为什么 GCC -Ofast 会使程序出错，但只有在它打印两次结果时才会出错？
[英]Why GCC -Ofast makes the program wrong but only when it prints the result twice?
问题描述
Recompiling an old program made it output the wrong result.
重新编译旧程序使其 output 结果错误。 I'd like to know why .我想知道为什么。
I know that -Ofast may "disregard strict standards compliance" but I'm curious about what happens under the hood.我知道-Ofast可能“无视严格的标准合规性”，但我很好奇幕后发生的事情。
I reduced the program to this minimal example foo1.c :我将程序简化为这个最小的示例foo1.c ：
#include <stdio.h>
double my_pow(double x, unsigned n)
{ /* returns x^n */
double y = 1;
while(n--) y *= x;
return y;
}
void foo(double small)
{ /* prints small^19 */
double x = my_pow(small,19);
printf("%E\n",x);
printf("%E\n",x);
}
int main(void)
{
foo(1-0.8-0.2);
return 0;
}
When compiled with -Ofast it gives a different output than with any other optimization level.当使用-Ofast编译时，它给出的 output 与任何其他优化级别不同。
gcc -Ofast foo1.c &&./a.out : gcc -Ofast foo1.c &&./a.out ：
-0.000000E+00
-0.000000E+00
gcc foo1.c &&./a.out : gcc foo1.c &&./a.out ：
-1.390671E-309
-1.390671E-309
A strange fact is that when one of the printf is commented out (file foo2.c ) this behavior doesn't replicate making it a sort of heisenbug.一个奇怪的事实是，当其中一个printf被注释掉（文件foo2.c ）时，这种行为不会复制，使其成为一种海森错误。
gcc -Ofast foo2.c &&./a.out : gcc -Ofast foo2.c &&./a.out ：
-1.390671E-309
gcc foo2.c &&./a.out : gcc foo2.c &&./a.out ：
-1.390671E-309
Informations that might be useful:可能有用的信息：
gcc -v : gcc -v ：
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/x86_64-redhat-linux/4.8.5/lto-wrapper
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-linker-build-id --with-linker-hash-style=gnu --enable-languages=c,c++,objc,obj-c++,java,fortran,ada,go,lto --enable-plugin --enable-initfini-array --disable-libgcj --with-isl=/builddir/build/BUILD/gcc-4.8.5-20150702/obj-x86_64-redhat-linux/isl-install --with-cloog=/builddir/build/BUILD/gcc-4.8.5-20150702/obj-x86_64-redhat-linux/cloog-install --enable-gnu-indirect-function --with-tune=generic --with-arch_32=x86-64 --build=x86_64-redhat-linux
Thread model: posix
gcc version 4.8.5 20150623 (Red Hat 4.8.5-39) (GCC)
gcc -Ofast foo1.c -S -o - : gcc -Ofast foo1.c -S -o - ：
.file "foo1.c"
.text
.p2align 4,,15
.globl my_pow
.type my_pow, @function
my_pow:
.LFB11:
.cfi_startproc
testl %edi, %edi
leal -1(%rdi), %edx
je .L10
movl %edi, %ecx
shrl %ecx
movl %ecx, %esi
addl %esi, %esi
je .L11
cmpl $9, %edi
jbe .L11
movapd %xmm0, %xmm1
movapd .LC0(%rip), %xmm2
xorl %eax, %eax
unpcklpd %xmm1, %xmm1
.L9:
addl $1, %eax
mulpd %xmm1, %xmm2
cmpl %eax, %ecx
ja .L9
movapd %xmm2, -24(%rsp)
subl %esi, %edx
cmpl %esi, %edi
movsd -16(%rsp), %xmm1
mulsd %xmm2, %xmm1
je .L2
testl %edx, %edx
mulsd %xmm0, %xmm1
je .L2
.L35:
cmpl $1, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $2, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $3, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $4, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $5, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $6, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $7, %edx
mulsd %xmm0, %xmm1
je .L2
mulsd %xmm0, %xmm1
.p2align 4,,10
.p2align 3
.L2:
movapd %xmm1, %xmm0
ret
.p2align 4,,10
.p2align 3
.L11:
movsd .LC1(%rip), %xmm1
testl %edx, %edx
mulsd %xmm0, %xmm1
je .L2
jmp .L35
.p2align 4,,10
.p2align 3
.L10:
movsd .LC1(%rip), %xmm1
jmp .L2
.cfi_endproc
.LFE11:
.size my_pow, .-my_pow
.section .rodata.str1.1,"aMS",@progbits,1
.LC2:
.string "%E\n"
.text
.p2align 4,,15
.globl foo
.type foo, @function
foo:
.LFB12:
.cfi_startproc
movapd %xmm0, %xmm2
subq $24, %rsp
.cfi_def_cfa_offset 32
movl $.LC2, %edi
movl $1, %eax
unpcklpd %xmm2, %xmm2
movapd %xmm2, %xmm1
mulpd %xmm2, %xmm1
mulpd %xmm1, %xmm1
mulpd %xmm1, %xmm1
mulpd %xmm2, %xmm1
movapd %xmm1, %xmm2
unpckhpd %xmm1, %xmm1
mulsd %xmm1, %xmm2
mulsd %xmm0, %xmm2
movapd %xmm2, %xmm0
movsd %xmm2, 8(%rsp)
call printf
movsd 8(%rsp), %xmm2
movl $.LC2, %edi
movl $1, %eax
addq $24, %rsp
.cfi_def_cfa_offset 8
movapd %xmm2, %xmm0
jmp printf
.cfi_endproc
.LFE12:
.size foo, .-foo
.section .text.startup,"ax",@progbits
.p2align 4,,15
.globl main
.type main, @function
main:
.LFB13:
.cfi_startproc
subq $8, %rsp
.cfi_def_cfa_offset 16
movsd .LC3(%rip), %xmm0
call foo
xorl %eax, %eax
addq $8, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE13:
.size main, .-main
.section .rodata.cst16,"aM",@progbits,16
.align 16
.LC0:
.long 0
.long 1072693248
.long 0
.long 1072693248
.section .rodata.cst8,"aM",@progbits,8
.align 8
.LC1:
.long 0
.long 1072693248
.align 8
.LC3:
.long 0
.long -1131413504
.ident "GCC: (GNU) 4.8.5 20150623 (Red Hat 4.8.5-39)"
.section .note.GNU-stack,"",@progbits
gcc foo1.c -S -o - : gcc foo1.c -S -o - ：
.file "foo1.c"
.text
.globl my_pow
.type my_pow, @function
my_pow:
.LFB0:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp
.cfi_def_cfa_register 6
movsd %xmm0, -24(%rbp)
movl %edi, -28(%rbp)
movabsq $4607182418800017408, %rax
movq %rax, -8(%rbp)
jmp .L2
.L3:
movsd -8(%rbp), %xmm0
mulsd -24(%rbp), %xmm0
movsd %xmm0, -8(%rbp)
.L2:
movl -28(%rbp), %eax
leal -1(%rax), %edx
movl %edx, -28(%rbp)
testl %eax, %eax
jne .L3
movq -8(%rbp), %rax
movq %rax, -40(%rbp)
movsd -40(%rbp), %xmm0
popq %rbp
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE0:
.size my_pow, .-my_pow
.section .rodata
.LC1:
.string "%E\n"
.text
.globl foo
.type foo, @function
foo:
.LFB1:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp
.cfi_def_cfa_register 6
subq $32, %rsp
movsd %xmm0, -24(%rbp)
movq -24(%rbp), %rax
movl $19, %edi
movq %rax, -32(%rbp)
movsd -32(%rbp), %xmm0
call my_pow
movsd %xmm0, -32(%rbp)
movq -32(%rbp), %rax
movq %rax, -8(%rbp)
movq -8(%rbp), %rax
movq %rax, -32(%rbp)
movsd -32(%rbp), %xmm0
movl $.LC1, %edi
movl $1, %eax
call printf
movq -8(%rbp), %rax
movq %rax, -32(%rbp)
movsd -32(%rbp), %xmm0
movl $.LC1, %edi
movl $1, %eax
call printf
leave
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE1:
.size foo, .-foo
.globl main
.type main, @function
main:
.LFB2:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp
.cfi_def_cfa_register 6
subq $16, %rsp
movabsq $-4859383997932765184, %rax
movq %rax, -8(%rbp)
movsd -8(%rbp), %xmm0
call foo
movl $0, %eax
leave
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE2:
.size main, .-main
.ident "GCC: (GNU) 4.8.5 20150623 (Red Hat 4.8.5-39)"
.section .note.GNU-stack,"",@progbits
gcc -Ofast foo2.c -S -o - : gcc -Ofast foo2.c -S -o - ：
.file "foo2.c"
.text
.p2align 4,,15
.globl my_pow
.type my_pow, @function
my_pow:
.LFB11:
.cfi_startproc
testl %edi, %edi
leal -1(%rdi), %edx
je .L10
movl %edi, %ecx
shrl %ecx
movl %ecx, %esi
addl %esi, %esi
je .L11
cmpl $9, %edi
jbe .L11
movapd %xmm0, %xmm1
movapd .LC0(%rip), %xmm2
xorl %eax, %eax
unpcklpd %xmm1, %xmm1
.L9:
addl $1, %eax
mulpd %xmm1, %xmm2
cmpl %eax, %ecx
ja .L9
movapd %xmm2, -24(%rsp)
subl %esi, %edx
cmpl %esi, %edi
movsd -16(%rsp), %xmm1
mulsd %xmm2, %xmm1
je .L2
testl %edx, %edx
mulsd %xmm0, %xmm1
je .L2
.L35:
cmpl $1, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $2, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $3, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $4, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $5, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $6, %edx
mulsd %xmm0, %xmm1
je .L2
cmpl $7, %edx
mulsd %xmm0, %xmm1
je .L2
mulsd %xmm0, %xmm1
.p2align 4,,10
.p2align 3
.L2:
movapd %xmm1, %xmm0
ret
.p2align 4,,10
.p2align 3
.L11:
movsd .LC1(%rip), %xmm1
testl %edx, %edx
mulsd %xmm0, %xmm1
je .L2
jmp .L35
.p2align 4,,10
.p2align 3
.L10:
movsd .LC1(%rip), %xmm1
jmp .L2
.cfi_endproc
.LFE11:
.size my_pow, .-my_pow
.section .rodata.str1.1,"aMS",@progbits,1
.LC2:
.string "%E\n"
.text
.p2align 4,,15
.globl foo
.type foo, @function
foo:
.LFB12:
.cfi_startproc
movapd %xmm0, %xmm2
movl $.LC2, %edi
movl $1, %eax
unpcklpd %xmm2, %xmm2
movapd %xmm2, %xmm1
mulpd %xmm2, %xmm1
mulpd %xmm1, %xmm1
mulpd %xmm1, %xmm1
mulpd %xmm2, %xmm1
movapd %xmm1, %xmm2
unpckhpd %xmm1, %xmm1
mulsd %xmm1, %xmm2
mulsd %xmm0, %xmm2
movapd %xmm2, %xmm0
jmp printf
.cfi_endproc
.LFE12:
.size foo, .-foo
.section .text.startup,"ax",@progbits
.p2align 4,,15
.globl main
.type main, @function
main:
.LFB13:
.cfi_startproc
subq $8, %rsp
.cfi_def_cfa_offset 16
movl $.LC2, %edi
movl $1, %eax
movsd .LC3(%rip), %xmm0
call printf
xorl %eax, %eax
addq $8, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE13:
.size main, .-main
.section .rodata.cst16,"aM",@progbits,16
.align 16
.LC0:
.long 0
.long 1072693248
.long 0
.long 1072693248
.section .rodata.cst8,"aM",@progbits,8
.align 8
.LC1:
.long 0
.long 1072693248
.align 8
.LC3:
.long 0
.long -2147418112
.ident "GCC: (GNU) 4.8.5 20150623 (Red Hat 4.8.5-39)"
.section .note.GNU-stack,"",@progbits
gcc foo2.c -S -o - : gcc foo2.c -S -o - ：
.file "foo2.c"
.text
.globl my_pow
.type my_pow, @function
my_pow:
.LFB0:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp
.cfi_def_cfa_register 6
movsd %xmm0, -24(%rbp)
movl %edi, -28(%rbp)
movabsq $4607182418800017408, %rax
movq %rax, -8(%rbp)
jmp .L2
.L3:
movsd -8(%rbp), %xmm0
mulsd -24(%rbp), %xmm0
movsd %xmm0, -8(%rbp)
.L2:
movl -28(%rbp), %eax
leal -1(%rax), %edx
movl %edx, -28(%rbp)
testl %eax, %eax
jne .L3
movq -8(%rbp), %rax
movq %rax, -40(%rbp)
movsd -40(%rbp), %xmm0
popq %rbp
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE0:
.size my_pow, .-my_pow
.section .rodata
.LC1:
.string "%E\n"
.text
.globl foo
.type foo, @function
foo:
.LFB1:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp
.cfi_def_cfa_register 6
subq $32, %rsp
movsd %xmm0, -24(%rbp)
movq -24(%rbp), %rax
movl $19, %edi
movq %rax, -32(%rbp)
movsd -32(%rbp), %xmm0
call my_pow
movsd %xmm0, -32(%rbp)
movq -32(%rbp), %rax
movq %rax, -8(%rbp)
movq -8(%rbp), %rax
movq %rax, -32(%rbp)
movsd -32(%rbp), %xmm0
movl $.LC1, %edi
movl $1, %eax
call printf
leave
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE1:
.size foo, .-foo
.globl main
.type main, @function
main:
.LFB2:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp
.cfi_def_cfa_register 6
subq $16, %rsp
movabsq $-4859383997932765184, %rax
movq %rax, -8(%rbp)
movsd -8(%rbp), %xmm0
call foo
movl $0, %eax
leave
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE2:
.size main, .-main
.ident "GCC: (GNU) 4.8.5 20150623 (Red Hat 4.8.5-39)"
.section .note.GNU-stack,"",@progbits
1 个解决方案
解决方案1
3 2019-12-14 21:03:20
Under -ffast-math (and it's siblings like -Ofast ) gcc links your app with a special startup code in crtfastmath.c which sets flush-to-zero flag:在-ffast-math （和它的兄弟姐妹一样-Ofast ） gcc 将您的应用程序与crtfastmath.c中的特殊启动代码链接起来，该代码设置刷新为零标志：
static void __attribute__((constructor))
set_fast_math (void)
{
#ifndef __x86_64__
...
#else
unsigned int mxcsr = __builtin_ia32_stmxcsr ();
mxcsr |= MXCSR_DAZ | MXCSR_FTZ;
__builtin_ia32_ldmxcsr (mxcsr);
#endif
}
声明:本站的技术帖子网页，遵循CC BY-SA 4.0协议，如果您需要转载，请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.
相关问题
gcc -Ofast -pg错误的gprof结果 - wrong gprof results with gcc -Ofast -pg
是什么让 Printf 打印两次？ - What makes Printf prints twice?
-Ofast 的 GCC 问题？ - GCC issue with -Ofast?
程序打印错误的输出 - Program prints the wrong output
-DNDEBUG with -Ofast比-Ofast慢 - -DNDEBUG with -Ofast is slower than only -Ofast
C中的Memmove（）将结果打印两次 - Memmove() in C prints the result twice
printf在gcc错误的结果 - printf in gcc wrong result
为什么这个程序打印 0？ - Why this program prints 0?
程序打印两次相同的打印语句 - Program prints out the same print statement twice
仅打印“程序中发生错误！” - Prints only “An error occurred in the program!”
相关标签