| 1 | //===-- clzsi2.c - Implement __clzsi2 -------------------------------------===// |
|---|
| 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 32bit arguments. |
|---|
| 10 | // |
|---|
| 11 | //===----------------------------------------------------------------------===// |
|---|
| 12 | |
|---|
| 13 | #include "../assembly.h" |
|---|
| 14 | |
|---|
| 15 | .syntax unified |
|---|
| 16 | .text |
|---|
| 17 | DEFINE_CODE_STATE |
|---|
| 18 | |
|---|
| 19 | .p2align 2 |
|---|
| 20 | DEFINE_COMPILERRT_FUNCTION(__clzsi2) |
|---|
| 21 | #ifdef __ARM_FEATURE_CLZ |
|---|
| 22 | clz r0, r0 |
|---|
| 23 | JMP(lr) |
|---|
| 24 | #else |
|---|
| 25 | // Assumption: n != 0 |
|---|
| 26 | |
|---|
| 27 | // r0: n |
|---|
| 28 | // r1: count of leading zeros in n + 1 |
|---|
| 29 | // r2: scratch register for shifted r0 |
|---|
| 30 | mov r1, 1 |
|---|
| 31 | |
|---|
| 32 | // Basic block: |
|---|
| 33 | // if ((r0 >> SHIFT) == 0) |
|---|
| 34 | // r1 += SHIFT; |
|---|
| 35 | // else |
|---|
| 36 | // r0 >>= SHIFT; |
|---|
| 37 | // for descending powers of two as SHIFT. |
|---|
| 38 | |
|---|
| 39 | #define BLOCK(shift) \ |
|---|
| 40 | lsrs r2, r0, shift; \ |
|---|
| 41 | movne r0, r2; \ |
|---|
| 42 | addeq r1, shift \ |
|---|
| 43 | |
|---|
| 44 | BLOCK(16) |
|---|
| 45 | BLOCK(8) |
|---|
| 46 | BLOCK(4) |
|---|
| 47 | BLOCK(2) |
|---|
| 48 | |
|---|
| 49 | // The basic block invariants at this point are (r0 >> 2) == 0 and |
|---|
| 50 | // r0 != 0. This means 1 <= r0 <= 3 and 0 <= (r0 >> 1) <= 1. |
|---|
| 51 | // |
|---|
| 52 | // r0 | (r0 >> 1) == 0 | (r0 >> 1) == 1 | -(r0 >> 1) | 1 - (r0 >> 1) |
|---|
| 53 | // ---+----------------+----------------+------------+-------------- |
|---|
| 54 | // 1 | 1 | 0 | 0 | 1 |
|---|
| 55 | // 2 | 0 | 1 | -1 | 0 |
|---|
| 56 | // 3 | 0 | 1 | -1 | 0 |
|---|
| 57 | // |
|---|
| 58 | // The r1's initial value of 1 compensates for the 1 here. |
|---|
| 59 | sub r0, r1, r0, lsr #1 |
|---|
| 60 | |
|---|
| 61 | JMP(lr) |
|---|
| 62 | #endif // __ARM_FEATURE_CLZ |
|---|
| 63 | END_COMPILERRT_FUNCTION(__clzsi2) |
|---|
| 64 | |
|---|
| 65 | NO_EXEC_STACK_DIRECTIVE |
|---|
| 66 | |
|---|
| 67 | |
|---|
