floating point emulation can fail to set FE_UNDERFLOW

[Paul Clarke]

Floating point emulation can fail to set FE_UNDERFLOW in some circumstances. This shows up often in glibc's "math" tests. A similar test is attached.

This is similar to bug #1841442, but not the same problem, and I don't think the fix will be in the same code.

On ppc64le native:
--
$ gcc -c -O2 fma.c
$ gcc -O2 test-fma.c fma.o -lm -o test-fma
$ ./test-fma $(./test-fma)
fma(0x1.ffffffffffffcp-1022, 0x1.0000000000001p-1, 0x0.0000000000001p-1022)
0x0

0xa000000
FE_INEXACT FE_UNDERFLOW
0x1p-1022
--

On qemu-system-ppc64:
--
$ ./test-fma $(./test-fma)
fma(0x1.ffffffffffffcp-1022, 0x1.0000000000001p-1, 0x0.0000000000001p-1022)
0x0

0x2000000
FE_INEXACT
0x1p-1022
--

QEMU versions vary, but not too much, and are pretty close to git HEAD:
- 586f3dced9 (HEAD -> master, origin/master, origin/HEAD) Merge remote-tracking branch 'remotes/cohuck/tags/s390x-20190822' into staging
- 864ab31 Update version for v4.1.0-rc4 release

There are worse symptoms on qemu-x86_64, but this is apparently not surprising per https://bugs.launchpad.net/qemu/+bug/1841442/comments/6.

[Paul Clarke (7-pc)]

Responding to the patch https://lists.nongnu.org/archive/html/qemu-ppc/2019-08/msg00404.html, it seems to work for "double", but not for "float". Test case attached.

[Richard Henderson (rth)]

The float test failure is part of a larger problem for target/powerpc in which all float routines are implemented incorrectly. They are all implemented as double operations with rounding to float as a second step. Which not only produces incorrect exceptions, as in this case, but incorrect numerical results from the double rounding.

This should probably be split to a separate bug...

[Thomas Huth (th-huth)]

A patch for this bug has been merged here:
https://git.qemu.org/?p=qemu.git;a=commitdiff;h=cbc65a8f22b29680f3
... can we close this ticket now or is there more to do?

[Paul Clarke (7-pc)]

Comment #5 suggested splitting the "float" issue to a separate bug, which was done some time ago (bug #1841592).

I think this ticket can be closed.

[Thomas Huth (th-huth)]

Ok, thanks for the pointer to the other bug! So I'm closing this one now.