[OpenACC] Implement 'collapse' for combined constructs.
[llvm-project.git] / compiler-rt / lib / builtins / hexagon / umoddi3.S
blob646ca128dd960f9ad8e5d18928543588ada17d18
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 //===----------------------------------------------------------------------===//
10         .macro FUNCTION_BEGIN name
11         .text
12         .p2align 5
13         .globl \name
14         .type  \name, @function
15 \name:
16         .endm
18         .macro FUNCTION_END name
19         .size  \name, . - \name
20         .endm
23 FUNCTION_BEGIN __hexagon_umoddi3
24         {
25                 r6 = cl0(r1:0)              // count leading 0's of dividend (numerator)
26                 r7 = cl0(r3:2)              // count leading 0's of divisor (denominator)
27                 r5:4 = r3:2                 // divisor moved into working registers
28                 r3:2 = r1:0                 // dividend is the initial remainder, r3:2 contains remainder
29         }
30         {
31                 r10 = sub(r7,r6)            // left shift count for bit & divisor
32                 r1:0 = #0                   // initialize quotient to 0
33                 r15:14 = #1                 // initialize bit to 1
34         }
35         {
36                 r11 = add(r10,#1)           // loop count is 1 more than shift count
37                 r13:12 = lsl(r5:4,r10)      // shift divisor msb into same bit position as dividend msb
38                 r15:14 = lsl(r15:14,r10)    // shift the bit left by same amount as divisor
39         }
40         {
41                 p0 = cmp.gtu(r5:4,r3:2)     // check if divisor > dividend
42                 loop0(1f,r11)               // register loop
43         }
44         {
45                 if (p0) jump .hexagon_umoddi3_return           // if divisor > dividend, we're done, so return
46         }
47         .falign
49         {
50                 p0 = cmp.gtu(r13:12,r3:2)   // set predicate reg if shifted divisor > current remainder
51         }
52         {
53                 r7:6 = sub(r3:2, r13:12)    // subtract shifted divisor from current remainder
54                 r9:8 = add(r1:0, r15:14)    // save current quotient to temp (r9:8)
55         }
56         {
57                 r1:0 = vmux(p0, r1:0, r9:8) // choose either current quotient or new quotient (r9:8)
58                 r3:2 = vmux(p0, r3:2, r7:6) // choose either current remainder or new remainder (r7:6)
59         }
60         {
61                 r15:14 = lsr(r15:14, #1)    // shift bit right by 1 for next iteration
62                 r13:12 = lsr(r13:12, #1)    // shift "shifted divisor" right by 1 for next iteration
63         }:endloop0
65 .hexagon_umoddi3_return:
66         {
67                 r1:0 = r3:2
68                 jumpr r31
69         }
70 FUNCTION_END __hexagon_umoddi3
72   .globl __qdsp_umoddi3
73   .set   __qdsp_umoddi3, __hexagon_umoddi3