[SLOF] [PATCH 4/4] fbuffer: Implement RFILL as an accelerated primitive

[Nikunj A Dadhania]

Thomas Huth <thuth at redhat.com> writes:

> By implementing RFILL as an primitive, we can get a huge
> speed-up of the screen erasing function. On board-js2x,
> it writes the pattern directly into the IO region, and on
> board-qemu it uses the KVMPPC_H_LOGICAL_MEMOP hypercall to
> copy the pattern from a temporary buffer into the IO region.
>
> Signed-off-by: Thomas Huth <thuth at redhat.com>

Reviewed-by: Nikunj A Dadhania <nikunj at linux.vnet.ibm.com>

> ---
>  include/ppc970/cache.h | 22 ++++++++++++++++++++++
>  include/ppcp7/cache.h  | 12 ++++++++++++
>  slof/fs/base.fs        |  2 --
>  slof/fs/rmove.fs       | 37 -------------------------------------
>  slof/prim.code         |  7 +++++++
>  slof/prim.in           |  1 +
>  6 files changed, 42 insertions(+), 39 deletions(-)
>  delete mode 100644 slof/fs/rmove.fs
>
> diff --git a/include/ppc970/cache.h b/include/ppc970/cache.h
> index 3d27f92..5592faf 100644
> --- a/include/ppc970/cache.h
> +++ b/include/ppc970/cache.h
> @@ -101,4 +101,26 @@ cache_inhibited_access(uint64_t, 64)
>  		default:		_MRMOVE(s, d, size, type_c); break; \
>  	}
>
> +/* fill IO memory with pattern */
> +#define _RFILL(dst, size, pat, t)	\
> +{ \
> +	t *d1 = (t *)dst; \
> +	register t tmp = 0; \
> +	int i = sizeof(t); \
> +	while (i > 0) { tmp <<= 8; tmp |= pat & 0xff; } \
> +	SET_CI; \
> +	while (size > 0) { \
> +		*d1++ = tmp; size -= sizeof(t); \
> +	} \
> +	CLR_CI; \
> +}
> +
> +#define _FASTRFILL(dst, size, pat) \
> +	switch (((type_u)dst | size) & (sizeof(type_u)-1)) { \
> +		case 0:			_RFILL(dst, size, pat, type_u); break; \
> +		case sizeof(type_l):	_RFILL(dst, size, pat, type_l); break; \
> +		case sizeof(type_w):	_RFILL(dst, size, pat, type_w); break; \
> +		default:		_RFILL(dst, size, pat, type_c); break; \
> +	}
> +
>  #endif
> diff --git a/include/ppcp7/cache.h b/include/ppcp7/cache.h
> index 69a69c3..8d4bcb7 100644
> --- a/include/ppcp7/cache.h
> +++ b/include/ppcp7/cache.h
> @@ -124,6 +124,18 @@ static inline void ci_rmove(void *dst, void *src, unsigned long esize,
>
>  #define _FASTMRMOVE(s, d, size) _FASTRMOVE(s, d, size)
>
> +#define _FASTRFILL(dst, size, pat) do {					   \
> +		type_u buf[64];						   \
> +		char *d = (char *)dst;					   \
> +		memset(buf, pat, size < sizeof(buf) ? size : sizeof(buf)); \
> +		while (size > sizeof(buf)) {				   \
> +			_FASTMRMOVE(buf, d, sizeof(buf));		   \
> +			d += sizeof(buf);				   \
> +			size -= sizeof(buf);				   \
> +		}							   \
> +		_FASTMRMOVE(buf, d, size);				   \
> +	} while(0)
> +
>  static inline uint16_t bswap16_load(uint64_t addr)
>  {
>  	unsigned int val;
> diff --git a/slof/fs/base.fs b/slof/fs/base.fs
> index e71e087..03e77e5 100644
> --- a/slof/fs/base.fs
> +++ b/slof/fs/base.fs
> @@ -579,8 +579,6 @@ defer cursor-off ( -- )
>  #include "debug.fs"
>  \ provide 7.5.3.1 Dictionary search
>  #include "dictionary.fs"
> -\ block data access for IO devices - ought to be implemented in engine
> -#include "rmove.fs"
>  \ provide a simple run time preprocessor
>  #include <preprocessor.fs>
>
> diff --git a/slof/fs/rmove.fs b/slof/fs/rmove.fs
> deleted file mode 100644
> index 3776ee2..0000000
> --- a/slof/fs/rmove.fs
> +++ /dev/null
> @@ -1,37 +0,0 @@
> -\ *****************************************************************************
> -\ * Copyright (c) 2004, 2008 IBM Corporation
> -\ * All rights reserved.
> -\ * This program and the accompanying materials
> -\ * are made available under the terms of the BSD License
> -\ * which accompanies this distribution, and is available at
> -\ * http://www.opensource.org/licenses/bsd-license.php
> -\ *
> -\ * Contributors:
> -\ *     IBM Corporation - initial implementation
> -\ ****************************************************************************/
> -
> -defer '(r@)
> -defer '(r!)
> -1 VALUE /(r)
> -
> -
> -\ The rest of the code already implemented in prim.in
> -\ In the end all of this should be moved over there and this file terminated
> -
> -: (rfill) ( addr size pattern 'r! /r -- )
> -	to /(r) to '(r!) ff and
> -	dup 8 lshift or dup 10 lshift or dup 20 lshift or
> -	-rot bounds ?do dup i '(r!) /(r) +loop drop
> -;
> -
> -: rfill ( addr size pattern -- )
> -	3dup drop or 7 AND CASE
> -		0 OF ['] rx! /x ENDOF
> -		4 OF ['] rl! /l ENDOF
> -		2 OF ['] rw! /w ENDOF
> -		dup OF ['] rb! /c ENDOF
> -	ENDCASE (rfill)
> -;
> -
> -
> -
> diff --git a/slof/prim.code b/slof/prim.code
> index 77e550d..4531d3e 100644
> --- a/slof/prim.code
> +++ b/slof/prim.code
> @@ -527,6 +527,13 @@ PRIM(MRMOVE)
>  	_FASTMRMOVE(s, d, size);
>  	MIRP
>
> +PRIM(RFILL)
> +	type_u pat = ((dp--)->u);
> +	type_u size = ((dp--)->u);
> +	type_u *dst = (type_u *)((dp--)->u);
> +	_FASTRFILL(dst, size, pat);
> +	MIRP
> +
>  // String compare, case insensitive:
>  // : string=ci  ( str1 len1 str2 len2 -- equal? )
>  PRIM(STRING_X3d_CI)
> diff --git a/slof/prim.in b/slof/prim.in
> index f323aed..855f592 100644
> --- a/slof/prim.in
> +++ b/slof/prim.in
> @@ -106,6 +106,7 @@ cod(EXECUTE)
>  cod(MOVE)
>  cod(RMOVE)
>  cod(MRMOVE)
> +cod(RFILL)
>  cod(ZCOUNT)
>  con(HASH-SIZE HASHSIZE)
>  cod(HASH)
> -- 
> 1.8.3.1