Fix T98672: Noise texture shows incorrect behaviour for large scales

This was a floating point precision issue - or, to be more precise,
an issue with how Cycles split floats into the integer and fractional
parts for Perlin noise.
For coordinates below -2^24, the integer could be wrong, leading to
the fractional part being outside of 0-1 range, which breaks all sorts
of other things. 2^24 sounds like a lot, but due to how the detail
octaves work, it's not that hard to reach when combined with a large
scale.

Since this code is originally based on OSL, I checked if they changed
it in the meantime, and sure enough, there's a fix for it:
https://github.com/OpenImageIO/oiio/commit/5c9dc68391e9

So, this basically just ports over that change to Cycles.

The original code mentions being faster, but as pointed out in the
linked commit, the performance impact is actually irrelevant.

I also checked in a simple scene with eight Noise textures at
detail 15 (with >90% of render time being spent on the noise), and
the render time went from 13.06sec to 13.05sec. So, yeah, no issue.

3 files changed, 14 insertions, 31 deletions

diff --git a/intern/cycles/util/math.h b/intern/cycles/util/math.h
index 0905b3ec5c9..3a2e0e074a2 100644
--- a/intern/cycles/util/math.h
+++ b/intern/cycles/util/math.h
@@ -417,15 +417,11 @@ ccl_device_inline int floor_to_int(float f)
   return float_to_int(floorf(f));
 }
 
-ccl_device_inline int quick_floor_to_int(float x)
-{
-  return float_to_int(x) - ((x < 0) ? 1 : 0);
-}
-
 ccl_device_inline float floorfrac(float x, ccl_private int *i)
 {
-  *i = quick_floor_to_int(x);
-  return x - *i;
+  float f = floorf(x);
+  *i = float_to_int(f);
+  return x - f;
 }
 
 ccl_device_inline int ceil_to_int(float f)
diff --git a/intern/cycles/util/math_float3.h b/intern/cycles/util/math_float3.h
index c408eadf195..eec7122b9dc 100644
--- a/intern/cycles/util/math_float3.h
+++ b/intern/cycles/util/math_float3.h
@@ -535,18 +535,6 @@ ccl_device_inline float3 pow(float3 v, float e)
   return make_float3(powf(v.x, e), powf(v.y, e), powf(v.z, e));
 }
 
-ccl_device_inline int3 quick_floor_to_int3(const float3 a)
-{
-#ifdef __KERNEL_SSE__
-  int3 b = int3(_mm_cvttps_epi32(a.m128));
-  int3 isneg = int3(_mm_castps_si128(_mm_cmplt_ps(a.m128, _mm_set_ps1(0.0f))));
-  /* Unsaturated add 0xffffffff is the same as subtract -1. */
-  return b + isneg;
-#else
-  return make_int3(quick_floor_to_int(a.x), quick_floor_to_int(a.y), quick_floor_to_int(a.z));
-#endif
-}
-
 ccl_device_inline bool isfinite_safe(float3 v)
 {
   return isfinite_safe(v.x) && isfinite_safe(v.y) && isfinite_safe(v.z);
diff --git a/intern/cycles/util/ssef.h b/intern/cycles/util/ssef.h
index a2fff94303e..7f9e332a997 100644
--- a/intern/cycles/util/ssef.h
+++ b/intern/cycles/util/ssef.h
@@ -5,6 +5,8 @@
 #ifndef __UTIL_SSEF_H__
 #define __UTIL_SSEF_H__
 
+#include <math.h>
+
 #include "util/ssei.h"
 
 CCL_NAMESPACE_BEGIN
@@ -534,6 +536,12 @@ __forceinline const ssef ceil(const ssef &a)
   return _mm_round_ps(a, _MM_FROUND_TO_POS_INF);
 #    endif
 }
+#  else
+/* Non-SSE4.1 fallback, needed for floorfrac. */
+__forceinline const ssef floor(const ssef &a)
+{
+  return _mm_set_ps(floorf(a.f[3]), floorf(a.f[2]), floorf(a.f[1]), floorf(a.f[0]));
+}
 #  endif
 
 __forceinline ssei truncatei(const ssef &a)
@@ -541,20 +549,11 @@ __forceinline ssei truncatei(const ssef &a)
   return _mm_cvttps_epi32(a.m128);
 }
 
-/* This is about 25% faster than straightforward floor to integer conversion
- * due to better pipelining.
- *
- * Unsaturated add 0xffffffff (a < 0) is the same as subtract -1.
- */
-__forceinline ssei floori(const ssef &a)
-{
-  return truncatei(a) + cast((a < 0.0f).m128);
-}
-
 __forceinline ssef floorfrac(const ssef &x, ssei *i)
 {
-  *i = floori(x);
-  return x - ssef(*i);
+  ssef f = floor(x);
+  *i = truncatei(f);
+  return x - f;
 }
 
 ////////////////////////////////////////////////////////////////////////////////