[Gmp-commit] /var/hg/gmp: 4 new changesets

Sun Feb 12 20:57:01 UTC 2017

details:   /var/hg/gmp/rev/b7c07428536a
changeset: 17262:b7c07428536a
user:      Torbjorn Granlund <tg at gmplib.org>
date:      Sun Feb 12 21:52:53 2017 +0100
description:
(LEA): Put code in lea_list instead of in EPILOGUE_cpu.
(EPILOGUE_cpu): Output lea_list, the zap it.

details:   /var/hg/gmp/rev/0b3806252384
changeset: 17263:0b3806252384
user:      Torbjorn Granlund <tg at gmplib.org>
date:      Sun Feb 12 21:55:15 2017 +0100
description:
For AIX, don't jump from mpn_bdiv_q_1 to middle of mpn_pi1_bdiv_q_1.
Streamline.

details:   /var/hg/gmp/rev/5ad2179a66f0
changeset: 17264:5ad2179a66f0
user:      Torbjorn Granlund <tg at gmplib.org>
date:      Sun Feb 12 21:56:26 2017 +0100
description:
(AIX): New define.

details:   /var/hg/gmp/rev/2ab28c0b9c4a
changeset: 17265:2ab28c0b9c4a
user:      Torbjorn Granlund <tg at gmplib.org>
date:      Sun Feb 12 21:56:48 2017 +0100
description:
ChangeLog

diffstat:

 ChangeLog                         |  10 +++++++++-
 mpn/powerpc64/aix.m4              |   2 ++
 mpn/powerpc64/darwin.m4           |  13 ++++++++-----
 mpn/powerpc64/mode64/bdiv_q_1.asm |  30 ++++++++++++++++--------------
 4 files changed, 35 insertions(+), 20 deletions(-)

diffs (152 lines):

diff -r df28d12d9692 -r 2ab28c0b9c4a ChangeLog

--- a/ChangeLog	Sun Feb 12 13:43:16 2017 +0100
+++ b/ChangeLog	Sun Feb 12 21:56:48 2017 +0100
@@ -1,5 +1,14 @@
 2017-02-12  TorbjÃ¶rn Granlund  <tg at gmplib.org>
 
+	* mpn/powerpc64/aix.m4 (AIX): New define.
+
+	* mpn/powerpc64/mode64/bdiv_q_1.asm: For AIX, don't jump from
+	mpn_bdiv_q_1 to middle of mpn_pi1_bdiv_q_1.  Streamline.
+
+	* mpn/powerpc64/darwin.m4 (LEA): Put code in lea_list instead of in
+	EPILOGUE_cpu.
+	(EPILOGUE_cpu): Output lea_list, the zap it.
+
 	* mpn/sparc64/ultrasparct3/bdiv_q_1.asm: New file, based on dive_1.asm.
 
 2017-02-11  TorbjÃ¶rn Granlund  <tg at gmplib.org>
@@ -23,7 +32,6 @@
 
 2017-02-10  TorbjÃ¶rn Granlund  <tg at gmplib.org>
 
-
 	* mpn/arm/arm-defs.m4 (EPILOGUE_cpu): Zap lea_list to avoid repetition.
 
 	* mpn/x86_64/bdiv_q_1.asm: Rewrite, base on atom/dive_1.asm.
diff -r df28d12d9692 -r 2ab28c0b9c4a mpn/powerpc64/aix.m4
--- a/mpn/powerpc64/aix.m4	Sun Feb 12 13:43:16 2017 +0100
+++ b/mpn/powerpc64/aix.m4	Sun Feb 12 21:56:48 2017 +0100
@@ -29,6 +29,8 @@
 dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
 dnl  see https://www.gnu.org/licenses/.
 
+define(`AIX')
+
 define(`ASM_START',
 	`.machine	"any"
 	.toc')
diff -r df28d12d9692 -r 2ab28c0b9c4a mpn/powerpc64/darwin.m4
--- a/mpn/powerpc64/darwin.m4	Sun Feb 12 13:43:16 2017 +0100
+++ b/mpn/powerpc64/darwin.m4	Sun Feb 12 21:56:48 2017 +0100
@@ -29,14 +29,14 @@
 dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
 dnl  see https://www.gnu.org/licenses/.
 
+define(`DARWIN')
+
 define(`ASM_START',`')
 
 dnl  Called: PROLOGUE_cpu(GSYM_PREFIX`'foo[,toc])
 dnl          EPILOGUE_cpu(GSYM_PREFIX`'foo)
 dnl
 
-define(`DARWIN')
-
 define(`PROLOGUE_cpu',
 m4_assert_numargs_range(1,2)
 `ifelse(`$2',toc,,
@@ -46,8 +46,7 @@
 	.align	5
 $1:')
 
-define(`EPILOGUE_cpu',
-m4_assert_numargs(1))
+define(`lea_list', `')
 
 dnl  LEAL -- Load Effective Address Local.  This is to be used for symbols
 dnl  defined in the same file.  It will not work for externally defined
@@ -74,7 +73,7 @@
 define(`LEA',
 m4_assert_numargs(2)
 `ifdef(`PIC',
-`define(`EPILOGUE_cpu',
+`define(`lea_list',
 `	.non_lazy_symbol_pointer
 `L'$2`'$non_lazy_ptr:
 	.indirect_symbol $2
@@ -112,6 +111,10 @@
 define(`CALL',
 	`bl	GSYM_PREFIX`'$1')
 
+define(`EPILOGUE_cpu',
+`lea_list'
+`define(`lea_list', `')')
+
 define(`ASM_END', `dnl')
 
 define(`EXTRA_REGISTER', r2)
diff -r df28d12d9692 -r 2ab28c0b9c4a mpn/powerpc64/mode64/bdiv_q_1.asm
--- a/mpn/powerpc64/mode64/bdiv_q_1.asm	Sun Feb 12 13:43:16 2017 +0100
+++ b/mpn/powerpc64/mode64/bdiv_q_1.asm	Sun Feb 12 21:56:48 2017 +0100
@@ -55,44 +55,46 @@
 EXTERN(binvert_limb_table)
 
 PROLOGUE(mpn_bdiv_q_1,toc)
-	addic	r7, n, -1
+	addi	r7, n, -1
 	cmpi	cr1, n, 1
 	ld	r12, 0(up)
-	rldicl.	r0, d, 0, 63
 	li	cnt, 0
-	bne	cr0, L(7)
 	neg	r0, d
 	and	r0, d, r0
 	cntlzd	r0, r0
-	subfic	r0, r0, 63
-	rldicl	cnt, r0, 0, 32
-	srd	d, d, r0
+	subfic	cnt, r0, 63
+	srd	d, d, cnt
 L(7):
 	mtctr	r7
-	LEA(	r5, binvert_limb_table)
+	LEA(	r10, binvert_limb_table)
 	rldicl	r11, d, 63, 57
-	lbzx	r0, r5, r11
+	lbzx	r0, r10, r11
 	mulld	r9, r0, r0
 	sldi	r0, r0, 1
 	mulld	r9, d, r9
 	subf	r0, r9, r0
-	mulld	r5, r0, r0
+	mulld	r10, r0, r0
 	sldi	r0, r0, 1
-	mulld	r5, d, r5
-	subf	r0, r5, r0
+	mulld	r10, d, r10
+	subf	r0, r10, r0
 	mulld	r9, r0, r0
 	sldi	r0, r0, 1
 	mulld	r9, d, r9
 	subf	di, r9, r0		C di = 1/d mod 2^64
-
+ifdef(`AIX',
+`	C For AIX it is not clear how to jump into another function.
+	b	.mpn_pi1_bdiv_q_1
+',`
+	C For non-AIX, dispatch into the pi1 variant.
 	bne	cr0, L(norm)
 	b	L(unorm)
+')
 EPILOGUE()
 
-PROLOGUE(mpn_pi1_bdiv_q_1,toc)
+PROLOGUE(mpn_pi1_bdiv_q_1)
 	cmpi	cr0, cnt, 0
 	ld	r12, 0(up)
-	addic	r0, n, -1
+	addic	r0, n, -1		C set carry as side effect
 	cmpi	cr1, n, 1
 	mtctr	r0
 	beq	cr0, L(norm)
