fixed division, undefined behaviors and some improvements (#33)

* testing using ctest

* emove old testing script

* added github workflow CI

* updated CI

* added unit test for macros

* F16() and F16C() are both rounding allways, so fix16_from_dbl should as well

* added tests for strign operations, but these functions are in my opinion unreliable and tests are failing

* removed old unittests

* removed old unittests from cmake

* problem with division using gcc

* improved benchmark

* clarification of problem with division

* attempt to fix

* fixed some undefined behaviors, fixed division

1 files changed, 16 insertions, 29 deletions

diff --git a/benchmarks/benchmark.c b/benchmarks/benchmark.c
index 70a2aeb..3942abc 100644
--- a/benchmarks/benchmark.c
+++ b/benchmarks/benchmark.c
@@ -9,15 +9,6 @@
 /* Autogenerated testcases */
 #include "testcases.c"
 
-/* Tools for profiling */
-
-typedef struct {
-    uint32_t min;
-    uint32_t max;
-    uint32_t sum;
-    uint32_t count;
-} cyclecount_t;
-
 // Initializer for cyclecount_t structure.
 // Max is initialized to 0 and min is 2^32-1 so that first call to cyclecount_update will set them.
 #define CYCLECOUNT_INIT {0xFFFFFFFF, 0, 0, 0}
@@ -40,12 +31,6 @@ static void cyclecount_update(cyclecount_t *data, uint32_t cycles)
     cyclecount_update(&variable, end_timing()); \
 }
 
-#define PRINT(variable, label) { \
-    print_value(label " min", variable.min); \
-    print_value(label " max", variable.max); \
-    print_value(label " avg", variable.sum / variable.count); \
-}
-
 static cyclecount_t exp_cycles = CYCLECOUNT_INIT;
 static cyclecount_t sqrt_cycles = CYCLECOUNT_INIT;
 static cyclecount_t add_cycles = CYCLECOUNT_INIT;
@@ -53,13 +38,12 @@ static cyclecount_t sub_cycles = CYCLECOUNT_INIT;
 static cyclecount_t div_cycles = CYCLECOUNT_INIT;
 static cyclecount_t mul_cycles = CYCLECOUNT_INIT;
 
-#ifndef NO_FLOAT
 static cyclecount_t float_sqrtf_cycles = CYCLECOUNT_INIT;
+static cyclecount_t float_expf_cycles = CYCLECOUNT_INIT;
 static cyclecount_t float_add_cycles = CYCLECOUNT_INIT;
 static cyclecount_t float_sub_cycles = CYCLECOUNT_INIT;
 static cyclecount_t float_div_cycles = CYCLECOUNT_INIT;
 static cyclecount_t float_mul_cycles = CYCLECOUNT_INIT;
-#endif
 
 static fix16_t delta(fix16_t result, fix16_t expected)
 {
@@ -119,8 +103,6 @@ int main()
             print_value("Failed EXP, expected", expected);
         }
     }
-    PRINT(sqrt_cycles, "fix16_sqrt");
-    PRINT(exp_cycles, "fix16_exp");
 
     for (i = 0; i < TESTCASES2_COUNT; i++)
     {
@@ -175,10 +157,6 @@ int main()
             }
         }
     }
-    PRINT(add_cycles, "fix16_add");
-    PRINT(sub_cycles, "fix16_sub");
-    PRINT(mul_cycles, "fix16_mul");
-    PRINT(div_cycles, "fix16_div");
     
     /* Compare with floating point performance */
 #ifndef NO_FLOAT
@@ -187,8 +165,8 @@ int main()
         float input = fix16_to_float(testcases1[i].a);
         volatile float result;
         MEASURE(float_sqrtf_cycles, result = sqrtf(input));
+        MEASURE(float_expf_cycles, result = expf(input));
     }
-    PRINT(float_sqrtf_cycles, "float sqrtf");
     
     for (i = 0; i < TESTCASES2_COUNT; i++)
     {
@@ -204,11 +182,20 @@ int main()
             MEASURE(float_div_cycles, result = a / b);
         }
     }
-    PRINT(float_add_cycles, "float add");
-    PRINT(float_sub_cycles, "float sub");
-    PRINT(float_mul_cycles, "float mul");
-    PRINT(float_div_cycles, "float div");
-#endif    
+#endif 
+
+    print("fix16_sqrt", &sqrt_cycles);
+    print("float sqrtf", &float_sqrtf_cycles);
+    print("fix16_exp", &exp_cycles);
+    print("float expf", &float_expf_cycles);
+    print("fix16_add", &add_cycles);
+    print("float add", &float_add_cycles);
+    print("fix16_sub", &sub_cycles);
+    print("float sub", &float_sub_cycles);
+    print("fix16_mul", &mul_cycles);
+    print("float mul", &float_mul_cycles);
+    print("fix16_div", &div_cycles);
+    print("float div", &float_div_cycles);
 
     return 0;
 }