Update structure of proj and fix some of fp16

Author: Rinzii

mim/include/float16.hpp

@@ -0,0 +1,59 @@
+// 16 bit floating point data structure
+
+#pragma once
+
+#include <cstdint>
+#include "internal/float16compressor.hpp"
+
+namespace mim
+{
+    struct float16
+    {
+        uint16_t value;
+
+        float16() = default;
+
+        inline float16(const float16& other) : value(other.value) {}
+        inline explicit float16(const uint16_t other) : value(other) {}
+
+        inline float16& operator=(const float16& other)
+        {
+            value = other.value;
+            return *this;
+        }
+
+        inline explicit float16(const float other)
+        {
+            value = Float16Compressor::Compress(other);
+        }
+
+        inline explicit operator float() const
+        {
+            uint16_t temp;
+            std::memcpy(&temp, &value, sizeof(temp));
+            return Float16Compressor::Decompress(temp);
+        }
+
+        inline float ToFloat() const
+        {
+            uint16_t temp;
+            std::memcpy(&temp, &value, sizeof(temp));
+            return Float16Compressor::Decompress(temp);
+        }
+
+        inline bool operator==(const float16& other) const
+        {
+            return value == other.value;
+        }
+
+        inline bool operator!=(const float16& other) const
+        {
+            return value != other.value;
+        }
+
+        static const float16 Zero() { return float16(static_cast<uint16_t>(0)); }
+    };
+
+    // Validate that the float16 structure is packed correctly.
+    static_assert(sizeof(float16) == sizeof(uint16_t), "float16 structure is not packed correctly.");
+}

mim/include/mim/float16.hpp

@@ -18,28 +18,28 @@ namespace mim
 			uint32_t uint;
 		};
 
-		static int32_t const shift = 13;	 				// mantissa bits: 13, exponent bits: 16
-		static int32_t const shiftSign = 16; 				// sign bits: 1, left shift amount: 15
+		static int32_t constexpr shift = 13;	 				// mantissa bits: 13, exponent bits: 16
+		static int32_t constexpr shiftSign = 16; 				// sign bits: 1, left shift amount: 15
 
-		static int32_t const inf32 = 0x7F800000;  			// FP32 infinity bit pattern
-		static int32_t const max32 = 0x477FE000;  			// max FP16 normal as a FP32 bit pattern
-		static int32_t const min32 = 0x38800000;  			// min FP16 normal as a FP32 bit pattern
-		static int32_t const sign32 = 0x80000000; 			// FP32 sign bit pattern
+		static int32_t constexpr inf32 = 0x7F800000;  			// FP32 infinity bit pattern
+		static int32_t constexpr max32 = 0x477FE000;  			// max FP16 normal as a FP32 bit pattern
+		static int32_t constexpr min32 = 0x38800000;  			// min FP16 normal as a FP32 bit pattern
+		static int32_t constexpr sign32 = 0x80000000; 			// FP32 sign bit pattern
 
-		static int32_t const inf16 = inf32 >> shift;	   	// FP16 infinity bit pattern
-		static int32_t const nan16 = (inf16 + 1) << shift; 	// min FP16 NaN as a FP32 bit pattern
-		static int32_t const max16 = max32 >> shift;	   	// max FP16 normal bit pattern
-		static int32_t const min16 = min32 >> shift;	   	// min FP16 normal bit pattern
-		static int32_t const sign16 = sign32 >> shiftSign; 	// FP16 sign bit pattern
+		static int32_t constexpr inf16 = inf32 >> shift;	   	// FP16 infinity bit pattern
+		static int32_t constexpr nan16 = (inf16 + 1) << shift; 	// min FP16 NaN as a FP32 bit pattern
+		static int32_t constexpr max16 = max32 >> shift;	   	// max FP16 normal bit pattern
+		static int32_t constexpr min16 = min32 >> shift;	   	// min FP16 normal bit pattern
+		static int32_t constexpr sign16 = sign32 >> shiftSign; 	// FP16 sign bit pattern
 
-		static int32_t const mul32to16 = 0x52000000; 		// multiplier to convert FP32 to FP16
-		static int32_t const mul16to32 = 0x33800000; 		// multiplier to convert FP16 to FP32
+		static int32_t constexpr mul32to16 = 0x52000000; 		// multiplier to convert FP32 to FP16
+		static int32_t constexpr mul16to32 = 0x33800000; 		// multiplier to convert FP16 to FP32
 
-		static int32_t const sub32 = 0x003FF; 				// maximum float32 subnormal value, down shifted
-		static int32_t const nor32 = 0x00400; 				// minimum float32 normal value, down shifted
+		static int32_t constexpr sub32 = 0x003FF; 				// maximum float32 subnormal value, down shifted
+		static int32_t constexpr nor32 = 0x00400; 				// minimum float32 normal value, down shifted
 
-		static int32_t const maxDiff = inf16 - max16 - 1; 	// diff. between max FP16 normal and infinity
-		static int32_t const minDiff = min16 - sub32 - 1;  	// diff. between min FP16 normal and max float32 subnormal
+		static int32_t constexpr maxDiff = inf16 - max16 - 1; 	// diff. between max FP16 normal and infinity
+		static int32_t constexpr minDiff = min16 - sub32 - 1;  	// diff. between min FP16 normal and max float32 subnormal
 
 
 	public:
@@ -55,7 +55,7 @@ namespace mim
 
 			// Calculate the shifted bits of the floating-point number.
 			shift_bits.sint = mul32to16;
-			shift_bits.sint = (int32_t)(shift_bits.fl * value_bits.fl); // correct subnormals
+			shift_bits.sint = static_cast<int32_t>(shift_bits.fl * value_bits.fl); // correct subnormals
 
 			// Adjust the floating-point number's bits based on certain conditions.
 			value_bits.sint ^= (shift_bits.sint ^ value_bits.sint) & -(min32 > value_bits.sint);
@@ -70,7 +70,7 @@ namespace mim
 			value_bits.sint ^= ((value_bits.sint - minDiff) ^ value_bits.sint) & -(value_bits.sint > sub32);
 
 			// Combine the sign and bits of the compressed number and return it.
-			return (uint16_t)(value_bits.uint | sign_bits);
+			return static_cast<uint16_t>(value_bits.uint | sign_bits);
 		}
 
 		static float Decompress(uint16_t compressed_value)