Hi,
When analyzing some disassembled code I observed an issue which seems
like a bug to me. I'll try to properly describe it below. However I'm
wondering if this comes from something that went wrong while building
the cross-compiler, from the linker or from some option I'm not using
rightly.
----------------------------------------- ISSUE
When compiling codes that use the libm (I didn't get to reproduce the
test with other libs, I didn't try that much either), many functions
appear quite a few times in the disassembled file. I wrote this
minimalist test case to reproduce the issue with ISNAN but this extends
to several functions in benchmarks like Paranoia.
----------------------------------------- GCC
This test has been done with GCC 4.5.1 and mainline built for SPARCV8.
/opt/sparc-elf-4.5.1/bin/sparc-elf-gcc test.c -Os -flto -fwhole-program
-finline -ffunction-sections -fdata-sections -e main -Wl,--gc-sections
-nostdlib -lm -lgcc
----------------------------------------- C code
#include <math.h>
int main(){
int i, a;
for(i=-10;i<=10;i++){a+=isnan(i);}
return a;
}
----------------------------------------- Bytecode
40000000 <main>:
4000003c <isnan>: <------------------------------------ First
instantiation
4000003c: 9d e3 bf a0 save %sp, -96, %sp
40000040: f0 27 a0 44 st %i0, [ %fp + 0x44 ]
40000044: c2 07 a0 44 ld [ %fp + 0x44 ], %g1
40000048: c2 00 40 00 ld [ %g1 ], %g1
4000004c: 82 18 60 00 xor %g1, 0, %g1
40000050: 80 a0 00 01 cmp %g0, %g1
40000054: 82 60 3f ff subx %g0, -1, %g1
40000058: 80 a0 60 00 cmp %g1, 0
4000005c: 12 80 00 0a bne 40000084 <isnan+0x48>
40000060: 01 00 00 00 nop
40000064: c2 07 a0 44 ld [ %fp + 0x44 ], %g1
40000068: c2 00 40 00 ld [ %g1 ], %g1
4000006c: 82 18 60 01 xor %g1, 1, %g1
40000070: 80 a0 00 01 cmp %g0, %g1
40000074: 82 60 3f ff subx %g0, -1, %g1
40000078: 80 a0 60 00 cmp %g1, 0
4000007c: 02 80 00 05 be 40000090 <isnan+0x54>
40000080: 01 00 00 00 nop
40000084: 82 10 20 01 mov 1, %g1 ! 1 <main-0x3fffffff>
40000088: 10 80 00 03 b 40000094 <isnan+0x58>
4000008c: 01 00 00 00 nop
40000090: 82 10 20 00 clr %g1 ! 0 <main-0x40000000>
40000094: b0 10 00 01 mov %g1, %i0
40000098: 81 e8 00 00 restore
4000009c: 81 c3 e0 08 retl
400000a0: 01 00 00 00 nop
400000a4 <__unorddf2>:
40000150 <clzusi>:
40000178 <__floatsidf>:
400002d8 <__clzsi2>:
40000398 <isnan>: <------------------------------------ Second
instantiation
40000398: 9d e3 bf a0 save %sp, -96, %sp
4000039c: f0 27 a0 44 st %i0, [ %fp + 0x44 ]
400003a0: c2 07 a0 44 ld [ %fp + 0x44 ], %g1
400003a4: c2 00 40 00 ld [ %g1 ], %g1
400003a8: 82 18 60 00 xor %g1, 0, %g1
400003ac: 80 a0 00 01 cmp %g0, %g1
400003b0: 82 60 3f ff subx %g0, -1, %g1
400003b4: 80 a0 60 00 cmp %g1, 0
400003b8: 12 80 00 0a bne 400003e0 <isnan+0x48>
400003bc: 01 00 00 00 nop
400003c0: c2 07 a0 44 ld [ %fp + 0x44 ], %g1
400003c4: c2 00 40 00 ld [ %g1 ], %g1
400003c8: 82 18 60 01 xor %g1, 1, %g1
400003cc: 80 a0 00 01 cmp %g0, %g1
400003d0: 82 60 3f ff subx %g0, -1, %g1
400003d4: 80 a0 60 00 cmp %g1, 0
400003d8: 02 80 00 05 be 400003ec <isnan+0x54>
400003dc: 01 00 00 00 nop
400003e0: 82 10 20 01 mov 1, %g1 ! 1 <main-0x3fffffff>
400003e4: 10 80 00 03 b 400003f0 <isnan+0x58>
400003e8: 01 00 00 00 nop
400003ec: 82 10 20 00 clr %g1 ! 0 <main-0x40000000>
400003f0: b0 10 00 01 mov %g1, %i0
400003f4: 81 e8 00 00 restore
400003f8: 81 c3 e0 08 retl
400003fc: 01 00 00 00 nop
40000400 <isinf>:
4000042c <iszero>:
40000458 <__pack_d>:
40000960 <__unpack_d>:
40000bd0 <___CTOR_LIST__>:
40000bd8 <___CTOR_END__>:
40000be0 <___DTOR_END__>:
40000be8 <__clz_tab>:
I'll appreciate any feedback you provide,
Thanks in advance,
Jorge
