1//===-- clzdi2.c - Implement __clzdi2 -------------------------------------===//
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// This file implements count leading zeros for 64bit arguments.
10//
11//===----------------------------------------------------------------------===//
12
13#include "../assembly.h"
14
15	.syntax unified
16	.text
17	DEFINE_CODE_STATE
18
19	.p2align	2
20DEFINE_COMPILERRT_FUNCTION(__clzdi2)
21#ifdef __ARM_FEATURE_CLZ
22#ifdef __ARMEB__
23	cmp	r0, 0
24	itee ne
25	clzne	r0, r0
26	clzeq	r0, r1
27	addeq	r0, r0, 32
28#else
29	cmp	r1, 0
30	itee ne
31	clzne	r0, r1
32	clzeq	r0, r0
33	addeq	r0, r0, 32
34#endif
35	JMP(lr)
36#else
37	// Assumption: n != 0
38
39	// r0: n
40	// r1: upper half of n, overwritten after check
41	// r1: count of leading zeros in n + 1
42	// r2: scratch register for shifted r0
43#ifdef __ARMEB__
44	cmp	r0, 0
45	moveq	r0, r1
46#else
47	cmp	r1, 0
48	movne	r0, r1
49#endif
50	movne	r1, 1
51	moveq	r1, 33
52
53	// Basic block:
54	// if ((r0 >> SHIFT) == 0)
55	//   r1 += SHIFT;
56	// else
57	//   r0 >>= SHIFT;
58	// for descending powers of two as SHIFT.
59#define BLOCK(shift) \
60	lsrs	r2, r0, shift; \
61	movne	r0, r2; \
62	addeq	r1, shift \
63
64	BLOCK(16)
65	BLOCK(8)
66	BLOCK(4)
67	BLOCK(2)
68
69	// The basic block invariants at this point are (r0 >> 2) == 0 and
70	// r0 != 0. This means 1 <= r0 <= 3 and 0 <= (r0 >> 1) <= 1.
71	//
72	// r0 | (r0 >> 1) == 0 | (r0 >> 1) == 1 | -(r0 >> 1) | 1 - (r0 >> 1)
73	// ---+----------------+----------------+------------+--------------
74	// 1  | 1              | 0              | 0          | 1
75	// 2  | 0              | 1              | -1         | 0
76	// 3  | 0              | 1              | -1         | 0
77	//
78	// The r1's initial value of 1 compensates for the 1 here.
79	sub	r0, r1, r0, lsr #1
80
81	JMP(lr)
82#endif // __ARM_FEATURE_CLZ
83END_COMPILERRT_FUNCTION(__clzdi2)
84
85NO_EXEC_STACK_DIRECTIVE
86
87