[Gmp-commit] /var/hg/gmp: Provide alderlake mul_basecase.asm.

mercurial at gmplib.org mercurial at gmplib.org
Thu Mar 3 21:50:36 CET 2022 

details:   /var/hg/gmp/rev/1cbf890d4bff
changeset: 18318:1cbf890d4bff
user:      Torbjorn Granlund <tg at gmplib.org>
date:      Thu Mar 03 21:50:34 2022 +0100
description:
Provide alderlake mul_basecase.asm.

diffstat:

 mpn/x86_64/alderlake/mul_basecase.asm |  474 ++++++++++++++++++++++++++++++++++
 1 files changed, 474 insertions(+), 0 deletions(-)

diffs (truncated from 478 to 300 lines):

diff -r 7ba46ff69b6b -r 1cbf890d4bff mpn/x86_64/alderlake/mul_basecase.asm
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/x86_64/alderlake/mul_basecase.asm	Thu Mar 03 21:50:34 2022 +0100
@@ -0,0 +1,474 @@
+dnl  AMD64 mpn_mul_basecase.
+
+dnl  Contributed to the GNU project by Torbjörn Granlund.
+
+dnl  Copyright 2012, 2013, 2022 Free Software Foundation, Inc.
+
+dnl  This file is part of the GNU MP Library.
+dnl
+dnl  The GNU MP Library is free software; you can redistribute it and/or modify
+dnl  it under the terms of either:
+dnl
+dnl    * the GNU Lesser General Public License as published by the Free
+dnl      Software Foundation; either version 3 of the License, or (at your
+dnl      option) any later version.
+dnl
+dnl  or
+dnl
+dnl    * the GNU General Public License as published by the Free Software
+dnl      Foundation; either version 2 of the License, or (at your option) any
+dnl      later version.
+dnl
+dnl  or both in parallel, as here.
+dnl
+dnl  The GNU MP Library is distributed in the hope that it will be useful, but
+dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+dnl  for more details.
+dnl
+dnl  You should have received copies of the GNU General Public License and the
+dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
+dnl  see https://www.gnu.org/licenses/.
+
+include(`../config.m4')
+
+C	     cycles/limb
+C AMD K8,K9	 -
+C AMD K10	 -
+C AMD bd1	 -
+C AMD bd2	 -
+C AMD bd3	 -
+C AMD bd4	 -
+C AMD zn1	 ?
+C AMD zn2	 ?
+C AMD zn3	 ?
+C AMD bt1	 -
+C AMD bt2	 -
+C Intel P4	 -
+C Intel CNR	 -
+C Intel PNR	 -
+C Intel NHM	 -
+C Intel WSM	 -
+C Intel SBR	 -
+C Intel IBR	 -
+C Intel HWL	 -
+C Intel BWL	 ?
+C Intel SKL	 ?
+C Intel RKL	 ?
+C Intel ALD	 1.29
+C Intel atom	 -
+C Intel SLM	 -
+C Intel GLM	 -
+C VIA nano	 -
+
+C TODO
+C  * Do overlapped software pipelining.
+C  * Try shallower pipeline, which would result in using fewer registers.
+C  * There are false dependencies on CF/OF between iterations.  Try breaking
+C    them to see if it helps.
+
+define(`rp',	`%rdi')	dnl rcx
+define(`up',	`%rsi')	dnl rdx
+define(`un_arg',`%rdx')	dnl r8
+define(`vp_arg',`%rcx')	dnl r9
+define(`vn_arg',`%r8')	dnl stack
+
+define(`un',	`%r14')
+define(`vp',	`%r15')
+define(`vn',	`%rbp')
+
+define(`n',	`%rcx')
+define(`v0',	`%rdx')
+
+
+ASM_START()
+	TEXT
+	ALIGN(16)
+PROLOGUE(mpn_mul_basecase)
+	cmp	$2, un_arg
+	ja	L(gen)
+	mov	(vp_arg), %rdx
+	mulx(	(up), %rax, %r9)
+	mov	%rax, (rp)
+	je	L(s2x)
+
+	mov	%r9, 8(rp)
+	ret
+
+L(s2x):	mulx(	8,(up), %rax, %r10)
+	add	%r9, %rax
+	adc	$0, %r10
+	cmp	$2, R32(vn_arg)
+	je	L(s22)
+
+L(s21):	mov	%rax, 8(rp)
+	mov	%r10, 16(rp)
+	ret
+
+L(s22):	mov	8(vp_arg), %rdx
+	mulx(	(up), %r8, %r9)
+	add	%r8, %rax
+	adc	%r10, %r9
+	mov	%rax, 8(rp)
+	mulx(	8,(up), %rax, %r10)
+	adc	$0, %r10
+	adc	%r9, %rax
+	mov	%rax, 16(rp)
+	adc	$0, %r10
+	mov	%r10, 24(rp)
+	ret
+
+L(gen):	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+
+	mov	un_arg, un
+	neg	un
+	shl	$3, un
+	mov	vp_arg, vp
+	mov	vn_arg, vn
+
+	test	$1, R8(un_arg)
+	mov	(vp), %rdx
+	jz	L(bx0)
+
+L(bx1):	test	$16, R8(un)
+	jnz	L(b01)
+
+L(b11):	lea	24(un), n
+	mulx(	(up), %r11, %r10)
+	mulx(	8,(up), %r13, %r12)
+	mulx(	16,(up), %rbx, %rax)
+	lea	8(rp), rp
+	lea	24(up), up
+	jrcxz	L(med3)
+L(mtp3):mulx(	(up), %r9, %r8)
+	adcx(	%r10, %r13)
+	mov	%r11, -8(rp)
+	mulx(	8,(up), %r11, %r10)
+	adcx(	%r12, %rbx)
+	mov	%r13, (rp)
+	mulx(	16,(up), %r13, %r12)
+	adcx(	%rax, %r9)
+	mov	%rbx, 8(rp)
+	mulx(	24,(up), %rbx, %rax)
+	adcx(	%r8, %r11)
+	mov	%r9, 16(rp)
+	lea	32(up), up
+	lea	32(rp), rp
+	lea	32(n), n
+	jrcxz	L(med3)
+	jmp	L(mtp3)
+L(med3):adcx(	%r10, %r13)
+	mov	%r11, -8(rp)
+	adcx(	%r12, %rbx)
+	mov	%r13, (rp)
+	adcx(	%rcx, %rax)
+	mov	%rbx, 8(rp)
+	mov	%rax, 16(rp)
+	dec	vn
+	jz	L(ret)
+L(out3):lea	32(rp,un), rp
+	lea	24(up,un), up
+	lea	8(vp), vp
+	xor	R32(%rdx), R32(%rdx)
+	mov	(vp), %rdx
+	mulx(	-24,(up), %r11, %r10)
+	mulx(	-16,(up), %r13, %r12)
+	mulx(	-8,(up), %rbx, %rax)
+	lea	24(un), n
+	adox(	-8,(rp), %r11)
+	jrcxz	L(ed3)
+L(tp3):	mulx(	(up), %r9, %r8)
+	adcx(	%r10, %r13)
+	mov	%r11, -8(rp)
+	adox(	(rp), %r13)
+	mulx(	8,(up), %r11, %r10)
+	adcx(	%r12, %rbx)
+	mov	%r13, (rp)
+	adox(	8,(rp), %rbx)
+	mulx(	16,(up), %r13, %r12)
+	adcx(	%rax, %r9)
+	mov	%rbx, 8(rp)
+	adox(	16,(rp), %r9)
+	mulx(	24,(up), %rbx, %rax)
+	adcx(	%r8, %r11)
+	mov	%r9, 16(rp)
+	adox(	24,(rp), %r11)
+	lea	32(up), up
+	lea	32(rp), rp
+	lea	32(n), n
+	jrcxz	L(ed3)
+	jmp	L(tp3)
+L(ed3):	adcx(	%r10, %r13)
+	mov	%r11, -8(rp)
+	adox(	(rp), %r13)
+	adcx(	%r12, %rbx)
+	mov	%r13, (rp)
+	adox(	8,(rp), %rbx)
+	adcx(	%rcx, %rax)
+	adox(	%rcx, %rax)
+	mov	%rbx, 8(rp)
+	mov	%rax, 16(rp)
+	dec	vn
+	jnz	L(out3)
+	jmp	L(ret)
+
+
+L(b01):	mulx(	(up), %rbx, %rax)
+	lea	8(un), n
+	mulx(	8,(up), %r9, %r8)
+	mulx(	16,(up), %r11, %r10)
+	lea	8(up), up
+	lea	-8(rp), rp
+	jmp	L(ml1)
+L(mtp1):mulx(	(up), %r9, %r8)
+	adcx(	%r10, %r13)
+	mov	%r11, -8(rp)
+	mulx(	8,(up), %r11, %r10)
+	adcx(	%r12, %rbx)
+	mov	%r13, (rp)
+L(ml1):	mulx(	16,(up), %r13, %r12)
+	adcx(	%rax, %r9)
+	mov	%rbx, 8(rp)
+	mulx(	24,(up), %rbx, %rax)
+	adcx(	%r8, %r11)
+	mov	%r9, 16(rp)
+	lea	32(up), up
+	lea	32(rp), rp
+	lea	32(n), n
+	jrcxz	L(med1)
+	jmp	L(mtp1)
+L(med1):adcx(	%r10, %r13)
+	mov	%r11, -8(rp)
+	adcx(	%r12, %rbx)
+	mov	%r13, (rp)
+	adcx(	%rcx, %rax)
+	mov	%rbx, 8(rp)
+	mov	%rax, 16(rp)
+	dec	vn
+	jz	L(ret)
+L(out1):lea	16(rp,un), rp
+	lea	8(up,un), up
+	lea	8(vp), vp
+	xor	R32(%rdx), R32(%rdx)
+	mov	(vp), %rdx
+	lea	8(un), n
+	mulx(	-8,(up), %rbx, %rax)
+	mulx(	(up), %r9, %r8)
+	mulx(	8,(up), %r11, %r10)
+	jmp	L(lo1)
+L(tp1):	mulx(	(up), %r9, %r8)
+	adcx(	%r10, %r13)
+	mov	%r11, -8(rp)
+	adox(	(rp), %r13)
+	mulx(	8,(up), %r11, %r10)
+	adcx(	%r12, %rbx)
+	mov	%r13, (rp)
+L(lo1):	adox(	8,(rp), %rbx)
+	mulx(	16,(up), %r13, %r12)
+	adcx(	%rax, %r9)
+	mov	%rbx, 8(rp)
+	adox(	16,(rp), %r9)
+	mulx(	24,(up), %rbx, %rax)
+	adcx(	%r8, %r11)
+	mov	%r9, 16(rp)
+	adox(	24,(rp), %r11)
+	lea	32(up), up
+	lea	32(rp), rp
+	lea	32(n), n
+	jrcxz	L(ed1)
+	jmp	L(tp1)
+L(ed1):	adcx(	%r10, %r13)
+	mov	%r11, -8(rp)
+	adox(	(rp), %r13)
+	adcx(	%r12, %rbx)
+	mov	%r13, (rp)
+	adox(	8,(rp), %rbx)
+	adcx(	%rcx, %rax)
+	adox(	%rcx, %rax)
+	mov	%rbx, 8(rp)
+	mov	%rax, 16(rp)
+	dec	vn
+	jnz	L(out1)

More information about the gmp-commit mailing list