| 1 | //===-- comparesf2.S - Implement single-precision soft-float comparisons --===// |
| 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 the following soft-fp_t comparison routines: |
| 10 | // |
| 11 | // __eqsf2 __gesf2 __unordsf2 |
| 12 | // __lesf2 __gtsf2 |
| 13 | // __ltsf2 |
| 14 | // __nesf2 |
| 15 | // |
| 16 | // The semantics of the routines grouped in each column are identical, so there |
| 17 | // is a single implementation for each, with multiple names. |
| 18 | // |
| 19 | // The routines behave as follows: |
| 20 | // |
| 21 | // __lesf2(a,b) returns -1 if a < b |
| 22 | // 0 if a == b |
| 23 | // 1 if a > b |
| 24 | // 1 if either a or b is NaN |
| 25 | // |
| 26 | // __gesf2(a,b) returns -1 if a < b |
| 27 | // 0 if a == b |
| 28 | // 1 if a > b |
| 29 | // -1 if either a or b is NaN |
| 30 | // |
| 31 | // __unordsf2(a,b) returns 0 if both a and b are numbers |
| 32 | // 1 if either a or b is NaN |
| 33 | // |
| 34 | // Note that __lesf2( ) and __gesf2( ) are identical except in their handling of |
| 35 | // NaN values. |
| 36 | // |
| 37 | //===----------------------------------------------------------------------===// |
| 38 | |
| 39 | #include "../assembly.h" |
| 40 | |
| 41 | .syntax unified |
| 42 | .text |
| 43 | DEFINE_CODE_STATE |
| 44 | |
| 45 | .macro COMPARESF2_FUNCTION_BODY handle_nan:req |
| 46 | #if defined(COMPILER_RT_ARMHF_TARGET) |
| 47 | vmov r0, s0 |
| 48 | vmov r1, s1 |
| 49 | #endif |
| 50 | // Make copies of a and b with the sign bit shifted off the top. These will |
| 51 | // be used to detect zeros and NaNs. |
| 52 | #if defined(USE_THUMB_1) |
| 53 | push {r6, lr} |
| 54 | lsls r2, r0, #1 |
| 55 | lsls r3, r1, #1 |
| 56 | #else |
| 57 | mov r2, r0, lsl #1 |
| 58 | mov r3, r1, lsl #1 |
| 59 | #endif |
| 60 | |
| 61 | // We do the comparison in three stages (ignoring NaN values for the time |
| 62 | // being). First, we orr the absolute values of a and b; this sets the Z |
| 63 | // flag if both a and b are zero (of either sign). The shift of r3 doesn't |
| 64 | // effect this at all, but it *does* make sure that the C flag is clear for |
| 65 | // the subsequent operations. |
| 66 | #if defined(USE_THUMB_1) |
| 67 | lsrs r6, r3, #1 |
| 68 | orrs r6, r2 |
| 69 | #else |
| 70 | orrs r12, r2, r3, lsr #1 |
| 71 | #endif |
| 72 | // Next, we check if a and b have the same or different signs. If they have |
| 73 | // opposite signs, this eor will set the N flag. |
| 74 | #if defined(USE_THUMB_1) |
| 75 | beq 1f |
| 76 | movs r6, r0 |
| 77 | eors r6, r1 |
| 78 | 1: |
| 79 | #else |
| 80 | it ne |
| 81 | eorsne r12, r0, r1 |
| 82 | #endif |
| 83 | |
| 84 | // If a and b are equal (either both zeros or bit identical; again, we're |
| 85 | // ignoring NaNs for now), this subtract will zero out r0. If they have the |
| 86 | // same sign, the flags are updated as they would be for a comparison of the |
| 87 | // absolute values of a and b. |
| 88 | #if defined(USE_THUMB_1) |
| 89 | bmi 1f |
| 90 | subs r0, r2, r3 |
| 91 | 1: |
| 92 | #else |
| 93 | it pl |
| 94 | subspl r0, r2, r3 |
| 95 | #endif |
| 96 | |
| 97 | // If a is smaller in magnitude than b and both have the same sign, place |
| 98 | // the negation of the sign of b in r0. Thus, if both are negative and |
| 99 | // a > b, this sets r0 to 0; if both are positive and a < b, this sets |
| 100 | // r0 to -1. |
| 101 | // |
| 102 | // This is also done if a and b have opposite signs and are not both zero, |
| 103 | // because in that case the subtract was not performed and the C flag is |
| 104 | // still clear from the shift argument in orrs; if a is positive and b |
| 105 | // negative, this places 0 in r0; if a is negative and b positive, -1 is |
| 106 | // placed in r0. |
| 107 | #if defined(USE_THUMB_1) |
| 108 | bhs 1f |
| 109 | // Here if a and b have the same sign and absA < absB, the result is thus |
| 110 | // b < 0 ? 1 : -1. Same if a and b have the opposite sign (ignoring Nan). |
| 111 | movs r0, #1 |
| 112 | lsrs r1, #31 |
| 113 | bne LOCAL_LABEL(CHECK_NAN\@) |
| 114 | negs r0, r0 |
| 115 | b LOCAL_LABEL(CHECK_NAN\@) |
| 116 | 1: |
| 117 | #else |
| 118 | it lo |
| 119 | mvnlo r0, r1, asr #31 |
| 120 | #endif |
| 121 | |
| 122 | // If a is greater in magnitude than b and both have the same sign, place |
| 123 | // the sign of b in r0. Thus, if both are negative and a < b, -1 is placed |
| 124 | // in r0, which is the desired result. Conversely, if both are positive |
| 125 | // and a > b, zero is placed in r0. |
| 126 | #if defined(USE_THUMB_1) |
| 127 | bls 1f |
| 128 | // Here both have the same sign and absA > absB. |
| 129 | movs r0, #1 |
| 130 | lsrs r1, #31 |
| 131 | beq LOCAL_LABEL(CHECK_NAN\@) |
| 132 | negs r0, r0 |
| 133 | 1: |
| 134 | #else |
| 135 | it hi |
| 136 | movhi r0, r1, asr #31 |
| 137 | #endif |
| 138 | |
| 139 | // If you've been keeping track, at this point r0 contains -1 if a < b and |
| 140 | // 0 if a >= b. All that remains to be done is to set it to 1 if a > b. |
| 141 | // If a == b, then the Z flag is set, so we can get the correct final value |
| 142 | // into r0 by simply or'ing with 1 if Z is clear. |
| 143 | // For Thumb-1, r0 contains -1 if a < b, 0 if a > b and 0 if a == b. |
| 144 | #if !defined(USE_THUMB_1) |
| 145 | it ne |
| 146 | orrne r0, r0, #1 |
| 147 | #endif |
| 148 | |
| 149 | // Finally, we need to deal with NaNs. If either argument is NaN, replace |
| 150 | // the value in r0 with 1. |
| 151 | #if defined(USE_THUMB_1) |
| 152 | LOCAL_LABEL(CHECK_NAN\@): |
| 153 | movs r6, #0xff |
| 154 | lsls r6, #24 |
| 155 | cmp r2, r6 |
| 156 | bhi 1f |
| 157 | cmp r3, r6 |
| 158 | 1: |
| 159 | bls 2f |
| 160 | \handle_nan |
| 161 | 2: |
| 162 | pop {r6, pc} |
| 163 | #else |
| 164 | cmp r2, #0xff000000 |
| 165 | ite ls |
| 166 | cmpls r3, #0xff000000 |
| 167 | \handle_nan |
| 168 | JMP(lr) |
| 169 | #endif |
| 170 | .endm |
| 171 | |
| 172 | @ int __eqsf2(float a, float b) |
| 173 | |
| 174 | .p2align 2 |
| 175 | DEFINE_COMPILERRT_FUNCTION(__eqsf2) |
| 176 | |
| 177 | .macro __eqsf2_handle_nan |
| 178 | #if defined(USE_THUMB_1) |
| 179 | movs r0, #1 |
| 180 | #else |
| 181 | movhi r0, #1 |
| 182 | #endif |
| 183 | .endm |
| 184 | |
| 185 | COMPARESF2_FUNCTION_BODY __eqsf2_handle_nan |
| 186 | |
| 187 | END_COMPILERRT_FUNCTION(__eqsf2) |
| 188 | |
| 189 | DEFINE_COMPILERRT_FUNCTION_ALIAS(__lesf2, __eqsf2) |
| 190 | DEFINE_COMPILERRT_FUNCTION_ALIAS(__ltsf2, __eqsf2) |
| 191 | DEFINE_COMPILERRT_FUNCTION_ALIAS(__nesf2, __eqsf2) |
| 192 | |
| 193 | #if defined(__ELF__) |
| 194 | // Alias for libgcc compatibility |
| 195 | DEFINE_COMPILERRT_FUNCTION_ALIAS(__cmpsf2, __lesf2) |
| 196 | #endif |
| 197 | |
| 198 | @ int __gtsf2(float a, float b) |
| 199 | |
| 200 | .p2align 2 |
| 201 | DEFINE_COMPILERRT_FUNCTION(__gtsf2) |
| 202 | |
| 203 | .macro __gtsf2_handle_nan |
| 204 | #if defined(USE_THUMB_1) |
| 205 | movs r0, #1 |
| 206 | negs r0, r0 |
| 207 | #else |
| 208 | movhi r0, #-1 |
| 209 | #endif |
| 210 | .endm |
| 211 | |
| 212 | COMPARESF2_FUNCTION_BODY __gtsf2_handle_nan |
| 213 | |
| 214 | END_COMPILERRT_FUNCTION(__gtsf2) |
| 215 | |
| 216 | DEFINE_COMPILERRT_FUNCTION_ALIAS(__gesf2, __gtsf2) |
| 217 | |
| 218 | @ int __unordsf2(float a, float b) |
| 219 | |
| 220 | .p2align 2 |
| 221 | DEFINE_COMPILERRT_FUNCTION(__unordsf2) |
| 222 | |
| 223 | #if defined(COMPILER_RT_ARMHF_TARGET) |
| 224 | vmov r0, s0 |
| 225 | vmov r1, s1 |
| 226 | #endif |
| 227 | // Return 1 for NaN values, 0 otherwise. |
| 228 | lsls r2, r0, #1 |
| 229 | lsls r3, r1, #1 |
| 230 | movs r0, #0 |
| 231 | #if defined(USE_THUMB_1) |
| 232 | movs r1, #0xff |
| 233 | lsls r1, #24 |
| 234 | cmp r2, r1 |
| 235 | bhi 1f |
| 236 | cmp r3, r1 |
| 237 | 1: |
| 238 | bls 2f |
| 239 | movs r0, #1 |
| 240 | 2: |
| 241 | #else |
| 242 | cmp r2, #0xff000000 |
| 243 | ite ls |
| 244 | cmpls r3, #0xff000000 |
| 245 | movhi r0, #1 |
| 246 | #endif |
| 247 | JMP(lr) |
| 248 | END_COMPILERRT_FUNCTION(__unordsf2) |
| 249 | |
| 250 | #if defined(COMPILER_RT_ARMHF_TARGET) |
| 251 | DEFINE_COMPILERRT_FUNCTION(__aeabi_fcmpun) |
| 252 | vmov s0, r0 |
| 253 | vmov s1, r1 |
| 254 | b SYMBOL_NAME(__unordsf2) |
| 255 | END_COMPILERRT_FUNCTION(__aeabi_fcmpun) |
| 256 | #else |
| 257 | DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_fcmpun, __unordsf2) |
| 258 | #endif |
| 259 | |
| 260 | NO_EXEC_STACK_DIRECTIVE |
| 261 | |
| 262 | |