<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/155119>155119</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            -fexcess-precision=standard does not limit precision of x87 math
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            clang
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          mstorsjo
      </td>
    </tr>
</table>

<pre>
    According to the documentation, x87 math should have the precision limited to the nominal precision of the type:

https://clang.llvm.org/docs/UsersManual.html

> For example, even on pre-SSE X86 targets where `float` and `double` computations must be performed in the 80-bit X87 format, Clang rounds all intermediate results correctly for their type.

Despite this, for a test function like this, the full x87 precision is retained even if both intermediates are assigned to float types (the same also holds if using explicit casts).

```c
int math(float a, float b, float c) {
    float prod = a * b;
    float sum = prod + c;
    return (int)sum;
}
```

```
$ clang --target=i686-linux-gnu -mno-sse -O2 -fexcess-precision=standard math.c -S -o -
```

This differs from GCC, where `-fexcess-precision=standard` does limit the precision to the nominal precision of type type - see https://godbolt.org/z/arMsje11T. Here, GCC produces
```
        fstp DWORD PTR [esp]
        fld     DWORD PTR [esp]
```
after both the multiplication and the addition, while Clang does nothing to limit the intermediate precision.
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJx8VE1v4zYQ_TX0ZSBDpvwhHXRI7PX2sthik6J7pciRxZQiBQ6ZTfrrC1JKnKSLGAZEcYbDeU_vjSDSF4vYst0t251WIobB-Xak4Dw9uFXn1HN7I6XzStsLBAdhQFBOxhFtEEE7y_gRnuoDjCIMQIOLRsEgHjFnTh6lJu0sGD3qgOqlhHWjtsK8SXB9DoTnCVl1w8r0H0KYKL3xM-NnaYS9rI15HNfOXxg_KyeJ8fNfhJ6-CRuFWQ9hNPNZVn2Bs_OAT2KcDKY28REtOJsuLe7uvsDPeg9B-AsGgl8DegS2L3vjRGD7EoRV6V252BlMG9KNU5xBE4yRAnQIE_re-REVaJsB1GXR6QA_6wOkgAjp5mNqHbyLVhEIY0DbgOmUFgHBI0UTCKTzHmUwz-lkKqZ95mM9IzohTTokYjWloilJQEAK0Ecrw0zzP9eE1E4fjcnf58q0JvAYhLaoZkp0D50Lw7umCIRHWPSRP1smJvdDwHidipMYEYQhB4MzilKhSEko-DQZLXUAKSgQ480Cge3L-S9ZeaNtyKphvJ5ri4wqL7vrUjLeADvcsvIGAJbNyTsFrDqBAMZvoGPV-zjFMYfnPH4L8prhMURvEwZtA-MNxXEOssPpbY8fWk5rvoWsQiiKWTisOul9vS-MtvGpuNgIxWhdQYRQfOdQ9Pgkkah4ZZ9VJwrCKuFVBr-WUNxB4aD4_9X3gyZQuu_RE_TejfD1eEzEvIr10_pJs8ohzd774MdPffg8zUaEAggR3tvw4lTnTFg8-C_jZ-G_0QNuNvdr-AN9ttrX4zFzHyXSB2Cw_HoKE5z-_v7jBH_e_wC2u0Wa2O70NsWo_Px91tuiog_oZxUnWGM0QScJZrdmK6dtoZR-mVm_Bm1wMWZmybowLEPuStg7n76StF6ptlJN1YgVtpvDbtfUfLNvVkPb8abaqro8yJqLA695hRWv637XbERfNbjSLS_5rqx5xcttw5u13B6aTVdWqt5s95Irti1xFNq8DrqVJorYbna7zaZZGdGhoTyvOc9aZJyn0e3bdKDo4oXYtjSaAl1LBB0Mtp-r8YWEBf07RbwM-FX0pv0gBx2G2K2lGxk_pwuXRzF594AyMH7OANKkXjA8tvy_AAAA__98fxkZ">