1/*
  2 * mul.S:       This routine was taken from glibc-1.09 and is covered
  3 *              by the GNU Library General Public License Version 2.
  4 */
  5
  6/*
  7 * Signed multiply, from Appendix E of the Sparc Version 8
  8 * Architecture Manual.
  9 */
 10
 11/*
 12 * Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the upper 32 bits of
 13 * the 64-bit product).
 14 *
 15 * This code optimizes short (less than 13-bit) multiplies.
 16 */
 17
 18	.globl .mul
 19	.globl _Mul
 20.mul:
 21_Mul:	/* needed for export */
 22	mov	%o0, %y		! multiplier -> Y
 23	andncc	%o0, 0xfff, %g0	! test bits 12..31
 24	be	Lmul_shortway	! if zero, can do it the short way
 25	 andcc	%g0, %g0, %o4	! zero the partial product and clear N and V
 26
 27	/*
 28	 * Long multiply.  32 steps, followed by a final shift step.
 29	 */
 30	mulscc	%o4, %o1, %o4	! 1
 31	mulscc	%o4, %o1, %o4	! 2
 32	mulscc	%o4, %o1, %o4	! 3
 33	mulscc	%o4, %o1, %o4	! 4
 34	mulscc	%o4, %o1, %o4	! 5
 35	mulscc	%o4, %o1, %o4	! 6
 36	mulscc	%o4, %o1, %o4	! 7
 37	mulscc	%o4, %o1, %o4	! 8
 38	mulscc	%o4, %o1, %o4	! 9
 39	mulscc	%o4, %o1, %o4	! 10
 40	mulscc	%o4, %o1, %o4	! 11
 41	mulscc	%o4, %o1, %o4	! 12
 42	mulscc	%o4, %o1, %o4	! 13
 43	mulscc	%o4, %o1, %o4	! 14
 44	mulscc	%o4, %o1, %o4	! 15
 45	mulscc	%o4, %o1, %o4	! 16
 46	mulscc	%o4, %o1, %o4	! 17
 47	mulscc	%o4, %o1, %o4	! 18
 48	mulscc	%o4, %o1, %o4	! 19
 49	mulscc	%o4, %o1, %o4	! 20
 50	mulscc	%o4, %o1, %o4	! 21
 51	mulscc	%o4, %o1, %o4	! 22
 52	mulscc	%o4, %o1, %o4	! 23
 53	mulscc	%o4, %o1, %o4	! 24
 54	mulscc	%o4, %o1, %o4	! 25
 55	mulscc	%o4, %o1, %o4	! 26
 56	mulscc	%o4, %o1, %o4	! 27
 57	mulscc	%o4, %o1, %o4	! 28
 58	mulscc	%o4, %o1, %o4	! 29
 59	mulscc	%o4, %o1, %o4	! 30
 60	mulscc	%o4, %o1, %o4	! 31
 61	mulscc	%o4, %o1, %o4	! 32
 62	mulscc	%o4, %g0, %o4	! final shift
 63
 64	! If %o0 was negative, the result is
 65	!	(%o0 * %o1) + (%o1 << 32))
 66	! We fix that here.
 67
 68#if 0
 69	tst	%o0
 70	bge	1f
 71	 rd	%y, %o0
 72
 73	! %o0 was indeed negative; fix upper 32 bits of result by subtracting 
 74	! %o1 (i.e., return %o4 - %o1 in %o1).
 75	retl
 76	 sub	%o4, %o1, %o1
 77
 781:
 79	retl
 80	 mov	%o4, %o1
 81#else
 82	/* Faster code adapted from tege@sics.se's code for umul.S.  */
 83	sra	%o0, 31, %o2	! make mask from sign bit
 84	and	%o1, %o2, %o2	! %o2 = 0 or %o1, depending on sign of %o0
 85	rd	%y, %o0		! get lower half of product
 86	retl
 87	 sub	%o4, %o2, %o1	! subtract compensation 
 88				!  and put upper half in place
 89#endif
 90
 91Lmul_shortway:
 92	/*
 93	 * Short multiply.  12 steps, followed by a final shift step.
 94	 * The resulting bits are off by 12 and (32-12) = 20 bit positions,
 95	 * but there is no problem with %o0 being negative (unlike above).
 96	 */
 97	mulscc	%o4, %o1, %o4	! 1
 98	mulscc	%o4, %o1, %o4	! 2
 99	mulscc	%o4, %o1, %o4	! 3
100	mulscc	%o4, %o1, %o4	! 4
101	mulscc	%o4, %o1, %o4	! 5
102	mulscc	%o4, %o1, %o4	! 6
103	mulscc	%o4, %o1, %o4	! 7
104	mulscc	%o4, %o1, %o4	! 8
105	mulscc	%o4, %o1, %o4	! 9
106	mulscc	%o4, %o1, %o4	! 10
107	mulscc	%o4, %o1, %o4	! 11
108	mulscc	%o4, %o1, %o4	! 12
109	mulscc	%o4, %g0, %o4	! final shift
110
111	/*
112	 *  %o4 has 20 of the bits that should be in the low part of the
113	 * result; %y has the bottom 12 (as %y's top 12).  That is:
114	 *
115	 *	  %o4		    %y
116	 * +----------------+----------------+
117	 * | -12- |   -20-  | -12- |   -20-  |
118	 * +------(---------+------)---------+
119	 *  --hi-- ----low-part----
120	 *
121	 * The upper 12 bits of %o4 should be sign-extended to form the
122	 * high part of the product (i.e., highpart = %o4 >> 20).
123	 */
124
125	rd	%y, %o5
126	sll	%o4, 12, %o0	! shift middle bits left 12
127	srl	%o5, 20, %o5	! shift low bits right 20, zero fill at left
128	or	%o5, %o0, %o0	! construct low part of result
129	retl
130	 sra	%o4, 20, %o1	! ... and extract high part of result
131
132	.globl	.mul_patch
133.mul_patch:
134	smul	%o0, %o1, %o0
135	retl
136	 rd	%y, %o1
137	nop