OLD | NEW |
| (Empty) |
1 dnl Alpha mpn_divrem_2 -- Divide an mpn number by a normalized 2-limb number. | |
2 | |
3 dnl Copyright 2007, 2008 Free Software Foundation, Inc. | |
4 | |
5 dnl This file is part of the GNU MP Library. | |
6 | |
7 dnl The GNU MP Library is free software; you can redistribute it and/or modify | |
8 dnl it under the terms of the GNU Lesser General Public License as published | |
9 dnl by the Free Software Foundation; either version 3 of the License, or (at | |
10 dnl your option) any later version. | |
11 | |
12 dnl The GNU MP Library is distributed in the hope that it will be useful, but | |
13 dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY | |
14 dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public | |
15 dnl License for more details. | |
16 | |
17 dnl You should have received a copy of the GNU Lesser General Public License | |
18 dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/. | |
19 | |
20 include(`../config.m4') | |
21 | |
22 C norm frac | |
23 C ev4 | |
24 C ev5 70 70 | |
25 C ev6 29 29 | |
26 | |
27 C TODO | |
28 C * Perhaps inline mpn_invert_limb, that would allow us to not save/restore | |
29 C any registers (thus save ~10 cycles per call). | |
30 C * Use negated d1 and/or d0 to speed carry propagation. Might save a cycle | |
31 C or two. | |
32 C * Check cluster delays (for ev6). We very likely could save some cycles. | |
33 C * Use branch-free code for computing di. | |
34 C * CAVEAT: We rely on r19 not being clobbered by mpn_invert_limb call. | |
35 | |
36 C INPUT PARAMETERS | |
37 define(`qp', `r16') | |
38 define(`fn', `r17') | |
39 define(`up_param', `r18') | |
40 define(`un_param', `r19') | |
41 define(`dp', `r20') | |
42 | |
43 ASM_START() | |
44 PROLOGUE(mpn_divrem_2) | |
45 ldgp r29, 0(r27) | |
46 lda r30, -80(r30) | |
47 stq r26, 0(r30) | |
48 stq r9, 8(r30) | |
49 stq r10, 16(r30) | |
50 stq r11, 24(r30) | |
51 stq r12, 32(r30) | |
52 stq r13, 40(r30) | |
53 C stq r14, 48(r30) | |
54 stq r15, 56(r30) | |
55 .prologue 1 | |
56 stq r16, 64(r30) | |
57 bis r31, r17, r15 | |
58 s8addq r19, r18, r13 | |
59 lda r13, -24(r13) | |
60 ldq r12, 8(r20) | |
61 ldq r10, 0(r20) | |
62 ldq r11, 16(r13) | |
63 ldq r9, 8(r13) | |
64 | |
65 bis r31, r31, r3 C most_significant_q_limb = 0 | |
66 cmpult r11, r12, r1 | |
67 bne r1, L(L8) | |
68 cmpule r11, r12, r1 | |
69 cmpult r9, r10, r2 | |
70 and r1, r2, r1 | |
71 bne r1, L(L8) | |
72 subq r11, r12, r11 | |
73 subq r11, r2, r11 | |
74 subq r9, r10, r9 | |
75 lda r3, 1(r31) C most_significant_q_limb = 1 | |
76 L(L8): stq r3, 72(r30) | |
77 | |
78 addq r15, r19, r19 | |
79 lda r19, -3(r19) | |
80 blt r19, L(L10) | |
81 bis r31, r12, r16 | |
82 jsr r26, mpn_invert_limb | |
83 ldgp r29, 0(r26) | |
84 mulq r0, r12, r4 C t0 = LO(di * d1) | |
85 umulh r0, r10, r2 C s1 = HI(di * d0) | |
86 addq r4, r10, r4 C t0 += d0 | |
87 cmpule r10, r4, r7 C (t0 < d0) | |
88 addq r4, r2, r4 C t0 += s1 | |
89 cmpult r4, r2, r1 | |
90 subq r1, r7, r7 C t1 (-1, 0, or 1) | |
91 blt r7, L(L42) | |
92 L(L22): | |
93 lda r0, -1(r0) C di-- | |
94 cmpult r4, r12, r1 C cy for: t0 -= d1 (below) | |
95 subq r7, r1, r7 C t1 -= cy | |
96 subq r4, r12, r4 C t0 -= d1 | |
97 bge r7, L(L22) | |
98 L(L42): | |
99 ldq r16, 64(r30) | |
100 s8addq r19, r16, r16 | |
101 ALIGN(16) | |
102 L(loop): | |
103 mulq r11, r0, r5 C q0 (early) | |
104 umulh r11, r0, r6 C q (early) | |
105 addq r5, r9, r8 C q0 += n1 | |
106 addq r6, r11, r6 C q += n2 | |
107 cmpult r8, r5, r1 C cy for: q0 += n1 | |
108 addq r6, r1, r6 C q += cy | |
109 unop | |
110 mulq r12, r6, r1 C LO(d1 * q) | |
111 umulh r10, r6, r7 C t1 = HI(d0 * q) | |
112 subq r9, r1, r9 C n1 -= LO(d1 * q) | |
113 mulq r10, r6, r4 C t0 = LO(d0 * q) | |
114 unop | |
115 cmple r15, r19, r5 C condition and n0... | |
116 beq r5, L(L31) | |
117 ldq r5, 0(r13) | |
118 lda r13, -8(r13) | |
119 L(L31): subq r9, r12, r9 C n1 -= d1 | |
120 cmpult r5, r10, r1 C | |
121 subq r9, r1, r9 C | |
122 subq r5, r10, r5 C n0 -= d0 | |
123 subq r9, r7, r9 C n1 -= t0 | |
124 cmpult r5, r4, r1 C | |
125 subq r9, r1, r2 C | |
126 subq r5, r4, r5 C n0 -= t1 | |
127 cmpult r2, r8, r1 C (n1 < q0) | |
128 addq r6, r1, r6 C q += cond | |
129 lda r1, -1(r1) C -(n1 >= q0) | |
130 and r1, r10, r4 C | |
131 addq r5, r4, r9 C n0 += mask & d0 | |
132 and r1, r12, r1 C | |
133 cmpult r9, r5, r11 C cy for: n0 += mask & d0 | |
134 addq r2, r1, r1 C n1 += mask & d1 | |
135 addq r1, r11, r11 C n1 += cy | |
136 cmpult r11, r12, r1 C | |
137 beq r1, L(fix) C | |
138 L(bck): stq r6, 0(r16) | |
139 lda r16, -8(r16) | |
140 lda r19, -1(r19) | |
141 bge r19, L(loop) | |
142 | |
143 L(L10): stq r9, 8(r13) | |
144 stq r11, 16(r13) | |
145 ldq r0, 72(r30) | |
146 ldq r26, 0(r30) | |
147 ldq r9, 8(r30) | |
148 ldq r10, 16(r30) | |
149 ldq r11, 24(r30) | |
150 ldq r12, 32(r30) | |
151 ldq r13, 40(r30) | |
152 C ldq r14, 48(r30) | |
153 ldq r15, 56(r30) | |
154 lda r30, 80(r30) | |
155 ret r31, (r26), 1 | |
156 | |
157 L(fix): cmpule r11, r12, r1 | |
158 cmpult r9, r10, r2 | |
159 and r1, r2, r1 | |
160 bne r1, L(bck) | |
161 subq r11, r12, r11 | |
162 subq r11, r2, r11 | |
163 subq r9, r10, r9 | |
164 lda r6, 1(r6) | |
165 br L(bck) | |
166 EPILOGUE() | |
167 ASM_END() | |
OLD | NEW |