This is the mail archive of the
cgen@sources.redhat.com
mailing list for the CGEN project.
play.cpu or parallel semantics broken
- From: Hans-Peter Nilsson <hans-peter dot nilsson at axis dot com>
- To: cgen at sources dot redhat dot com
- Date: Thu, 21 Mar 2002 00:56:45 +0100
- Subject: play.cpu or parallel semantics broken
#guile
guile> (load "dev.scm")
[...]
guile> (load-sim)
[...]
guile> (cload #:arch "cpu/play")
[...]
guile> (cgen-sem-switch.c)
Generating cpuf sem-switch.c ...
Analyzing instruction set ...
[...]
Processing semantic switch case for "add $dr,$sr" ...
[ Note that this is actually addv2; they confusingly have the
same "syntax". ]
Backtrace:
32* [#<procedure (e)> (parallel () DFLT ...)]
33* [rtl-c-with-estate #(# #) #(# #) (parallel () DFLT () ...)]
34 [cx:c ...
35* [rtl-c-get #(# #) #(# #) ...
36* [rtx-eval-with-estate (parallel () DFLT () ...) #(# #) #(# #)]
37 (if (pair? expr) (let* (# #) (if fn # ...)) ...)
...
38 [s-parallel #(#("object" # #f ...) (# #f # ...)) (set () DFLT ...) ...]
39 (let* ((set-dests #) (temps #)) (set! -par-temp-list temps) ...)
40* [string-map #<procedure (e)> ...
41* [-par-replace-set-dests #(# #) (# # #)]
42 (let ((sets (list # # # #))) (letrec ((replace #)) (map replace exprs)))
...
43 [map #<procedure replace (expr)> (# # #)]
44* [replace (set () DFLT ...)]
45* (let (# # #) (if # # #))
46 (if (memq name sets) (list name options ...) ...)
47 [list set () DFLT ...
48* [-par-new-temp! ...
49* (if (mode:eq? # mode) (rtx-lvalue-mode-name estate #) mode)
50 [rtx-lvalue-mode-name #(# #) (local () DFLT tmp1)]
...
51 [error "rtx-lvalue-mode-name: not an operand or hardware reference:" $
./rtl.scm:437:6: In procedure error in expression (error "rtx-lvalue-mode-name: not an operand or hardware reference:" x):
./rtl.scm:437:6: rtx-lvalue-mode-name: not an operand or hardware reference: (local () DFLT tmp1)
ABORT: (misc-error)
Also happens with (cgen-semantics.c). And this also happens
without my recent pmacros.scm patch. ;-)
It looks like a parallel does not inherit the variables of a
surrounding sequence; in this case tmp1. Not that I need that
feature, but I think play.cpu shouldn't fail. An easy way out
is to just remove addv2.
brgds, H-P