powerpc: Optimized memcmp for power10

This patch was based on the __memcmp_power8 and the recent
__strlen_power10.

Improvements from __memcmp_power8:

1. Don't need alignment code.

   On POWER10 lxvp and lxvl do not generate alignment interrupts, so
they are safe for use on caching-inhibited memory.  Notice that the
comparison on the main loop will wait for both VSR to be ready.
Therefore aligning one of the input address does not improve
performance.  In order to align both registers a vperm is necessary
which add too much overhead.

2. Uses new POWER10 instructions

   This code uses lxvp to decrease contention on load by loading 32 bytes
per instruction.
   The vextractbm is used to have a smaller tail code for calculating the
return value.

3. Performance improvement

   This version has around 35% better performance on average. I saw no
performance regressions for any length or alignment.

Thanks Matheus for helping me out with some details.

Co-authored-by: Matheus Castanho <msc@linux.ibm.com>
Reviewed-by: Raphael M Zinsly <rzinsly@linux.ibm.com>

1 files changed, 179 insertions, 0 deletions

diff --git a/sysdeps/powerpc/powerpc64/le/power10/memcmp.S b/sysdeps/powerpc/powerpc64/le/power10/memcmp.S
new file mode 100644
index 0000000000..52f244e7e7
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/le/power10/memcmp.S
@@ -0,0 +1,179 @@
+/* Optimized memcmp implementation for POWER10.
+   Copyright (C) 2021 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
+   <https://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+
+/* TODO: Replace macros by the actual instructions when minimum binutils becomes
+   >= 2.35.  This is used to keep compatibility with older versions.  */
+#define VEXTRACTBM(rt,vrb)	 \
+	.long(((4)<<(32-6))	 \
+	      | ((rt)<<(32-11))  \
+	      | ((8)<<(32-16))	 \
+	      | ((vrb)<<(32-21)) \
+	      | 1602)
+
+#define LXVP(xtp,dq,ra)			   \
+	.long(((6)<<(32-6))		   \
+	      | ((((xtp)-32)>>1)<<(32-10)) \
+	      | ((1)<<(32-11))		   \
+	      | ((ra)<<(32-16))		   \
+	      | dq)
+
+/* Compare 32 bytes.  */
+#define COMPARE_32(vr1,vr2,offset,tail_1,tail_2)\
+	LXVP(32+vr1,offset,r3);		\
+	LXVP(32+vr2,offset,r4);		\
+	vcmpneb.  v5,vr1+1,vr2+1;	\
+	bne	  cr6,L(tail_2);	\
+	vcmpneb.  v4,vr1,vr2;		\
+	bne	  cr6,L(tail_1);	\
+
+#define TAIL(v_res,s1,s2)	\
+	vctzlsbb  r7,v_res;	\
+	vextubrx  r8,r7,s1;	\
+	vextubrx  r9,r7,s2;	\
+	subf	  r3,r9,r8;	\
+	blr;			\
+
+/* int [r3] memcmp (const char *s1 [r3], const char *s2 [r4],
+					size_t size [r5])  */
+
+#ifndef MEMCMP
+# define MEMCMP memcmp
+#endif
+	.machine  power9
+ENTRY_TOCLESS (MEMCMP, 4)
+	CALL_MCOUNT 3
+
+	cmpldi	cr6,r5,64
+	bgt	cr6,L(loop_head)
+
+/* Compare 64 bytes. This section is used for lengths <= 64 and for the last
+   bytes for larger lengths.  */
+L(last_compare):
+	li	r8,16
+
+	sldi	r9,r5,56
+	sldi	r8,r8,56
+	addi	r6,r3,16
+	addi	r7,r4,16
+
+	/* Align up to 16 bytes.  */
+	lxvl	32+v0,r3,r9
+	lxvl	32+v2,r4,r9
+
+	/* The sub. and vcmpneb. results are concatenated by the crnand in order
+	   to do a single branch. It's doing a NOT(CR0.GT AND CR6.EQ) then
+	   loading to CR0.LT.  That means r9 is not bigger than 0 and v4 is not
+	   all equal to 0.  */
+	sub.	  r9,r9,r8
+	vcmpneb.  v4,v0,v2
+	crnand	  4*cr0+lt,4*cr0+gt,4*cr6+eq
+	bt	  4*cr0+lt,L(tail1)
+
+	addi	  r3,r3,32
+	addi	  r4,r4,32
+
+	lxvl	  32+v1,r6,r9
+	lxvl	  32+v3,r7,r9
+	sub.	  r9,r9,r8
+	vcmpneb.  v5,v1,v3
+	crnand	  4*cr0+lt,4*cr0+gt,4*cr6+eq
+	bt	  4*cr0+lt,L(tail2)
+
+	addi	  r6,r3,16
+	addi	  r7,r4,16
+
+	lxvl	  32+v6,r3,r9
+	lxvl	  32+v8,r4,r9
+	sub.	  r9,r9,r8
+	vcmpneb.  v4,v6,v8
+	crnand	  4*cr0+lt,4*cr0+gt,4*cr6+eq
+	bt	  4*cr0+lt,L(tail3)
+
+	lxvl	  32+v7,r6,r9
+	lxvl	  32+v9,r7,r9
+	vcmpneb.  v5,v7,v9
+	bne	  cr6,L(tail4)
+
+L(finish):
+	/* The contents are equal.  */
+	li	r3,0
+	blr
+
+L(loop_head):
+	/* Calculate how many loops to run.  */
+	srdi.	r8,r5,7
+	beq	L(loop_tail)
+	mtctr	r8
+
+/* Main loop.  Compares 128 bytes each loop.  */
+	.p2align 5
+L(loop_128):
+	COMPARE_32(v0,v2,0,tail1,tail2)
+	COMPARE_32(v6,v8,32,tail3,tail4)
+	COMPARE_32(v10,v12,64,tail5,tail6)
+	COMPARE_32(v14,v16,96,tail7,tail8)
+
+	addi	r3,r3,128
+	addi	r4,r4,128
+	bdnz	L(loop_128)
+
+	/* Account loop comparisons.  */
+	clrldi.  r5,r5,57
+	beq	 L(finish)
+
+/* Compares 64 bytes if length is still bigger than 64 bytes.  */
+	.p2align 5
+L(loop_tail):
+	cmpldi	r5,64
+	ble	L(last_compare)
+	COMPARE_32(v0,v2,0,tail1,tail2)
+	COMPARE_32(v6,v8,32,tail3,tail4)
+	addi	r3,r3,64
+	addi	r4,r4,64
+	subi	r5,r5,64
+	b	L(last_compare)
+
+L(tail1):
+	TAIL(v4,v0,v2)
+
+L(tail2):
+	TAIL(v5,v1,v3)
+
+L(tail3):
+	TAIL(v4,v6,v8)
+
+L(tail4):
+	TAIL(v5,v7,v9)
+
+L(tail5):
+	TAIL(v4,v10,v12)
+
+L(tail6):
+	TAIL(v5,v11,v13)
+
+L(tail7):
+	TAIL(v4,v14,v16)
+
+L(tail8):
+	TAIL(v5,v15,v17)
+
+END (MEMCMP)
+libc_hidden_builtin_def (memcmp)
+weak_alias (memcmp, bcmp)