<html>
<head>
<base href="https://llvm.org/bugs/" />
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW --- - clang generates 1.5 slower loop code than gcc"
href="https://llvm.org/bugs/show_bug.cgi?id=23269">23269</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>clang generates 1.5 slower loop code than gcc
</td>
</tr>
<tr>
<th>Product</th>
<td>clang
</td>
</tr>
<tr>
<th>Version</th>
<td>trunk
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Linux
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>normal
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>LLVM Codegen
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedclangbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>dvyukov@google.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvmbugs@cs.uiuc.edu
</td>
</tr>
<tr>
<th>Classification</th>
<td>Unclassified
</td>
</tr></table>
<p>
<div>
<pre>$ clang++ -v
clang version 3.7.0 (trunk 234143)
Target: x86_64-unknown-linux-gnu
$ g++ -v
Target: x86_64-linux-gnu
gcc version 4.8.2 (Ubuntu 4.8.2-19ubuntu1)
Below is the test program.
Processor is Intel(R) Xeon(R) CPU E5-2690 0 @ 2.90GHz.
Build the program with:
$ g++/clang++ test.cc -Wall -O3 -msse3 -g
g++ compiled binary runs 2.813s.
g++ compiled binary runs 4.353s.
===========
#include <stdlib.h>
typedef unsigned char byte;
byte* volatile arr1;
byte* volatile arr2;
__attribute__((noinline)) void compare(byte* p1, byte* p2, bool* f1, bool *f2)
{
bool cnt = false;
for (int i = 0; i < 1<<16; i++) {
byte v1 = p1[i];
byte v2 = p2[i];
if (__builtin_expect(v1 == 0 && v2 != 0, 0)) {
*f1 = true;
*f2 = true;
return;
}
if (__builtin_expect(v1 < v2, 0)) {
cnt = true;
}
}
*f1 = false;
*f2 = cnt;
}
int main() {
arr1 = (byte*)calloc(1<<16, 1);
arr2 = (byte*)calloc(1<<16, 1);
for (int i = 0; i < 1000; i++) {
int idx = rand() % (1<<16);
arr1[idx] = 100;
arr2[idx] = 100;
}
int x = 0;
for (int i = 0; i < 50000; i++) {
bool f1, f2;
compare(arr1, arr2, &f1, &f2);
x += f1;
x += f2;
}
return x;
}
=====
g++-compiled binary profile:
│ 0000000000400630 <compare(unsigned char*, unsigned char*, bool*,
bool*)>:
│ xor %eax,%eax
│ xor %r9d,%r9d
│ mov $0x1,%r11d
│ nop
0.02 │10: movzbl (%rsi,%rax,1),%r8d
8.05 │ movzbl (%rdi,%rax,1),%r10d
29.73 │ test %r8b,%r8b
│ jne 39
6.48 │1f: cmp %r8b,%r10b
9.31 │ cmovb %r11d,%r9d
43.24 │ add $0x1,%rax
0.03 │ cmp $0x10000,%rax
│ jne 10
│ movb $0x0,(%rdx)
│ mov %r9b,(%rcx)
│ retq
3.14 │39: test %r10b,%r10b
│ jne 1f
│ movb $0x1,(%rdx)
│ movb $0x1,(%rcx)
│ retq
clang++-compiled profile:
│ 0000000000400640 <compare(unsigned char*, unsigned char*, bool*,
bool*)>: ▒
│ xor %r11d,%r11d
▒
│ xor %r8d,%r8d
▒
│ nop
▒
4.14 │10: mov (%rdi,%r11,1),%r9b
▒
12.93 │ mov (%rsi,%r11,1),%r10b
▒
6.93 │ test %r9b,%r9b
▒
│ jne 22
▒
4.11 │ test %r10b,%r10b
▒
│ jne 76
▒
5.82 │22: movzbl %r10b,%r10d
▒
1.43 │ movzbl %r9b,%eax
▒
4.40 │ cmp %r10d,%eax
▒
8.51 │ mov $0x1,%r9b
▒
1.44 │ ┌──jb 35
◆
7.17 │ │ mov %r8b,%r9b
▒
1.63 │35:└─ mov 0x1(%rdi,%r11,1),%r8b
▒
2.84 │ mov 0x1(%rsi,%r11,1),%r10b
▒
8.69 │ test %r8b,%r8b
▒
│ jne 49
▒
1.80 │ test %r10b,%r10b
▒
│ jne 76
▒
3.94 │49: inc %r11
▒
5.29 │ movzbl %r10b,%r10d
▒
2.14 │ movzbl %r8b,%eax
▒
1.18 │ cmp %r10d,%eax
▒
4.64 │ mov $0x1,%r8b
▒
3.93 │ jb 5f
▒
3.83 │ mov %r9b,%r8b
▒
1.58 │5f: inc %r11
▒
1.62 │ cmp $0x10000,%r11
▒
│ jl 10
▒
│ movb $0x0,(%rdx)
▒
│ and $0x1,%r8b
▒
│ mov %r8b,(%rcx)
▒
│ retq
▒
│76: movb $0x1,(%rdx)
▒
│ mov $0x1,%r8b
▒
│ mov %r8b,(%rcx)
▒
│ retq</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>