This is the mail archive of the
cgen@sourceware.org
mailing list for the CGEN project.
Problem in cgen-generated sim insn decoder.
- From: Dmitry Eremin-Solenikov <dbaryshkov at gmail dot com>
- To: cgen at sourceware dot org
- Date: Tue, 29 Dec 2009 02:46:50 +0300
- Subject: Problem in cgen-generated sim insn decoder.
Hello,
Just to remind, I'm trying to do a cgen-based m68hc08 description.
Currently I've stumbled upon sim insn decoder. All the extract_ifmt_*
parts of @cpu@_decode expect to have entire insn bits in entire_insn
aligned to the LSB bit of that value. The problem is that
instructions of my MCU are of variable length and I use lsb0=#f setting.
I found two solutions for my case:
* First one is to teach extract() part of mloop.in more details about
insn lengths, so that it finds on it's own what is the length of the
insn and pushes the correct entire_insn to @cpu@_decode.
I find this pre-parsing clumsy and unclean.
* Another one is to change the API of @cpu@_decode to require
entire_insn to be msb-aligned if the ISA is lsb0=#f.
The patch is really simple (see attach), however it can (and most
probably will) break other lsb0=#f arches (m32r/m32c).
Could you please comment on this topic?
--
With best wishes
Dmitry
diff --git a/cgen/sim-decode.scm b/cgen/sim-decode.scm
index 0c6d48c..42c41ec 100644
--- a/cgen/sim-decode.scm
+++ b/cgen/sim-decode.scm
@@ -437,7 +437,10 @@ void
(string-append
" CGEN_INSN_WORD insn = "
(if (adata-integral-insn? CURRENT-ARCH)
- "entire_insn;\n"
+ (if (current-arch-insn-lsb0?)
+ "entire_insn;\n"
+ (string-append
+ "entire_insn >> (sizeof(UINT) * 8 - " (number->string (sfmt-length sfmt)) ");\n" ))
"base_insn;\n"))
"")
(gen-define-field-macro sfmt)