powerpc: PPCA2/A2 optimized memcpy function

1 files changed, 501 insertions, 0 deletions

diff --git a/sysdeps/powerpc/powerpc64/a2/memcpy.S b/sysdeps/powerpc/powerpc64/a2/memcpy.S
new file mode 100644
index 0000000000..ac95171aa8
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/a2/memcpy.S
@@ -0,0 +1,501 @@
+/* Optimized memcpy implementation for PowerPC A2.
+   Copyright (C) 2010 Free Software Foundation, Inc.
+   Contributed by Michael Brutman <brutman@us.ibm.com>.
+   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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include <bp-sym.h>
+#include <bp-asm.h>
+
+#define PREFETCH_AHEAD 4        /* no cache lines SRC prefetching ahead  */
+#define ZERO_AHEAD 2            /* no cache lines DST zeroing ahead  */
+
+	.section        ".toc","aw"
+.LC0:
+	.tc __cache_line_size[TC],__cache_line_size
+	.section        ".text"
+	.align 2
+
+
+	.machine  a2
+EALIGN (BP_SYM (memcpy), 5, 0)
+	CALL_MCOUNT 3
+
+	dcbt    0,r4            /* Prefetch ONE SRC cacheline  */
+	cmpldi  cr1,r5,16       /* is size < 16 ?  */
+	mr      r6,r3           /* Copy dest reg to r6; */
+	blt+    cr1,L(shortcopy)
+
+
+	/* Big copy (16 bytes or more)
+
+	   Figure out how far to the nearest quadword boundary, or if we are
+	   on one already.  Also get the cache line size.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	*/
+
+	neg     r8,r3           /* LS 4 bits = # bytes to 8-byte dest bdry  */
+	ld      r9,.LC0@toc(r2) /* Get cache line size (part 1) */
+	clrldi  r8,r8,64-4      /* align to 16byte boundary  */
+	sub     r7,r4,r3        /* compute offset to src from dest */
+	lwz     r9,0(r9)        /* Get cache line size (part 2) */
+	cmpldi  cr0,r8,0        /* Were we aligned on a 16 byte bdy? */
+	addi    r10,r9,-1       /* Cache line mask */
+	beq+    L(dst_aligned)
+
+
+
+	/* Destination is not aligned on quadword boundary.  Get us to one.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	   r7 - offset to src from dest
+	   r8 - number of bytes to quadword boundary
+	*/
+
+	mtcrf   0x01,r8         /* put #bytes to boundary into cr7  */
+	subf    r5,r8,r5        /* adjust remaining len */
+
+	bf      cr7*4+3,1f
+	lbzx    r0,r7,r6        /* copy 1 byte addr */
+	stb     r0,0(r6)
+	addi    r6,r6,1
+1:
+	bf      cr7*4+2,2f
+	lhzx    r0,r7,r6        /* copy 2 byte addr */
+	sth     r0,0(r6)
+	addi    r6,r6,2
+2:
+	bf      cr7*4+1,4f
+	lwzx    r0,r7,r6        /* copy 4 byte addr */
+	stw     r0,0(r6)
+	addi    r6,r6,4
+4:
+	bf      cr7*4+0,8f
+	ldx     r0,r7,r6        /* copy 8 byte addr */
+	std     r0,0(r6)
+	addi    r6,r6,8
+8:
+	add     r4,r7,r6        /* update src addr */
+
+
+
+	/* Dest is quadword aligned now.
+
+	   Lots of decisions to make.  If we are copying less than a cache
+	   line we won't be here long.  If we are not on a cache line
+	   boundary we need to get there.  And then we need to figure out
+	   how many cache lines ahead to pre-touch.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	*/
+
+
+	.align 4
+L(dst_aligned):
+
+
+	cmpd   cr5,r5,r10       /* Less than a cacheline to go? */
+
+	neg     r7,r6           /* How far to next cacheline bdy? */
+
+	addi    r6,r6,-8        /* prepare for stdu  */
+	cmpdi   cr0,r9,128
+	addi    r4,r4,-8        /* prepare for ldu  */
+
+
+	ble+    cr5,L(lessthancacheline)
+
+	beq-    cr0,L(big_lines) /* 128 byte line code */
+
+
+
+	/* More than a cacheline left to go, and using 64 byte cachelines */
+
+	clrldi  r7,r7,64-6      /* How far to next cacheline bdy? */
+
+	cmpldi  cr6,r7,0        /* Are we on a cacheline bdy already? */
+
+	/* Reduce total len by what it takes to get to the next cache line */
+	subf    r5,r7,r5
+	srdi    r7,r7,4         /* How many qws to get to the line bdy? */
+
+	/* How many full cache lines to copy after getting to a line bdy? */
+	srdi    r10,r5,6
+
+	cmpldi  r10,0           /* If no full cache lines to copy ... */
+	li      r11,0           /* number cachelines to copy with prefetch  */
+	beq     L(nocacheprefetch)
+
+
+	/* We are here because we have at least one full cache line to copy,
+	   and therefore some pre-touching to do. */
+
+	cmpldi  r10,PREFETCH_AHEAD
+	li      r12,64+8        /* prefetch distance  */
+	ble     L(lessthanmaxprefetch)
+
+	/* We can only do so much pre-fetching.  R11 will have the count of
+	   lines left to prefetch after the initial batch of prefetches
+	   are executed. */
+
+	subi    r11,r10,PREFETCH_AHEAD
+	li      r10,PREFETCH_AHEAD
+
+L(lessthanmaxprefetch):
+	mtctr   r10
+
+	/* At this point r10/ctr hold the number of lines to prefetch in this
+	   initial batch, and r11 holds any remainder. */
+
+L(prefetchSRC):
+	dcbt    r12,r4
+	addi    r12,r12,64
+	bdnz    L(prefetchSRC)
+
+
+	/* Prefetching is done, or was not needed.
+
+	   cr6 - are we on a cacheline boundary already?
+	   r7  - number of quadwords to the next cacheline boundary
+	*/
+
+L(nocacheprefetch):
+	mtctr   r7
+
+	cmpldi  cr1,r5,64   /* Less than a cache line to copy? */
+
+	/* How many bytes are left after we copy whatever full
+	   cache lines we can get? */
+	clrldi  r5,r5,64-6
+
+	beq     cr6,L(cachelinealigned)
+
+
+	/* Copy quadwords up to the next cacheline boundary */
+
+L(aligntocacheline):
+	ld      r9,0x08(r4)
+	ld      r7,0x10(r4)
+	addi    r4,r4,0x10
+	std     r9,0x08(r6)
+	stdu    r7,0x10(r6)
+	bdnz    L(aligntocacheline)
+
+
+	.align 4
+L(cachelinealigned):            /* copy while cache lines  */
+
+	blt-    cr1,L(lessthancacheline) /* size <64  */
+
+L(outerloop):
+	cmpdi   r11,0
+	mtctr   r11
+	beq-    L(endloop)
+
+	li      r11,64*ZERO_AHEAD +8    /* DCBZ dist  */
+
+	.align  4
+	/* Copy whole cachelines, optimized by prefetching SRC cacheline  */
+L(loop):                        /* Copy aligned body  */
+	dcbt    r12,r4          /* PREFETCH SOURCE some cache lines ahead  */
+	ld      r9, 0x08(r4)
+	dcbz    r11,r6
+	ld      r7, 0x10(r4)
+	ld      r8, 0x18(r4)
+	ld      r0, 0x20(r4)
+	std     r9, 0x08(r6)
+	std     r7, 0x10(r6)
+	std     r8, 0x18(r6)
+	std     r0, 0x20(r6)
+	ld      r9, 0x28(r4)
+	ld      r7, 0x30(r4)
+	ld      r8, 0x38(r4)
+	ld      r0, 0x40(r4)
+	addi    r4, r4,0x40
+	std     r9, 0x28(r6)
+	std     r7, 0x30(r6)
+	std     r8, 0x38(r6)
+	stdu    r0, 0x40(r6)
+
+	bdnz    L(loop)
+
+
+L(endloop):
+	cmpdi   r10,0
+	beq-    L(endloop2)
+	mtctr   r10
+
+L(loop2):                       /* Copy aligned body  */
+	ld      r9, 0x08(r4)
+	ld      r7, 0x10(r4)
+	ld      r8, 0x18(r4)
+	ld      r0, 0x20(r4)
+	std     r9, 0x08(r6)
+	std     r7, 0x10(r6)
+	std     r8, 0x18(r6)
+	std     r0, 0x20(r6)
+	ld      r9, 0x28(r4)
+	ld      r7, 0x30(r4)
+	ld      r8, 0x38(r4)
+	ld      r0, 0x40(r4)
+	addi    r4, r4,0x40
+	std     r9, 0x28(r6)
+	std     r7, 0x30(r6)
+	std     r8, 0x38(r6)
+	stdu    r0, 0x40(r6)
+
+	bdnz    L(loop2)
+L(endloop2):
+
+
+	.align 4
+L(lessthancacheline):           /* Was there less than cache to do ?  */
+	cmpldi  cr0,r5,16
+	srdi    r7,r5,4         /* divide size by 16  */
+	blt-    L(do_lt16)
+	mtctr   r7
+
+L(copy_remaining):
+	ld      r8,0x08(r4)
+	ld      r7,0x10(r4)
+	addi    r4,r4,0x10
+	std     r8,0x08(r6)
+	stdu    r7,0x10(r6)
+	bdnz    L(copy_remaining)
+
+L(do_lt16):                     /* less than 16 ?  */
+	cmpldi  cr0,r5,0        /* copy remaining bytes (0-15)  */
+	beqlr+                  /* no rest to copy  */
+	addi    r4,r4,8
+	addi    r6,r6,8
+
+L(shortcopy):                   /* SIMPLE COPY to handle size =< 15 bytes  */
+	mtcrf   0x01,r5
+	sub     r7,r4,r6
+	bf-     cr7*4+0,8f
+	ldx     r0,r7,r6        /* copy 8 byte  */
+	std     r0,0(r6)
+	addi    r6,r6,8
+8:
+	bf      cr7*4+1,4f
+	lwzx    r0,r7,r6        /* copy 4 byte  */
+	stw     r0,0(r6)
+	addi    r6,r6,4
+4:
+	bf      cr7*4+2,2f
+	lhzx    r0,r7,r6        /* copy 2 byte  */
+	sth     r0,0(r6)
+	addi    r6,r6,2
+2:
+	bf      cr7*4+3,1f
+	lbzx    r0,r7,r6        /* copy 1 byte  */
+	stb     r0,0(r6)
+1:
+	blr
+
+
+
+
+
+	/* Similar to above, but for use with 128 byte lines. */
+
+
+L(big_lines):
+
+	clrldi  r7,r7,64-7      /* How far to next cacheline bdy? */
+
+	cmpldi  cr6,r7,0        /* Are we on a cacheline bdy already? */
+
+	/* Reduce total len by what it takes to get to the next cache line */
+	subf    r5,r7,r5
+	srdi    r7,r7,4         /* How many qws to get to the line bdy? */
+
+	/* How many full cache lines to copy after getting to a line bdy? */
+	srdi    r10,r5,7
+
+	cmpldi  r10,0           /* If no full cache lines to copy ... */
+	li      r11,0           /* number cachelines to copy with prefetch  */
+	beq     L(nocacheprefetch_128)
+
+
+	/* We are here because we have at least one full cache line to copy,
+	   and therefore some pre-touching to do. */
+
+	cmpldi  r10,PREFETCH_AHEAD
+	li      r12,128+8       /* prefetch distance  */
+	ble     L(lessthanmaxprefetch_128)
+
+	/* We can only do so much pre-fetching.  R11 will have the count of
+	   lines left to prefetch after the initial batch of prefetches
+	   are executed. */
+
+	subi    r11,r10,PREFETCH_AHEAD
+	li      r10,PREFETCH_AHEAD
+
+L(lessthanmaxprefetch_128):
+	mtctr   r10
+
+	/* At this point r10/ctr hold the number of lines to prefetch in this
+	   initial batch, and r11 holds any remainder. */
+
+L(prefetchSRC_128):
+	dcbt    r12,r4
+	addi    r12,r12,128
+	bdnz    L(prefetchSRC_128)
+
+
+	/* Prefetching is done, or was not needed.
+
+	   cr6 - are we on a cacheline boundary already?
+	   r7  - number of quadwords to the next cacheline boundary
+	*/
+
+L(nocacheprefetch_128):
+	mtctr   r7
+
+	cmpldi  cr1,r5,128  /* Less than a cache line to copy? */
+
+	/* How many bytes are left after we copy whatever full
+	   cache lines we can get? */
+	clrldi  r5,r5,64-7
+
+	beq     cr6,L(cachelinealigned_128)
+
+
+	/* Copy quadwords up to the next cacheline boundary */
+
+L(aligntocacheline_128):
+	ld      r9,0x08(r4)
+	ld      r7,0x10(r4)
+	addi    r4,r4,0x10
+	std     r9,0x08(r6)
+	stdu    r7,0x10(r6)
+	bdnz    L(aligntocacheline_128)
+
+
+L(cachelinealigned_128):        /* copy while cache lines  */
+
+	blt-    cr1,L(lessthancacheline) /* size <128  */
+
+L(outerloop_128):
+	cmpdi   r11,0
+	mtctr   r11
+	beq-    L(endloop_128)
+
+	li      r11,128*ZERO_AHEAD +8    /* DCBZ dist  */
+
+	.align  4
+	/* Copy whole cachelines, optimized by prefetching SRC cacheline  */
+L(loop_128):                    /* Copy aligned body  */
+	dcbt    r12,r4          /* PREFETCH SOURCE some cache lines ahead  */
+	ld      r9, 0x08(r4)
+	dcbz    r11,r6
+	ld      r7, 0x10(r4)
+	ld      r8, 0x18(r4)
+	ld      r0, 0x20(r4)
+	std     r9, 0x08(r6)
+	std     r7, 0x10(r6)
+	std     r8, 0x18(r6)
+	std     r0, 0x20(r6)
+	ld      r9, 0x28(r4)
+	ld      r7, 0x30(r4)
+	ld      r8, 0x38(r4)
+	ld      r0, 0x40(r4)
+	std     r9, 0x28(r6)
+	std     r7, 0x30(r6)
+	std     r8, 0x38(r6)
+	std     r0, 0x40(r6)
+	ld      r9, 0x48(r4)
+	ld      r7, 0x50(r4)
+	ld      r8, 0x58(r4)
+	ld      r0, 0x60(r4)
+	std     r9, 0x48(r6)
+	std     r7, 0x50(r6)
+	std     r8, 0x58(r6)
+	std     r0, 0x60(r6)
+	ld      r9, 0x68(r4)
+	ld      r7, 0x70(r4)
+	ld      r8, 0x78(r4)
+	ld      r0, 0x80(r4)
+	addi    r4, r4,0x80
+	std     r9, 0x68(r6)
+	std     r7, 0x70(r6)
+	std     r8, 0x78(r6)
+	stdu    r0, 0x80(r6)
+
+	bdnz    L(loop_128)
+
+
+L(endloop_128):
+	cmpdi   r10,0
+	beq-    L(endloop2_128)
+	mtctr   r10
+
+L(loop2_128):                       /* Copy aligned body  */
+	ld      r9, 0x08(r4)
+	ld      r7, 0x10(r4)
+	ld      r8, 0x18(r4)
+	ld      r0, 0x20(r4)
+	std     r9, 0x08(r6)
+	std     r7, 0x10(r6)
+	std     r8, 0x18(r6)
+	std     r0, 0x20(r6)
+	ld      r9, 0x28(r4)
+	ld      r7, 0x30(r4)
+	ld      r8, 0x38(r4)
+	ld      r0, 0x40(r4)
+	std     r9, 0x28(r6)
+	std     r7, 0x30(r6)
+	std     r8, 0x38(r6)
+	std     r0, 0x40(r6)
+	ld      r9, 0x48(r4)
+	ld      r7, 0x50(r4)
+	ld      r8, 0x58(r4)
+	ld      r0, 0x60(r4)
+	std     r9, 0x48(r6)
+	std     r7, 0x50(r6)
+	std     r8, 0x58(r6)
+	std     r0, 0x60(r6)
+	ld      r9, 0x68(r4)
+	ld      r7, 0x70(r4)
+	ld      r8, 0x78(r4)
+	ld      r0, 0x80(r4)
+	addi    r4, r4,0x80
+	std     r9, 0x68(r6)
+	std     r7, 0x70(r6)
+	std     r8, 0x78(r6)
+	stdu    r0, 0x80(r6)
+
+	bdnz    L(loop2_128)
+L(endloop2_128):
+
+	b       L(lessthancacheline)
+
+
+END_GEN_TB (BP_SYM (memcpy),TB_TOCLESS)
+libc_hidden_builtin_def (memcpy)