xref: /freebsd/contrib/llvm-project/compiler-rt/lib/builtins/hexagon/sfsqrt_opt.S (revision e8d8bef961a50d4dc22501cde4fb9fb0be1b2532)
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#define RIN r0
22#define S r0
23#define H r1
24#define D r2
25#define E r3
26#define HALF r4
27#define R r5
28
29#define Q6_ALIAS(TAG) .global __qdsp_##TAG ; .set __qdsp_##TAG, __hexagon_##TAG
30#define FAST_ALIAS(TAG) .global __hexagon_fast_##TAG ; .set __hexagon_fast_##TAG, __hexagon_##TAG
31#define FAST2_ALIAS(TAG) .global __hexagon_fast2_##TAG ; .set __hexagon_fast2_##TAG, __hexagon_##TAG
32
33FUNCTION_BEGIN __hexagon_sqrtf
34  {
35    E,p0 = sfinvsqrta(RIN)
36    R = sffixupr(RIN)
37    HALF = ##0x3f000000   // 0.5
38    r1:0 = combine(#0,#0)   // clear S/H
39  }
40  {
41    S += sfmpy(E,R):lib   // S0
42    H += sfmpy(E,HALF):lib    // H0
43    D = HALF
44    E = R
45  }
46  {
47    D -= sfmpy(S,H):lib   // d0
48    p1 = sfclass(R,#1)    // is zero?
49    //E -= sfmpy(S,S):lib   // e0
50  }
51  {
52    S += sfmpy(S,D):lib   // S1
53    H += sfmpy(H,D):lib   // H1
54    D = HALF
55    E = R
56  }
57  {
58    D -= sfmpy(S,H):lib   // d0
59    E -= sfmpy(S,S):lib   // e0
60  }
61  {
62    S += sfmpy(H,E):lib   // S2
63    H += sfmpy(H,D):lib   // H2
64    D = HALF
65    E = R
66  }
67  {
68    //D -= sfmpy(S,H):lib   // d2
69    E -= sfmpy(S,S):lib   // e2
70    if (p1) r0 = or(r0,R)     // sqrt(-0.0) = -0.0
71  }
72  {
73    S += sfmpy(H,E,p0):scale  // S3
74    jumpr r31
75  }
76
77FUNCTION_END __hexagon_sqrtf
78
79Q6_ALIAS(sqrtf)
80FAST_ALIAS(sqrtf)
81FAST2_ALIAS(sqrtf)
82