1 /*===-- umodsi3.S - 32-bit unsigned integer modulus -----------------------===// 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 the __umodsi3 (32-bit unsigned integer modulus) 11 * function for the ARM 32-bit architecture. 12 * 13 *===----------------------------------------------------------------------===*/ 14 15 #include "../assembly.h" 16 17 .syntax unified 18 .text 19 20 .p2align 2 21 DEFINE_COMPILERRT_FUNCTION(__umodsi3) 22 #if __ARM_ARCH_EXT_IDIV__ 23 tst r1, r1 24 beq LOCAL_LABEL(divby0) 25 mov r3, r0 26 udiv r0, r3, r1 27 mls r1, r0, r1, r3 28 str r1, [r2] 29 bx lr 30 #else 31 cmp r1, #1 32 bcc LOCAL_LABEL(divby0) 33 moveq r0, #0 34 JMPc(lr, eq) 35 cmp r0, r1 36 JMPc(lr, cc) 37 /* 38 * Implement division using binary long division algorithm. 39 * 40 * r0 is the numerator, r1 the denominator. 41 * 42 * The code before JMP computes the correct shift I, so that 43 * r0 and (r1 << I) have the highest bit set in the same position. 44 * At the time of JMP, ip := .Ldiv0block - 8 * I. 45 * This depends on the fixed instruction size of block. 46 * 47 * block(shift) implements the test-and-update-quotient core. 48 * It assumes (r0 << shift) can be computed without overflow and 49 * that (r0 << shift) < 2 * r1. The quotient is stored in r3. 50 */ 51 52 # ifdef __ARM_FEATURE_CLZ 53 clz ip, r0 54 clz r3, r1 55 /* r0 >= r1 implies clz(r0) <= clz(r1), so ip <= r3. */ 56 sub r3, r3, ip 57 adr ip, LOCAL_LABEL(div0block) 58 sub ip, ip, r3, lsl #3 59 bx ip 60 # else 61 mov r2, r0 62 adr ip, LOCAL_LABEL(div0block) 63 64 lsr r3, r2, #16 65 cmp r3, r1 66 movhs r2, r3 67 subhs ip, ip, #(16 * 8) 68 69 lsr r3, r2, #8 70 cmp r3, r1 71 movhs r2, r3 72 subhs ip, ip, #(8 * 8) 73 74 lsr r3, r2, #4 75 cmp r3, r1 76 movhs r2, r3 77 subhs ip, #(4 * 8) 78 79 lsr r3, r2, #2 80 cmp r3, r1 81 movhs r2, r3 82 subhs ip, ip, #(2 * 8) 83 84 /* Last block, no need to update r2 or r3. */ 85 cmp r1, r2, lsr #1 86 subls ip, ip, #(1 * 8) 87 88 JMP(ip) 89 # endif 90 91 #define IMM # 92 93 #define block(shift) \ 94 cmp r0, r1, lsl IMM shift; \ 95 subhs r0, r0, r1, lsl IMM shift 96 97 block(31) 98 block(30) 99 block(29) 100 block(28) 101 block(27) 102 block(26) 103 block(25) 104 block(24) 105 block(23) 106 block(22) 107 block(21) 108 block(20) 109 block(19) 110 block(18) 111 block(17) 112 block(16) 113 block(15) 114 block(14) 115 block(13) 116 block(12) 117 block(11) 118 block(10) 119 block(9) 120 block(8) 121 block(7) 122 block(6) 123 block(5) 124 block(4) 125 block(3) 126 block(2) 127 block(1) 128 LOCAL_LABEL(div0block): 129 block(0) 130 JMP(lr) 131 #endif /* __ARM_ARCH_EXT_IDIV__ */ 132 133 LOCAL_LABEL(divby0): 134 mov r0, #0 135 #ifdef __ARM_EABI__ 136 b __aeabi_idiv0 137 #else 138 JMP(lr) 139 #endif 140 141 END_COMPILERRT_FUNCTION(__umodsi3) 142