it seems the code which notices things like assert(x > 0) to help
strength-reduce division to shifts doesn't help if the divide is inside a loop.
for example:
% cat idiv-assert.c
void fail(void) __attribute__((noreturn));
int bar(int);
int foo(int x) {
int i;
int s = 0;
if (x <= 0) fail();
for (i = 0; i < x; ++i) {
s += bar(i/4);
}
return s;
}
% gcc -g -O3 -Wall -c -o idiv-assert.o idiv-assert.c
% objdump -dr idiv-assert.o
idiv-assert.o: file format elf64-x86-64
Disassembly of section .text:
0000000000000000 <foo>:
0: 41 54 push %r12
2: 41 89 fc mov %edi,%r12d
5: 55 push %rbp
6: 31 ed xor %ebp,%ebp
8: 53 push %rbx
9: 31 db xor %ebx,%ebx
b: 85 ff test %edi,%edi
d: 7e 22 jle 31 <foo+0x31>
f: 90 nop
10: 8d 7b 03 lea 0x3(%rbx),%edi
13: 85 db test %ebx,%ebx
15: 0f 49 fb cmovns %ebx,%edi
18: 83 c3 01 add $0x1,%ebx
1b: c1 ff 02 sar $0x2,%edi
1e: e8 00 00 00 00 callq 23 <foo+0x23>
1f: R_X86_64_PC32 bar+0xfffffffffffffffc
23: 01 c5 add %eax,%ebp
25: 41 39 dc cmp %ebx,%r12d
28: 7f e6 jg 10 <foo+0x10>
2a: 5b pop %rbx
2b: 89 e8 mov %ebp,%eax
2d: 5d pop %rbp
2e: 41 5c pop %r12
30: c3 retq
31: e8 00 00 00 00 callq 36 <foo+0x36>
32: R_X86_64_PC32 fail+0xfffffffffffffffc
% gcc -v
Using built-in specs.
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc/configure --prefix=/home/odo/gcc --disable-multilib
--disable-biarch x86_64-unknown-linux-gnu --enable-languages=c
Thread model: posix
gcc version 4.3.0 20070217 (experimental)
note that gcc handles a more trivial case just fine:
if (x <= 0) fail();
return x / 4;
-dean
--
Summary: strength-reduce idiv given assertion numerator is
positive
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: pending
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dean at arctic dot org
GCC build triplet: x86_64-unknown-linux-gnu
GCC host triplet: x86_64-unknown-linux-gnu
GCC target triplet: x86_64-unknown-linux-gnu
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31521
