Added autotests

Fixed compilation for STM32 (no constexpr abs...)

Added autotests for opposite

Fixed auto tests for STM32?

Changed compilation guards for autotests. More tests

Fixed compilation of autotests

Fixed compilation for stm32

Fix of auto again

Author: tomcombriat

src/FixMath.h

@@ -157,7 +157,7 @@ namespace FixMathPrivate {
   // Alias declaration for a UFix type with the suitable NI count given RANGE and NF
   template<int8_t NF, uint64_t RANGE> using UFixByRange_t=UFix<NIcount<RANGE>()-NF, NF, RANGE>;
   // Alias declaration for an SFix type with the suitable NI count given RANGE and NF
-  template<int8_t NF, uint64_t RANGE> using SFixByRange_t=SFix<NIcount<RANGE>()-NF, NF, RANGE>;
+  template<int8_t NF, uint64_t RANGE> using SFixByRange_t=SFix<NIcount<RANGE-1>()-NF, NF, RANGE>;
 }
 
 /** Instanciate an unsigned fixed point math number.
@@ -1394,8 +1394,8 @@ constexpr SFix<sizeof(T)*8-1,0> toSInt(T val) {
     @endcode
 */
 template<int64_t value>
-constexpr const FixMathPrivate::SFixByRange_t<0, value < 0 ? abs(-value-1) : value> SFixAuto() {
-  return FixMathPrivate::SFixByRange_t<0, value < 0 ? abs(-value-1) : value>::fromRaw(value);
+constexpr const FixMathPrivate::SFixByRange_t<0, value < 0 ? -value : value+1> SFixAuto() {
+  return FixMathPrivate::SFixByRange_t<0, value < 0 ? -value : value+1>::fromRaw(value);
 }
 
 #include "FixMath_Autotests.h"

src/FixMath_Autotests.h

@@ -37,13 +37,27 @@ namespace FixMathPrivate {
       static_assert((e + e).getNF() == 2, "test fail");
       static_assert((d + e).getNI() == 11, "test fail");
 
-      // Two's complement peculiar additions
+      // Two's complement peculiar additions and opposite
       static_assert(SFixAuto<-128>().getNI() == 7, "test fail");
       static_assert(SFixAuto<127>().getNI() == 7, "test fail");
       static_assert(SFixAuto<128>().getNI() == 8, "test fail");
 
-      constexpr auto s = UFix<7,0>(-128);
+#if (__cplusplus >= 202002L)
+      constexpr auto s = SFix<7,0>(-128);
       static_assert((s+s).getNI() == 8, "test fail");
+      static_assert((-s).getNI() == 8, "test fail");
+      
+      constexpr auto zero = SFix<7,0>(0);
+      constexpr auto negone = SFix<1,0>(-1);
+      static_assert((zero - s).getNI() == 8, "test fail");
+      static_assert(-(zero - s) == s, "test fail");
+      static_assert((s*negone).getNI() == 8, "test fail");
+#endif
+      static_assert((-SFixAuto<-127>()).getNI() == 7, "test fail");
+      //static_assert((-SFixAuto<127>()).getNI() == 7, "test fail"); // if someone manages that I am very grateful
+      static_assert((-UFixAuto<127>()).getNI() == 7, "test fail");
+      
+
 
       // the point of this block is to ascertain that addtion does not overflow, internally, where the internal_type of the operands is too small to hold the result
       constexpr auto large = UFix<32,0>(1LL << 31, true);
@@ -58,7 +72,7 @@ namespace FixMathPrivate {
       static_assert(b + a - b == a, "test fail");
       static_assert(b + a + (-b) == a, "test fail");  // same with unary minus
       static_assert(-(-a) == a, "test fail");
-#if __cplusplus >= 202002L
+#if (__cplusplus >= 202002L)
       // These here involve shifts of negative numbers, which used to be "implementation defined" before C++-20.
       // It doesn't cause a real-world problem, but the compiler won't accept it in a constexpr
       static_assert(UFix<43,9>(0) - b - a == -(a+b), "test fail");