X-Git-Url: http://git.mutantstargoat.com/user/nuclear/?a=blobdiff_plain;f=src%2Futil.h;h=69d9d1e73a507e94846599f55719f587c5f6db4c;hb=387c4948b144c51c5b6fcfb8f558c3becf324f70;hp=d1f81eb45e0665054c30ff04b807029c4ba5ff95;hpb=b651991a8cb4cf8a1e64c66175f27091c805fdf6;p=retrobench

diff --git a/src/util.h b/src/util.h
index d1f81eb..69d9d1e 100644
--- a/src/util.h
+++ b/src/util.h
@@ -1,6 +1,7 @@
 #ifndef UTIL_H_
 #define UTIL_H_
 
+
 #ifdef NO_STDINT_H
 typedef char int8_t;
 typedef unsigned char uint8_t;
@@ -8,10 +9,32 @@ typedef short int16_t;
 typedef unsigned short uint16_t;
 typedef int int32_t;
 typedef unsigned int uint32_t;
+typedef unsigned long intptr_t;
 #else
 #include <stdint.h>
 #endif
 
+#ifdef __GNUC__
+#define INLINE __inline
+#define PACKED __attribute__((packed))
+
+#elif defined(__WATCOMC__)
+#define INLINE __inline
+#define PACKED
+
+#else
+#define INLINE
+#define PACKED
+#endif
+
+#define BSWAP16(x)	((((x) >> 8) & 0xff) | (((x) & 0xff) << 8))
+#define BSWAP32(x)	\
+	((((x) >> 24) & 0xff) | \
+	 (((x) >> 8) & 0xff00) | \
+	 (((x) << 8) & 0xff0000) | \
+	 ((x) << 24))
+
+
 extern int sinlut[];
 
 #define SIN(x) sinlut[(x) & 0x3ff]
@@ -19,4 +42,198 @@ extern int sinlut[];
 
 int mask_to_shift(unsigned int mask);
 
+#if defined(__i386__) || defined(__x86_64__) || defined(__386__) || defined(MSDOS)
+/* fast conversion of double -> 32bit int
+ * for details see:
+ *  - http://chrishecker.com/images/f/fb/Gdmfp.pdf
+ *  - http://stereopsis.com/FPU.html#convert
+ */
+static INLINE int32_t cround64(double val)
+{
+	val += 6755399441055744.0;
+	return *(int32_t*)&val;
+}
+#else
+#define cround64(x)	((int32_t)(x))
+#endif
+
+static INLINE float rsqrt(float x)
+{
+	float xhalf = x * 0.5f;
+	int32_t i = *(int32_t*)&x;
+	i = 0x5f3759df - (i >> 1);
+	x = *(float*)&i;
+	x = x * (1.5f - xhalf * x * x);
+	return x;
+}
+
+extern uint32_t perf_start_count, perf_interval_count;
+
+#ifdef __WATCOMC__
+void memset16(void *dest, uint16_t val, int count);
+#pragma aux memset16 = \
+	"cld" \
+	"test ecx, 1" \
+	"jz memset16_dwords" \
+	"rep stosw" \
+	"jmp memset16_done" \
+	"memset16_dwords:" \
+	"shr ecx, 1" \
+	"push ax" \
+	"shl eax, 16" \
+	"pop ax" \
+	"rep stosd" \
+	"memset16_done:" \
+	parm[edi][ax][ecx];
+
+#ifdef USE_MMX
+void memcpy64(void *dest, void *src, int count);
+#pragma aux memcpy64 = \
+	"cploop:" \
+	"movq mm0, [edx]" \
+	"movq [ebx], mm0" \
+	"add edx, 8" \
+	"add ebx, 8" \
+	"dec ecx" \
+	"jnz cploop" \
+	"emms" \
+	parm[ebx][edx][ecx] \
+	modify[8087];
+#else
+#define memcpy64(dest, src, count)	memcpy(dest, src, (count) << 3)
+#endif
+
+void perf_start(void);
+#pragma aux perf_start = \
+	"xor eax, eax" \
+	"cpuid" \
+	"rdtsc" \
+	"mov [perf_start_count], eax" \
+	modify[eax ebx ecx edx];
+
+void perf_end(void);
+#pragma aux perf_end = \
+	"xor eax, eax" \
+	"cpuid" \
+	"rdtsc" \
+	"sub eax, [perf_start_count]" \
+	"mov [perf_interval_count], eax" \
+	modify [eax ebx ecx edx];
+
+void debug_break(void);
+#pragma aux debug_break = "int 3";
+#endif
+
+#ifdef __GNUC__
+#if defined(__i386__) || defined(__x86_64__)
+#define memset16(dest, val, count) asm volatile ( \
+	"cld\n\t" \
+	"test $1, %2\n\t" \
+	"jz 0f\n\t" \
+	"rep stosw\n\t" \
+	"jmp 1f\n\t" \
+	"0:\n\t" \
+	"shr $1, %2\n\t" \
+	"push %%ax\n\t" \
+	"shl $16, %%eax\n\t" \
+	"pop %%ax\n\t" \
+	"rep stosl\n\t" \
+	"1:\n\t"\
+	:: "D"(dest), "a"((uint16_t)(val)), "c"(count) \
+	: "memory")
+#else
+static void INLINE memset16(void *dest, uint16_t val, int count)
+{
+	uint16_t *ptr = dest;
+	while(count--) *ptr++ = val;
+}
+#endif
+
+#ifdef USE_MMX
+#define memcpy64(dest, src, count) asm volatile ( \
+	"0:\n\t" \
+	"movq (%1), %%mm0\n\t" \
+	"movq %%mm0, (%0)\n\t" \
+	"add $8, %1\n\t" \
+	"add $8, %0\n\t" \
+	"dec %2\n\t" \
+	"jnz 0b\n\t" \
+	"emms\n\t" \
+	:: "r"(dest), "r"(src), "r"(count) \
+	: "%mm0")
+#else
+#define memcpy64(dest, src, count)	memcpy(dest, src, (count) << 3)
+#endif
+
+#define perf_start()  asm volatile ( \
+	"xor %%eax, %%eax\n" \
+	"cpuid\n" \
+	"rdtsc\n" \
+	"mov %%eax, %0\n" \
+	: "=m"(perf_start_count) \
+	:: "%eax", "%ebx", "%ecx", "%edx")
+
+#define perf_end() asm volatile ( \
+	"xor %%eax, %%eax\n" \
+	"cpuid\n" \
+	"rdtsc\n" \
+	"sub %1, %%eax\n" \
+	"mov %%eax, %0\n" \
+	: "=m"(perf_interval_count) \
+	: "m"(perf_start_count) \
+	: "%eax", "%ebx", "%ecx", "%edx")
+
+#define debug_break() \
+	asm volatile ("int $3")
+#endif
+
+#ifdef _MSC_VER
+void __inline memset16(void *dest, uint16_t val, int count)
+{
+	__asm {
+		cld
+		mov ax, val
+		mov edi, dest
+		mov ecx, count
+		test ecx, 1
+		jz memset16_dwords
+		rep stosw
+		jmp memset16_done
+		memset16_dwords:
+		shr ecx, 1
+		push ax
+		shl eax, 16
+		pop ax
+		rep stosd
+		memset16_done:
+	}
+}
+
+#define perf_start() \
+	do { \
+		__asm { \
+			xor eax, eax \
+			cpuid \
+			rdtsc \
+			mov [perf_start_count], eax \
+		} \
+	} while(0)
+
+#define perf_end() \
+	do { \
+		__asm { \
+			xor eax, eax \
+			cpuid \
+			rdtsc \
+			sub eax, [perf_start_count] \
+			mov [perf_interval_count], eax \
+		} \
+	} while(0)
+
+#define debug_break() \
+	do { \
+		__asm { int 3 } \
+	} while(0)
+#endif
+
 #endif	/* UTIL_H_ */