+    }
 
-		// ToDo -- non-default GPU operations
+    // Check if offload threshold has been achieved for each GPU offload type.
+    // M can only be 0 when structure is newly initialised
+    if ((cpuGpu_once_.M == 0) && cpuResult.gflops < gpuResult_once.gflops) {
+      cpuGpu_once_.cpuGflops = cpuResult.gflops;
+      cpuGpu_once_.gpuGflops = gpuResult_once.gflops;
+      cpuGpu_once_.probSize_kib = probSize;
+      cpuGpu_once_.M = M;
+      cpuGpu_once_.N = N;
+      cpuGpu_once_.K = K;
+    }
+    if ((cpuGpu_always_.M == 0) && cpuResult.gflops < gpuResult_always.gflops) {
+      cpuGpu_always_.cpuGflops = cpuResult.gflops;
+      cpuGpu_always_.gpuGflops = gpuResult_always.gflops;
+      cpuGpu_always_.probSize_kib = probSize;
+      cpuGpu_always_.M = M;
+      cpuGpu_always_.N = N;
+      cpuGpu_always_.K = K;
+    }
+    if ((cpuGpu_unified_.M == 0) &&
+        cpuResult.gflops < gpuResult_unified.gflops) {
+      cpuGpu_unified_.cpuGflops = cpuResult.gflops;
+      cpuGpu_unified_.gpuGflops = gpuResult_unified.gflops;
+      cpuGpu_unified_.probSize_kib = probSize;
+      cpuGpu_unified_.M = M;
+      cpuGpu_unified_.N = N;
+      cpuGpu_unified_.K = K;
+    }
+#endif
 
 		// Write lines to CSV file
 		writeLineToCsv(csvFile, "cpu", kernelName, N, N, N, probSize, iterations_,