diff --git a/unitTest/checkRounding/checkRounding.cxx b/unitTest/checkRounding/checkRounding.cxx
index e3559e2..212ca2d 100644
--- a/unitTest/checkRounding/checkRounding.cxx
+++ b/unitTest/checkRounding/checkRounding.cxx
@@ -445,6 +445,67 @@ class testInvariantProdDivm:public test<REALTYPE>{
   }
 
 
+  inline double myFms(const double& a, const double& b, const double& c){
+    double d;
+#ifdef TEST_FMA
+#if defined(__x86_64__)
+    __m128d ai, bi,ci,di;
+    ai = _mm_load_sd(&a);
+    bi = _mm_load_sd(&b);
+    ci = _mm_load_sd(&c);
+    di=_mm_fmsub_sd(ai,bi,ci);
+    d=_mm_cvtsd_f64(di);
+#elif  defined(__aarch64__)
+  const float64x1_t ai=vld1_f64(&a);
+  const float64x1_t bi=vld1_f64(&b);
+  const float64x1_t ci=vld1_f64(&c);
+  const float64x1_t di=vfms_f64(ci,ai,bi);// warning strange argument order
+  // cf doc : https://developer.arm.com/architectures/instruction-set/intrinsics/#q=vfma
+  vst1_f64(&d, di);
+#else
+#error "not yet implemented for this architecture"
+#endif
+#else
+    d=a*b-c;
+#endif
+    return d;
+  }
+
+
+  inline float myFms(const float& a, const float& b, const float& c){
+    float d;
+#ifdef TEST_FMA
+#if defined(__x86_64__)
+    __m128 ai, bi,ci,di;
+    ai = _mm_load_ss(&a);
+    bi = _mm_load_ss(&b);
+    ci = _mm_load_ss(&c);
+    di=_mm_fmsub_ss(ai,bi,ci);
+    d=_mm_cvtss_f32(di);
+#elif  defined(__aarch64__)
+  float av[2]={a,0};
+  float bv[2]={b,0};
+  float cv[2]={c,0};
+
+  float32x2_t ap=vld1_f32(av);
+  float32x2_t bp=vld1_f32(bv);
+  float32x2_t cp=vld1_f32(cv);
+
+  float32x2_t resp= vfms_f32(cp,ap,bp); // warning strange argument order
+  // cf doc : https://developer.arm.com/architectures/instruction-set/intrinsics/#q=vfma
+  float res[2];
+  vst1_f32(res, resp);
+  d=res[0];
+#else
+#error "not yet implemented for this architecture"
+#endif
+#else
+    d=a*b-c;
+#endif
+    return d;
+  }
+
+
 
 template<class REALTYPE>
 class testFma:public test<REALTYPE>{
@@ -476,6 +537,37 @@ class testFma:public test<REALTYPE>{
 
 };
 
+template<class REALTYPE>
+class testFms:public test<REALTYPE>{
+ public:
+  testFms():test<REALTYPE>(10000),
+    size(1000000),
+    value(0.1),
+    init(0.){}
+
+  std::string name(){
+    return std::string("testFms");
+  }
+
+
+  REALTYPE compute(){
+    REALTYPE acc=init;
+    volatile REALTYPE mone=-1.;
+    for(int i=0; i<size; i++){
+      acc=myFms(value,value,mone*acc);
+    }
+    return acc;
+  }
+
+ private:
+  const int size;
+  const REALTYPE value;
+  const REALTYPE init;
+
+};
+
+
+
 
 template<class REALTYPE>
 class testFmam:public test<REALTYPE>{
@@ -1155,6 +1247,7 @@ int main(int argc, char** argv){
     testInvariantProdDivm<RealType> t4m; t4m.run();
     testFma<RealType> t5; t5.run();
     testFmam<RealType> t5m; t5m.run();
+    testFms<RealType> t5s; t5s.run();
     testMixSseLlo<RealType> t6; t6.run();
     testMixSseLlom<RealType> t6m; t6m.run();
     testCast<RealType,double> t7; t7.run();
@@ -1182,6 +1275,7 @@ int main(int argc, char** argv){
     testInvariantProdDivm<RealType> t4m; t4m.run();
     testFma<RealType> t5; t5.run();
     testFmam<RealType> t5m; t5m.run();
+    testFms<RealType> t5s; t5s.run();
     testMixSseLlo<RealType> t6; t6.run();
     testMixSseLlom<RealType> t6m; t6m.run();
     testCast<RealType,double> t7; t7.run();
diff --git a/unitTest/checkRounding/runCheck.py b/unitTest/checkRounding/runCheck.py
index 9c0a7bf..0235ca7 100644
--- a/unitTest/checkRounding/runCheck.py
+++ b/unitTest/checkRounding/runCheck.py
@@ -695,6 +695,24 @@ def checkScomdet(allResult, testPairList, typeTab=["<double>", "<float>"]):
 
     return errorCounter(ok,ko,0)
 
+def checkEqualSymDet(allResult, testPairList, typeTab=["<double>", "<float>"]):
+    ok=0
+    ko=0
+    roundingList=["random_scomdet", "average_scomdet", "sr_smonotonic", "nearest", "farthest"]
+    for (code1,code2) in testPairList:
+        for RealType in typeTab:
+            testName1=code1+RealType
+            testName2=code2+RealType
+
+            for rounding in roundingList:
+                if assertCmpTest(testName1, rounding, testName2, rounding):
+                    ok+=1
+                else:
+                    ko+=1
+
+    return errorCounter(ok,ko,0)
+
+
 
 if __name__=='__main__':
     cmdHandler=cmdPrepare(os.path.join(os.curdir,sys.argv[1]))
@@ -717,7 +735,7 @@ def checkScomdet(allResult, testPairList, typeTab=["<double>", "<float>"]):
     eCount+=checkTestNegativeAndOptimistRandomVerrou(allResult, testList=["testInc0d1m", "testIncSquare0d1m", "testIncDiv10m"], typeTab=typeTab)
     eCount+=checkTestPositive(allResult, testList=["testInvariantProdDiv"], typeTab=typeTab)
     eCount+=checkTestNegative(allResult, testList=["testInvariantProdDivm"], typeTab=typeTab)
-    eCount+=checkTestPositiveAndOptimistRandomVerrou(allResult, testList=["testFma"], typeTab=["<double>", "<float>"])
+    eCount+=checkTestPositiveAndOptimistRandomVerrou(allResult, testList=["testFma", "testFms"], typeTab=["<double>", "<float>"])
     eCount+=checkTestNegativeAndOptimistRandomVerrou(allResult, testList=["testFmam"], typeTab=["<double>", "<float>"])
 
     eCount+=checkTestPositive(allResult, testList=["testMixSseLlo", "testMixAvxLlo", "testFmaMixSseLlo", "testFmaMixAvxLlo"],  typeTab=["<double>", "<float>"], statNumber="low")
@@ -735,5 +753,7 @@ def checkScomdet(allResult, testPairList, typeTab=["<double>", "<float>"]):
                                     ("testMixSseLlo", "testMixSseLlom",True  ),("testMixAvxLlo", "testMixAvxLlom",True  ),
                                     ("testFmaMixSseLlo", "testFmaMixSseLlom",True  ),("testFmaMixAvxLlo", "testFmaMixAvxLlom",True  )])
 
+    eCount+=checkEqualSymDet(allResult,[("testFma","testFms")])
+
     eCount.printSummary()
     sys.exit(eCount.ko+eCount.warn)