Discussion:
[patch][rfa] Representation of ISA Attribute in CGEN
(too old to reply)
Dave Brolley
2005-09-22 20:53:55 UTC
Permalink
Hi,

These changes have been in our local tree for a few years now and were
developed for an internal port which requires *much* more than 32 ISAs
or even 64 ISAs. These changes could be of benifit to others, so I would
like to submit them for approval.

Currently, ISA is represented as an integer, like all the other
non-boolean attributes. It is a bit mask with each bit representing
whether a particular ISA is supported. Our port requires that there be
no fixed limit on the number of ISAs, and so, we developed an open-ended
representation for a bitset using a bitstring and a length. It is called
CGEN_BITSET and is declared in include/opcode/cgen-bitset.h and
supported by several new functions in opcodes/cgen-opc.c. All
manipulation of these bitsets is done using these functions which hide
the internal representation. See cgen-bitset.h (attached) for a
description of the implementation

The patch has 3 parts:
1) Extension of CGEN_ATTR_VALUE_TYPE to be a union allowing the use of
CGEN_BITSET as well as its supporting macros.

2) Changes to CGEN so that it generates code to correctly access and
initialize the new definition of CGEN_ATTR_VALUE_TYPE and to use the
proper cover functions when manipulating ISAs.

3) Changes to hand written portions of existing opcodes, sim and sid
ports as in 2). Fortunately, the use of existing CGEN macros made these
changes minimal.

I have also included in the patch the regenerated source for the frv
port so that you can see the effect on the generated code. None of the
generated code for existing sid ports is affected.

One natural extension of this work would be use the same representation
for all bitset attributes in CGEN. I believe that MACH is the only other
one at this time.

Seeking comments and approval to commit.

Thanks,
Dave
Dave Brolley
2005-09-23 18:27:55 UTC
Permalink
Hi,

These changes have been in our local tree for a few years now and were
developed for an internal port which requires *much* more than 32 ISAs
or even 64 ISAs. These changes could be of benifit to others, so I would
like to submit them for approval.

Currently, ISA is represented as an integer, like all the other
non-boolean attributes. It is a bit mask with each bit representing
whether a particular ISA is supported. Our port requires that there be
no fixed limit on the number of ISAs, and so, we developed an open-ended
representation for a bitset using a bitstring and a length. It is called
CGEN_BITSET and is declared in include/opcode/cgen-bitset.h and
supported by several new functions in opcodes/cgen-opc.c. All
manipulation of these bitsets is done using these functions which hide
the internal representation. See cgen-bitset.h (attached) for a
description of the implementation

The patch has 3 parts:
1) Extension of CGEN_ATTR_VALUE_TYPE to be a union allowing the use of
CGEN_BITSET as well as its supporting macros.

2) Changes to CGEN so that it generates code to correctly access and
initialize the new definition of CGEN_ATTR_VALUE_TYPE and to use the
proper cover functions when manipulating ISAs.

3) Changes to hand written portions of existing opcodes, sim and sid
ports as in 2). Fortunately, the use of existing CGEN macros made these
changes minimal.

I have also included in the patch the regenerated source for the frv
port so that you can see the effect on the generated code. None of the
generated code for existing sid ports is affected.

One natural extension of this work would be use the same representation
for all bitset attributes in CGEN. I believe that MACH is the only other
one at this time.

Seeking comments and approval to commit.

Thanks,
Dave
Daniel Jacobowitz
2005-09-23 19:40:06 UTC
Permalink
Post by Dave Brolley
Hi,
These changes have been in our local tree for a few years now and were
developed for an internal port which requires *much* more than 32 ISAs
or even 64 ISAs. These changes could be of benifit to others, so I would
like to submit them for approval.
MAINTAINERS says:

cgen/; cgen parts of opcodes/, sim/ & include/
cgen: http://sources.redhat.com/cgen/
Patches to ***@sources.redhat.com
May need separate opcodes/ or sim/ approval for
commits of regenerated files there.

Is that list still alive? If so, asking there may be more useful. I
have no idea who can approve patches for cgen, but I think the gdb
maintainers won't.
--
Daniel Jacobowitz
CodeSourcery, LLC
Dave Brolley
2005-09-23 20:24:25 UTC
Permalink
Post by Daniel Jacobowitz
cgen/; cgen parts of opcodes/, sim/ & include/
cgen: http://sources.redhat.com/cgen/
May need separate opcodes/ or sim/ approval for
commits of regenerated files there.
Is that list still alive? If so, asking there may be more useful. I
have no idea who can approve patches for cgen, but I think the gdb
maintainers won't.
Yes, thanks. I have submitted this patch for approval to cgen and sid
separately.

Dave
Ben Elliston
2005-09-26 03:43:25 UTC
Permalink
Post by Daniel Jacobowitz
Post by Dave Brolley
These changes have been in our local tree for a few years now and were
developed for an internal port which requires *much* more than 32 ISAs
or even 64 ISAs. These changes could be of benifit to others, so I would
like to submit them for approval.
Is that list still alive? If so, asking there may be more useful. I
have no idea who can approve patches for cgen, but I think the gdb
maintainers won't.
I see from the CGEN web page (http://sources.redhat.com/cgen) that Dave is still
listed as a maintainer ..

Cheers, Ben
Dave Brolley
2005-10-28 20:35:18 UTC
Permalink
This patch has now been committed.....
Post by Dave Brolley
Hi,
These changes have been in our local tree for a few years now and were
developed for an internal port which requires *much* more than 32 ISAs
or even 64 ISAs. These changes could be of benifit to others, so I
would like to submit them for approval.
Currently, ISA is represented as an integer, like all the other
non-boolean attributes. It is a bit mask with each bit representing
whether a particular ISA is supported. Our port requires that there be
no fixed limit on the number of ISAs, and so, we developed an
open-ended representation for a bitset using a bitstring and a length.
It is called CGEN_BITSET and is declared in
include/opcode/cgen-bitset.h and supported by several new functions in
opcodes/cgen-opc.c. All manipulation of these bitsets is done using
these functions which hide the internal representation. See
cgen-bitset.h (attached) for a description of the implementation
1) Extension of CGEN_ATTR_VALUE_TYPE to be a union allowing the use of
CGEN_BITSET as well as its supporting macros.
2) Changes to CGEN so that it generates code to correctly access and
initialize the new definition of CGEN_ATTR_VALUE_TYPE and to use the
proper cover functions when manipulating ISAs.
3) Changes to hand written portions of existing opcodes, sim and sid
ports as in 2). Fortunately, the use of existing CGEN macros made
these changes minimal.
I have also included in the patch the regenerated source for the frv
port so that you can see the effect on the generated code. None of the
generated code for existing sid ports is affected.
One natural extension of this work would be use the same
representation for all bitset attributes in CGEN. I believe that MACH
is the only other one at this time.
Seeking comments and approval to commit.
Thanks,
Dave
Loading...