[dmantipov@yandex.ru: Re: Your Emacs changes]

[dmantipov@yandex.ru: Re: Your Emacs changes]

[Richard Stallman]

Would someone please install this in the trunk, then ack?

------- Start of forwarded message -------
X-Spam-Status: No, score=0.0 required=5.0 tests=none autolearn=failed 
	version=3.1.0
Date:   Tue, 21 Aug 2007 18:42:20 +0400
From:   Dmitry Antipov <dmantipov@yandex.ru>
MIME-Version: 1.0
To:     rms@gnu.org
CC:     emacs-devel@gnu.org
Subject: Re: Your Emacs changes
In-Reply-To: <E1IM8Gf-0005ZW-Qc@fencepost.gnu.org>
Content-Type: multipart/mixed;
 boundary="------------040802050101020103000601"

This is a multi-part message in MIME format.
- --------------040802050101020103000601
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit

Richard Stallman wrote:

> Now that we have your papers, what changes should we install now?

Probably the thing described at http://lists.gnu.org/archive/html/emacs-devel/2007-07/msg00094.html.
Attached is the latest stuff on this.

Note the same optimization may be performed on sweeping floats.  But, since an amount of allocated
floats is too small in comparison with conses, an effect is expected to be very negligible.

Dmitry

- --------------040802050101020103000601
Content-Type: text/x-patch;
 name="fast_cons_sweep.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="fast_cons_sweep.patch"

Index: alloc.c
===================================================================
RCS file: /sources/emacs/emacs/src/alloc.c,v
retrieving revision 1.414
diff -u -r1.414 alloc.c
- --- alloc.c	19 Aug 2007 00:15:26 -0000	1.414
+++ alloc.c	21 Aug 2007 14:31:37 -0000
@@ -5974,23 +5974,53 @@
 
     for (cblk = cons_block; cblk; cblk = *cprev)
       {
- -	register int i;
+	register int i = 0;
 	int this_free = 0;
- -	for (i = 0; i < lim; i++)
- -	  if (!CONS_MARKED_P (&cblk->conses[i]))
- -	    {
- -	      this_free++;
- -	      cblk->conses[i].u.chain = cons_free_list;
- -	      cons_free_list = &cblk->conses[i];
+
+	while (1)
+	  {
+	    if (cblk->gcmarkbits[i] == -1)
+	      {
+		/* Fast path - everything is marked.  */
+		cblk->gcmarkbits[i++] = 0;
+		num_used += BITS_PER_INT;
+	      }
+	    else
+	      {
+		/* Slow path - scan over each bit, from the beginning
+		   of current word to 'min (word boundary, LIM)'.  */
+		int start, stop;
+
+		start = i * BITS_PER_INT;
+		stop = lim - start;
+		if (stop > BITS_PER_INT)
+		  stop = BITS_PER_INT;
+		stop += start;
+
+		while (start < stop)
+		  {
+		    if (!CONS_MARKED_P (&cblk->conses[start]))
+		      {
+			this_free++;
+			cblk->conses[start].u.chain = cons_free_list;
+			cons_free_list = &cblk->conses[start];
 #if GC_MARK_STACK
- -	      cons_free_list->car = Vdead;
+			cons_free_list->car = Vdead;
 #endif
- -	    }
- -	  else
- -	    {
- -	      num_used++;
- -	      CONS_UNMARK (&cblk->conses[i]);
- -	    }
+		      }
+		    else
+		      {
+			num_used++;
+			CONS_UNMARK (&cblk->conses[start]);
+		      }
+		    start++;
+		  }
+		if (stop < (++i) * BITS_PER_INT)
+		  /* Whole bitmap is scanned or LIM is reached.  */
+		  break;
+	      }
+	  }
+
 	lim = CONS_BLOCK_SIZE;
 	/* If this block contains only free conses and we have already
 	   seen more than two blocks worth of free conses then deallocate

- --------------040802050101020103000601--
------- End of forwarded message -------

[dmantipov@yandex.ru: Re: Your Emacs changes]

[Richard Stallman]

[I sent this message a week ago but did not get a response.]

Would someone please install this in the trunk, then ack?

------- Start of forwarded message -------
X-Spam-Status: No, score=0.0 required=5.0 tests=none autolearn=failed 
	version=3.1.0
Date:   Tue, 21 Aug 2007 18:42:20 +0400
From:   Dmitry Antipov <dmantipov@yandex.ru>
MIME-Version: 1.0
To:     rms@gnu.org
CC:     emacs-devel@gnu.org
Subject: Re: Your Emacs changes
In-Reply-To: <E1IM8Gf-0005ZW-Qc@fencepost.gnu.org>
Content-Type: multipart/mixed;
 boundary="------------040802050101020103000601"

This is a multi-part message in MIME format.
- --------------040802050101020103000601
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit

Richard Stallman wrote:

> Now that we have your papers, what changes should we install now?

Probably the thing described at http://lists.gnu.org/archive/html/emacs-devel/2007-07/msg00094.html.
Attached is the latest stuff on this.

Note the same optimization may be performed on sweeping floats.  But, since an amount of allocated
floats is too small in comparison with conses, an effect is expected to be very negligible.

Dmitry

- --------------040802050101020103000601
Content-Type: text/x-patch;
 name="fast_cons_sweep.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="fast_cons_sweep.patch"

Index: alloc.c
===================================================================
RCS file: /sources/emacs/emacs/src/alloc.c,v
retrieving revision 1.414
diff -u -r1.414 alloc.c
- --- alloc.c	19 Aug 2007 00:15:26 -0000	1.414
+++ alloc.c	21 Aug 2007 14:31:37 -0000
@@ -5974,23 +5974,53 @@
 
     for (cblk = cons_block; cblk; cblk = *cprev)
       {
- -	register int i;
+	register int i = 0;
 	int this_free = 0;
- -	for (i = 0; i < lim; i++)
- -	  if (!CONS_MARKED_P (&cblk->conses[i]))
- -	    {
- -	      this_free++;
- -	      cblk->conses[i].u.chain = cons_free_list;
- -	      cons_free_list = &cblk->conses[i];
+
+	while (1)
+	  {
+	    if (cblk->gcmarkbits[i] == -1)
+	      {
+		/* Fast path - everything is marked.  */
+		cblk->gcmarkbits[i++] = 0;
+		num_used += BITS_PER_INT;
+	      }
+	    else
+	      {
+		/* Slow path - scan over each bit, from the beginning
+		   of current word to 'min (word boundary, LIM)'.  */
+		int start, stop;
+
+		start = i * BITS_PER_INT;
+		stop = lim - start;
+		if (stop > BITS_PER_INT)
+		  stop = BITS_PER_INT;
+		stop += start;
+
+		while (start < stop)
+		  {
+		    if (!CONS_MARKED_P (&cblk->conses[start]))
+		      {
+			this_free++;
+			cblk->conses[start].u.chain = cons_free_list;
+			cons_free_list = &cblk->conses[start];
 #if GC_MARK_STACK
- -	      cons_free_list->car = Vdead;
+			cons_free_list->car = Vdead;
 #endif
- -	    }
- -	  else
- -	    {
- -	      num_used++;
- -	      CONS_UNMARK (&cblk->conses[i]);
- -	    }
+		      }
+		    else
+		      {
+			num_used++;
+			CONS_UNMARK (&cblk->conses[start]);
+		      }
+		    start++;
+		  }
+		if (stop < (++i) * BITS_PER_INT)
+		  /* Whole bitmap is scanned or LIM is reached.  */
+		  break;
+	      }
+	  }
+
 	lim = CONS_BLOCK_SIZE;
 	/* If this block contains only free conses and we have already
 	   seen more than two blocks worth of free conses then deallocate

- --------------040802050101020103000601--
------- End of forwarded message -------

Re: [dmantipov@yandex.ru: Re: Your Emacs changes]

[Glenn Morris]

Please could you provide a ChangeLog entry for this.


> Probably the thing described at
> http://lists.gnu.org/archive/html/emacs-devel/2007-07/msg00094.html.
> Attached is the latest stuff on this.
[...]
> Index: alloc.c
> ===================================================================
> RCS file: /sources/emacs/emacs/src/alloc.c,v
> retrieving revision 1.414
> diff -u -r1.414 alloc.c
> - --- alloc.c	19 Aug 2007 00:15:26 -0000	1.414
> +++ alloc.c	21 Aug 2007 14:31:37 -0000
> @@ -5974,23 +5974,53 @@
>  
>      for (cblk = cons_block; cblk; cblk = *cprev)
>        {
> - -	register int i;
> +	register int i = 0;
>  	int this_free = 0;
> - -	for (i = 0; i < lim; i++)
> - -	  if (!CONS_MARKED_P (&cblk->conses[i]))
> - -	    {
> - -	      this_free++;
> - -	      cblk->conses[i].u.chain = cons_free_list;
> - -	      cons_free_list = &cblk->conses[i];
> +
> +	while (1)
> +	  {
> +	    if (cblk->gcmarkbits[i] == -1)
> +	      {
> +		/* Fast path - everything is marked.  */
> +		cblk->gcmarkbits[i++] = 0;
> +		num_used += BITS_PER_INT;
> +	      }
> +	    else
> +	      {
> +		/* Slow path - scan over each bit, from the beginning
> +		   of current word to 'min (word boundary, LIM)'.  */
> +		int start, stop;
> +
> +		start = i * BITS_PER_INT;
> +		stop = lim - start;
> +		if (stop > BITS_PER_INT)
> +		  stop = BITS_PER_INT;
> +		stop += start;
> +
> +		while (start < stop)
> +		  {
> +		    if (!CONS_MARKED_P (&cblk->conses[start]))
> +		      {
> +			this_free++;
> +			cblk->conses[start].u.chain = cons_free_list;
> +			cons_free_list = &cblk->conses[start];
>  #if GC_MARK_STACK
> - -	      cons_free_list->car = Vdead;
> +			cons_free_list->car = Vdead;
>  #endif
> - -	    }
> - -	  else
> - -	    {
> - -	      num_used++;
> - -	      CONS_UNMARK (&cblk->conses[i]);
> - -	    }
> +		      }
> +		    else
> +		      {
> +			num_used++;
> +			CONS_UNMARK (&cblk->conses[start]);
> +		      }
> +		    start++;
> +		  }
> +		if (stop < (++i) * BITS_PER_INT)
> +		  /* Whole bitmap is scanned or LIM is reached.  */
> +		  break;
> +	      }
> +	  }
> +
>  	lim = CONS_BLOCK_SIZE;
>  	/* If this block contains only free conses and we have already
>  	   seen more than two blocks worth of free conses then deallocate
>
> - --------------040802050101020103000601--

Re: [dmantipov@yandex.ru: Re: Your Emacs changes]

[Glenn Morris]

Glenn Morris wrote:

> Please could you provide a ChangeLog entry for this.

Pretty please? I'm asking so I can install it for you.

Re: [dmantipov@yandex.ru: Re: Your Emacs changes]

[Glenn Morris]

Glenn Morris wrote:

> Please could you provide a ChangeLog entry for this.

Since the original author apparently can't be bothered, can anyone
else write a ChangeLog entry for this? Though I'm not really sure it
should be installed under these grounds...

Re: [dmantipov@yandex.ru: Re: Your Emacs changes]

[Richard Stallman]

    Since the original author apparently can't be bothered, can anyone
    else write a ChangeLog entry for this? Though I'm not really sure it
    should be installed under these grounds...

What is the uncertainty?  We have the legal papers.  If the change is
useful, why not install it?

Re: [dmantipov@yandex.ru: Re: Your Emacs changes]

[Glenn Morris]

Richard Stallman wrote:

> What is the uncertainty? We have the legal papers. If the change is
> useful, why not install it?

I don't understand this code well enough to write a useful ChangeLog
entry, so I can't do it.

Re: [dmantipov@yandex.ru: Re: Your Emacs changes]

[Richard Stallman]

    I don't understand this code well enough to write a useful ChangeLog
    entry, so I can't do it.

I will do it.