gcc/libgcc/config/c6x/lib1funcs.S
2022-01-03 10:42:10 +01:00

439 lines
9.4 KiB
ArmAsm

/* Copyright (C) 2010-2022 Free Software Foundation, Inc.
Contributed by Bernd Schmidt <bernds@codesourcery.com>.
This file is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 3, or (at your option) any
later version.
This file 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
General Public License for more details.
Under Section 7 of GPL version 3, you are granted additional
permissions described in the GCC Runtime Library Exception, version
3.1, as published by the Free Software Foundation.
You should have received a copy of the GNU General Public License and
a copy of the GCC Runtime Library Exception along with this program;
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
<http://www.gnu.org/licenses/>. */
;; ABI considerations for the divide functions
;; The following registers are call-used:
;; __c6xabi_divi A0,A1,A2,A4,A6,B0,B1,B2,B4,B5
;; __c6xabi_divu A0,A1,A2,A4,A6,B0,B1,B2,B4
;; __c6xabi_remi A1,A2,A4,A5,A6,B0,B1,B2,B4
;; __c6xabi_remu A1,A4,A5,A7,B0,B1,B2,B4
;;
;; In our implementation, divu and remu are leaf functions,
;; while both divi and remi call into divu.
;; A0 is not clobbered by any of the functions.
;; divu does not clobber B2 either, which is taken advantage of
;; in remi.
;; divi uses B5 to hold the original return address during
;; the call to divu.
;; remi uses B2 and A5 to hold the input values during the
;; call to divu. It stores B3 in on the stack.
#ifdef L_divsi3
.text
.align 2
.global __c6xabi_divi
.hidden __c6xabi_divi
.type __c6xabi_divi, STT_FUNC
__c6xabi_divi:
call .s2 __c6xabi_divu
|| mv .d2 B3, B5
|| cmpgt .l1 0, A4, A1
|| cmpgt .l2 0, B4, B1
[A1] neg .l1 A4, A4
|| [B1] neg .l2 B4, B4
|| xor .s1x A1, B1, A1
#ifdef _TMS320C6400
[A1] addkpc .s2 1f, B3, 4
#else
[A1] mvkl .s2 1f, B3
[A1] mvkh .s2 1f, B3
nop 2
#endif
1:
neg .l1 A4, A4
|| mv .l2 B3,B5
|| ret .s2 B5
nop 5
#endif
#if defined L_modsi3 || defined L_divmodsi4
.align 2
#ifdef L_modsi3
#define MOD_OUTPUT_REG A4
.global __c6xabi_remi
.hidden __c6xabi_remi
.type __c6xabi_remi, STT_FUNC
#else
#define MOD_OUTPUT_REG A5
.global __c6xabi_divremi
.hidden __c6xabi_divremi
.type __c6xabi_divremi, STT_FUNC
__c6xabi_divremi:
#endif
__c6xabi_remi:
stw .d2t2 B3, *B15--[2]
|| cmpgt .l1 0, A4, A1
|| cmpgt .l2 0, B4, B2
|| mv .s1 A4, A5
|| call .s2 __c6xabi_divu
[A1] neg .l1 A4, A4
|| [B2] neg .l2 B4, B4
|| xor .s2x B2, A1, B0
|| mv .d2 B4, B2
#ifdef _TMS320C6400
[B0] addkpc .s2 1f, B3, 1
[!B0] addkpc .s2 2f, B3, 1
nop 2
#else
[B0] mvkl .s2 1f,B3
[!B0] mvkl .s2 2f,B3
[B0] mvkh .s2 1f,B3
[!B0] mvkh .s2 2f,B3
#endif
1:
neg .l1 A4, A4
2:
ldw .d2t2 *++B15[2], B3
#ifdef _TMS320C6400_PLUS
mpy32 .m1x A4, B2, A6
nop 3
ret .s2 B3
sub .l1 A5, A6, MOD_OUTPUT_REG
nop 4
#else
mpyu .m1x A4, B2, A1
nop 1
mpylhu .m1x A4, B2, A6
|| mpylhu .m2x B2, A4, B2
nop 1
add .l1x A6, B2, A6
|| ret .s2 B3
shl .s1 A6, 16, A6
add .d1 A6, A1, A6
sub .l1 A5, A6, MOD_OUTPUT_REG
nop 2
#endif
#endif
#if defined L_udivsi3 || defined L_udivmodsi4
.align 2
#ifdef L_udivsi3
.global __c6xabi_divu
.hidden __c6xabi_divu
.type __c6xabi_divu, STT_FUNC
__c6xabi_divu:
#else
.global __c6xabi_divremu
.hidden __c6xabi_divremu
.type __c6xabi_divremu, STT_FUNC
__c6xabi_divremu:
#endif
;; We use a series of up to 31 subc instructions. First, we find
;; out how many leading zero bits there are in the divisor. This
;; gives us both a shift count for aligning (shifting) the divisor
;; to the, and the number of times we have to execute subc.
;; At the end, we have both the remainder and most of the quotient
;; in A4. The top bit of the quotient is computed first and is
;; placed in A2.
;; Return immediately if the dividend is zero. Setting B4 to 1
;; is a trick to allow us to leave the following insns in the jump
;; delay slot without affecting the result.
mv .s2x A4, B1
#ifndef _TMS320C6400
[!b1] mvk .s2 1, B4
#endif
[b1] lmbd .l2 1, B4, B1
||[!b1] b .s2 B3 ; RETURN A
#ifdef _TMS320C6400
||[!b1] mvk .d2 1, B4
#endif
#ifdef L_udivmodsi4
||[!b1] zero .s1 A5
#endif
mv .l1x B1, A6
|| shl .s2 B4, B1, B4
;; The loop performs a maximum of 28 steps, so we do the
;; first 3 here.
cmpltu .l1x A4, B4, A2
[!A2] sub .l1x A4, B4, A4
|| shru .s2 B4, 1, B4
|| xor .s1 1, A2, A2
shl .s1 A2, 31, A2
|| [b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
;; RETURN A may happen here (note: must happen before the next branch)
0:
cmpgt .l2 B1, 7, B0
|| [b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
|| [b0] b .s1 0b
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
;; loop backwards branch happens here
ret .s2 B3
|| mvk .s1 32, A1
sub .l1 A1, A6, A6
#ifdef L_udivmodsi4
|| extu .s1 A4, A6, A5
#endif
shl .s1 A4, A6, A4
shru .s1 A4, 1, A4
|| sub .l1 A6, 1, A6
or .l1 A2, A4, A4
shru .s1 A4, A6, A4
nop
#endif
#ifdef L_umodsi3
.align 2
.global __c6xabi_remu
.hidden __c6xabi_remu
.type __c6xabi_remu, STT_FUNC
__c6xabi_remu:
;; The ABI seems designed to prevent these functions calling each other,
;; so we duplicate most of the divsi3 code here.
mv .s2x A4, B1
#ifndef _TMS320C6400
[!b1] mvk .s2 1, B4
#endif
lmbd .l2 1, B4, B1
||[!b1] b .s2 B3 ; RETURN A
#ifdef _TMS320C6400
||[!b1] mvk .d2 1, B4
#endif
mv .l1x B1, A7
|| shl .s2 B4, B1, B4
cmpltu .l1x A4, B4, A1
[!a1] sub .l1x A4, B4, A4
shru .s2 B4, 1, B4
0:
cmpgt .l2 B1, 7, B0
|| [b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
;; RETURN A may happen here (note: must happen before the next branch)
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
|| [b0] b .s1 0b
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
;; loop backwards branch happens here
ret .s2 B3
[b1] subc .l1x A4,B4,A4
|| [b1] add .s2 -1, B1, B1
[b1] subc .l1x A4,B4,A4
extu .s1 A4, A7, A4
nop 2
#endif
#if defined L_strasgi_64plus && defined _TMS320C6400_PLUS
.align 2
.global __c6xabi_strasgi_64plus
.hidden __c6xabi_strasgi_64plus
.type __c6xabi_strasgi_64plus, STT_FUNC
__c6xabi_strasgi_64plus:
shru .s2x a6, 2, b31
|| mv .s1 a4, a30
|| mv .d2 b4, b30
add .s2 -4, b31, b31
sploopd 1
|| mvc .s2 b31, ilc
ldw .d2t2 *b30++, b31
nop 4
mv .s1x b31,a31
spkernel 6, 0
|| stw .d1t1 a31, *a30++
ret .s2 b3
nop 5
#endif
#ifdef L_strasgi
.global __c6xabi_strasgi
.type __c6xabi_strasgi, STT_FUNC
__c6xabi_strasgi:
;; This is essentially memcpy, with alignment known to be at least
;; 4, and the size a multiple of 4 greater than or equal to 28.
ldw .d2t1 *B4++, A0
|| mvk .s2 16, B1
ldw .d2t1 *B4++, A1
|| mvk .s2 20, B2
|| sub .d1 A6, 24, A6
ldw .d2t1 *B4++, A5
ldw .d2t1 *B4++, A7
|| mv .l2x A6, B7
ldw .d2t1 *B4++, A8
ldw .d2t1 *B4++, A9
|| mv .s2x A0, B5
|| cmpltu .l2 B2, B7, B0
0:
stw .d1t2 B5, *A4++
||[b0] ldw .d2t1 *B4++, A0
|| mv .s2x A1, B5
|| mv .l2 B7, B6
[b0] sub .d2 B6, 24, B7
||[b0] b .s2 0b
|| cmpltu .l2 B1, B6, B0
[b0] ldw .d2t1 *B4++, A1
|| stw .d1t2 B5, *A4++
|| mv .s2x A5, B5
|| cmpltu .l2 12, B6, B0
[b0] ldw .d2t1 *B4++, A5
|| stw .d1t2 B5, *A4++
|| mv .s2x A7, B5
|| cmpltu .l2 8, B6, B0
[b0] ldw .d2t1 *B4++, A7
|| stw .d1t2 B5, *A4++
|| mv .s2x A8, B5
|| cmpltu .l2 4, B6, B0
[b0] ldw .d2t1 *B4++, A8
|| stw .d1t2 B5, *A4++
|| mv .s2x A9, B5
|| cmpltu .l2 0, B6, B0
[b0] ldw .d2t1 *B4++, A9
|| stw .d1t2 B5, *A4++
|| mv .s2x A0, B5
|| cmpltu .l2 B2, B7, B0
;; loop back branch happens here
cmpltu .l2 B1, B6, B0
|| ret .s2 b3
[b0] stw .d1t1 A1, *A4++
|| cmpltu .l2 12, B6, B0
[b0] stw .d1t1 A5, *A4++
|| cmpltu .l2 8, B6, B0
[b0] stw .d1t1 A7, *A4++
|| cmpltu .l2 4, B6, B0
[b0] stw .d1t1 A8, *A4++
|| cmpltu .l2 0, B6, B0
[b0] stw .d1t1 A9, *A4++
;; return happens here
#endif
#ifdef _TMS320C6400_PLUS
#ifdef L_push_rts
.align 2
.global __c6xabi_push_rts
.hidden __c6xabi_push_rts
.type __c6xabi_push_rts, STT_FUNC
__c6xabi_push_rts:
stw .d2t2 B14, *B15--[2]
stdw .d2t1 A15:A14, *B15--
|| b .s2x A3
stdw .d2t2 B13:B12, *B15--
stdw .d2t1 A13:A12, *B15--
stdw .d2t2 B11:B10, *B15--
stdw .d2t1 A11:A10, *B15--
stdw .d2t2 B3:B2, *B15--
#endif
#ifdef L_pop_rts
.align 2
.global __c6xabi_pop_rts
.hidden __c6xabi_pop_rts
.type __c6xabi_pop_rts, STT_FUNC
__c6xabi_pop_rts:
lddw .d2t2 *++B15, B3:B2
lddw .d2t1 *++B15, A11:A10
lddw .d2t2 *++B15, B11:B10
lddw .d2t1 *++B15, A13:A12
lddw .d2t2 *++B15, B13:B12
lddw .d2t1 *++B15, A15:A14
|| b .s2 B3
ldw .d2t2 *++B15[2], B14
nop 4
#endif
#ifdef L_call_stub
.align 2
.global __c6xabi_call_stub
.type __c6xabi_call_stub, STT_FUNC
__c6xabi_call_stub:
stw .d2t1 A2, *B15--[2]
stdw .d2t1 A7:A6, *B15--
|| call .s2 B31
stdw .d2t1 A1:A0, *B15--
stdw .d2t2 B7:B6, *B15--
stdw .d2t2 B5:B4, *B15--
stdw .d2t2 B1:B0, *B15--
stdw .d2t2 B3:B2, *B15--
|| addkpc .s2 1f, B3, 0
1:
lddw .d2t2 *++B15, B3:B2
lddw .d2t2 *++B15, B1:B0
lddw .d2t2 *++B15, B5:B4
lddw .d2t2 *++B15, B7:B6
lddw .d2t1 *++B15, A1:A0
lddw .d2t1 *++B15, A7:A6
|| b .s2 B3
ldw .d2t1 *++B15[2], A2
nop 4
#endif
#endif