[Bug c++/60747] New: the tree-vrp compilation flag produce wrong assembly
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60747
Bug ID: 60747
Summary: the tree-vrp compilation flag produce wrong assembly
Product: gcc
Version: 4.4.7
Status: UNCONFIRMED
Severity: major
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: techsgin at gmail dot com
Hi
At optimization level O2 the results of return value of short if are miss
calculated i.e wrong results. see example
Solution:
Don't use the tree-vrp compilation flag i.e add -fno-tree-vrp
Example code:
#include
#include
int hash_it(const char* d, const int prime) __attribute__((noinline));
int hash_it(const char* d, const int prime)
{
int n = strlen(d);
int h = n;
for (int i = 0; i < n; i++, d++)
h = (777*h) + (*d);
return ((h >= 0) ? (h % prime) : (-h % prime));
}
int main ()
{
int lret = 0;
const char* lsdata = "ABCD";
lret = hash_it(lsdata, 10);
printf("%d, %s\n", lret, lsdata);
return 0;
}
Results:
correct result (i.e NO tree vrp) :
$ g++ -O2 -fno-tree-vrp main.cpp -o main_2_no_tree_vrp
$ ./main_2_no_tree_vrp
2, ABCD
incorrect result (i.e tree vrp in use) :
$ g++ -O2 main.cpp -o main_o2_tree_vrp
$ ./main_o2_tree_vrp
-2, ABCD
Version details:
$ cat /etc/*-release
CentOS release 6.5 (Final)
LSB_VERSION=base-4.0-amd64:base-4.0-noarch:core-4.0-amd64:core-4.0-noarch:graphics-4.0-amd64:graphics-4.0-noarch:printing-4.0-amd64:printing-4.0-noarch
CentOS release 6.5 (Final)
CentOS release 6.5 (Final)
$ uname -a
Linux mastercos6 2.6.32-431.3.1.el6.x86_64 #1 SMP Fri Jan 3 21:39:27 UTC 2014
x86_64 x86_64 x86_64 GNU/Linux
$ gcc -v
Using built-in specs.
Target: x86_64-redhat-linux
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man
--infodir=/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla
--enable-bootstrap --enable-shared --enable-threads=posix
--enable-checking=release --with-system-zlib --enable-__cxa_atexit
--disable-libunwind-exceptions --enable-gnu-unique-object
--enable-languages=c,c++,objc,obj-c++,java,fortran,ada --enable-java-awt=gtk
--disable-dssi --with-java-home=/usr/lib/jvm/java-1.5.0-gcj-1.5.0.0/jre
--enable-libgcj-multifile --enable-java-maintainer-mode
--with-ecj-jar=/usr/share/java/eclipse-ecj.jar --disable-libjava-multilib
--with-ppl --with-cloog --with-tune=generic --with-arch_32=i686
--build=x86_64-redhat-linux
Thread model: posix
gcc version 4.4.7 20120313 (Red Hat 4.4.7-4) (GCC)
[Bug c++/60747] the tree-vrp compilation flag produce wrong assembly
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60747 --- Comment #2 from tech --- Created attachment 32533 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32533&action=edit Assembly of O2 no tree-vrp Assembly of O2 no tree-vrp .file"main.cpp" .text .p2align 4,,15 .globl _Z7hash_itPKci .type_Z7hash_itPKci, @function _Z7hash_itPKci: .LFB26: .cfi_startproc .cfi_personality 0x3,__gxx_personality_v0 pushq%rbp .cfi_def_cfa_offset 16 .cfi_offset 6, -16 movl%esi, %ebp pushq%rbx .cfi_def_cfa_offset 24 .cfi_offset 3, -24 movq%rdi, %rbx subq$8, %rsp .cfi_def_cfa_offset 32 callstrlen testl%eax, %eax movl%eax, %edx jle.L3 leaq1(%rbx), %rcx leal-1(%rax), %esi leaq(%rcx,%rsi), %rsi jmp.L4 .p2align 4,,10 .p2align 3 .L11: addq$1, %rcx .L4: movsbl(%rbx), %ebx imull$777, %edx, %edx addl%ebx, %edx cmpq%rsi, %rcx movq%rcx, %rbx jne.L11 .L3: testl%edx, %edx jns.L9 negl%edx .L9: movl%edx, %eax sarl$31, %edx addq$8, %rsp .cfi_def_cfa_offset 24 idivl%ebp popq%rbx .cfi_def_cfa_offset 16 popq%rbp .cfi_def_cfa_offset 8 movl%edx, %eax ret
[Bug c++/60747] the tree-vrp compilation flag produce wrong assembly
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60747 --- Comment #3 from tech --- Created attachment 32534 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32534&action=edit Assembly of O2 WITH tree-vrp Assembly of O2 WITH tree-vrp .file"main.cpp" .text .p2align 4,,15 .globl _Z7hash_itPKci .type_Z7hash_itPKci, @function _Z7hash_itPKci: .LFB26: .cfi_startproc .cfi_personality 0x3,__gxx_personality_v0 pushq%rbp .cfi_def_cfa_offset 16 .cfi_offset 6, -16 movl%esi, %ebp pushq%rbx .cfi_def_cfa_offset 24 .cfi_offset 3, -24 movq%rdi, %rbx subq$8, %rsp .cfi_def_cfa_offset 32 callstrlen cmpl$0, %eax movl%eax, %edx jle.L2 leaq1(%rbx), %rcx leal-1(%rax), %esi leaq(%rcx,%rsi), %rsi jmp.L3 .p2align 4,,10 .p2align 3 .L11: addq$1, %rcx .L3: movsbl(%rbx), %ebx imull$777, %edx, %edx addl%ebx, %edx cmpq%rsi, %rcx movq%rcx, %rbx jne.L11 .L9: movl%edx, %eax sarl$31, %edx addq$8, %rsp .cfi_remember_state .cfi_def_cfa_offset 24 idivl%ebp popq%rbx .cfi_def_cfa_offset 16 popq%rbp .cfi_def_cfa_offset 8 movl%edx, %eax ret .L2: .cfi_restore_state jge.L12 negl%edx addq$8, %rsp .cfi_remember_state .cfi_def_cfa_offset 24 movl%edx, %eax sarl$31, %edx idivl%ebp popq%rbx .cfi_def_cfa_offset 16 popq%rbp .cfi_def_cfa_offset 8 movl%edx, %eax ret .L12: .cfi_restore_state movl%eax, %edx jmp.L9 .cfi_endproc .LFE26: .size_Z7hash_itPKci, .-_Z7hash_itPKci
[Bug c++/60747] the tree-vrp compilation flag produce wrong assembly
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60747 tech changed: What|Removed |Added Resolution|INVALID |FIXED --- Comment #6 from tech --- We have already changed our code to use unsigned. We agree the the variable should be unsigned. What we don't agree on is the code should behave the same regardless to the optimization level. Take care thanks for your time. I have spent 3 days to find this, While converting a working code on AIX and Solaris to Linux. I still think it is a gcc bug, take care.
