From 20a46e4098469c47fd124a5610f49e3f928759fc Mon Sep 17 00:00:00 2001
From: Karthik Gururaj <karthik.gururaj@intel.com>
Date: Fri, 7 Feb 2014 15:19:55 -0800
Subject: [PATCH] Check only for SSE 4.1 (rather than SSE 4.2) when trying to
 use the SSE implementation of PairHMM

---
 .../utils/pairhmm/VectorLoglessPairHMM.java    |  5 +++--
 public/c++/VectorPairHMM/utils.cc              | 18 ++++++++++++++++--
 public/c++/VectorPairHMM/utils.h               |  3 ++-
 3 files changed, 21 insertions(+), 5 deletions(-)

diff --git a/protected/java/src/org/broadinstitute/sting/utils/pairhmm/VectorLoglessPairHMM.java b/protected/java/src/org/broadinstitute/sting/utils/pairhmm/VectorLoglessPairHMM.java
index aebe3cf95..29d3a729d 100644
--- a/protected/java/src/org/broadinstitute/sting/utils/pairhmm/VectorLoglessPairHMM.java
+++ b/protected/java/src/org/broadinstitute/sting/utils/pairhmm/VectorLoglessPairHMM.java
@@ -77,8 +77,9 @@ import java.io.OutputStream;
 public class VectorLoglessPairHMM extends JNILoglessPairHMM {
 
     //For machine capabilities
-    public static final long sse42Mask = 1;
-    public static final long avxMask = 2;
+    public static final long sse41Mask = 1;
+    public static final long sse42Mask = 2;
+    public static final long avxMask = 4;
     public static final long enableAll = 0xFFFFFFFFFFFFFFFFl;
 
     //Used to copy references to byteArrays to JNI from reads
diff --git a/public/c++/VectorPairHMM/utils.cc b/public/c++/VectorPairHMM/utils.cc
index 9974c5ace..8fe20234e 100644
--- a/public/c++/VectorPairHMM/utils.cc
+++ b/public/c++/VectorPairHMM/utils.cc
@@ -22,6 +22,18 @@ bool is_avx_supported()
   return ((ecx >> 28)&1) == 1;
 }
 
+bool is_sse41_supported()
+{
+  int ecx = 0, edx = 0, ebx = 0;
+  __asm__("cpuid"
+      : "=b" (ebx),
+      "=c" (ecx),
+      "=d" (edx)
+      : "a" (1)
+      );
+  return ((ecx >> 19)&1) == 1;
+}
+
 bool is_sse42_supported()
 {
   int ecx = 0, edx = 0, ebx = 0;
@@ -41,6 +53,8 @@ uint64_t get_machine_capabilities()
     machine_mask |= (1 << AVX_CUSTOM_IDX);
   if(is_sse42_supported())
     machine_mask |= (1 << SSE42_CUSTOM_IDX);
+  if(is_sse41_supported())
+    machine_mask |= (1 << SSE41_CUSTOM_IDX);
   return machine_mask;
 }
 
@@ -54,9 +68,9 @@ void initialize_function_pointers(uint64_t mask)
     g_compute_full_prob_double = compute_full_prob_avxd<double>;
   }
   else
-    if(is_sse42_supported() && (mask & (1<< SSE42_CUSTOM_IDX)))
+    if(is_sse41_supported() && (mask & (1<< SSE41_CUSTOM_IDX)))
     {
-      cout << "Using SSE4.2 accelerated implementation of PairHMM\n";
+      cout << "Using SSE4.1 accelerated implementation of PairHMM\n";
       g_compute_full_prob_float = compute_full_prob_sses<float>;
       g_compute_full_prob_double = compute_full_prob_ssed<double>;
     }
diff --git a/public/c++/VectorPairHMM/utils.h b/public/c++/VectorPairHMM/utils.h
index 092132473..501b48fbc 100644
--- a/public/c++/VectorPairHMM/utils.h
+++ b/public/c++/VectorPairHMM/utils.h
@@ -32,7 +32,8 @@ uint64_t diff_time(struct timespec& prev_time);
 //bit 0 is sse4.2, bit 1 is AVX
 enum ProcessorCapabilitiesEnum
 {
-  SSE42_CUSTOM_IDX=0,
+  SSE41_CUSTOM_IDX=0,
+  SSE42_CUSTOM_IDX,
   AVX_CUSTOM_IDX
 };
 #define ENABLE_ALL_HARDWARE_FEATURES 0xFFFFFFFFFFFFFFFFull