diff --git a/intern/cycles/kernel/kernel_jitter.h b/intern/cycles/kernel/kernel_jitter.h
index f7270a14940..e59d8946950 100644
--- a/intern/cycles/kernel/kernel_jitter.h
+++ b/intern/cycles/kernel/kernel_jitter.h
@@ -38,43 +38,13 @@ ccl_device_inline int cmj_fast_mod_pow2(int a, int b)
 ccl_device_inline int cmj_fast_div_pow2(int a, int b)
 {
   kernel_assert(b > 1);
-#if defined(__KERNEL_SSE2__)
-#  ifdef _MSC_VER
-  unsigned long ctz;
-  _BitScanForward(&ctz, b);
-  return a >> ctz;
-#  else
-  return a >> __builtin_ctz(b);
-#  endif
-#elif defined(__KERNEL_CUDA__)
-  return a >> (__ffs(b) - 1);
-#else
-  return a / b;
-#endif
+  return a >> count_trailing_zeros(b);
 }
 
 ccl_device_inline uint cmj_w_mask(uint w)
 {
   kernel_assert(w > 1);
-#if defined(__KERNEL_SSE2__)
-#  ifdef _MSC_VER
-  unsigned long leading_zero;
-  _BitScanReverse(&leading_zero, w);
-  return ((1 << (1 + leading_zero)) - 1);
-#  else
-  return ((1 << (32 - __builtin_clz(w))) - 1);
-#  endif
-#elif defined(__KERNEL_CUDA__)
-  return ((1 << (32 - __clz(w))) - 1);
-#else
-  w |= w >> 1;
-  w |= w >> 2;
-  w |= w >> 4;
-  w |= w >> 8;
-  w |= w >> 16;
-
-  return w;
-#endif
+  return ((1 << (32 - count_leading_zeros(w))) - 1);
 }
 
 ccl_device_inline uint cmj_permute(uint i, uint l, uint p)
diff --git a/intern/cycles/util/util_math.h b/intern/cycles/util/util_math.h
index dde0d31f467..9faf7149ce2 100644
--- a/intern/cycles/util/util_math.h
+++ b/intern/cycles/util/util_math.h
@@ -617,6 +617,57 @@ ccl_device float bits_to_01(uint bits)
   return bits * (1.0f / (float)0xFFFFFFFF);
 }
 
+ccl_device_inline uint count_leading_zeros(uint x)
+{
+  assert(x != 0);
+#if defined(__KERNEL_CUDA__) || defined(__KERNEL_OPTIX__)
+  return __clz(x);
+#elif defined(__KERNEL_OPENCL__)
+  return clz(x);
+#else
+#  ifdef _MSC_VER
+  unsigned long leading_zero = 0;
+  _BitScanReverse(&leading_zero, x);
+  return (31 - leading_zero);
+#  else
+  return __builtin_clz(x);
+#  endif
+#endif
+}
+
+ccl_device_inline uint count_trailing_zeros(uint x)
+{
+  assert(x != 0);
+#if defined(__KERNEL_CUDA__) || defined(__KERNEL_OPTIX__)
+  return (__ffs(x) - 1);
+#elif defined(__KERNEL_OPENCL__)
+  return (31 - count_leading_zeros(x & -x));
+#else
+#  ifdef _MSC_VER
+  unsigned long ctz = 0;
+  _BitScanForward(&ctz, x);
+  return ctz;
+#  else
+  return __builtin_ctz(x);
+#  endif
+#endif
+}
+
+ccl_device_inline uint find_first_set(uint x)
+{
+#if defined(__KERNEL_CUDA__) || defined(__KERNEL_OPTIX__)
+  return __ffs(x);
+#elif defined(__KERNEL_OPENCL__)
+  return (x != 0) ? (32 - count_leading_zeros(x & (-x))) : 0;
+#else
+#  ifdef _MSC_VER
+  return (x != 0) ? (32 - count_leading_zeros(x & (-x))) : 0;
+#  else
+  return __builtin_ffs(x);
+#  endif
+#endif
+}
+
 /* projections */
 ccl_device_inline float2 map_to_tube(const float3 co)
 {