target/5985: fix_truncMN2 (e.g. fix_truncdfdi) and fixuns_truncMN2 incorrect on a29k, alpha, arm, c4x, dsp16xx, i370, i386, ia64, m88k, mips, ns32k, pj, romp, rs6000, s390, sh, xtensa

[hp at bitrange dot com]

>Number:         5985
>Category:       target
>Synopsis:       fix_truncMN2 (e.g. fix_truncdfdi) and fixuns_truncMN2 incorrect on a29k, alpha, arm, c4x, dsp16xx, i370, i386, ia64, m88k, mips, ns32k, pj, romp, rs6000, s390, sh, xtensa
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sun Mar 17 05:46:01 PST 2002
>Closed-Date:
>Last-Modified:
>Originator:     Hans-Peter Nilsson
>Release:        trunk and 3.1 branch and everywhere else
>Organization:
>Environment:

>Description:
>From <URL:http://gcc.gnu.org/ml/gcc-patches/2002-03/msg01023.html>, edited:
While checking on the floating point operations and comparing
with the manual and other targets, it seems several targets
(briefly: a29k, alpha, arm, c4x, dsp16xx, i370, i386, ia64,
m88k, mips, ns32k, pj, romp, rs6000, s390, sh, xtensa) have fix_truncMN2
patterns that are incorrectly expressed.  For example for M=df
N=di, the input operand should be (fix:DI (fix:DF (DF-operand))
not just (fix:DI (DF-operand)), since the latter expression
requires/promises that the operand is already truncated to an
integer in a float-mode, which is the input operand for the
fixdfdi2 pattern.  Similarly for fixuns_truncMN2.
>How-To-Repeat:
Read .md for target, specifically fix_truncMN2 and fixuns_truncMN2
patterns.  Compare with docs and other targets, like sparc.md, m68k.md.
Marvel at how it doesn't seem to affect anything.
>Fix:
See sparc.md, m68k.md, mmix.md.
>Release-Note:
>Audit-Trail:
>Unformatted: