libnum: some additional functions

FindAzur.cmake

@@ -97,7 +97,8 @@ if(Azur_FOUND)
   set_target_properties(Azur::Azur PROPERTIES
     IMPORTED_LOCATION "${AZUR_LIB}"
     INTERFACE_INCLUDE_DIRECTORIES "${AZUR_INCLUDE}"
-    INTERFACE_COMPILE_OPTIONS "-DAZUR_PLATFORM=${AZUR_PLATFORM}")
+    INTERFACE_COMPILE_OPTIONS "-DAZUR_PLATFORM=${AZUR_PLATFORM}"
+    INTERFACE_LINK_LIBRARIES -lnum)
 
   if(AZUR_PLATFORM STREQUAL linux)
     pkg_check_modules(sdl2 REQUIRED sdl2 IMPORTED_TARGET)

libnum/CMakeLists.txt

@@ -8,6 +8,7 @@
 include(CTest)
 
 add_library(num STATIC
+  src/num.cpp
   src/str.cpp)
 
 target_include_directories(num PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

libnum/include/num/num.h

@@ -117,9 +117,15 @@ struct num8
         v %= other.v;
         return *this;
     }
+    inline constexpr num8 ifloor() {
+        return 0;
+    }
     inline constexpr num8 floor() {
         return num8(0);
     }
+    inline constexpr num8 iceil() {
+        return v != 0;
+    }
     /* Warning: num8::ceil() always overflows! */
     inline constexpr num8 ceil() {
         return num8(0);
@@ -195,6 +201,9 @@ struct num16
     /* num16 x num16 -> num32 multiplication
        This is efficiently implemented with a muls.w instruction. */
     static constexpr num32 dmul(num16 const &x, num16 const &y);
+    /* num16 / num16 -> num16 division for positive numbers
+       This bypasses some sign tests, which saves a bit of time. */
+    static constexpr num16 div_positive(num16 const &x, num16 const &y);
 
     /* Basic arithmetic */
 
@@ -218,11 +227,17 @@ struct num16
         v %= other.v;
         return *this;
     }
+    inline constexpr int ifloor() {
+        return v >> 8;
+    }
     inline constexpr num16 floor() {
         num16 x;
         x.v = v & 0xff00;
         return x;
     }
+    inline constexpr int iceil() {
+        return (v + 0xff) >> 8;
+    }
     inline constexpr num16 ceil() {
         num16 x;
         x.v = ((v - 1) | 0xff) + 1;
@@ -330,11 +345,17 @@ struct num32
         v %= other.v;
         return *this;
     }
+    inline constexpr int ifloor() {
+        return v >> 16;
+    }
     inline constexpr num32 floor() {
         num32 x;
         x.v = v & 0xffff0000;
         return x;
     }
+    inline constexpr int iceil() {
+        return (v + 0xffff) >> 16;
+    }
     inline constexpr num32 ceil() {
         num32 x;
         x.v = ((v - 1) | 0xffff) + 1;
@@ -345,6 +366,7 @@ struct num32
         x.v = v & 0xffff;
         return x;
     }
+    num32 sqrt() const;
 
     /* Comparisons with int */
 
@@ -447,11 +469,17 @@ struct num64
         v %= other.v;
         return *this;
     }
+    inline constexpr int ifloor() {
+        return v >> 32;
+    }
     inline constexpr num64 floor() {
         num64 x;
         x.v = v & 0xffffffff00000000ull;
         return x;
     }
+    inline constexpr int iceil() {
+        return (v >> 32) + ((uint32_t)v != 0);
+    }
     inline constexpr num64 ceil() {
         num64 x;
         x.v = ((v - 1) | 0xffffffffull) + 1;
@@ -569,6 +597,26 @@ inline constexpr T operator-(T const &op) {
     return T(0) - op;
 }
 
+/* Internal minima, maxima and clamp */
+
+template<typename T> requires(is_num<T>)
+inline constexpr T min(T const &left, T const &right)
+{
+    return (left < right) ? left : right;
+}
+
+template<typename T> requires(is_num<T>)
+inline constexpr T max(T const &left, T const &right)
+{
+    return (left > right) ? left : right;
+}
+
+template<typename T> requires(is_num<T>)
+inline constexpr T clamp(T const &val, T const &lower, T const &upper)
+{
+    return max(lower, min(val, upper));
+}
+
 /* Other specific operations */
 
 inline constexpr num32 num16::dmul(num16 const &x, num16 const &y)
@@ -578,6 +626,13 @@ inline constexpr num32 num16::dmul(num16 const &x, num16 const &y)
     return n;
 }
 
+inline constexpr num16 num16::div_positive(num16 const &x, num16 const &y)
+{
+    num16 r;
+    r.v = ((uint32_t)(uint16_t)x.v << 8) / (uint16_t)y.v;
+    return r;
+}
+
 inline constexpr num64 num32::dmul(num32 const &x, num32 const &y)
 {
     num64 n;
@@ -585,4 +640,12 @@ inline constexpr num64 num32::dmul(num32 const &x, num32 const &y)
     return n;
 }
 
+/* Floor modulo. We provide an optimized version for constants, which optimizes
+   away the main condition. */
+template<typename T> requires(is_num<T>)
+inline constexpr T modf(T const &x, T const &y) {
+    T r = x % y;
+    return (r.v && (r.v ^ y.v) < 0) ? r + y : r;
+}
+
 } /* namespace libnum */

libnum/include/num/vec.h

@@ -189,6 +189,10 @@ inline constexpr vec<T,N> operator*(T const &lhs, vec<T,N> rhs) {
         rhs[i] *= lhs;
     return rhs;
 }
+template<typename T, int N>
+inline constexpr vec<T,N> operator/(vec<T,N> lhs, T const &rhs) {
+    return lhs /= rhs;
+}
 
 /* Comparisons */
 

libnum/src/num.cpp

@@ -0,0 +1,21 @@
+#include <num/num.h>
+using namespace libnum;
+
+/* Integer square root (rather slow) */
+static int64_t sqrtll(int64_t n)
+{
+    if(n < 4)
+        return (n > 0);
+
+    int64_t low_bound = sqrtll(n / 4) * 2;
+    int64_t high_bound = low_bound + 1;
+
+    return (high_bound * high_bound <= n) ? high_bound : low_bound;
+}
+
+num32 num32::sqrt() const
+{
+    num32 r;
+    r.v = sqrtll((int64_t)v << 16);
+    return r;
+}

libnum/test/isel.py

@@ -306,7 +306,7 @@ class ExprParser:
         return decorate
 
     def binaryOpsRight(ctor, ops):
-        return binaryOpsRight(ctor, ops, rassoc=True)
+        return binaryOps(ctor, ops, rassoc=True)
 
     def unaryOps(ctor, ops, assoc=True):
         def decorate(f):