Index: SingleSource/Benchmarks/Misc/CMakeLists.txt
===================================================================
--- SingleSource/Benchmarks/Misc/CMakeLists.txt
+++ SingleSource/Benchmarks/Misc/CMakeLists.txt
@@ -23,6 +23,7 @@
   oourafft.c
   perlin.c
   pi.c
+  revertBits.c
   richards_benchmark.c
   salsa20.c
   whetstone.c
Index: SingleSource/Benchmarks/Misc/LICENSE.TXT
===================================================================
--- SingleSource/Benchmarks/Misc/LICENSE.TXT
+++ SingleSource/Benchmarks/Misc/LICENSE.TXT
@@ -15,6 +15,10 @@
 This program is licensed under the LLVM license.  It was written by Don Shull
 and Mark Riordan.
 
+revertBits
+------------------------------------------------------------------------------
+This program is licensed under the LLVM license.  It was written by Tony Jiang.
+
 evalloop
 ------------------------------------------------------------------------------
 This is licensed under the Python Software Foundation License Version 2.
Index: SingleSource/Benchmarks/Misc/revertBits.c
===================================================================
--- /dev/null
+++ SingleSource/Benchmarks/Misc/revertBits.c
@@ -0,0 +1,48 @@
+/*
+ * A performance regression test for bit reverse. The bit reverse algorithm on
+ * PPC was O(N) and had lots of dependency between contiguous instructions. We
+ * replace it with a faster O(lgN) algorithm in the following two patches:
+ * https://reviews.llvm.org/D33572
+ * https://reviews.llvm.org/D34908
+ * This test case tests both the 32-bit and 64-bit bit reverse to ensure the
+ * new implementation is functionally correct and has a better performance.
+ */
+#include <stdio.h>
+#define NUM 0x1000000
+int __attribute__((noinline)) ReverseBits32(int n) {
+  n = ((n >> 1) & 0x55555555) | ((n & 0x55555555) << 1);
+  n = ((n >> 2) & 0x33333333) | ((n & 0x33333333) << 2);
+  n = ((n >> 4) & 0x0F0F0F0F) | ((n & 0x0F0F0F0F) << 4);
+  return ((n & 0xff000000u) >> 24) | ((n & 0x00ff0000u) >> 8) |
+         ((n & 0x0000ff00u) << 8) | ((n & 0x000000ffu) << 24);
+}
+
+long long __attribute__((noinline)) ReverseBits64(long long n) {
+  n = ((n >> 1) & 0x5555555555555555L) | ((n & 0x5555555555555555L) << 1);
+  n = ((n >> 2) & 0x3333333333333333L) | ((n & 0x3333333333333333L) << 2);
+  n = ((n >> 4) & 0x0F0F0F0F0F0F0F0FL) | ((n & 0x0F0F0F0F0F0F0F0FL) << 4);
+  return ((n & 0xFF00000000000000L) >> 56) |
+         ((n & 0x00FF000000000000L) >> 40) |
+         ((n & 0x0000FF0000000000L) >> 24) |
+         ((n & 0x000000FF00000000L) >> 8)  |
+         ((n & 0x00000000000000FFL) << 56) |
+         ((n & 0x000000000000FF00L) << 40) |
+         ((n & 0x0000000000FF0000L) << 24) |
+         ((n & 0x00000000FF000000L) << 8);
+}
+
+int main (void) {
+  long long sum = 0;
+  for (int i = 0; i < NUM; ++i) {
+    sum += ReverseBits32(__builtin_bitreverse32(i));
+    sum -= ReverseBits64(__builtin_bitreverse64(i));
+  }
+  printf("Sum1 = %llx, ", sum);
+
+  for (int i = 0; i < NUM; ++i) {
+    sum -= ReverseBits64(ReverseBits64(0x100000000 + i));
+  }
+  printf("Sum2 = %llx\n", sum);
+
+  return 0;
+}
Index: SingleSource/Benchmarks/Misc/revertBits.reference_output
===================================================================
--- /dev/null
+++ SingleSource/Benchmarks/Misc/revertBits.reference_output
@@ -0,0 +1,2 @@
+Sum1 = 0, Sum2 = feff800000800000
+exit 0