Correct issue #287

Author: christophe0606

Include/dsp/utils.h

@@ -219,6 +219,13 @@ __STATIC_INLINE  void arm_norm_64_to_32u(uint64_t in, int32_t * normalized, int3
     }
 }
 
+/**
+ * arm_div_int64_to_int32 function performs division of a 64-bit integer by a 32-bit integer
+ * and returns a 32-bit integer result. The function saturates the result to fit within the 32-bit integer range.
+ * @param num The 64-bit integer numerator.
+ * @param den The 32-bit integer denominator.
+ * @return The saturated 32-bit integer result of the division.
+ */
 __STATIC_INLINE int32_t arm_div_int64_to_int32(int64_t num, int32_t den)
 {
     int32_t   result;
@@ -230,9 +237,9 @@ __STATIC_INLINE int32_t arm_div_int64_to_int32(int64_t num, int32_t den)
      * if sum fits in 32bits
      * avoid costly 64-bit division
      */
-    if (num == (int64_t)LONG_MIN)
+    if (num == LLONG_MIN)
     {
-        absNum = LONG_MAX;
+        absNum = LLONG_MAX;
     }
     else
     {
@@ -248,7 +255,7 @@ __STATIC_INLINE int32_t arm_div_int64_to_int32(int64_t num, int32_t den)
         /*
          * 64-bit division
          */
-        result = (int32_t) (num / den);
+        result = clip_q63_to_q31(num / den);
 
     return result;
 }

Testing/PatternGeneration/FastMath.py

@@ -232,6 +232,13 @@ def cdiv(a,b):
 
     return(d)
 
+def sat_q63_to_q31(v):
+    if v > 0x7FFFFFFF:
+        return(0x7FFFFFFF)
+    if v < -0x80000000:
+        return(-0x80000000)
+    return(v)
+
 def testInt64(config):
     theInput=[0x1000000080000000,
                  0x0000000080000000,
@@ -264,13 +271,14 @@ def testInt64(config):
     ( -0x0000000080000000,2)
     ]
 
-    res = [tocint32(cdiv(x,y))  for (x,y) in allCombinations]
+    res = [sat_q63_to_q31(cdiv(x,y))  for (x,y) in allCombinations]
 
     allCombinations=np.array(allCombinations,dtype=np.int64).flatten()
+    config.setOverwrite(True)
     config.writeInputS64(1,allCombinations[0::2],"DivDenInput")
     config.writeInputS32(1,allCombinations[1::2],"DivNumInput")
 
-    config.writeReferenceU32(1, res,"DivRef")
+    config.writeReferenceS32(1, res,"DivRef")
     config.setOverwrite(False)
 
 

Testing/Patterns/DSP/FastMath/FastMathQ63/DivRef1_u32.txt renamed to Testing/Patterns/DSP/FastMath/FastMathQ63/DivRef1_s32.txt

@@ -1,26 +1,26 @@
 W
 12
-// 4294967295
-0xFFFFFFFF
-// 0
-0x00000000
-// 2147483649
-0x80000001
 // 2147483647
 0x7FFFFFFF
+// -2147483648
+0x80000000
+// 2147483647
+0x7FFFFFFF
+// -2147483648
+0x80000000
 // 1073741824
 0x40000000
-// 3221225472
+// -1073741824
 0xC0000000
 // 536870912
 0x20000000
-// 3758096384
+// -536870912
 0xE0000000
 // 1073741824
 0x40000000
-// 3221225472
+// -1073741824
 0xC0000000
 // 536870912
 0x20000000
-// 3221225472
+// -1073741824
 0xC0000000

Testing/Source/Tests/FastMathQ63.cpp

@@ -13,6 +13,7 @@ Reference patterns are generated with
 a double precision computation.
 
 */
+
 #define ABS_ERROR ((q63_t)0)
 
     void FastMathQ63::test_norm_64_to_32u()

Testing/desc.txt

@@ -1955,7 +1955,7 @@ group Root {
 
               Pattern DIV_DEN_INPUT1_S64_ID : DivDenInput1_s64.txt
               Pattern DIV_NUM_INPUT1_S32_ID : DivNumInput1_s32.txt
-              Pattern DIV_REF_S32_ID : DivRef1_u32.txt
+              Pattern DIV_REF_S32_ID : DivRef1_s32.txt
 
               Output  OUT_S32_ID : Output
               Output  NORMS_S16_ID : Output