|
| 1 | +// Compiler runtime builtins needed for cross-compilation from macOS to Linux. |
| 2 | +// When cross-compiling with clang, 128-bit integer operations (e.g. in mbedtls bignum.c) |
| 3 | +// generate calls to __udivti3/__umodti3 which are normally provided by libgcc or compiler-rt. |
| 4 | +// Since the linuxroot sysroot doesn't include these libraries, we provide minimal |
| 5 | +// implementations here. |
| 6 | + |
| 7 | +#ifdef __SIZEOF_INT128__ |
| 8 | + |
| 9 | +typedef unsigned __int128 tu_int; |
| 10 | +typedef unsigned long long du_int; |
| 11 | + |
| 12 | +typedef union { |
| 13 | + tu_int all; |
| 14 | + struct { |
| 15 | +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
| 16 | + du_int low; |
| 17 | + du_int high; |
| 18 | +#else |
| 19 | + du_int high; |
| 20 | + du_int low; |
| 21 | +#endif |
| 22 | + } s; |
| 23 | +} utwords; |
| 24 | + |
| 25 | +static inline unsigned __clzdi2(du_int a) { |
| 26 | + unsigned n = 0; |
| 27 | + if (a == 0) return 64; |
| 28 | + if ((a & 0xFFFFFFFF00000000ULL) == 0) { n += 32; a <<= 32; } |
| 29 | + if ((a & 0xFFFF000000000000ULL) == 0) { n += 16; a <<= 16; } |
| 30 | + if ((a & 0xFF00000000000000ULL) == 0) { n += 8; a <<= 8; } |
| 31 | + if ((a & 0xF000000000000000ULL) == 0) { n += 4; a <<= 4; } |
| 32 | + if ((a & 0xC000000000000000ULL) == 0) { n += 2; a <<= 2; } |
| 33 | + if ((a & 0x8000000000000000ULL) == 0) { n += 1; } |
| 34 | + return n; |
| 35 | +} |
| 36 | + |
| 37 | +tu_int __udivmodti4(tu_int a, tu_int b, tu_int *rem) { |
| 38 | + const unsigned n_udword_bits = sizeof(du_int) * 8; |
| 39 | + const unsigned n_utword_bits = sizeof(tu_int) * 8; |
| 40 | + utwords n, d, q, r; |
| 41 | + unsigned sr; |
| 42 | + n.all = a; |
| 43 | + d.all = b; |
| 44 | + |
| 45 | + // Special cases |
| 46 | + if (n.s.high == 0) { |
| 47 | + if (d.s.high == 0) { |
| 48 | + if (rem) *rem = n.s.low % d.s.low; |
| 49 | + return n.s.low / d.s.low; |
| 50 | + } |
| 51 | + if (rem) *rem = n.s.low; |
| 52 | + return 0; |
| 53 | + } |
| 54 | + // n.s.high != 0 |
| 55 | + if (d.s.low == 0) { |
| 56 | + if (d.s.high == 0) { |
| 57 | + // Division by zero (undefined behavior, just return 0) |
| 58 | + if (rem) *rem = 0; |
| 59 | + return 0; |
| 60 | + } |
| 61 | + if (n.s.low == 0) { |
| 62 | + if (rem) { |
| 63 | + r.s.high = n.s.high % d.s.high; |
| 64 | + r.s.low = 0; |
| 65 | + *rem = r.all; |
| 66 | + } |
| 67 | + return n.s.high / d.s.high; |
| 68 | + } |
| 69 | + // n.s.high != 0 && n.s.low != 0 && d.s.low == 0 && d.s.high != 0 |
| 70 | + if ((d.s.high & (d.s.high - 1)) == 0) { |
| 71 | + // d is a power of 2 |
| 72 | + if (rem) { |
| 73 | + r.s.low = n.s.low; |
| 74 | + r.s.high = n.s.high & (d.s.high - 1); |
| 75 | + *rem = r.all; |
| 76 | + } |
| 77 | + return n.s.high >> __clzdi2(d.s.high); |
| 78 | + } |
| 79 | + sr = __clzdi2(d.s.high) - __clzdi2(n.s.high); |
| 80 | + if (sr > n_udword_bits - 2) { |
| 81 | + if (rem) *rem = n.all; |
| 82 | + return 0; |
| 83 | + } |
| 84 | + ++sr; |
| 85 | + // 1 <= sr <= n_udword_bits - 1 |
| 86 | + q.s.low = 0; |
| 87 | + q.s.high = n.s.low << (n_udword_bits - sr); |
| 88 | + r.s.high = n.s.high >> sr; |
| 89 | + r.s.low = (n.s.high << (n_udword_bits - sr)) | (n.s.low >> sr); |
| 90 | + goto shift_loop; |
| 91 | + } |
| 92 | + // d.s.low != 0 |
| 93 | + if (d.s.high == 0) { |
| 94 | + if ((d.s.low & (d.s.low - 1)) == 0) { |
| 95 | + if (rem) *rem = n.s.low & (d.s.low - 1); |
| 96 | + if (d.s.low == 1) return n.all; |
| 97 | + sr = __builtin_ctzll(d.s.low); |
| 98 | + q.s.high = n.s.high >> sr; |
| 99 | + q.s.low = (n.s.high << (n_udword_bits - sr)) | (n.s.low >> sr); |
| 100 | + return q.all; |
| 101 | + } |
| 102 | + sr = 1 + n_udword_bits + __clzdi2(d.s.low) - __clzdi2(n.s.high); |
| 103 | + if (sr == n_udword_bits) { |
| 104 | + q.s.low = 0; |
| 105 | + q.s.high = 0; |
| 106 | + r.s.high = 0; |
| 107 | + r.s.low = n.s.high; |
| 108 | + } else if (sr < n_udword_bits) { |
| 109 | + q.s.low = 0; |
| 110 | + q.s.high = n.s.low << (n_udword_bits - sr); |
| 111 | + r.s.high = n.s.high >> sr; |
| 112 | + r.s.low = (n.s.high << (n_udword_bits - sr)) | (n.s.low >> sr); |
| 113 | + } else { |
| 114 | + q.s.low = n.s.low << (n_utword_bits - sr); |
| 115 | + q.s.high = (n.s.high << (n_utword_bits - sr)) | (n.s.low >> (sr - n_udword_bits)); |
| 116 | + r.s.high = 0; |
| 117 | + r.s.low = n.s.high >> (sr - n_udword_bits); |
| 118 | + } |
| 119 | + goto shift_loop; |
| 120 | + } |
| 121 | + // d.s.high != 0 && d.s.low != 0 |
| 122 | + sr = __clzdi2(d.s.high) - __clzdi2(n.s.high); |
| 123 | + if (sr > n_udword_bits - 1) { |
| 124 | + if (rem) *rem = n.all; |
| 125 | + return 0; |
| 126 | + } |
| 127 | + ++sr; |
| 128 | + q.s.low = 0; |
| 129 | + if (sr == n_udword_bits) { |
| 130 | + q.s.high = 0; |
| 131 | + r.s.high = 0; |
| 132 | + r.s.low = n.s.high; |
| 133 | + } else { |
| 134 | + r.s.high = n.s.high >> sr; |
| 135 | + r.s.low = (n.s.high << (n_udword_bits - sr)) | (n.s.low >> sr); |
| 136 | + q.s.high = n.s.low << (n_udword_bits - sr); |
| 137 | + } |
| 138 | + |
| 139 | +shift_loop: |
| 140 | + // The main shift-and-subtract loop |
| 141 | + for (; sr > 0; --sr) { |
| 142 | + r.s.high = (r.s.high << 1) | (r.s.low >> (n_udword_bits - 1)); |
| 143 | + r.s.low = (r.s.low << 1) | (q.s.high >> (n_udword_bits - 1)); |
| 144 | + q.s.high = (q.s.high << 1) | (q.s.low >> (n_udword_bits - 1)); |
| 145 | + q.s.low = (q.s.low << 1); |
| 146 | + // Compute (r - d) and conditionally subtract |
| 147 | + const long long s = (long long)(d.all - r.all - 1) >> (n_utword_bits - 1); |
| 148 | + q.s.low |= s & 1; |
| 149 | + r.all -= d.all & s; |
| 150 | + } |
| 151 | + if (rem) *rem = r.all; |
| 152 | + return q.all; |
| 153 | +} |
| 154 | + |
| 155 | +tu_int __udivti3(tu_int a, tu_int b) { |
| 156 | + return __udivmodti4(a, b, 0); |
| 157 | +} |
| 158 | + |
| 159 | +tu_int __umodti3(tu_int a, tu_int b) { |
| 160 | + tu_int r; |
| 161 | + __udivmodti4(a, b, &r); |
| 162 | + return r; |
| 163 | +} |
| 164 | + |
| 165 | +#endif // __SIZEOF_INT128__ |
0 commit comments