/* Optimized memset implementation for PowerPC64.

   Copyright (C) 1997-2018 Free Software Foundation, Inc.

   This file is part of the GNU C Library.

   The GNU C Library is free software; you can redistribute it and/or

   modify it under the terms of the GNU Lesser General Public

   License as published by the Free Software Foundation; either

   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,

   but WITHOUT ANY WARRANTY; without even the implied warranty of

   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public

   License along with the GNU C Library; if not, see

   <http://www.gnu.org/licenses/>.  */

#include <sysdep.h>

/* void * [r3] memset (void *s [r3], int c [r4], size_t n [r5]));

   Returns 's'.

   The memset is done in three sizes: byte (8 bits), word (32 bits),

   cache line (1024 bits). There is a special case for setting cache lines

   to 0, to take advantage of the dcbz instruction.  */

	.machine power4

EALIGN (memset, 5, 0)

	CALL_MCOUNT

#define rTMP	r0

#define rRTN	r3	/* Initial value of 1st argument.  */

#define rMEMP0	r3	/* Original value of 1st arg.  */

#define rCHR	r4	/* Char to set in each byte.  */

#define rLEN	r5	/* Length of region to set.  */

#define rMEMP	r6	/* Address at which we are storing.  */

#define rALIGN	r7	/* Number of bytes we are setting now (when aligning). */

#define rMEMP2	r8

#define rNEG64	r8	/* Constant -64 for clearing with dcbz.  */

#define rCLS	r8	/* Cache line size (known to be 128).  */

#define rCLM	r9	/* Cache line size mask to check for cache alignment.  */

L(_memset):

/* Take care of case for size <= 4.  */

	cmplwi	cr1, rLEN, 4

	andi.	rALIGN, rMEMP0, 3

	mr	rMEMP, rMEMP0

	ble-	cr1, L(small)

/* Align to word boundary.  */

	cmplwi	cr5, rLEN, 31

	insrwi	rCHR, rCHR, 8, 16     /* Replicate byte to halfword.  */

	beq+	L(aligned)

	mtcrf	0x01, rMEMP0

	subfic	rALIGN, rALIGN, 4

	add	rMEMP, rMEMP, rALIGN

	sub	rLEN, rLEN, rALIGN

	bf+	31, L(g0)

	stb	rCHR, 0(rMEMP0)

	bt	30, L(aligned)

L(g0):

	sth	rCHR, -2(rMEMP)

/* Handle the case of size < 31.  */

L(aligned):

	mtcrf	0x01, rLEN

	insrwi	rCHR, rCHR, 16, 0    /* Replicate halfword to word.  */

	ble	cr5, L(medium)

/* Align to 32-byte boundary.  */

	andi.	rALIGN, rMEMP, 0x1C

	subfic	rALIGN, rALIGN, 0x20

	beq	L(caligned)

	mtcrf	0x01, rALIGN

	add	rMEMP, rMEMP, rALIGN

	sub	rLEN, rLEN, rALIGN

	cmplwi	cr1, rALIGN, 0x10

	mr	rMEMP2, rMEMP

	bf	28, L(a1)

        stw     rCHR, -4(rMEMP2)

	stwu	rCHR, -8(rMEMP2)

L(a1):	blt	cr1, L(a2)

        stw     rCHR, -4(rMEMP2)

	stw	rCHR, -8(rMEMP2)

	stw	rCHR, -12(rMEMP2)

	stwu	rCHR, -16(rMEMP2)

L(a2):  bf      29, L(caligned)

        stw     rCHR, -4(rMEMP2)

/* Now aligned to a 32 byte boundary.  */

L(caligned):

	cmplwi	cr1, rCHR, 0

	clrrwi.	rALIGN, rLEN, 5

	mtcrf	0x01, rLEN

	beq	cr1, L(zloopstart) /* Special case for clearing memory using dcbz.  */

L(nondcbz):

	srwi	rTMP, rALIGN, 5

	mtctr	rTMP

	beq	L(medium)	/* We may not actually get to do a full line.  */

	clrlwi.	rLEN, rLEN, 27

	add	rMEMP, rMEMP, rALIGN

	li	rNEG64, -0x40

	bdz	L(cloopdone)

        .align 4

L(c3): 	dcbtst	rNEG64, rMEMP

        stw     rCHR, -4(rMEMP)

	stw	rCHR, -8(rMEMP)

        stw     rCHR, -12(rMEMP)

	stw	rCHR, -16(rMEMP)

        stw     rCHR, -20(rMEMP)

	stw	rCHR, -24(rMEMP)

        stw     rCHR, -28(rMEMP)

	stwu	rCHR, -32(rMEMP)

	bdnz	L(c3)

L(cloopdone):

        stw     rCHR, -4(rMEMP)

	stw	rCHR, -8(rMEMP)

        stw     rCHR, -12(rMEMP)

	stw	rCHR, -16(rMEMP)

	cmplwi	cr1, rLEN, 16

        stw     rCHR, -20(rMEMP)

	stw	rCHR, -24(rMEMP)

        stw     rCHR, -28(rMEMP)

	stwu	rCHR, -32(rMEMP)

	beqlr

	add	rMEMP, rMEMP, rALIGN

	b	L(medium_tail2)

	.align 5

/* Clear lines of memory in 128-byte chunks.  */

L(zloopstart):

/* If the remaining length is less the 32 bytes, don't bother getting

	 the cache line size.  */

	beq	L(medium)

	li      rCLS,128  /* cache line size is 128 */

	dcbt	0,rMEMP

L(getCacheAligned):

	cmplwi	cr1,rLEN,32

	andi.	rTMP,rMEMP,127

	blt	cr1,L(handletail32)

	beq	L(cacheAligned)

	addi	rMEMP,rMEMP,32

	addi	rLEN,rLEN,-32

	stw	rCHR,-32(rMEMP)

        stw     rCHR,-28(rMEMP)

	stw	rCHR,-24(rMEMP)

	stw     rCHR,-20(rMEMP)

	stw	rCHR,-16(rMEMP)

        stw     rCHR,-12(rMEMP)

	stw	rCHR,-8(rMEMP)

        stw     rCHR,-4(rMEMP)

	b	L(getCacheAligned)

/* Now we are aligned to the cache line and can use dcbz.  */

        .align 4

L(cacheAligned):

	cmplw	cr1,rLEN,rCLS

	blt	cr1,L(handletail32)

	dcbz	0,rMEMP

	subf	rLEN,rCLS,rLEN

	add	rMEMP,rMEMP,rCLS

	b	L(cacheAligned)

/* We are here because the cache line size was set and the remainder

  (rLEN) is less than the actual cache line size.

   So set up the preconditions for L(nondcbz) and go there.  */

L(handletail32):

	clrrwi.	rALIGN, rLEN, 5

	b		L(nondcbz)

	.align 5

L(small):

/* Memset of 4 bytes or less.  */

	cmplwi	cr5, rLEN, 1

	cmplwi	cr1, rLEN, 3

	bltlr	cr5

	stb	rCHR, 0(rMEMP)

	beqlr	cr5

	stb	rCHR, 1(rMEMP)

	bltlr	cr1

	stb	rCHR, 2(rMEMP)

	beqlr	cr1

	stb	rCHR, 3(rMEMP)

	blr

/* Memset of 0-31 bytes.  */

	.align 5

L(medium):

	cmplwi	cr1, rLEN, 16

L(medium_tail2):

	add	rMEMP, rMEMP, rLEN

L(medium_tail):

	bt-	31, L(medium_31t)

	bt-	30, L(medium_30t)

L(medium_30f):

	bt-	29, L(medium_29t)

L(medium_29f):

	bge-	cr1, L(medium_27t)

	bflr-	28

        stw     rCHR, -4(rMEMP)

	stw	rCHR, -8(rMEMP)

	blr

L(medium_31t):

	stbu	rCHR, -1(rMEMP)

	bf-	30, L(medium_30f)

L(medium_30t):

	sthu	rCHR, -2(rMEMP)

	bf-	29, L(medium_29f)

L(medium_29t):

	stwu	rCHR, -4(rMEMP)

	blt-	cr1, L(medium_27f)

L(medium_27t):

        stw     rCHR, -4(rMEMP)

	stw	rCHR, -8(rMEMP)

        stw     rCHR, -12(rMEMP)

	stwu	rCHR, -16(rMEMP)

L(medium_27f):

	bflr-	28

L(medium_28t):

        stw     rCHR, -4(rMEMP)

	stw	rCHR, -8(rMEMP)

	blr

END (memset)

libc_hidden_builtin_def (memset)