strange symbol in the object file

Li__Yang · ‎11-03-2017

Hi,all.

When i link to generate executable file after compiled the whole project in 32bits mode, the linker tell me there is a symbol at one object file can not be found! I work at the mac OSX 10.12 and use the intel compiler of version 2018.

The symbol is '___cvtu64todp_sse2'. I am sure I not define or use the symbol in my whole project. I use the objdump tool to get more info of the object file.

First, I open the symbol table. I find the strange symbol and it is tagged "UNDEFINED".

......

00000000 *UND* _g_thread_functions_for_glib_use

00000000 *UND* ___gxx_personality_v0

00000000 *UND* _g_threads_got_initialized

00000000 *UND* ___cxa_call_unexpected

00000000 *UND* __Unwind_Resume

00000000 *UND* ___cvtu64todp_sse2

00000000 *UND* __Z11GetLogLevelv

00000000 *UND* __Z17printf_log_helper13AudioLogLevelPKcz

00000000 *UND* _printf

Then I open the relocation table, There are a lot of entry of this symbol. A segment of the relocation table:

...

00000391 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
000003b0 GENERIC_RELOC_LOCAL_SECTDIFF 0x40f8-0x2ff
000003d9 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
000003ea GENERIC_RELOC_LOCAL_SECTDIFF 0x4100-0x2ff
00000403 GENERIC_RELOC_LOCAL_SECTDIFF 0x4150-0x2ff
000004ce GENERIC_RELOC_LOCAL_SECTDIFF 0x4098-0x2ff
0000063c GENERIC_RELOC_LOCAL_SECTDIFF 0x40f8-0x2ff
00000657 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
000006a5 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000790 GENERIC_RELOC_LOCAL_SECTDIFF __const-0x2ff
00000797 GENERIC_RELOC_LOCAL_SECTDIFF 0x4140-0x2ff
00000810 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000838 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
0000096c GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
000009df GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000a42 GENERIC_RELOC_LOCAL_SECTDIFF 0x40f8-0x2ff
00000b25 GENERIC_RELOC_LOCAL_SECTDIFF __const-0x2ff
00000b63 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000b86 GENERIC_RELOC_LOCAL_SECTDIFF 0x4140-0x2ff
00000b9a GENERIC_RELOC_LOCAL_SECTDIFF 0x4118-0x2ff
00000bf5 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000c1d GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000cae GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000cd6 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000d67 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000d8f GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000e20 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000e48 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000f36 GENERIC_RELOC_LOCAL_SECTDIFF __const-0x2ff
00000f67 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00000f92 GENERIC_RELOC_LOCAL_SECTDIFF 0x4140-0x2ff
00000fa6 GENERIC_RELOC_LOCAL_SECTDIFF 0x4118-0x2ff
0000100e GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00001036 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00001113 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
00001157 GENERIC_RELOC_VANILLA ___cvtu64todp_sse2
0000119c GENERIC_RELOC_LOCAL_SECTDIFF 0x4150-0x2ff

...

Last, I disassembling the object file and display the symbol in the assembly code. I find there are 77 site. Moreover the symbol appear at a almost same code:

......

6a4: e8 57 f9 ff ff calll -1705 <__ZN16ClockrateManagerC1Ev>

000006a5: GENERIC_RELOC_VANILLA ___cvtu64todp_sse2

......

The symbol, '__ZN16ClockrateManagerC1Ev', is a construction function in my source file. But I not call it so many times

So many strange things happened. I want to know the reason. Is it a defect of intel compile of 2018. There more than one strange object file just like this in my project

I can post the command used to compile the file:

"/usr/local/bin/icpc -o clockratemanager.o -c -arch i386 -std=c++11 -stdlib=libc++ -isysroot /Applications/... -mmacosx-version-min=10.7 -arch i386 -I/opt/intel/composerxe/ipp/include -I/opt/intel/composerxe/include -g -O3 -pipe -wd1,161,172,177,181,188,383,593,869,981,1418,1419,1572,1599,2259,3199,18000 -restrict -fasm-blocks -ip -MD -MP -m32, -DENV_DARWIN=1..."

On the other hand, I compile the same source file at the same system and 2017 intel compile. I open the above object file, and find there a no the strange symbol in the symbol table:

......

00000000 *UND* _g_thread_functions_for_glib_use

00000000 *UND* ___gxx_personality_v0

00000000 *UND* _g_threads_got_initialized

00000000 *UND* ___cxa_call_unexpected

00000000 *UND* __Unwind_Resume

00000000 *UND* __Z11GetLogLevelv

00000000 *UND* __Z17printf_log_helper13AudioLogLevelPKcz

00000000 *UND* _printf

You can see that the 2018 compiler just generate one more symbol, the strange one, ___cvtu64todp_sse2.

In the assembly code, every thing seem normal. The construct function is just called 6 times. Build the whole project in 32bits, successfully.

I go back to the 2018 compiler and compile the file and whole project in 64bits, Every thing also is OK.

so any one encounter the same problem? I need help!

jimdempseyatthecove · ‎11-06-2017

Try -arch IA32

Jim Dempsey

Li__Yang · ‎11-06-2017

jimdempseyatthecove wrote:

Try -arch IA32

Jim Dempsey

Thank you to reply

I try it. But the compiler tell me :

icpc: command line warning #10159: invalid argument for option '-arch'

so nothing chang

jimdempseyatthecove · ‎11-07-2017

Sorry, I notice you are on MAC

-arch is listed (v17 docs) as not being a Linux and macOS option

Windows has /arch:IA32, you could try -arch=IA32 (with = in place of the space)

-march=processor is supported (with available processor options)
*** however i386 is not an available processor option.

You do have -mcode (e.g. -mia32), but then this generates the x86/x87 (old Floating Point format)

ifort has (had) two versions of the compiler: Intel64 and IA32. Using the 32-bit compiler should default to generating 32-bit code.

Did you source the 32-bit compiler?

Jim Dempsey

Li__Yang · ‎11-07-2017

jimdempseyatthecove wrote:

Sorry, I notice you are on MAC

-arch is listed (v17 docs) as not being a Linux and macOS option

Windows has /arch:IA32, you could try -arch=IA32 (with = in place of the space)

-march=processor is supported (with available processor options)
*** however i386 is not an available processor option.

You do have -mcode (e.g. -mia32), but then this generates the x86/x87 (old Floating Point format)

ifort has (had) two versions of the compiler: Intel64 and IA32. Using the 32-bit compiler should default to generating 32-bit code.

Did you source the 32-bit compiler?

Jim Dempsey

Thank you very much!

I think i should chang the arguments option

-arch is not available for macOS，i will throw it.

-march = corei7 is suitable i think

"You do have -mcode (e.g. -mia32), but then this generates the x86/x87 (old Floating Point format)"

what is the meaning? -m32 should not be used? i find the docs :

-m32 or /Qm32

Tells the compiler to generate code for IA-32 architecture.

"ifort has (had) two versions of the compiler: Intel64 and IA32. Using the 32-bit compiler should default to generating 32-bit code."

Dose the above line means the intel 2018 compiler has tow version, for 64bit and for 32bits?

I think i can only find a dmg for intel 2018 compiler. m_ccompxe_2018.0.026.dmg

Dose i have to use different compiler for different build?

jimdempseyatthecove · ‎11-07-2017

I haven't installed V18 yet as I produce production code and I would rather wait for V18 update 1 or 2 before updating. I do use V17 update 5 on Linux and Windows. On Linux you issue something like

source /< install-dir >/bin/psxevars.sh ia32

or

source /< install-dir >/bin/compilervars.sh ia32

Where < install-dir > is where you installed the Intel code. The psxevars.sh may be in a different folder from the compilervars.sh

And the .sh files are for bash shell. You may have a different shell and may need to specify a different environment setup method.

Your documentation should state where and how to do this.

For 64-bit compiler, replace ia32 with intel64

Jim Dempsey