Cycles: use LIKELY/UNLIKELY macros

Gives overall ~3% speedup in own tests for BMW scene.

intern/cycles/kernel/osl/osl_services.cpp

@@ -50,7 +50,11 @@ CCL_NAMESPACE_BEGIN
 
 /* RenderServices implementation */
 
-#define COPY_MATRIX44(m1, m2) memcpy(m1, m2, sizeof(*m2))
+#define COPY_MATRIX44(m1, m2)  { \
+	CHECK_TYPE(m1, OSL::Matrix44*); \
+	CHECK_TYPE(m2, Transform*); \
+	memcpy(m1, m2, sizeof(*m2)); \
+} (void)0
 
 /* static ustrings */
 ustring OSLRenderServices::u_distance("distance");

intern/cycles/util/util_color.h

@@ -61,22 +61,22 @@ ccl_device float3 rgb_to_hsv(float3 rgb)
 		h = 0.0f;
 	}
 
-	if(s == 0.0f) {
+	if(s != 0.0f) {
-		h = 0.0f;
-	}
-	else {
 		float3 cmax3 = make_float3(cmax, cmax, cmax);
 		c = (cmax3 - rgb)/cdelta;
 
-		if(rgb.x == cmax) h = c.z - c.y;
+		if     (rgb.x == cmax) h =        c.z - c.y;
-		else if(rgb.y == cmax) h = 2.0f + c.x -  c.z;
+		else if(rgb.y == cmax) h = 2.0f + c.x - c.z;
-		else h = 4.0f + c.y - c.x;
+		else                   h = 4.0f + c.y - c.x;
 
 		h /= 6.0f;
 
 		if(h < 0.0f)
 			h += 1.0f;
 	}
+	else {
+		h = 0.0f;
+	}
 
 	return make_float3(h, s, v);
 }
@@ -90,13 +90,10 @@ ccl_device float3 hsv_to_rgb(float3 hsv)
 	s = hsv.y;
 	v = hsv.z;
 
-	if(s == 0.0f) {
+	if(s != 0.0f) {
-		rgb = make_float3(v, v, v);
-	}
-	else {
 		if(h == 1.0f)
 			h = 0.0f;
-		
+
 		h *= 6.0f;
 		i = floorf(h);
 		f = h - i;
@@ -104,13 +101,16 @@ ccl_device float3 hsv_to_rgb(float3 hsv)
 		p = v*(1.0f-s);
 		q = v*(1.0f-(s*f));
 		t = v*(1.0f-(s*(1.0f-f)));
-		
+
-		if(i == 0.0f) rgb = make_float3(v, t, p);
+		if     (i == 0.0f) rgb = make_float3(v, t, p);
 		else if(i == 1.0f) rgb = make_float3(q, v, p);
 		else if(i == 2.0f) rgb = make_float3(p, v, t);
 		else if(i == 3.0f) rgb = make_float3(p, q, v);
 		else if(i == 4.0f) rgb = make_float3(t, p, v);
-		else rgb = make_float3(v, p, q);
+		else               rgb = make_float3(v, p, q);
+	}
+	else {
+		rgb = make_float3(v, v, v);
 	}
 
 	return rgb;

intern/cycles/util/util_math.h

@@ -1237,7 +1237,7 @@ ccl_device float compatible_powf(float x, float y)
 
 ccl_device float safe_powf(float a, float b)
 {
-	if(a < 0.0f && b != float_to_int(b))
+	if(UNLIKELY(a < 0.0f && b != float_to_int(b)))
 		return 0.0f;
 
 	return compatible_powf(a, b);
@@ -1245,7 +1245,7 @@ ccl_device float safe_powf(float a, float b)
 
 ccl_device float safe_logf(float a, float b)
 {
-	if(a < 0.0f || b < 0.0f)
+	if(UNLIKELY(a < 0.0f || b < 0.0f))
 		return 0.0f;
 
 	return logf(a)/logf(b);
@@ -1305,7 +1305,7 @@ ccl_device bool ray_aligned_disk_intersect(
 	float3 disk_N = normalize_len(ray_P - disk_P, &disk_t);
 	float div = dot(ray_D, disk_N);
 
-	if(div == 0.0f)
+	if(UNLIKELY(div == 0.0f))
 		return false;
 
 	/* compute t to intersection point */
@@ -1335,7 +1335,7 @@ ccl_device bool ray_triangle_intersect(
 	float3 s1 = cross(ray_D, e2);
 
 	const float divisor = dot(s1, e1);
-	if(divisor == 0.0f)
+	if(UNLIKELY(divisor == 0.0f))
 		return false;
 
 	const float invdivisor = 1.0f/divisor;
@@ -1378,7 +1378,7 @@ ccl_device bool ray_triangle_intersect_uv(
 	float3 s1 = cross(ray_D, e2);
 
 	const float divisor = dot(s1, e1);
-	if(divisor == 0.0f)
+	if(UNLIKELY(divisor == 0.0f))
 		return false;
 
 	const float invdivisor = 1.0f/divisor;

intern/cycles/util/util_transform.cpp

@@ -75,7 +75,7 @@ static bool transform_matrix4_gj_inverse(float R[][4], float M[][4])
 			}
 		}
 
-		if(pivotsize == 0)
+		if(UNLIKELY(pivotsize == 0.0f))
 			return false;
 
 		if(pivot != i) {
@@ -106,7 +106,7 @@ static bool transform_matrix4_gj_inverse(float R[][4], float M[][4])
 	for(int i = 3; i >= 0; --i) {
 		float f;
 
-		if((f = M[i][i]) == 0)
+		if(UNLIKELY((f = M[i][i]) == 0.0f))
 			return false;
 
 		for(int j = 0; j < 4; j++) {
@@ -135,15 +135,16 @@ Transform transform_inverse(const Transform& tfm)
 	memcpy(R, &tfmR, sizeof(R));
 	memcpy(M, &tfm, sizeof(M));
 
-	if(!transform_matrix4_gj_inverse(R, M)) {
+	if(UNLIKELY(!transform_matrix4_gj_inverse(R, M))) {
 		/* matrix is degenerate (e.g. 0 scale on some axis), ideally we should
 		 * never be in this situation, but try to invert it anyway with tweak */
 		M[0][0] += 1e-8f;
 		M[1][1] += 1e-8f;
 		M[2][2] += 1e-8f;
 
-		if(!transform_matrix4_gj_inverse(R, M))
+		if(UNLIKELY(!transform_matrix4_gj_inverse(R, M))) {
 			return transform_identity();
+		}
 	}
 
 	memcpy(&tfmR, R, sizeof(R));

intern/cycles/util/util_types.h

@@ -456,6 +456,43 @@ enum InterpolationType {
 	INTERPOLATION_SMART = 3,
 };
 
+
+/* macros */
+
+/* hints for branch pradiction, only use in code that runs a _lot_ */
+#ifdef __GNUC__
+#  define LIKELY(x)       __builtin_expect(!!(x), 1)
+#  define UNLIKELY(x)     __builtin_expect(!!(x), 0)
+#else
+#  define LIKELY(x)       (x)
+#  define UNLIKELY(x)     (x)
+#endif
+
+/* Causes warning:
+ * incompatible types when assigning to type 'Foo' from type 'Bar'
+ * ... the compiler optimizes away the temp var */
+#ifdef __GNUC__
+#define CHECK_TYPE(var, type)  {  \
+	__typeof(var) *__tmp;         \
+	__tmp = (type *)NULL;         \
+	(void)__tmp;                  \
+} (void)0
+
+#define CHECK_TYPE_PAIR(var_a, var_b)  {  \
+	__typeof(var_a) *__tmp;               \
+	__tmp = (__typeof(var_b) *)NULL;      \
+	(void)__tmp;                          \
+} (void)0
+#else
+#  define CHECK_TYPE(var, type)
+#  define CHECK_TYPE_PAIR(var_a, var_b)
+#endif
+
+/* can be used in simple macros */
+#define CHECK_TYPE_INLINE(val, type) \
+	((void)(((type)0) != (val)))
+
+
 CCL_NAMESPACE_END
 
 #endif /* __UTIL_TYPES_H__ */