[thirdparty/gcc.git] / libgcc / config / bfin / lib1funcs.S

/* libgcc functions for Blackfin.
   Copyright (C) 2005-2023 Free Software Foundation, Inc.
   Contributed by Analog Devices.

This file is part of GCC.

GCC 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.

GCC 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/>.  */

#ifdef L_divsi3
.text
.align 2
.global ___divsi3;
.type ___divsi3, STT_FUNC;

___divsi3:
        [--SP]= RETS;
	[--SP] = R7;

	R2 = -R0;
        CC = R0 < 0;
	IF CC R0 = R2;
	R7 = CC;

	R2 = -R1;
        CC = R1 < 0;
	IF CC R1 = R2;
	R2 = CC;
	R7 = R7 ^ R2;

        CALL ___udivsi3;

	CC = R7;
	R1 = -R0;
	IF CC R0 = R1;

	R7 = [SP++];
        RETS = [SP++];
        RTS;
#endif

#ifdef L_modsi3	
.align 2
.global ___modsi3;
.type ___modsi3, STT_FUNC;

___modsi3:
	[--SP] = RETS;
	[--SP] = R0;
	[--SP] = R1;
	CALL ___divsi3;
	R2 = [SP++];
	R1 = [SP++];
	R2 *= R0;
	R0 = R1 - R2;
	RETS = [SP++];
	RTS; 
#endif

#ifdef L_udivsi3
.align 2
.global ___udivsi3;
.type ___udivsi3, STT_FUNC;

___udivsi3:
        P0 = 32;
        LSETUP (0f, 1f) LC0 = P0;
	/* upper half of dividend */
        R3 = 0;
0:
	/* The first time round in the loop we shift in garbage, but since we
	   perform 33 shifts, it doesn't matter.  */
	R0 = ROT R0 BY 1;
	R3 = ROT R3 BY 1;
	R2 = R3 - R1;
        CC = R3 < R1 (IU);
1:
	/* Last instruction of the loop.  */
	IF ! CC R3 = R2;

	/* Shift in the last bit.  */
	R0 = ROT R0 BY 1;
	/* R0 is the result, R3 contains the remainder.  */
	R0 = ~ R0;
        RTS;
#endif

#ifdef L_umodsi3
.align 2
.global ___umodsi3;
.type ___umodsi3, STT_FUNC;

___umodsi3:
	[--SP] = RETS;
	CALL ___udivsi3;
	R0 = R3;
	RETS = [SP++]; 
	RTS;
#endif

#ifdef L_umulsi3_highpart
.align 2
.global ___umulsi3_highpart;
.type ___umulsi3_highpart, STT_FUNC;

___umulsi3_highpart:
	A1 = R1.L * R0.L (FU);
	A1 = A1 >> 16;
	A0 = R1.H * R0.H, A1 += R1.L * R0.H (FU);
	A1 += R0.L * R1.H (FU);
	A1 = A1 >> 16;
	A0 += A1;
	R0 = A0 (FU);
	RTS;
#endif

#ifdef L_smulsi3_highpart
.align 2
.global ___smulsi3_highpart;
.type ___smulsi3_highpart, STT_FUNC;

___smulsi3_highpart:
	A1 = R1.L * R0.L (FU);
	A1 = A1 >> 16;
	A0 = R0.H * R1.H, A1 += R0.H * R1.L (IS,M);
	A1 += R1.H * R0.L (IS,M);
	A1 = A1 >>> 16;
	R0 = (A0 += A1);
	RTS;
#endif

#ifdef L_muldi3
.align 2
.global ___muldi3;
.type ___muldi3, STT_FUNC;

/*
	   R1:R0 * R3:R2
	 = R1.h:R1.l:R0.h:R0.l * R3.h:R3.l:R2.h:R2.l
[X]	 = (R1.h * R3.h) * 2^96
[X]	   + (R1.h * R3.l + R1.l * R3.h) * 2^80
[X]	   + (R1.h * R2.h + R1.l * R3.l + R3.h * R0.h) * 2^64
[T1]	   + (R1.h * R2.l + R3.h * R0.l + R1.l * R2.h + R3.l * R0.h) * 2^48
[T2]	   + (R1.l * R2.l + R3.l * R0.l + R0.h * R2.h) * 2^32
[T3]	   + (R0.l * R2.h + R2.l * R0.h) * 2^16
[T4]	   + (R0.l * R2.l)

	We can discard the first three lines marked "X" since we produce
	only a 64 bit result.  So, we need ten 16-bit multiplies.

	Individual mul-acc results:
[E1]	 =  R1.h * R2.l + R3.h * R0.l + R1.l * R2.h + R3.l * R0.h
[E2]	 =  R1.l * R2.l + R3.l * R0.l + R0.h * R2.h
[E3]	 =  R0.l * R2.h + R2.l * R0.h
[E4]	 =  R0.l * R2.l

	We also need to add high parts from lower-level results to higher ones:
	E[n]c = E[n] + (E[n+1]c >> 16), where E4c := E4

	One interesting property is that all parts of the result that depend
	on the sign of the multiplication are discarded.  Those would be the
	multiplications involving R1.h and R3.h, but only the top 16 bit of
	the 32 bit result depend on the sign, and since R1.h and R3.h only
	occur in E1, the top half of these results is cut off.
	So, we can just use FU mode for all of the 16-bit multiplies, and
	ignore questions of when to use mixed mode.  */

___muldi3:
	/* [SP] technically is part of the caller's frame, but we can
	   use it as scratch space.  */
	A0 = R2.H * R1.L, A1 = R2.L * R1.H (FU) || R3 = [SP + 12];	/* E1 */
	A0 += R3.H * R0.L, A1 += R3.L * R0.H (FU) || [SP] = R4;		/* E1 */
	A0 += A1;							/* E1 */
	R4 = A0.w;
	A0 = R0.l * R3.l (FU);						/* E2 */
	A0 += R2.l * R1.l (FU);						/* E2 */

	A1 = R2.L * R0.L (FU);						/* E4 */
	R3 = A1.w;
	A1 = A1 >> 16;							/* E3c */
	A0 += R2.H * R0.H, A1 += R2.L * R0.H (FU);			/* E2, E3c */
	A1 += R0.L * R2.H (FU);						/* E3c */
	R0 = A1.w;
	A1 = A1 >> 16;							/* E2c */
	A0 += A1;							/* E2c */
	R1 = A0.w;

	/* low(result) = low(E3c):low(E4) */
	R0 = PACK (R0.l, R3.l);
	/* high(result) = E2c + (E1 << 16) */
	R1.h = R1.h + R4.l (NS) || R4 = [SP];
	RTS;

.size ___muldi3, .-___muldi3
#endif
Commit	Line	Data
0d4a78eb	1	/* libgcc functions for Blackfin.
83ffe9cd	2	Copyright (C) 2005-2023 Free Software Foundation, Inc.
0d4a78eb BS	3	Contributed by Analog Devices.
	4
	5	This file is part of GCC.
	6
	7	GCC is free software; you can redistribute it and/or modify
	8	it under the terms of the GNU General Public License as published by
748086b7	9	the Free Software Foundation; either version 3, or (at your option)
0d4a78eb BS	10	any later version.
	11
	12	GCC is distributed in the hope that it will be useful,
	13	but WITHOUT ANY WARRANTY; without even the implied warranty of
	14	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	15	GNU General Public License for more details.
	16
748086b7 JJ	17	Under Section 7 of GPL version 3, you are granted additional
	18	permissions described in the GCC Runtime Library Exception, version
	19	3.1, as published by the Free Software Foundation.
0d4a78eb	20
748086b7 JJ	21	You should have received a copy of the GNU General Public License and
	22	a copy of the GCC Runtime Library Exception along with this program;
	23	see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
	24	<http://www.gnu.org/licenses/>. */
0d4a78eb BS	25
	26	#ifdef L_divsi3
	27	.text
	28	.align 2
	29	.global ___divsi3;
	30	.type ___divsi3, STT_FUNC;
	31
	32	___divsi3:
	33	[--SP]= RETS;
	34	[--SP] = R7;
	35
	36	R2 = -R0;
	37	CC = R0 < 0;
	38	IF CC R0 = R2;
	39	R7 = CC;
	40
	41	R2 = -R1;
	42	CC = R1 < 0;
	43	IF CC R1 = R2;
	44	R2 = CC;
	45	R7 = R7 ^ R2;
	46
	47	CALL ___udivsi3;
	48
	49	CC = R7;
	50	R1 = -R0;
	51	IF CC R0 = R1;
	52
	53	R7 = [SP++];
	54	RETS = [SP++];
	55	RTS;
	56	#endif
	57
	58	#ifdef L_modsi3
	59	.align 2
	60	.global ___modsi3;
	61	.type ___modsi3, STT_FUNC;
	62
	63	___modsi3:
dcef6aba BS	64	[--SP] = RETS;
	65	[--SP] = R0;
	66	[--SP] = R1;
	67	CALL ___divsi3;
	68	R2 = [SP++];
	69	R1 = [SP++];
0d4a78eb BS	70	R2 *= R0;
	71	R0 = R1 - R2;
	72	RETS = [SP++];
dcef6aba	73	RTS;
0d4a78eb BS	74	#endif
	75
	76	#ifdef L_udivsi3
	77	.align 2
	78	.global ___udivsi3;
	79	.type ___udivsi3, STT_FUNC;
	80
	81	___udivsi3:
	82	P0 = 32;
	83	LSETUP (0f, 1f) LC0 = P0;
	84	/* upper half of dividend */
	85	R3 = 0;
	86	0:
	87	/* The first time round in the loop we shift in garbage, but since we
	88	perform 33 shifts, it doesn't matter. */
	89	R0 = ROT R0 BY 1;
	90	R3 = ROT R3 BY 1;
	91	R2 = R3 - R1;
	92	CC = R3 < R1 (IU);
	93	1:
	94	/* Last instruction of the loop. */
	95	IF ! CC R3 = R2;
	96
	97	/* Shift in the last bit. */
	98	R0 = ROT R0 BY 1;
	99	/* R0 is the result, R3 contains the remainder. */
	100	R0 = ~ R0;
	101	RTS;
	102	#endif
	103
	104	#ifdef L_umodsi3
	105	.align 2
	106	.global ___umodsi3;
	107	.type ___umodsi3, STT_FUNC;
	108
	109	___umodsi3:
a94aa272 BS	110	[--SP] = RETS;
a94aa272 BS	111	CALL ___udivsi3;
0d4a78eb	112	R0 = R3;
a94aa272 BS	113	RETS = [SP++];
a94aa272 BS	114	RTS;
0d4a78eb BS	115	#endif
0d4a78eb BS	116
01e7cd6e BS	117	#ifdef L_umulsi3_highpart
	118	.align 2
	119	.global ___umulsi3_highpart;
	120	.type ___umulsi3_highpart, STT_FUNC;
	121
	122	___umulsi3_highpart:
3fbee523 BS	123	A1 = R1.L * R0.L (FU);
	124	A1 = A1 >> 16;
	125	A0 = R1.H * R0.H, A1 += R1.L * R0.H (FU);
	126	A1 += R0.L * R1.H (FU);
	127	A1 = A1 >> 16;
314f9913 BS	128	A0 += A1;
314f9913 BS	129	R0 = A0 (FU);
01e7cd6e BS	130	RTS;
	131	#endif
	132
	133	#ifdef L_smulsi3_highpart
	134	.align 2
	135	.global ___smulsi3_highpart;
	136	.type ___smulsi3_highpart, STT_FUNC;
	137
	138	___smulsi3_highpart:
3fbee523 BS	139	A1 = R1.L * R0.L (FU);
	140	A1 = A1 >> 16;
	141	A0 = R0.H * R1.H, A1 += R0.H * R1.L (IS,M);
	142	A1 += R1.H * R0.L (IS,M);
	143	A1 = A1 >>> 16;
	144	R0 = (A0 += A1);
01e7cd6e BS	145	RTS;
01e7cd6e BS	146	#endif
f0b568f3 SH	147
	148	#ifdef L_muldi3
	149	.align 2
	150	.global ___muldi3;
	151	.type ___muldi3, STT_FUNC;
	152
	153	/*
	154	R1:R0 * R3:R2
	155	= R1.h:R1.l:R0.h:R0.l * R3.h:R3.l:R2.h:R2.l
	156	[X] = (R1.h * R3.h) * 2^96
	157	[X] + (R1.h * R3.l + R1.l * R3.h) * 2^80
	158	[X] + (R1.h * R2.h + R1.l * R3.l + R3.h * R0.h) * 2^64
	159	[T1] + (R1.h * R2.l + R3.h * R0.l + R1.l * R2.h + R3.l * R0.h) * 2^48
	160	[T2] + (R1.l * R2.l + R3.l * R0.l + R0.h * R2.h) * 2^32
	161	[T3] + (R0.l * R2.h + R2.l * R0.h) * 2^16
	162	[T4] + (R0.l * R2.l)
	163
	164	We can discard the first three lines marked "X" since we produce
	165	only a 64 bit result. So, we need ten 16-bit multiplies.
	166
	167	Individual mul-acc results:
	168	[E1] = R1.h * R2.l + R3.h * R0.l + R1.l * R2.h + R3.l * R0.h
	169	[E2] = R1.l * R2.l + R3.l * R0.l + R0.h * R2.h
	170	[E3] = R0.l * R2.h + R2.l * R0.h
	171	[E4] = R0.l * R2.l
	172
	173	We also need to add high parts from lower-level results to higher ones:
	174	E[n]c = E[n] + (E[n+1]c >> 16), where E4c := E4
	175
	176	One interesting property is that all parts of the result that depend
	177	on the sign of the multiplication are discarded. Those would be the
	178	multiplications involving R1.h and R3.h, but only the top 16 bit of
	179	the 32 bit result depend on the sign, and since R1.h and R3.h only
	180	occur in E1, the top half of these results is cut off.
	181	So, we can just use FU mode for all of the 16-bit multiplies, and
	182	ignore questions of when to use mixed mode. */
	183
	184	___muldi3:
	185	/* [SP] technically is part of the caller's frame, but we can
	186	use it as scratch space. */
	187	A0 = R2.H * R1.L, A1 = R2.L * R1.H (FU) \|\| R3 = [SP + 12]; /* E1 */
	188	A0 += R3.H * R0.L, A1 += R3.L * R0.H (FU) \|\| [SP] = R4; /* E1 */
	189	A0 += A1; /* E1 */
	190	R4 = A0.w;
	191	A0 = R0.l * R3.l (FU); /* E2 */
	192	A0 += R2.l * R1.l (FU); /* E2 */
	193
	194	A1 = R2.L * R0.L (FU); /* E4 */
	195	R3 = A1.w;
	196	A1 = A1 >> 16; /* E3c */
	197	A0 += R2.H * R0.H, A1 += R2.L * R0.H (FU); /* E2, E3c */
	198	A1 += R0.L * R2.H (FU); /* E3c */
	199	R0 = A1.w;
	200	A1 = A1 >> 16; /* E2c */
	201	A0 += A1; /* E2c */
	202	R1 = A0.w;
	203
	204	/* low(result) = low(E3c):low(E4) */
	205	R0 = PACK (R0.l, R3.l);
	206	/* high(result) = E2c + (E1 << 16) */
	207	R1.h = R1.h + R4.l (NS) \|\| R4 = [SP];
	208	RTS;
	209
	210	.size ___muldi3, .-___muldi3
211	#endif