Home | History | Annotate | Download | only in arm
      1 /* ===-- clzsi2.c - Implement __clzsi2 -------------------------------------===
      2  *
      3  *               The LLVM Compiler Infrastructure
      4  *
      5  * This file is dual licensed under the MIT and the University of Illinois Open
      6  * Source Licenses. See LICENSE.TXT for details.
      7  *
      8  * ===----------------------------------------------------------------------===
      9  *
     10  * This file implements count leading zeros for 32bit arguments.
     11  *
     12  * ===----------------------------------------------------------------------===
     13  */
     14 #include "../assembly.h"
     15 
     16 	.syntax unified
     17 	.text
     18 #if __ARM_ARCH_ISA_THUMB == 2
     19 	.thumb
     20 #endif
     21 
     22 	.p2align	2
     23 DEFINE_COMPILERRT_FUNCTION(__clzsi2)
     24 #ifdef __ARM_FEATURE_CLZ
     25 	clz	r0, r0
     26 	JMP(lr)
     27 #else
     28 	/* Assumption: n != 0 */
     29 
     30 	/*
     31 	 * r0: n
     32 	 * r1: count of leading zeros in n + 1
     33 	 * r2: scratch register for shifted r0
     34 	 */
     35 	mov	r1, 1
     36 
     37 	/*
     38 	 * Basic block:
     39 	 * if ((r0 >> SHIFT) == 0)
     40 	 *   r1 += SHIFT;
     41 	 * else
     42 	 *   r0 >>= SHIFT;
     43 	 * for descending powers of two as SHIFT.
     44 	 */
     45 
     46 #define BLOCK(shift) \
     47 	lsrs	r2, r0, shift; \
     48 	movne	r0, r2; \
     49 	addeq	r1, shift \
     50 
     51 	BLOCK(16)
     52 	BLOCK(8)
     53 	BLOCK(4)
     54 	BLOCK(2)
     55 
     56 	/*
     57 	 * The basic block invariants at this point are (r0 >> 2) == 0 and
     58 	 * r0 != 0. This means 1 <= r0 <= 3 and 0 <= (r0 >> 1) <= 1.
     59 	 *
     60 	 * r0 | (r0 >> 1) == 0 | (r0 >> 1) == 1 | -(r0 >> 1) | 1 - (r0 >> 1)
     61 	 * ---+----------------+----------------+------------+--------------
     62 	 * 1  | 1              | 0              | 0          | 1
     63 	 * 2  | 0              | 1              | -1         | 0
     64 	 * 3  | 0              | 1              | -1         | 0
     65 	 *
     66 	 * The r1's initial value of 1 compensates for the 1 here.
     67 	 */
     68 	sub	r0, r1, r0, lsr #1
     69 
     70 	JMP(lr)
     71 #endif // __ARM_FEATURE_CLZ
     72 END_COMPILERRT_FUNCTION(__clzsi2)
     73