From a035261089403de259e74ce4dd196e2715138ed2 Mon Sep 17 00:00:00 2001
From: Thomas Martitz <kugel@rockbox.org>
Date: Sat, 7 Jan 2012 19:56:09 +0100
Subject: Move optimized memcpy and friends and strlen to firmware/asm, using
 the new automatic-asm-picking infrastructure.

---
 firmware/asm/m68k/memset.S | 152 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 152 insertions(+)
 create mode 100644 firmware/asm/m68k/memset.S

(limited to 'firmware/asm/m68k/memset.S')

diff --git a/firmware/asm/m68k/memset.S b/firmware/asm/m68k/memset.S
new file mode 100644
index 0000000000..839b305a05
--- /dev/null
+++ b/firmware/asm/m68k/memset.S
@@ -0,0 +1,152 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2004 by Jens Arnold
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
+ * KIND, either express or implied.
+ *
+ ****************************************************************************/
+#include "config.h"
+
+    .section    .icode,"ax",@progbits
+
+    .align      2
+    .global     memset
+    .type       memset,@function
+
+/* Fills a memory region with specified byte value
+ * This version is optimized for speed
+ *
+ * arguments:
+ *  (4,%sp)  - start address
+ *  (8,%sp)  - data
+ *  (12,%sp) - length
+ *
+ * return value:
+ *  %d0 - start address (like ANSI version)
+ *
+ * register usage:
+ *  %d0 - data (spread to all 4 bytes when using long stores)
+ *  %d1 - temporary / data (for burst transfer)
+ *  %d2 - data (for burst transfer)
+ *  %d3 - data (for burst transfer)
+ *  %a0 - start address
+ *  %a1 - current address (runs down from end to start)
+ *
+ * For maximum speed this routine uses both long stores and burst mode,
+ * storing whole lines with movem.l. The routine fills memory from end
+ * to start in order to ease returning the start address.
+ */
+memset:
+    move.l  (4,%sp),%a0     /* start address */
+    move.l  (8,%sp),%d0     /* data */
+    move.l  (12,%sp),%a1    /* length */
+    add.l   %a0,%a1         /* %a1 = end address */
+
+    move.l  %a0,%d1
+    addq.l  #7,%d1
+    and.l   #0xFFFFFFFC,%d1 /* %d1 = first long bound + 4 */
+    cmp.l   %d1,%a1         /* at least one aligned longword to fill? */
+    blo.b   .no_longs       /* no, jump directly to byte loop */
+
+    and.l   #0xFF,%d0       /* start: spread data to all 4 bytes */
+    move.l  %d0,%d1
+    lsl.l   #8,%d1
+    or.l    %d1,%d0         /* data now in 2 lower bytes of %d0 */
+    move.l  %d0,%d1
+    swap    %d0
+    or.l    %d1,%d0         /* data now in all 4 bytes of %d0 */
+
+    move.l  %a1,%d1
+    and.l   #0xFFFFFFFC,%d1 /* %d1 = last long bound */
+    cmp.l   %d1,%a1         /* any bytes to set? */
+    bls.b   .end_b1         /* no: skip byte loop */
+
+    /* leading byte loop: sets 0..3 bytes */
+.loop_b1:
+    move.b  %d0,-(%a1)      /* store byte */
+    cmp.l   %d1,%a1         /* runs %a1 down to last long bound */
+    bhi.b   .loop_b1
+    
+.end_b1:
+    moveq.l #31,%d1
+    add.l   %a0,%d1
+    and.l   #0xFFFFFFF0,%d1 /* %d1 = first line bound + 16 */
+    cmp.l   %d1,%a1         /* at least one full line to fill? */
+    blo.b   .no_lines       /* no, jump to longword loop */
+
+    mov.l   %a1,%d1
+    and.l   #0xFFFFFFF0,%d1 /* %d1 = last line bound */
+    cmp.l   %d1,%a1         /* any longwords to set? */
+    bls.b   .end_l1         /* no: skip longword loop */
+
+    /* leading longword loop: sets 0..3 longwords */
+.loop_l1:
+    move.l  %d0,-(%a1)      /* store longword */
+    cmp.l   %d1,%a1         /* runs %a1 down to last line bound */
+    bhi.b   .loop_l1
+
+.end_l1:
+    move.l  %d2,-(%sp)      /* free some registers */
+    move.l  %d3,-(%sp)
+
+    move.l  %d0,%d1         /* spread data to 4 data registers */
+    move.l  %d0,%d2
+    move.l  %d0,%d3
+    lea.l   (15,%a0),%a0    /* start address += 15, acct. for trl. data */
+    
+    /* main loop: set whole lines utilising burst mode */
+.loop_line:
+    lea.l   (-16,%a1),%a1   /* pre-decrement */
+    movem.l %d0-%d3,(%a1)   /* store line */
+    cmp.l   %a0,%a1         /* runs %a1 down to first line bound */
+    bhi.b   .loop_line
+
+    lea.l   (-15,%a0),%a0   /* correct start address */
+    move.l  (%sp)+,%d3      /* restore registers */
+    move.l  (%sp)+,%d2
+
+    move.l  %a0,%d1         /* %d1 = start address ... */
+    addq.l  #3,%d1          /* ... +3, account for possible trailing bytes */
+    cmp.l   %d1,%a1         /* any longwords left */
+    bhi.b   .loop_l2        /* yes: jump to longword loop */
+    bra.b   .no_longs       /* no: skip loop */
+
+.no_lines:
+    move.l  %a0,%d1         /* %d1 = start address ... */
+    addq.l  #3,%d1          /* ... +3, account for possible trailing bytes */
+
+    /* trailing longword loop */
+.loop_l2:
+    move.l  %d0,-(%a1)      /* store longword */
+    cmp.l   %d1,%a1         /* runs %a1 down to first long bound */
+    bhi.b   .loop_l2
+
+.no_longs:
+    cmp.l   %a0,%a1         /* any bytes left? */
+    bls.b   .end_b2         /* no: skip loop */
+
+    /* trailing byte loop */
+.loop_b2:
+    move.b  %d0,-(%a1)      /* store byte */
+    cmp.l   %a0,%a1         /* runs %a1 down to start address */
+    bhi.b   .loop_b2
+
+.end_b2:
+    move.l  %a0,%d0         /* return start address */
+    rts
+
+.end:
+    .size   memset,.end-memset
-- 
cgit v1.2.3