why cosmocc: fatal error: -march=native can't be used when building fat binaries

[liudonghua123]

I just want to compile a portable binary which runable on linux, windows and macos. I download the toolchain and use CC=/root/code/cosmocc/bin/cosmocc make to compile it on linux.

But it failed with the following errors.

[root@ha-master-1 gguf-tools]# CC=/root/code/cosmocc/bin/cosmocc make 
/root/code/cosmocc/bin/cosmocc gguf-tools.c gguflib.c sds.c fp16.c \
	-march=native -ffast-math \
	-g -ggdb -Wall -W -pedantic -O3 -o gguf-tools
cosmocc: fatal error: -march=native can't be used when building fat binaries
compilation terminated.
make: *** [Makefile:4: gguf-tools] Error 1
[root@ha-master-1 gguf-tools]#

I have to use x86_64-unknown-cosmo-cc instead of cosmocc.

[root@ha-master-1 gguf-tools]# CC=/root/code/cosmocc/bin/x86_64-unknown-cosmo-cc make
/root/code/cosmocc/bin/x86_64-unknown-cosmo-cc gguf-tools.c gguflib.c sds.c fp16.c \
	-march=native -ffast-math \
	-g -ggdb -Wall -W -pedantic -O3 -o gguf-tools
In function 'sdsll2str',
    inlined from 'sdsfromlonglong' at sds.c:516:15:
sds.c:473:12: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
  473 |         *s = *p;
      |         ~~~^~~~
......
[root@ha-master-1 gguf-tools]# /root/code/cosmocc/bin/x86_64-linux-cosmo-objcopy -SO binary gguf-tools gguf-tools.com
[root@ha-master-1 gguf-tools]# file gguf-tools gguf-tools.com
gguf-tools:     ELF 64-bit LSB executable, x86-64, version 1 (GNU/Linux), statically linked, with debug_info, not stripped
gguf-tools.com: DOS/MBR boot sector; partition 1 : ID=0x7f, active, start-CHS (0x0,0,1), end-CHS (0x3ff,255,63), startsector 0, 4294967295 sectors
[root@ha-master-1 gguf-tools]# ./gguf-tools.com -h
Usage: ./gguf-tools.com <subcommand> [arguments...] [options...]
Subcommands:
  show <filename> -- show GGUF model keys and tensors.
  inspect-tensor <filename> <tensor-name> [count] -- show tensor weights.
  compare <file1> <file2> -- avg weights diff for matching tensor names.
  split-mixtral <ids...> mixtral.gguf out.gguf -- extract expert.
Options:
  --verbose       :With 'show', print full arrays (e.g. token lists)
  --diffable      :Don't show tensor file offsets and sizes
Example:
  split-mixtral 65230776370407150546470161412165 mixtral.gguf out.gguf
[root@ha-master-1 gguf-tools]#