Added Agner's asmlib (#87)

regamedll/build.gradle

@@ -130,6 +130,8 @@ void setupToolchain(NativeBinarySpec b)
 			cfg.linkerOptions.randomizedBaseAddress = false
 			cfg.linkerOptions.baseAddress = '0x4970000'
 		}
+		cfg.projectLibpath(project, '/lib')
+		cfg.extraLibs 'libacof32.lib'
 	}
 	else if (cfg instanceof GccToolchainConfig)
 	{
@@ -153,7 +155,8 @@ void setupToolchain(NativeBinarySpec b)
 
 		cfg.linkerOptions.args '-no-opt-class-analysis'
 		cfg.compilerOptions.args '-Qoption,cpp,--treat_func_as_string_literal_cpp', '-g0', '-fno-rtti'
-		cfg.extraLibs 'dl', 'm', 'stdc++'
+		cfg.projectLibpath(project, '/lib/linux32')
+		cfg.extraLibs 'dl', 'm', 'stdc++', 'aelf32'
 	}
 
 	if (mpLib && GradleCppUtils.windows && !unitTestExecutable) {

regamedll/common/mathlib.h

@@ -106,24 +106,61 @@ inline T& operator&= (T& a, T b) { return (T&)((type&)a &= (type)b); }
 template<class T, class type=typename std::underlying_type<T>::type>
 inline T& operator^= (T& a, T b) { return (T&)((type&)a ^= (type)b); }
 
-inline double M_sqrt(int value) {
-	return sqrt(value);
-}
-
 inline float M_sqrt(float value) {
 	return _mm_cvtss_f32(_mm_sqrt_ss(_mm_load_ss(&value)));
 }
 
 inline double M_sqrt(double value) {
-	double ret;
 	auto v = _mm_load_sd(&value);
-	_mm_store_sd(&ret, _mm_sqrt_sd(v, v));
-	return ret;
+	return _mm_cvtsd_f64(_mm_sqrt_sd(v, v));
+}
+
+template<typename T>
+inline double M_sqrt(T value) {
+	return sqrt(value);
+}
+
+inline float M_min(float a, float b) {
+	return _mm_cvtss_f32(_mm_min_ss(_mm_load_ss(&a), _mm_load_ss(&b)));
+}
+
+inline double M_min(double a, double b) {
+	return _mm_cvtsd_f64(_mm_min_sd(_mm_load_sd(&a), _mm_load_sd(&b)));
+}
+
+template<typename T>
+inline T M_min(T a, T b) {
+	return min(a, b);
+}
+
+inline float M_max(float a, float b) {
+	return _mm_cvtss_f32(_mm_max_ss(_mm_load_ss(&a), _mm_load_ss(&b)));
+}
+
+inline double M_max(double a, double b) {
+	return _mm_cvtsd_f64(_mm_max_sd(_mm_load_sd(&a), _mm_load_sd(&b)));
+}
+
+template<typename T>
+inline T M_max(T a, T b) {
+	return max(a, b);
+}
+
+inline float M_clamp(float a, float min, float max) {
+	return _mm_cvtss_f32(_mm_min_ss(_mm_max_ss(_mm_load_ss(&a), _mm_load_ss(&min)), _mm_load_ss(&max)));
+}
+
+inline double M_clamp(double a, double min, double max) {
+	return _mm_cvtsd_f64(_mm_min_sd(_mm_max_sd(_mm_load_sd(&a), _mm_load_sd(&min)), _mm_load_sd(&max)));
+}
+
+template<typename T>
+inline T M_clamp(T a, T min, T max) {
+	return clamp(a, min, max);
 }
 
 template<class T>
-inline void SWAP(T &first, T &second)
-{
+inline void SWAP(T &first, T &second) {
 	T temp = first;
 	first = second;
 	second = temp;

regamedll/msvc/ReGameDLL.vcxproj

@@ -1075,6 +1075,7 @@
     <ClInclude Include="..\pm_shared\pm_movevars.h" />
     <ClInclude Include="..\pm_shared\pm_shared.h" />
     <ClInclude Include="..\public\archtypes.h" />
+    <ClInclude Include="..\public\asmlib.h" />
     <ClInclude Include="..\public\basetypes.h" />
     <ClInclude Include="..\public\commonmacros.h" />
     <ClInclude Include="..\public\FileSystem.h" />

regamedll/msvc/ReGameDLL.vcxproj.filters

@@ -1127,6 +1127,9 @@
     <ClInclude Include="..\dlls\mapinfo.h">
       <Filter>dlls</Filter>
     </ClInclude>
+    <ClInclude Include="..\public\asmlib.h">
+      <Filter>public</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <None Include="..\linux\appversion.sh">

regamedll/public/asmlib.h

@@ -0,0 +1,123 @@
+/*************************** asmlib.h ***************************************
+* Author:        Agner Fog
+* Date created:  2003-12-12
+* Last modified: 2013-10-04
+* Project:       asmlib.zip
+* Source URL:    www.agner.org/optimize
+*
+* Description:
+* Header file for the asmlib function library.
+* This library is available in many versions for different platforms.
+* See asmlib-instructions.pdf for details.
+*
+* (c) Copyright 2003 - 2013 by Agner Fog. 
+* GNU General Public License http://www.gnu.org/licenses/gpl.html
+*****************************************************************************/
+
+
+#ifndef ASMLIB_H
+#define ASMLIB_H
+
+
+/***********************************************************************
+Define compiler-specific types and directives
+***********************************************************************/
+
+// Define type size_t
+#ifndef _SIZE_T_DEFINED
+#include "stddef.h"
+#endif
+
+// Define integer types with known size: int32_t, uint32_t, int64_t, uint64_t.
+// If this doesn't work then insert compiler-specific definitions here:
+#if defined(__GNUC__) || (defined(_MSC_VER) && _MSC_VER >= 1600)
+  // Compilers supporting C99 or C++0x have stdint.h defining these integer types
+  #include <stdint.h>
+  #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers
+#elif defined(_MSC_VER)
+  // Older Microsoft compilers have their own definition
+  typedef signed   __int16  int16_t;
+  typedef unsigned __int16 uint16_t;
+  typedef signed   __int32  int32_t;
+  typedef unsigned __int32 uint32_t;
+  typedef signed   __int64  int64_t;
+  typedef unsigned __int64 uint64_t;
+  #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers
+#else
+  // This works with most compilers
+  typedef signed   short int  int16_t;
+  typedef unsigned short int uint16_t;
+  typedef signed   int        int32_t;
+  typedef unsigned int       uint32_t;
+  typedef long long           int64_t;
+  typedef unsigned long long uint64_t;
+  #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers
+#endif
+
+
+// Turn off name mangling
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/***********************************************************************
+Function prototypes, memory and string functions
+***********************************************************************/
+void * A_memcpy (void * dest, const void * src, size_t count); // Copy count bytes from src to dest
+void * A_memmove(void * dest, const void * src, size_t count); // Same as memcpy, allows overlap between src and dest
+void * A_memset (void * dest, int c, size_t count);            // Set count bytes in dest to (char)c
+int    A_memcmp (const void * buf1, const void * buf2, size_t num); // Compares two blocks of memory
+size_t GetMemcpyCacheLimit(void);                              // Data blocks bigger than this will be copied uncached by memcpy and memmove
+void   SetMemcpyCacheLimit(size_t);                            // Change limit in GetMemcpyCacheLimit
+size_t GetMemsetCacheLimit(void);                              // Data blocks bigger than this will be stored uncached by memset
+void   SetMemsetCacheLimit(size_t);                            // Change limit in GetMemsetCacheLimit
+char * A_strcat (char * dest, const char * src);               // Concatenate strings dest and src. Store result in dest
+char * A_strcpy (char * dest, const char * src);               // Copy string src to dest
+size_t A_strlen (const char * str);                            // Get length of zero-terminated string
+int    A_strcmp (const char * a, const char * b);              // Compare strings. Case sensitive
+int    A_stricmp (const char *string1, const char *string2);   // Compare strings. Case insensitive for A-Z only
+char * A_strstr (char * haystack, const char * needle);        // Search for substring in string
+void   A_strtolower(char * string);                            // Convert string to lower case for A-Z only
+void   A_strtoupper(char * string);                            // Convert string to upper case for a-z only
+size_t A_substring(char * dest, const char * source, size_t pos, size_t len); // Copy a substring for source into dest
+size_t A_strspn (const char * str, const char * set);          // Find span of characters that belong to set
+size_t A_strcspn(const char * str, const char * set);          // Find span of characters that don't belong to set
+size_t strCountInSet(const char * str, const char * set);      // Count characters that belong to set
+size_t strcount_UTF8(const char * str);                        // Counts the number of characters in a UTF-8 encoded string
+
+
+/***********************************************************************
+Function prototypes, miscellaneous functions
+***********************************************************************/
+uint32_t A_popcount(uint32_t x);                               // Count 1-bits in 32-bit integer
+int    RoundD (double x);                                      // Round to nearest or even
+int    RoundF (float  x);                                      // Round to nearest or even
+int    InstructionSet(void);                                   // Tell which instruction set is supported
+char * ProcessorName(void);                                    // ASCIIZ text describing microprocessor
+void   CpuType(int * vendor, int * family, int * model);       // Get CPU vendor, family and model
+size_t DataCacheSize(int level);                               // Get size of data cache
+void   A_DebugBreak(void);                                     // Makes a debug breakpoint
+#ifdef INT64_SUPPORTED
+   uint64_t ReadTSC(void);                                     // Read microprocessor internal clock (64 bits)
+#else
+   uint32_t ReadTSC(void);                                     // Read microprocessor internal clock (only 32 bits supported by compiler)
+#endif
+void cpuid_ex (int abcd[4], int eax, int ecx);                 // call CPUID instruction
+static inline void cpuid_abcd (int abcd[4], int eax) {
+   cpuid_ex(abcd, eax, 0);}
+
+#ifdef __cplusplus
+}  // end of extern "C"
+
+// Define overloaded versions if compiling as C++
+
+static inline int Round (double x) {   // Overload name Round
+   return RoundD(x);}
+static inline int Round (float  x) {   // Overload name Round
+   return RoundF(x);}
+static inline const char * A_strstr(const char * haystack, const char * needle) {
+   return A_strstr((char*)haystack, needle);} // Overload A_strstr with const char * version
+
+#endif // __cplusplus
+
+#endif // ASMLIB_H

regamedll/public/regamedll/regamedll_common.h

@@ -41,6 +41,57 @@
 #define Q_alloca alloca
 #define Q_free free
 
+#ifdef REGAMEDLL_FIXES
+#define Q_sqrt M_sqrt
+#define Q_min M_min
+#define Q_max M_max
+#define Q_clamp M_clamp
+#define Q_access _access
+#define Q_close _close
+#define Q_write _write
+#define Q_memset A_memset
+#define Q_memcpy A_memcpy
+#define Q_memcmp A_memcmp
+#define Q_memmove A_memmove
+#define Q_strlen A_strlen
+#define Q_strcpy A_strcpy
+#define Q_strncpy strncpy
+#define Q_strrchr A_strrchr
+#define Q_strcat A_strcat
+#define Q_strncat strncat
+#define Q_strcmp A_strcmp
+#define Q_strncmp strncmp
+#define Q_sscanf sscanf
+#define Q_strdup _strdup
+#define Q_stricmp _stricmp
+#define Q_strnicmp _strnicmp
+#define Q_strstr A_strstr
+#define Q_strchr strchr
+#define Q_strrchr A_strrchr
+#define Q_strlwr A_strtolower
+#define Q_sprintf sprintf
+#define Q_snprintf _snprintf
+#define Q_atoi atoi
+#define Q_atof atof
+#define Q_toupper toupper
+#define Q_vsnprintf _vsnprintf
+#define Q_vsnwprintf _vsnwprintf
+#define Q_abs abs
+#define Q_fabs fabs
+#define Q_tan tan
+#define Q_atan atan
+#define Q_atan2 atan2
+#define Q_acos acos
+#define Q_cos cos
+#define Q_sin sin
+#define Q_pow pow
+#define Q_fmod fmod
+#define Q_fopen fopen
+#define Q_fwrite fwrite
+#define Q_fprintf fprintf
+#define Q_fclose fclose
+#else
+#define Q_sqrt sqrt
 #define Q_min min
 #define Q_max max
 #define Q_clamp clamp
@@ -49,8 +100,9 @@
 #define Q_write _write
 #define Q_memset memset
 #define Q_memcpy memcpy
-#define Q_strlen strlen
 #define Q_memcmp memcmp
+#define Q_memmove memmove
+#define Q_strlen strlen
 #define Q_strcpy strcpy
 #define Q_strncpy strncpy
 #define Q_strrchr strrchr
@@ -71,7 +123,6 @@
 #define Q_atoi atoi
 #define Q_atof atof
 #define Q_toupper toupper
-#define Q_memmove memmove
 #define Q_vsnprintf _vsnprintf
 #define Q_vsnwprintf _vsnwprintf
 #define Q_abs abs
@@ -88,9 +139,4 @@
 #define Q_fwrite fwrite
 #define Q_fprintf fprintf
 #define Q_fclose fclose
-
-#ifdef REGAMEDLL_FIXES
-#define Q_sqrt M_sqrt
-#else
-#define Q_sqrt sqrt
 #endif

regamedll/regamedll/precompiled.h

@@ -6,6 +6,7 @@
 #include "basetypes.h"
 #include "archtypes.h"
 #include "sse_mathfun.h"
+#include "asmlib.h"
 
 #include "MemPool.h"
 #include "engine.h"