clzsi2.S

//===-- clzsi2.c - Implement __clzsi2 -------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements count leading zeros for 32bit arguments.
//
//===----------------------------------------------------------------------===//

#include "../assembly.h"

	.syntax unified
	.text
	DEFINE_CODE_STATE

	.p2align	2
DEFINE_COMPILERRT_FUNCTION(__clzsi2)
#ifdef __ARM_FEATURE_CLZ
	clz	r0, r0
	JMP(lr)
#else
	// Assumption: n != 0

	// r0: n
	// r1: count of leading zeros in n + 1
	// r2: scratch register for shifted r0
	mov	r1, 1

	// Basic block:
	// if ((r0 >> SHIFT) == 0)
	//   r1 += SHIFT;
	// else
	//   r0 >>= SHIFT;
	// for descending powers of two as SHIFT.

#define BLOCK(shift) \
	lsrs	r2, r0, shift; \
	movne	r0, r2; \
	addeq	r1, shift \

	BLOCK(16)
	BLOCK(8)
	BLOCK(4)
	BLOCK(2)

	// The basic block invariants at this point are (r0 >> 2) == 0 and
	// r0 != 0. This means 1 <= r0 <= 3 and 0 <= (r0 >> 1) <= 1.
	//
	// r0 | (r0 >> 1) == 0 | (r0 >> 1) == 1 | -(r0 >> 1) | 1 - (r0 >> 1)
	// ---+----------------+----------------+------------+--------------
	// 1  | 1              | 0              | 0          | 1
	// 2  | 0              | 1              | -1         | 0
	// 3  | 0              | 1              | -1         | 0
	//
	// The r1's initial value of 1 compensates for the 1 here.
	sub	r0, r1, r0, lsr #1

	JMP(lr)
#endif // __ARM_FEATURE_CLZ
END_COMPILERRT_FUNCTION(__clzsi2)

NO_EXEC_STACK_DIRECTIVE