1 | //===----------------------Hexagon builtin routine ------------------------===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | |
9 | .macro FUNCTION_BEGIN name |
10 | .text |
11 | .p2align 5 |
12 | .globl \name |
13 | .type \name, @function |
14 | \name: |
15 | .endm |
16 | |
17 | .macro FUNCTION_END name |
18 | .size \name, . - \name |
19 | .endm |
20 | |
21 | |
22 | FUNCTION_BEGIN __hexagon_divdi3 |
23 | { |
24 | p2 = tstbit(r1,#31) |
25 | p3 = tstbit(r3,#31) |
26 | } |
27 | { |
28 | r1:0 = abs(r1:0) |
29 | r3:2 = abs(r3:2) |
30 | } |
31 | { |
32 | r6 = cl0(r1:0) // count leading 0's of dividend (numerator) |
33 | r7 = cl0(r3:2) // count leading 0's of divisor (denominator) |
34 | r5:4 = r3:2 // divisor moved into working registers |
35 | r3:2 = r1:0 // dividend is the initial remainder, r3:2 contains remainder |
36 | } |
37 | { |
38 | p3 = xor(p2,p3) |
39 | r10 = sub(r7,r6) // left shift count for bit & divisor |
40 | r1:0 = #0 // initialize quotient to 0 |
41 | r15:14 = #1 // initialize bit to 1 |
42 | } |
43 | { |
44 | r11 = add(r10,#1) // loop count is 1 more than shift count |
45 | r13:12 = lsl(r5:4,r10) // shift divisor msb into same bit position as dividend msb |
46 | r15:14 = lsl(r15:14,r10) // shift the bit left by same amount as divisor |
47 | } |
48 | { |
49 | p0 = cmp.gtu(r5:4,r3:2) // check if divisor > dividend |
50 | loop0(1f,r11) // register loop |
51 | } |
52 | { |
53 | if (p0) jump .hexagon_divdi3_return // if divisor > dividend, we're done, so return |
54 | } |
55 | .falign |
56 | 1: |
57 | { |
58 | p0 = cmp.gtu(r13:12,r3:2) // set predicate reg if shifted divisor > current remainder |
59 | } |
60 | { |
61 | r7:6 = sub(r3:2, r13:12) // subtract shifted divisor from current remainder |
62 | r9:8 = add(r1:0, r15:14) // save current quotient to temp (r9:8) |
63 | } |
64 | { |
65 | r1:0 = vmux(p0, r1:0, r9:8) // choose either current quotient or new quotient (r9:8) |
66 | r3:2 = vmux(p0, r3:2, r7:6) // choose either current remainder or new remainder (r7:6) |
67 | } |
68 | { |
69 | r15:14 = lsr(r15:14, #1) // shift bit right by 1 for next iteration |
70 | r13:12 = lsr(r13:12, #1) // shift "shifted divisor" right by 1 for next iteration |
71 | }:endloop0 |
72 | |
73 | .hexagon_divdi3_return: |
74 | { |
75 | r3:2 = neg(r1:0) |
76 | } |
77 | { |
78 | r1:0 = vmux(p3,r3:2,r1:0) |
79 | jumpr r31 |
80 | } |
81 | FUNCTION_END __hexagon_divdi3 |
82 | |
83 | .globl __qdsp_divdi3 |
84 | .set __qdsp_divdi3, __hexagon_divdi3 |
85 | |