[PATCH] D22440: Use a faster implementation of maxUIntN.

[Justin Lebar via llvm-commits]

jlebar created this revision.
jlebar added a reviewer: rnk.
jlebar added subscribers: llvm-commits, mkuper, dylanmckay.

On x86-64 with clang 3.8, before:

   mov     edx, 1
   mov     cl, dil
   shl     rdx, cl
   cmp     rdi, 64
   mov     rax, -1
   cmovne  rax, rdx
   ret

after:

  mov     ecx, 64
  sub     ecx, edi
  mov     rax, -1
  shr     rax, cl
  ret

https://reviews.llvm.org/D22440

Files:
  include/llvm/Support/MathExtras.h

Index: include/llvm/Support/MathExtras.h
===================================================================
--- include/llvm/Support/MathExtras.h
+++ include/llvm/Support/MathExtras.h
@@ -316,10 +316,11 @@
 inline uint64_t maxUIntN(uint64_t N) {
   assert(N > 0 && N <= 64 && "integer width out of range");
 
-  // uint64_t(1) << 64 is undefined behavior.
-  if (N == 64)
-    return std::numeric_limits<uint64_t>::max();
-  return (UINT64_C(1) << N) - 1;
+  // uint64_t(1) << 64 is undefined behavior, so we can't do
+  //   (uint64_t(1) << N) - 1
+  // without checking first that N != 64.  But this works and doesn't have a
+  // branch.
+  return ~UINT64_C(0) >> (64 - N);
 }
 
 /// Gets the minimum value for a N-bit signed integer.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D22440.64232.patch
Type: text/x-patch
Size: 739 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160716/b115e0e4/attachment.bin>