Ghidra: x87 FPU isn't handled correctly by the decompiler

[GovanifY]

Describe the bug
The decompiler fails to analyze correctly code that is making use of the x87 FPU subset of instructions.

To Reproduce
Import the binary attached(it is a standard PE), several mistakes can be seen at the following adresses:

• 0x40d770, decompiler output being:

    if ((((bullet->state != 0) && (bullet->state != 5)) &&
        (bullet->sprites[0].additional_infos != (something_56_bytes *)0x0)) &&
       ((!in_PF && (!in_PF)))) {

while the correct output should be something more along the lines of

if bullet->state != 0 && bullet->state != 5 && (bullet->sprites[0].additional_infos) != 0x0 && (30. <= (bullet->sprites[0].additional_infos)->height) && bullet->speed == 0. {

• 0x40dbc0, where the decompiler is showing a fsincos being unused, while it is actually used in the binary.

Expected behavior
The decompiler analyzing correctly the file

Attachments
binary.txt

Environment (please complete the following information):

• OS: Gentoo Linux
• Java Version: 11.0.4
• Ghidra Version: 9.1 2019-Oct-22

[caheckman]

The first issue is resolved now in the master branch with the addition of full support for the x86 parity flag. The fsincos problem is a different issue with the detection of stack aliases but is also resolved in master; thanks for reporting it. Fixes should be present in the 9.2 release.