S390: Add arch13 memmove ifunc variant.

This patch introduces the new arch13 ifunc variant for memmove.
For the forward or non-overlapping case it is just using memcpy.
For the backward case it relies on the new instruction mvcrl.
The instruction copies up to 256 bytes at once.
In case of an overlap, it copies the bytes like copying them
one by one starting from right to left.

ChangeLog:

	* sysdeps/s390/ifunc-memcpy.h (HAVE_MEMMOVE_ARCH13, MEMMOVE_ARCH13
	HAVE_MEMMOVE_IFUNC_AND_ARCH13_SUPPORT): New defines.
	* sysdeps/s390/memcpy-z900.S: Add arch13 memmove implementation.
	* sysdeps/s390/memmove.c (memmove): Add arch13 variant in
	ifunc selector.
	* sysdeps/s390/multiarch/ifunc-impl-list.c
	(__libc_ifunc_impl_list): Add ifunc variant for arch13 memmove.
	* sysdeps/s390/multiarch/ifunc-resolve.h (S390_STFLE_BITS_ARCH13_MIE3,
	S390_IS_ARCH13_MIE3): New defines.

6 files changed, 111 insertions, 5 deletions

diff --git a/ChangeLog b/ChangeLog
index 35237ab..36af218 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,17 @@
 2019-03-22  Stefan Liebler  <stli@linux.ibm.com>
 
+	* sysdeps/s390/ifunc-memcpy.h (HAVE_MEMMOVE_ARCH13, MEMMOVE_ARCH13
+	HAVE_MEMMOVE_IFUNC_AND_ARCH13_SUPPORT): New defines.
+	* sysdeps/s390/memcpy-z900.S: Add arch13 memmove implementation.
+	* sysdeps/s390/memmove.c (memmove): Add arch13 variant in
+	ifunc selector.
+	* sysdeps/s390/multiarch/ifunc-impl-list.c
+	(__libc_ifunc_impl_list): Add ifunc variant for arch13 memmove.
+	* sysdeps/s390/multiarch/ifunc-resolve.h (S390_STFLE_BITS_ARCH13_MIE3,
+	S390_IS_ARCH13_MIE3): New defines.
+
+2019-03-22  Stefan Liebler  <stli@linux.ibm.com>
+
 	* config.h.in (HAVE_S390_MIN_ARCH13_ZARCH_ASM_SUPPORT,
 	HAVE_S390_ARCH13_ASM_SUPPORT): New undefine.
 	* sysdeps/s390/configure.ac: Add checks for arch13 support.
diff --git a/sysdeps/s390/ifunc-memcpy.h b/sysdeps/s390/ifunc-memcpy.h
index b83ae73..1badb30 100644
--- a/sysdeps/s390/ifunc-memcpy.h
+++ b/sysdeps/s390/ifunc-memcpy.h
@@ -44,7 +44,7 @@
 #endif
 
 #if defined SHARED && defined USE_MULTIARCH && IS_IN (libc)	\
-  && ! defined HAVE_S390_MIN_Z13_ZARCH_ASM_SUPPORT
+  && ! defined HAVE_S390_MIN_ARCH13_ZARCH_ASM_SUPPORT
 # define HAVE_MEMMOVE_IFUNC	1
 #else
 # define HAVE_MEMMOVE_IFUNC	0
@@ -56,14 +56,27 @@
 # define HAVE_MEMMOVE_IFUNC_AND_VX_SUPPORT 0
 #endif
 
-#if defined HAVE_S390_MIN_Z13_ZARCH_ASM_SUPPORT
+#ifdef HAVE_S390_ARCH13_ASM_SUPPORT
+# define HAVE_MEMMOVE_IFUNC_AND_ARCH13_SUPPORT HAVE_MEMMOVE_IFUNC
+#else
+# define HAVE_MEMMOVE_IFUNC_AND_ARCH13_SUPPORT 0
+#endif
+
+#if defined HAVE_S390_MIN_ARCH13_ZARCH_ASM_SUPPORT
+# define MEMMOVE_DEFAULT	MEMMOVE_ARCH13
+# define HAVE_MEMMOVE_C		0
+# define HAVE_MEMMOVE_Z13	0
+# define HAVE_MEMMOVE_ARCH13	1
+#elif defined HAVE_S390_MIN_Z13_ZARCH_ASM_SUPPORT
 # define MEMMOVE_DEFAULT	MEMMOVE_Z13
 # define HAVE_MEMMOVE_C		0
 # define HAVE_MEMMOVE_Z13	1
+# define HAVE_MEMMOVE_ARCH13	HAVE_MEMMOVE_IFUNC_AND_ARCH13_SUPPORT
 #else
 # define MEMMOVE_DEFAULT	MEMMOVE_C
 # define HAVE_MEMMOVE_C		1
 # define HAVE_MEMMOVE_Z13	HAVE_MEMMOVE_IFUNC_AND_VX_SUPPORT
+# define HAVE_MEMMOVE_ARCH13	HAVE_MEMMOVE_IFUNC_AND_ARCH13_SUPPORT
 #endif
 
 #if HAVE_MEMCPY_Z900_G5
@@ -101,3 +114,9 @@
 #else
 # define MEMMOVE_Z13		NULL
 #endif
+
+#if HAVE_MEMMOVE_ARCH13
+# define MEMMOVE_ARCH13		__memmove_arch13
+#else
+# define MEMMOVE_ARCH13		NULL
+#endif
diff --git a/sysdeps/s390/memcpy-z900.S b/sysdeps/s390/memcpy-z900.S
index 90d5f7b..307332fc 100644
--- a/sysdeps/s390/memcpy-z900.S
+++ b/sysdeps/s390/memcpy-z900.S
@@ -277,6 +277,61 @@ ENTRY(MEMMOVE_Z13)
 END(MEMMOVE_Z13)
 #endif /* HAVE_MEMMOVE_Z13  */
 
+#if HAVE_MEMMOVE_ARCH13
+ENTRY(MEMMOVE_ARCH13)
+	.machine "arch13"
+	.machinemode "zarch_nohighgprs"
+# if ! defined __s390x__
+	/* Note: The 31bit dst and src pointers are prefixed with zeroes.  */
+	llgfr	%r4,%r4
+	llgfr	%r3,%r3
+	llgfr	%r2,%r2
+# endif /* ! defined __s390x__ */
+	sgrk	%r5,%r2,%r3
+	aghik	%r0,%r4,-1	/* Both vstl and mvcrl needs highest index.  */
+	clgijh	%r4,16,.L_MEMMOVE_ARCH13_LARGE
+.L_MEMMOVE_ARCH13_SMALL:
+	jl .L_MEMMOVE_ARCH13_END /* Return if len was zero (cc of aghik).  */
+	/* Store up to 16 bytes with vll/vstl (needs highest index).  */
+	vll	%v16,%r0,0(%r3)
+	vstl	%v16,%r0,0(%r2)
+.L_MEMMOVE_ARCH13_END:
+	br      %r14
+.L_MEMMOVE_ARCH13_LARGE:
+	lgr     %r1,%r2	/* For memcpy: r1: Use as dest ; r2: Return dest  */
+	/* The unsigned comparison (dst - src >= len) determines if we can
+	   execute the forward case with memcpy.  */
+#if ! HAVE_MEMCPY_Z196
+# error The arch13 variant of memmove needs the z196 variant of memcpy!
+#endif
+	/* Backward case.  */
+	clgrjhe %r5,%r4,.L_Z196_start2
+	clgijh	%r0,255,.L_MEMMOVE_ARCH13_LARGER_256B
+	/* Move up to 256bytes with mvcrl (move right to left).  */
+	mvcrl	0(%r1),0(%r3)	/* Move (r0 + 1) bytes from r3 to r1.  */
+	br      %r14
+.L_MEMMOVE_ARCH13_LARGER_256B:
+	/* First move the "remaining" block of up to 256 bytes at the end of
+	   src/dst buffers.  Then move blocks of 256bytes in a loop starting
+	   with the block at the end.
+	   (If src/dst pointers are aligned e.g. to 256 bytes, then the pointers
+	   passed to mvcrl instructions are aligned, too)  */
+	risbgn	%r5,%r0,8,128+63,56	/* r5 = r0 / 256  */
+	risbgn	%r0,%r0,56,128+63,0	/* r0 = r0 & 0xFF  */
+	slgr	%r4,%r0
+	lay	%r1,-1(%r4,%r1)
+	lay	%r3,-1(%r4,%r3)
+	mvcrl	0(%r1),0(%r3)	/* Move (r0 + 1) bytes from r3 to r1.  */
+	lghi	%r0,255		/* Always copy 256 bytes in the loop below!  */
+.L_MEMMOVE_ARCH13_LARGE_256B_LOOP:
+	aghi	%r1,-256
+	aghi	%r3,-256
+	mvcrl	0(%r1),0(%r3)	/* Move (r0 + 1) bytes from r3 to r1.  */
+	brctg	%r5,.L_MEMMOVE_ARCH13_LARGE_256B_LOOP
+	br      %r14
+END(MEMMOVE_ARCH13)
+#endif /* HAVE_MEMMOVE_ARCH13  */
+
 #if ! HAVE_MEMCPY_IFUNC
 /* If we don't use ifunc, define an alias for mem[p]cpy here.
    Otherwise see sysdeps/s390/mem[p]cpy.c.  */
diff --git a/sysdeps/s390/memmove.c b/sysdeps/s390/memmove.c
index fd4da37..fb6b69a 100644
--- a/sysdeps/s390/memmove.c
+++ b/sysdeps/s390/memmove.c
@@ -36,9 +36,19 @@ extern __typeof (__redirect_memmove) MEMMOVE_C attribute_hidden;
 extern __typeof (__redirect_memmove) MEMMOVE_Z13 attribute_hidden;
 # endif
 
+# if HAVE_MEMMOVE_ARCH13
+extern __typeof (__redirect_memmove) MEMMOVE_ARCH13 attribute_hidden;
+# endif
+
 s390_libc_ifunc_expr (__redirect_memmove, memmove,
-		      (HAVE_MEMMOVE_Z13 && (hwcap & HWCAP_S390_VX))
-		      ? MEMMOVE_Z13
-		      : MEMMOVE_DEFAULT
+		      ({
+			s390_libc_ifunc_expr_stfle_init ();
+			(HAVE_MEMMOVE_ARCH13
+			 && S390_IS_ARCH13_MIE3 (stfle_bits))
+			  ? MEMMOVE_ARCH13
+			  : (HAVE_MEMMOVE_Z13 && (hwcap & HWCAP_S390_VX))
+			  ? MEMMOVE_Z13
+			  : MEMMOVE_DEFAULT;
+		      })
 		      )
 #endif
diff --git a/sysdeps/s390/multiarch/ifunc-impl-list.c b/sysdeps/s390/multiarch/ifunc-impl-list.c
index b54c52a..d742d66 100644
--- a/sysdeps/s390/multiarch/ifunc-impl-list.c
+++ b/sysdeps/s390/multiarch/ifunc-impl-list.c
@@ -169,6 +169,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 
 #if HAVE_MEMMOVE_IFUNC
     IFUNC_IMPL (i, name, memmove,
+# if HAVE_MEMMOVE_ARCH13
+		IFUNC_IMPL_ADD (array, i, memmove,
+				S390_IS_ARCH13_MIE3 (stfle_bits),
+				MEMMOVE_ARCH13)
+# endif
 # if HAVE_MEMMOVE_Z13
 		IFUNC_IMPL_ADD (array, i, memmove,
 				dl_hwcap & HWCAP_S390_VX, MEMMOVE_Z13)
diff --git a/sysdeps/s390/multiarch/ifunc-resolve.h b/sysdeps/s390/multiarch/ifunc-resolve.h
index b833dfe..743de9e 100644
--- a/sysdeps/s390/multiarch/ifunc-resolve.h
+++ b/sysdeps/s390/multiarch/ifunc-resolve.h
@@ -22,6 +22,11 @@
 
 #define S390_STFLE_BITS_Z10  34 /* General instructions extension */
 #define S390_STFLE_BITS_Z196 45 /* Distinct operands, pop ... */
+#define S390_STFLE_BITS_ARCH13_MIE3 61 /* Miscellaneous-Instruction-Extensions
+					  Facility 3, e.g. mvcrl.  */
+
+#define S390_IS_ARCH13_MIE3(STFLE_BITS)			\
+  ((STFLE_BITS & (1ULL << (63 - S390_STFLE_BITS_ARCH13_MIE3))) != 0)
 
 #define S390_IS_Z196(STFLE_BITS)			\
   ((STFLE_BITS & (1ULL << (63 - S390_STFLE_BITS_Z196))) != 0)