Example source:
void f1(long foo, long bar) {/*empty*/}
void f2(long foo, ...) {/*empty*/}
void _start()
{
f1(-1, -2); //both correct
f2(-1, -2); //-1 correct, -2 becomes 0x00000000FFFFFFFE
f2(0xffffffffffffffff, 0xfffffffffffffffe); //both correct
}
$ amd64-elf-gcc -ffreestanding -nodefaultlibs -nostdlib -nostartfiles -o
standalone.bin standalone.c
$ amd64-elf-gcc -v
Using built-in specs.
Target: amd64-elf
Configured with: /elf/src/gcc-4.1.1/configure --target=amd64-elf --prefix=/elf
-
-exec-prefix=/elf/amd64-pc-cygwin --with-gnu-as --with-gnu-ld --without-headers
Thread model: single
gcc version 4.1.1
$ amd64-elf-objdump -d standalone.bin
<snipped f1 & f2>
000000000040014a <_start>:
40014a: 55 push %rbp
40014b: 48 89 e5 mov %rsp,%rbp
40014e: 48 c7 c6 fe ff ff ff mov $0xfffffffffffffffe,%rsi
400155: 48 c7 c7 ff ff ff ff mov $0xffffffffffffffff,%rdi
40015c: e8 4f ff ff ff callq 4000b0 <f1>
400161: be fe ff ff ff mov $0xfffffffe,%esi <-- should
be qword move
400166: 48 c7 c7 ff ff ff ff mov $0xffffffffffffffff,%rdi
40016d: b8 00 00 00 00 mov $0x0,%eax
400172: e8 47 ff ff ff callq 4000be <f2>
400177: 48 c7 c6 fe ff ff ff mov $0xfffffffffffffffe,%rsi
40017e: 48 c7 c7 ff ff ff ff mov $0xffffffffffffffff,%rdi
400185: b8 00 00 00 00 mov $0x0,%eax
40018a: e8 2f ff ff ff callq 4000be <f2>
40018f: c9 leaveq
400190: c3 retq
It should emit the 0x48 prefixed version as in the 3rd call. My guess is that
"movl" instead of "movq" or "mov ..., esi" instead of "mov ..., rsi" is being
passed to as.
--
Summary: Negative numbers in varargs incorrect on AMD64
Product: gcc
Version: 4.1.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: wulf at ignmail dot com
GCC build triplet: i686-pc-cygwin
GCC host triplet: i686-pc-cygwin
GCC target triplet: amd64-pc-elf
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28931
