Merge branch 'builtin_math' into 'master'

Prefer relying on built in codegen rather than relying on sh4_math.h

See merge request simulant/GLdc!141

GL/alloc/alloc.c

@@ -538,6 +538,6 @@ size_t alloc_count_continuous(void* pool) {
     if(free_bits && (free_bits * 256) > largest_block) {
         largest_block = (free_bits * 256);
     }
-
+printf("LARGEST: %d\n", largest_block);
     return largest_block;
 }

GL/draw.c

@@ -757,9 +757,7 @@ static void _readNormalData(ReadNormalFunc func, const GLuint first, const GLuin
 
         if(_glIsNormalizeEnabled()) {
             GLfloat* n = (GLfloat*) it->nxyz;
-            float temp = n[0] * n[0];
-            temp = MATH_fmac(n[1], n[1], temp);
-            temp = MATH_fmac(n[2], n[2], temp);
+            float temp = n[0] * n[0] + n[1] * n[1] + n[2] * n[2];
 
             float ilength = MATH_fsrra(temp);
             n[0] *= ilength;

GL/lighting.c

@@ -58,10 +58,10 @@ void _glPrecalcLightingValues(GLuint mask) {
     if((mask & AMBIENT_MASK) || (mask & EMISSION_MASK) || (mask & SCENE_AMBIENT_MASK)) {
         GLfloat* scene_ambient = _glLightModelSceneAmbient();
 
-        material->baseColour[0] = MATH_fmac(scene_ambient[0], material->ambient[0], material->emissive[0]);
-        material->baseColour[1] = MATH_fmac(scene_ambient[1], material->ambient[1], material->emissive[1]);
-        material->baseColour[2] = MATH_fmac(scene_ambient[2], material->ambient[2], material->emissive[2]);
-        material->baseColour[3] = MATH_fmac(scene_ambient[3], material->ambient[3], material->emissive[3]);
+        material->baseColour[0] = scene_ambient[0] * material->ambient[0] + material->emissive[0];
+        material->baseColour[1] = scene_ambient[1] * material->ambient[1] + material->emissive[1];
+        material->baseColour[2] = scene_ambient[2] * material->ambient[2] + material->emissive[2];
+        material->baseColour[3] = scene_ambient[3] * material->ambient[3] + material->emissive[3];
     }
 }
 

GL/matrix.c

@@ -372,11 +372,12 @@ static inline void vec3f_cross(const GLfloat* v1, const GLfloat* v2, GLfloat* re
 }
 
 GL_FORCE_INLINE void vec3f_normalize_sh4(float *v){
-    float length, ilength;
+    float lengthSq, ilength;
 
-    ilength = MATH_fsrra(v[0]*v[0] + v[1]*v[1] + v[2]*v[2]);
-    length = MATH_Fast_Invert(ilength);
-    if (length)
+    lengthSq = v[0]*v[0] + v[1]*v[1] + v[2]*v[2];
+    ilength  = MATH_fsrra(lengthSq);
+
+    if (lengthSq)
     {
         v[0] *= ilength;
         v[1] *= ilength;

GL/platforms/sh4.h

@@ -10,8 +10,6 @@
 #include "../types.h"
 #include "../private.h"
 
-#include "sh4_math.h"
-
 #ifndef NDEBUG
 #define PERF_WARNING(msg) printf("[PERF] %s\n", msg)
 #else
@@ -24,6 +22,42 @@
 #define GL_FORCE_INLINE static GL_INLINE_DEBUG
 #endif
 
+
+// ---- sh4_math.h - SH7091 Math Module ----
+//
+// This file is part of the DreamHAL project, a hardware abstraction library
+// primarily intended for use on the SH7091 found in hardware such as the SEGA
+// Dreamcast game console.
+//
+// This math module is hereby released into the public domain in the hope that it
+// may prove useful. Now go hit 60 fps! :)
+//
+// --Moopthehedgehog
+
+// 1/sqrt(x)
+GL_FORCE_INLINE float MATH_fsrra(float x)
+{
+  asm volatile ("fsrra %[one_div_sqrt]\n"
+  : [one_div_sqrt] "+f" (x) // outputs, "+" means r/w
+  : // no inputs
+  : // no clobbers
+  );
+
+  return x;
+}
+
+// 1/x = 1 / sqrt(x^2)
+GL_FORCE_INLINE float MATH_Fast_Invert(float x)
+{
+  int neg = x < 0.0f;
+
+  x = MATH_fsrra(x * x);
+
+  if (neg) x = -x;
+  return x;
+}
+// end of ---- sh4_math.h ----
+
 #define PREFETCH(addr) __builtin_prefetch((addr))
 
 GL_FORCE_INLINE void* memcpy_fast(void *dest, const void *src, size_t len) {

GL/platforms/software.h

@@ -7,9 +7,6 @@
 
 #define PREFETCH(addr) do {} while(0)
 
-#define MATH_Fast_Divide(n, d) (n / d)
-#define MATH_fmac(a, b, c) (a * b + c)
-#define MATH_Fast_Sqrt(x) sqrtf((x))
 #define MATH_fsrra(x) (1.0f / sqrtf((x)))
 #define MATH_Fast_Invert(x) (1.0f / (x))
 
@@ -28,7 +25,7 @@
     } while(0)
 
 #define VEC3_LENGTH(x, y, z, d) \
-    d = MATH_Fast_Sqrt((x) * (x) + (y) * (y) + (z) * (z))
+    d = sqrtf((x) * (x) + (y) * (y) + (z) * (z))
 
 #define VEC3_DOT(x1, y1, z1, x2, y2, z2, d) \
     d = (x1 * x2) + (y1 * y2) + (z1 * z2)