[i386][Float128][compiler-rt] compiler-rt lack of __subtf3 symbol

[zhangtianhao6]

The __subtf3 function is automatically generated by the subtraction compiler of the float 128 type. However, the compiler rt library of the i386 architecture does not provide the __subtf3 function. The libgcc provides the __subtf3 function. Why does this difference exist? Can we make up for it or do not allow the compiler to automatically generate the __subtf3 function?

case

#include <iostream>

int main() {
    // 创建两个 __float128 类型的变量
    __float128 a = 1.23456789012345678901234567890123456789Q;  // 精度较高
    __float128 b = 0.12345678901234567890123456789012345678Q;

    // 执行减法，编译器将自动使用 __subtf3
    __float128 result = a - b;

    return 0;
}

use compiler-rt error: -rtlib=compiler-rt -m32
use libgcc ok: -m32
https://godbolt.org/z/3YT3PbPEW

[zhangtianhao6]

cc @phoebewang
how about add ${GENERIC_TF_SOURCES} to i386_SOURCES in compiler-rt/lib/builtins/CMakeLists.txt

[hstk30-hw]

CC @topperc @pranavk @compnerd

[phoebewang]

cc @phoebewang how about add ${GENERIC_TF_SOURCES} to i386_SOURCES in compiler-rt/lib/builtins/CMakeLists.txt

I don't see any reason why not. Maybe create a patch for it?

[compnerd]

AFAIK, x86 does not support fp128 which is why it is not present. Windows, Darwin, and FreeBSD (unlike Linux) do not support FP80 which is possible with the x87 co-processor. But, it does make sense to match libgcc as the library is supposed to be ABI compatible.