From d081c19178bfd8fdfb78eaf7643b7179444b16b2 Mon Sep 17 00:00:00 2001
From: Karthik Gururaj <karthik.gururaj@intel.com>
Date: Sun, 9 Feb 2014 18:05:35 -0800
Subject: [PATCH] Minor: added support in C++ sandbox to choose implementation
 and check from command line

---
 public/c++/VectorPairHMM/pairhmm-1-base.cc | 26 ++++++++++--
 public/c++/VectorPairHMM/utils.cc          | 49 ++++++++++------------
 public/c++/VectorPairHMM/utils.h           |  2 +-
 3 files changed, 46 insertions(+), 31 deletions(-)

diff --git a/public/c++/VectorPairHMM/pairhmm-1-base.cc b/public/c++/VectorPairHMM/pairhmm-1-base.cc
index 8b686d5ea..78dbe296d 100644
--- a/public/c++/VectorPairHMM/pairhmm-1-base.cc
+++ b/public/c++/VectorPairHMM/pairhmm-1-base.cc
@@ -18,10 +18,28 @@ int main(int argc, char** argv)
   if(argc >= 3 && string(argv[2]) == "1")
     use_old_read_testcase = true;
   unsigned chunk_size = 10000;
-  if(argc >= 4)
-    chunk_size = strtol(argv[3],0,10);
-
-  do_compute(argv[1], use_old_read_testcase, chunk_size); 
+  bool do_check = true;
+  uint64_t mask = ~(0ull);
+  for(int i=3;i<argc;++i)
+  {
+    if(strncmp(argv[i], "-chunk_size", 15) == 0)
+    {
+      ++i;
+      chunk_size = strtol(argv[i],0,10);
+    }
+    else
+      if(strncmp(argv[i], "-mask", 15) == 0)
+      {
+        ++i;
+        mask = strtoll(argv[i],0,16);
+      }
+      else
+        if(strncmp(argv[i], "-no-check", 15) == 0)
+          do_check = false;
+  }
+  if(mask != (~0ull))
+    initialize_function_pointers(mask);
+  do_compute(argv[1], use_old_read_testcase, chunk_size, do_check); 
   return 0;  
 }
 
diff --git a/public/c++/VectorPairHMM/utils.cc b/public/c++/VectorPairHMM/utils.cc
index 8fe20234e..76cee7327 100644
--- a/public/c++/VectorPairHMM/utils.cc
+++ b/public/c++/VectorPairHMM/utils.cc
@@ -68,7 +68,7 @@ void initialize_function_pointers(uint64_t mask)
     g_compute_full_prob_double = compute_full_prob_avxd<double>;
   }
   else
-    if(is_sse41_supported() && (mask & (1<< SSE41_CUSTOM_IDX)))
+    if(is_sse41_supported() && (mask & ((1<< SSE41_CUSTOM_IDX) | (1<<SSE42_CUSTOM_IDX))))
     {
       cout << "Using SSE4.1 accelerated implementation of PairHMM\n";
       g_compute_full_prob_float = compute_full_prob_sses<float>;
@@ -299,10 +299,9 @@ uint64_t diff_time(struct timespec& prev_time)
 }
 
 //#define DUMP_COMPUTE_VALUES 1
-#define CHECK_VALUES 1
 #define BATCH_SIZE  10000
 #define RUN_HYBRID
-void do_compute(char* filename, bool use_old_read_testcase, unsigned chunk_size)
+void do_compute(char* filename, bool use_old_read_testcase, unsigned chunk_size, bool do_check)
 {
   FILE* fptr = 0;
   ifstream ifptr;
@@ -322,10 +321,7 @@ void do_compute(char* filename, bool use_old_read_testcase, unsigned chunk_size)
   uint64_t vector_compute_time = 0;
   uint64_t baseline_compute_time = 0;
   uint64_t num_double_calls = 0;
-  bool all_ok = true;
-#ifndef CHECK_VALUES
-  all_ok = false;
-#endif
+  bool all_ok = do_check ? true : false;
   while(1)
   {
     int break_value = use_old_read_testcase ? read_testcase(&tc, fptr) : read_mod_testcase(ifptr,&tc,true);
@@ -360,29 +356,30 @@ void do_compute(char* filename, bool use_old_read_testcase, unsigned chunk_size)
 	results_vec[i] = result;
       }
       vector_compute_time +=  diff_time(start_time);
-#ifdef CHECK_VALUES
-      get_time(&start_time);
+      if(do_check)
+      {
+        get_time(&start_time);
 #pragma omp parallel for schedule(dynamic,chunk_size)
-      for(unsigned i=0;i<tc_vector.size();++i)
-      {
-	testcase& tc = tc_vector[i];
-	double baseline_result = compute_full_prob<double>(&tc);
-	baseline_result = log10(baseline_result) - log10(ldexp(1.0, 1020.0));
-        baseline_results_vec[i] = baseline_result;
-      }
-      baseline_compute_time += diff_time(start_time);
-      for(unsigned i=0;i<tc_vector.size();++i)
-      {
-        double baseline_result = baseline_results_vec[i];
-	double abs_error = fabs(baseline_result-results_vec[i]);
-	double rel_error = (baseline_result != 0) ? fabs(abs_error/baseline_result) : 0;
-	if(abs_error > 1e-5 && rel_error > 1e-5)
+        for(unsigned i=0;i<tc_vector.size();++i)
         {
-	  cout << std::scientific << baseline_result << " "<<results_vec[i]<<"\n";
-          all_ok = false;
+          testcase& tc = tc_vector[i];
+          double baseline_result = compute_full_prob<double>(&tc);
+          baseline_result = log10(baseline_result) - log10(ldexp(1.0, 1020.0));
+          baseline_results_vec[i] = baseline_result;
+        }
+        baseline_compute_time += diff_time(start_time);
+        for(unsigned i=0;i<tc_vector.size();++i)
+        {
+          double baseline_result = baseline_results_vec[i];
+          double abs_error = fabs(baseline_result-results_vec[i]);
+          double rel_error = (baseline_result != 0) ? fabs(abs_error/baseline_result) : 0;
+          if(abs_error > 1e-5 && rel_error > 1e-5)
+          {
+            cout << std::scientific << baseline_result << " "<<results_vec[i]<<"\n";
+            all_ok = false;
+          }
         }
       }
-#endif
       for(unsigned i=0;i<tc_vector.size();++i)
       {
 	delete tc_vector[i].rs;
diff --git a/public/c++/VectorPairHMM/utils.h b/public/c++/VectorPairHMM/utils.h
index 501b48fbc..98115e283 100644
--- a/public/c++/VectorPairHMM/utils.h
+++ b/public/c++/VectorPairHMM/utils.h
@@ -39,5 +39,5 @@ enum ProcessorCapabilitiesEnum
 #define ENABLE_ALL_HARDWARE_FEATURES 0xFFFFFFFFFFFFFFFFull
 uint64_t get_machine_capabilities();
 void initialize_function_pointers(uint64_t mask=ENABLE_ALL_HARDWARE_FEATURES);
-void do_compute(char* filename, bool use_old_read_testcase=true, unsigned chunk_size=10000);
+void do_compute(char* filename, bool use_old_read_testcase=true, unsigned chunk_size=10000, bool do_check=true);
 #endif