From 26b23231db94c716ae8609234fc666014740df0a Mon Sep 17 00:00:00 2001
From: Swudu Susuwu <2002luvabbaluvu@gmail.com>
Date: Fri, 16 May 2025 10:21:53 -0700
Subject: [PATCH 1/2] `include/float8.h:NextPowerOfTwo`; infinite values

+`ml_dtypes/include/float8.h:MostSignificantBit`; `constexpr` version of `clz` opcode.
*`ml_dtypes/include/float8.h:NextPowerOfTwo`; replace long list of hardcoded values with `2 << MostSignificantBit<Size>`, which extends `NextPowerOfTwo` (was limited to integer sizes, now is less code plus is general use).
---
 ml_dtypes/include/float8.h | 27 ++++++++++-----------------
 1 file changed, 10 insertions(+), 17 deletions(-)

diff --git a/ml_dtypes/include/float8.h b/ml_dtypes/include/float8.h
index 7696c059..3a97d3e0 100644
--- a/ml_dtypes/include/float8.h
+++ b/ml_dtypes/include/float8.h
@@ -1194,26 +1194,19 @@ template <typename T>
 bool constexpr IsPowerOfTwo(T x) {
   return (x != 0) && ((x & (x - 1)) == 0);
 }
+template <unsigned int N>
+constexpr unsigned int MostSignificantBit() {
+  unsigned int result = 0;
+  unsigned int x = N;
+  while (x >>= 1) {
+    ++result;
+  }
+  return result; // return N == 0 ? 0 : (sizeof(long long) * 8 - 1 - __builtin_clz(N));
+}
 // Helper for getting a bytes size which is a power of two.
 template <int Size>
 struct NextPowerOfTwo {
-  static constexpr int value = Size;
-};
-template <>
-struct NextPowerOfTwo<3> {
-  static constexpr int value = 4;
-};
-template <>
-struct NextPowerOfTwo<5> {
-  static constexpr int value = 8;
-};
-template <>
-struct NextPowerOfTwo<6> {
-  static constexpr int value = 8;
-};
-template <>
-struct NextPowerOfTwo<7> {
-  static constexpr int value = 8;
+  static constexpr int value = IsPowerOfTwo(Size) ? Size : 2 << MostSignificantBit<Size>();
 };
 
 // Helper for getting a bit representation provided a byte size.

From 6291f38463653ecd709454aba4fb197d35fc2351 Mon Sep 17 00:00:00 2001
From: Swudu Susuwu <2002luvabbaluvu@gmail.com>
Date: Fri, 16 May 2025 10:33:02 -0700
Subject: [PATCH 2/2] `NextPowerOfTwo`; `s/unsigned int/long long/`

`MostSignificantBit`; `s/unsigned int/long long/`
`NextPowerOfTwo`; `s/int/long long/`
Ensures that those functions allowed the most possible amount of values.
---
 ml_dtypes/include/float8.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/ml_dtypes/include/float8.h b/ml_dtypes/include/float8.h
index 3a97d3e0..db9acdb8 100644
--- a/ml_dtypes/include/float8.h
+++ b/ml_dtypes/include/float8.h
@@ -1194,17 +1194,17 @@ template <typename T>
 bool constexpr IsPowerOfTwo(T x) {
   return (x != 0) && ((x & (x - 1)) == 0);
 }
-template <unsigned int N>
+template <long long N>
 constexpr unsigned int MostSignificantBit() {
   unsigned int result = 0;
-  unsigned int x = N;
+  long long x = N;
   while (x >>= 1) {
     ++result;
   }
   return result; // return N == 0 ? 0 : (sizeof(long long) * 8 - 1 - __builtin_clz(N));
 }
 // Helper for getting a bytes size which is a power of two.
-template <int Size>
+template <long long Size>
 struct NextPowerOfTwo {
   static constexpr int value = IsPowerOfTwo(Size) ? Size : 2 << MostSignificantBit<Size>();
 };