1*3cab2bb3Spatrick//===----------------------Hexagon builtin routine ------------------------===//
2*3cab2bb3Spatrick//
3*3cab2bb3Spatrick// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4*3cab2bb3Spatrick// See https://llvm.org/LICENSE.txt for license information.
5*3cab2bb3Spatrick// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6*3cab2bb3Spatrick//
7*3cab2bb3Spatrick//===----------------------------------------------------------------------===//
8*3cab2bb3Spatrick
9*3cab2bb3Spatrick
10*3cab2bb3Spatrick	.macro FUNCTION_BEGIN name
11*3cab2bb3Spatrick	.text
12*3cab2bb3Spatrick        .p2align 5
13*3cab2bb3Spatrick	.globl \name
14*3cab2bb3Spatrick	.type  \name, @function
15*3cab2bb3Spatrick\name:
16*3cab2bb3Spatrick	.endm
17*3cab2bb3Spatrick
18*3cab2bb3Spatrick	.macro FUNCTION_END name
19*3cab2bb3Spatrick	.size  \name, . - \name
20*3cab2bb3Spatrick	.endm
21*3cab2bb3Spatrick
22*3cab2bb3Spatrick
23*3cab2bb3SpatrickFUNCTION_BEGIN __hexagon_udivdi3
24*3cab2bb3Spatrick	{
25*3cab2bb3Spatrick		r6 = cl0(r1:0)              // count leading 0's of dividend (numerator)
26*3cab2bb3Spatrick		r7 = cl0(r3:2)              // count leading 0's of divisor (denominator)
27*3cab2bb3Spatrick		r5:4 = r3:2                 // divisor moved into working registers
28*3cab2bb3Spatrick		r3:2 = r1:0                 // dividend is the initial remainder, r3:2 contains remainder
29*3cab2bb3Spatrick	}
30*3cab2bb3Spatrick	{
31*3cab2bb3Spatrick		r10 = sub(r7,r6)            // left shift count for bit & divisor
32*3cab2bb3Spatrick		r1:0 = #0                   // initialize quotient to 0
33*3cab2bb3Spatrick		r15:14 = #1                 // initialize bit to 1
34*3cab2bb3Spatrick	}
35*3cab2bb3Spatrick	{
36*3cab2bb3Spatrick		r11 = add(r10,#1)           // loop count is 1 more than shift count
37*3cab2bb3Spatrick		r13:12 = lsl(r5:4,r10)      // shift divisor msb into same bit position as dividend msb
38*3cab2bb3Spatrick		r15:14 = lsl(r15:14,r10)    // shift the bit left by same amount as divisor
39*3cab2bb3Spatrick	}
40*3cab2bb3Spatrick	{
41*3cab2bb3Spatrick		p0 = cmp.gtu(r5:4,r3:2)     // check if divisor > dividend
42*3cab2bb3Spatrick		loop0(1f,r11)               // register loop
43*3cab2bb3Spatrick	}
44*3cab2bb3Spatrick	{
45*3cab2bb3Spatrick		if (p0) jumpr r31           // if divisor > dividend, we're done, so return
46*3cab2bb3Spatrick	}
47*3cab2bb3Spatrick	.falign
48*3cab2bb3Spatrick1:
49*3cab2bb3Spatrick	{
50*3cab2bb3Spatrick		p0 = cmp.gtu(r13:12,r3:2)   // set predicate reg if shifted divisor > current remainder
51*3cab2bb3Spatrick	}
52*3cab2bb3Spatrick	{
53*3cab2bb3Spatrick		r7:6 = sub(r3:2, r13:12)    // subtract shifted divisor from current remainder
54*3cab2bb3Spatrick		r9:8 = add(r1:0, r15:14)    // save current quotient to temp (r9:8)
55*3cab2bb3Spatrick	}
56*3cab2bb3Spatrick	{
57*3cab2bb3Spatrick		r1:0 = vmux(p0, r1:0, r9:8) // choose either current quotient or new quotient (r9:8)
58*3cab2bb3Spatrick		r3:2 = vmux(p0, r3:2, r7:6) // choose either current remainder or new remainder (r7:6)
59*3cab2bb3Spatrick	}
60*3cab2bb3Spatrick	{
61*3cab2bb3Spatrick		r15:14 = lsr(r15:14, #1)    // shift bit right by 1 for next iteration
62*3cab2bb3Spatrick		r13:12 = lsr(r13:12, #1)    // shift "shifted divisor" right by 1 for next iteration
63*3cab2bb3Spatrick	}:endloop0
64*3cab2bb3Spatrick	{
65*3cab2bb3Spatrick		jumpr r31                   // return
66*3cab2bb3Spatrick	}
67*3cab2bb3SpatrickFUNCTION_END __hexagon_udivdi3
68*3cab2bb3Spatrick
69*3cab2bb3Spatrick  .globl __qdsp_udivdi3
70*3cab2bb3Spatrick  .set   __qdsp_udivdi3, __hexagon_udivdi3
71