Refactor: replace Cycles sse/avx types by vectorized float4/int4/float8/int8

The distinction existed for legacy reasons, to easily port of Embree
intersection code without affecting the main vector types. However we are now
using SIMD for these types as well, so no good reason to keep the distinction.

Also more consistently pass these vector types by value in inline functions.
Previously it was partially changed for functions used by Metal to avoid having
to add address space qualifiers, simple to do it everywhere.

Also removes function declarations for vector math headers, serves no real
purpose.

Differential Revision: https://developer.blender.org/D16146

1 files changed, 78 insertions, 140 deletions

diff --git a/intern/cycles/util/math_float3.h b/intern/cycles/util/math_float3.h
index eec7122b9dc..79ee86d9c82 100644
--- a/intern/cycles/util/math_float3.h
+++ b/intern/cycles/util/math_float3.h
@@ -1,4 +1,5 @@
 /* SPDX-License-Identifier: Apache-2.0
+ * Copyright 2011-2013 Intel Corporation
  * Copyright 2011-2022 Blender Foundation */
 
 #ifndef __UTIL_MATH_FLOAT3_H__
@@ -10,73 +11,6 @@
 
 CCL_NAMESPACE_BEGIN
 
-/*******************************************************************************
- * Declaration.
- */
-
-#if !defined(__KERNEL_METAL__)
-ccl_device_inline float3 operator-(const float3 &a);
-ccl_device_inline float3 operator*(const float3 &a, const float3 &b);
-ccl_device_inline float3 operator*(const float3 &a, const float f);
-ccl_device_inline float3 operator*(const float f, const float3 &a);
-ccl_device_inline float3 operator/(const float f, const float3 &a);
-ccl_device_inline float3 operator/(const float3 &a, const float f);
-ccl_device_inline float3 operator/(const float3 &a, const float3 &b);
-ccl_device_inline float3 operator+(const float3 &a, const float f);
-ccl_device_inline float3 operator+(const float3 &a, const float3 &b);
-ccl_device_inline float3 operator-(const float3 &a, const float f);
-ccl_device_inline float3 operator-(const float3 &a, const float3 &b);
-ccl_device_inline float3 operator+=(float3 &a, const float3 &b);
-ccl_device_inline float3 operator-=(float3 &a, const float3 &b);
-ccl_device_inline float3 operator*=(float3 &a, const float3 &b);
-ccl_device_inline float3 operator*=(float3 &a, float f);
-ccl_device_inline float3 operator/=(float3 &a, const float3 &b);
-ccl_device_inline float3 operator/=(float3 &a, float f);
-
-ccl_device_inline bool operator==(const float3 &a, const float3 &b);
-ccl_device_inline bool operator!=(const float3 &a, const float3 &b);
-
-ccl_device_inline float distance(const float3 &a, const float3 &b);
-ccl_device_inline float dot(const float3 &a, const float3 &b);
-ccl_device_inline float dot_xy(const float3 &a, const float3 &b);
-ccl_device_inline float3 cross(const float3 &a, const float3 &b);
-ccl_device_inline float3 normalize(const float3 &a);
-ccl_device_inline float3 min(const float3 &a, const float3 &b);
-ccl_device_inline float3 max(const float3 &a, const float3 &b);
-ccl_device_inline float3 clamp(const float3 &a, const float3 &mn, const float3 &mx);
-ccl_device_inline float3 fabs(const float3 &a);
-ccl_device_inline float3 mix(const float3 &a, const float3 &b, float t);
-ccl_device_inline float3 rcp(const float3 &a);
-ccl_device_inline float3 sqrt(const float3 &a);
-ccl_device_inline float3 floor(const float3 &a);
-ccl_device_inline float3 ceil(const float3 &a);
-ccl_device_inline float3 reflect(const float3 incident, const float3 normal);
-#endif /* !defined(__KERNEL_METAL__) */
-
-ccl_device_inline float reduce_min(float3 a);
-ccl_device_inline float reduce_max(float3 a);
-ccl_device_inline float len(const float3 a);
-ccl_device_inline float len_squared(const float3 a);
-
-ccl_device_inline float3 project(const float3 v, const float3 v_proj);
-
-ccl_device_inline float3 safe_normalize(const float3 a);
-ccl_device_inline float3 normalize_len(const float3 a, ccl_private float *t);
-ccl_device_inline float3 safe_normalize_len(const float3 a, ccl_private float *t);
-ccl_device_inline float3 safe_divide(const float3 a, const float3 b);
-ccl_device_inline float3 safe_divide(const float3 a, const float b);
-ccl_device_inline float3 interp(float3 a, float3 b, float t);
-ccl_device_inline float3 sqr(float3 a);
-
-ccl_device_inline bool is_zero(const float3 a);
-ccl_device_inline float reduce_add(const float3 a);
-ccl_device_inline float average(const float3 a);
-ccl_device_inline bool isequal(const float3 a, const float3 b);
-
-/*******************************************************************************
- * Definition.
- */
-
 ccl_device_inline float3 zero_float3()
 {
 #ifdef __KERNEL_SSE__
@@ -109,7 +43,7 @@ ccl_device_inline float3 operator-(const float3 &a)
 #  endif
 }
 
-ccl_device_inline float3 operator*(const float3 &a, const float3 &b)
+ccl_device_inline float3 operator*(const float3 a, const float3 b)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_mul_ps(a.m128, b.m128));
@@ -118,7 +52,7 @@ ccl_device_inline float3 operator*(const float3 &a, const float3 &b)
 #  endif
 }
 
-ccl_device_inline float3 operator*(const float3 &a, const float f)
+ccl_device_inline float3 operator*(const float3 a, const float f)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_mul_ps(a.m128, _mm_set1_ps(f)));
@@ -127,7 +61,7 @@ ccl_device_inline float3 operator*(const float3 &a, const float f)
 #  endif
 }
 
-ccl_device_inline float3 operator*(const float f, const float3 &a)
+ccl_device_inline float3 operator*(const float f, const float3 a)
 {
 #  if defined(__KERNEL_SSE__)
   return float3(_mm_mul_ps(_mm_set1_ps(f), a.m128));
@@ -136,7 +70,7 @@ ccl_device_inline float3 operator*(const float f, const float3 &a)
 #  endif
 }
 
-ccl_device_inline float3 operator/(const float f, const float3 &a)
+ccl_device_inline float3 operator/(const float f, const float3 a)
 {
 #  if defined(__KERNEL_SSE__)
   return float3(_mm_div_ps(_mm_set1_ps(f), a.m128));
@@ -145,7 +79,7 @@ ccl_device_inline float3 operator/(const float f, const float3 &a)
 #  endif
 }
 
-ccl_device_inline float3 operator/(const float3 &a, const float f)
+ccl_device_inline float3 operator/(const float3 a, const float f)
 {
 #  if defined(__KERNEL_SSE__)
   return float3(_mm_div_ps(a.m128, _mm_set1_ps(f)));
@@ -154,7 +88,7 @@ ccl_device_inline float3 operator/(const float3 &a, const float f)
 #  endif
 }
 
-ccl_device_inline float3 operator/(const float3 &a, const float3 &b)
+ccl_device_inline float3 operator/(const float3 a, const float3 b)
 {
 #  if defined(__KERNEL_SSE__)
   return float3(_mm_div_ps(a.m128, b.m128));
@@ -163,12 +97,7 @@ ccl_device_inline float3 operator/(const float3 &a, const float3 &b)
 #  endif
 }
 
-ccl_device_inline float3 operator+(const float3 &a, const float f)
-{
-  return a + make_float3(f, f, f);
-}
-
-ccl_device_inline float3 operator+(const float3 &a, const float3 &b)
+ccl_device_inline float3 operator+(const float3 a, const float3 b)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_add_ps(a.m128, b.m128));
@@ -177,12 +106,12 @@ ccl_device_inline float3 operator+(const float3 &a, const float3 &b)
 #  endif
 }
 
-ccl_device_inline float3 operator-(const float3 &a, const float f)
+ccl_device_inline float3 operator+(const float3 a, const float f)
 {
-  return a - make_float3(f, f, f);
+  return a + make_float3(f, f, f);
 }
 
-ccl_device_inline float3 operator-(const float3 &a, const float3 &b)
+ccl_device_inline float3 operator-(const float3 a, const float3 b)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_sub_ps(a.m128, b.m128));
@@ -191,17 +120,22 @@ ccl_device_inline float3 operator-(const float3 &a, const float3 &b)
 #  endif
 }
 
-ccl_device_inline float3 operator+=(float3 &a, const float3 &b)
+ccl_device_inline float3 operator-(const float3 a, const float f)
+{
+  return a - make_float3(f, f, f);
+}
+
+ccl_device_inline float3 operator+=(float3 &a, const float3 b)
 {
   return a = a + b;
 }
 
-ccl_device_inline float3 operator-=(float3 &a, const float3 &b)
+ccl_device_inline float3 operator-=(float3 &a, const float3 b)
 {
   return a = a - b;
 }
 
-ccl_device_inline float3 operator*=(float3 &a, const float3 &b)
+ccl_device_inline float3 operator*=(float3 &a, const float3 b)
 {
   return a = a * b;
 }
@@ -211,7 +145,7 @@ ccl_device_inline float3 operator*=(float3 &a, float f)
   return a = a * f;
 }
 
-ccl_device_inline float3 operator/=(float3 &a, const float3 &b)
+ccl_device_inline float3 operator/=(float3 &a, const float3 b)
 {
   return a = a / b;
 }
@@ -223,7 +157,7 @@ ccl_device_inline float3 operator/=(float3 &a, float f)
 }
 
 #  if !(defined(__KERNEL_METAL__) || defined(__KERNEL_CUDA__))
-ccl_device_inline packed_float3 operator*=(packed_float3 &a, const float3 &b)
+ccl_device_inline packed_float3 operator*=(packed_float3 &a, const float3 b)
 {
   a = float3(a) * b;
   return a;
@@ -235,7 +169,7 @@ ccl_device_inline packed_float3 operator*=(packed_float3 &a, float f)
   return a;
 }
 
-ccl_device_inline packed_float3 operator/=(packed_float3 &a, const float3 &b)
+ccl_device_inline packed_float3 operator/=(packed_float3 &a, const float3 b)
 {
   a = float3(a) / b;
   return a;
@@ -248,7 +182,7 @@ ccl_device_inline packed_float3 operator/=(packed_float3 &a, float f)
 }
 #  endif
 
-ccl_device_inline bool operator==(const float3 &a, const float3 &b)
+ccl_device_inline bool operator==(const float3 a, const float3 b)
 {
 #  ifdef __KERNEL_SSE__
   return (_mm_movemask_ps(_mm_cmpeq_ps(a.m128, b.m128)) & 7) == 7;
@@ -257,17 +191,12 @@ ccl_device_inline bool operator==(const float3 &a, const float3 &b)
 #  endif
 }
 
-ccl_device_inline bool operator!=(const float3 &a, const float3 &b)
+ccl_device_inline bool operator!=(const float3 a, const float3 b)
 {
   return !(a == b);
 }
 
-ccl_device_inline float distance(const float3 &a, const float3 &b)
-{
-  return len(a - b);
-}
-
-ccl_device_inline float dot(const float3 &a, const float3 &b)
+ccl_device_inline float dot(const float3 a, const float3 b)
 {
 #  if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__)
   return _mm_cvtss_f32(_mm_dp_ps(a, b, 0x7F));
@@ -276,26 +205,62 @@ ccl_device_inline float dot(const float3 &a, const float3 &b)
 #  endif
 }
 
-ccl_device_inline float dot_xy(const float3 &a, const float3 &b)
+#endif
+
+ccl_device_inline float dot_xy(const float3 a, const float3 b)
 {
-#  if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__)
+#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__)
   return _mm_cvtss_f32(_mm_hadd_ps(_mm_mul_ps(a, b), b));
-#  else
+#else
   return a.x * b.x + a.y * b.y;
-#  endif
+#endif
+}
+
+ccl_device_inline float len(const float3 a)
+{
+#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__)
+  return _mm_cvtss_f32(_mm_sqrt_ss(_mm_dp_ps(a.m128, a.m128, 0x7F)));
+#else
+  return sqrtf(dot(a, a));
+#endif
+}
+
+ccl_device_inline float reduce_min(float3 a)
+{
+  return min(min(a.x, a.y), a.z);
+}
+
+ccl_device_inline float reduce_max(float3 a)
+{
+  return max(max(a.x, a.y), a.z);
+}
+
+ccl_device_inline float len_squared(const float3 a)
+{
+  return dot(a, a);
+}
+
+#ifndef __KERNEL_METAL__
+
+ccl_device_inline float distance(const float3 a, const float3 b)
+{
+  return len(a - b);
 }
 
-ccl_device_inline float3 cross(const float3 &a, const float3 &b)
+ccl_device_inline float3 cross(const float3 a, const float3 b)
 {
 #  ifdef __KERNEL_SSE__
-  return float3(shuffle<1, 2, 0, 3>(
-      msub(ssef(a), shuffle<1, 2, 0, 3>(ssef(b)), shuffle<1, 2, 0, 3>(ssef(a)) * ssef(b))));
+  const float4 x = float4(a.m128);
+  const float4 y = shuffle<1, 2, 0, 3>(float4(b.m128));
+  const float4 z = float4(_mm_mul_ps(shuffle<1, 2, 0, 3>(float4(a.m128)), float4(b.m128)));
+
+  return float3(shuffle<1, 2, 0, 3>(msub(x, y, z)).m128);
 #  else
   return make_float3(a.y * b.z - a.z * b.y, a.z * b.x - a.x * b.z, a.x * b.y - a.y * b.x);
 #  endif
 }
 
-ccl_device_inline float3 normalize(const float3 &a)
+ccl_device_inline float3 normalize(const float3 a)
 {
 #  if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__)
   __m128 norm = _mm_sqrt_ps(_mm_dp_ps(a.m128, a.m128, 0x7F));
@@ -305,7 +270,7 @@ ccl_device_inline float3 normalize(const float3 &a)
 #  endif
 }
 
-ccl_device_inline float3 min(const float3 &a, const float3 &b)
+ccl_device_inline float3 min(const float3 a, const float3 b)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_min_ps(a.m128, b.m128));
@@ -314,7 +279,7 @@ ccl_device_inline float3 min(const float3 &a, const float3 &b)
 #  endif
 }
 
-ccl_device_inline float3 max(const float3 &a, const float3 &b)
+ccl_device_inline float3 max(const float3 a, const float3 b)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_max_ps(a.m128, b.m128));
@@ -323,12 +288,12 @@ ccl_device_inline float3 max(const float3 &a, const float3 &b)
 #  endif
 }
 
-ccl_device_inline float3 clamp(const float3 &a, const float3 &mn, const float3 &mx)
+ccl_device_inline float3 clamp(const float3 a, const float3 mn, const float3 mx)
 {
   return min(max(a, mn), mx);
 }
 
-ccl_device_inline float3 fabs(const float3 &a)
+ccl_device_inline float3 fabs(const float3 a)
 {
 #  ifdef __KERNEL_SSE__
 #    ifdef __KERNEL_NEON__
@@ -342,7 +307,7 @@ ccl_device_inline float3 fabs(const float3 &a)
 #  endif
 }
 
-ccl_device_inline float3 sqrt(const float3 &a)
+ccl_device_inline float3 sqrt(const float3 a)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_sqrt_ps(a));
@@ -351,7 +316,7 @@ ccl_device_inline float3 sqrt(const float3 &a)
 #  endif
 }
 
-ccl_device_inline float3 floor(const float3 &a)
+ccl_device_inline float3 floor(const float3 a)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_floor_ps(a));
@@ -360,7 +325,7 @@ ccl_device_inline float3 floor(const float3 &a)
 #  endif
 }
 
-ccl_device_inline float3 ceil(const float3 &a)
+ccl_device_inline float3 ceil(const float3 a)
 {
 #  ifdef __KERNEL_SSE__
   return float3(_mm_ceil_ps(a));
@@ -369,12 +334,12 @@ ccl_device_inline float3 ceil(const float3 &a)
 #  endif
 }
 
-ccl_device_inline float3 mix(const float3 &a, const float3 &b, float t)
+ccl_device_inline float3 mix(const float3 a, const float3 b, float t)
 {
   return a + t * (b - a);
 }
 
-ccl_device_inline float3 rcp(const float3 &a)
+ccl_device_inline float3 rcp(const float3 a)
 {
 #  ifdef __KERNEL_SSE__
   /* Don't use _mm_rcp_ps due to poor precision. */
@@ -399,33 +364,6 @@ ccl_device_inline float3 log(float3 v)
   return make_float3(logf(v.x), logf(v.y), logf(v.z));
 }
 
-#endif /* !__KERNEL_METAL__ */
-
-ccl_device_inline float reduce_min(float3 a)
-{
-  return min(min(a.x, a.y), a.z);
-}
-
-ccl_device_inline float reduce_max(float3 a)
-{
-  return max(max(a.x, a.y), a.z);
-}
-
-ccl_device_inline float len(const float3 a)
-{
-#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__)
-  return _mm_cvtss_f32(_mm_sqrt_ss(_mm_dp_ps(a.m128, a.m128, 0x7F)));
-#else
-  return sqrtf(dot(a, a));
-#endif
-}
-
-ccl_device_inline float len_squared(const float3 a)
-{
-  return dot(a, a);
-}
-
-#if !defined(__KERNEL_METAL__)
 ccl_device_inline float3 reflect(const float3 incident, const float3 normal)
 {
   float3 unit_normal = normalize(normal);