Moving the u128_t struct may invoke calls to memcpy/memmove when compiled with tinycc, which would depend on libc. For example, some configure scripts get confused if they fail to build test programs involving long doubles. I'm not sure if this is the root cause, though.
Unified · Split
| @@ -18,7 +18,14 @@ typedef int int32_t; · 17 unmodified lines | ||
| 18 | 18 | typedef unsigned uint32_t; |
| 19 | 19 | typedef long long int64_t; |
| 20 | 20 | typedef unsigned long long uint64_t; |
| 21 | −void *memcpy(void*,void*,__SIZE_TYPE__); | |
| 21 | +static void *memcpy(void* d, void* s, __SIZE_TYPE__ c) { | |
| 22 | + char *d_, *s_; | |
| 23 | + d_ = d; s_ = s; | |
| 24 | + for (__SIZE_TYPE__ i = 0; i < c; ++i) { | |
| 25 | + d_[i] = s_[i]; | |
| 26 | + } | |
| 27 | + return d; | |
| 28 | +} | |
| 22 | 29 | #else |
| 23 | 30 | #include <stdint.h> |
| 24 | 31 | #include <string.h> |
| @@ -103,74 +110,72 @@ static void f3_unpack(int *sgn, int32_t *exp, u128_t *mnt, long double f) · 78 unmodified lines | ||
| 103 | 110 | x.x1 |= (uint64_t)1 << 48; |
| 104 | 111 | else |
| 105 | 112 | *exp = 1; |
| 106 | − *mnt = x; | |
| 113 | + memcpy(mnt, &x, 16); | |
| 107 | 114 | } |
| 108 | 115 | |
| 109 | −static u128_t f3_normalise(int32_t *exp, u128_t mnt) | |
| 116 | +static void f3_normalise(int32_t *exp, u128_t *mnt) | |
| 110 | 117 | { |
| 111 | 118 | int sh; |
| 112 | − if (!(mnt.x0 | mnt.x1)) | |
| 113 | − return mnt; | |
| 114 | − if (!mnt.x1) { | |
| 115 | − mnt.x1 = mnt.x0; | |
| 116 | − mnt.x0 = 0; | |
| 119 | + if (!(mnt->x0 | mnt->x1)) | |
| 120 | + return; | |
| 121 | + if (!mnt->x1) { | |
| 122 | + mnt->x1 = mnt->x0; | |
| 123 | + mnt->x0 = 0; | |
| 117 | 124 | *exp -= 64; |
| 118 | 125 | } |
| 119 | 126 | for (sh = 32; sh; sh >>= 1) { |
| 120 | − if (!(mnt.x1 >> (64 - sh))) { | |
| 121 | − mnt.x1 = mnt.x1 << sh | mnt.x0 >> (64 - sh); | |
| 122 | − mnt.x0 = mnt.x0 << sh; | |
| 127 | + if (!(mnt->x1 >> (64 - sh))) { | |
| 128 | + mnt->x1 = mnt->x1 << sh | mnt->x0 >> (64 - sh); | |
| 129 | + mnt->x0 = mnt->x0 << sh; | |
| 123 | 130 | *exp -= sh; |
| 124 | 131 | } |
| 125 | 132 | } |
| 126 | − return mnt; | |
| 127 | 133 | } |
| 128 | 134 | |
| 129 | −static u128_t f3_sticky_shift(int32_t sh, u128_t x) | |
| 135 | +static void f3_sticky_shift(int32_t sh, u128_t *x) | |
| 130 | 136 | { |
| 131 | 137 | if (sh >= 128) { |
| 132 | − x.x0 = !!(x.x0 | x.x1); | |
| 133 | − x.x1 = 0; | |
| 134 | − return x; | |
| 138 | + x->x0 = !!(x->x0 | x->x1); | |
| 139 | + x->x1 = 0; | |
| 140 | + return; | |
| 135 | 141 | } |
| 136 | 142 | if (sh >= 64) { |
| 137 | − x.x0 = x.x1 | !!x.x0; | |
| 138 | − x.x1 = 0; | |
| 143 | + x->x0 = x->x1 | !!x->x0; | |
| 144 | + x->x1 = 0; | |
| 139 | 145 | sh -= 64; |
| 140 | 146 | } |
| 141 | 147 | if (sh > 0) { |
| 142 | − x.x0 = x.x0 >> sh | x.x1 << (64 - sh) | !!(x.x0 << (64 - sh)); | |
| 143 | − x.x1 = x.x1 >> sh; | |
| 148 | + x->x0 = x->x0 >> sh | x->x1 << (64 - sh) | !!(x->x0 << (64 - sh)); | |
| 149 | + x->x1 = x->x1 >> sh; | |
| 144 | 150 | } |
| 145 | − return x; | |
| 146 | 151 | } |
| 147 | 152 | |
| 148 | −static long double f3_round(int sgn, int32_t exp, u128_t x) | |
| 153 | +static long double f3_round(int sgn, int32_t exp, u128_t *x) | |
| 149 | 154 | { |
| 150 | 155 | long double f; |
| 151 | 156 | int error; |
| 152 | 157 | |
| 153 | 158 | if (exp > 0) { |
| 154 | − x = f3_sticky_shift(13, x); | |
| 159 | + f3_sticky_shift(13, x); | |
| 155 | 160 | } |
| 156 | 161 | else { |
| 157 | − x = f3_sticky_shift(14 - exp, x); | |
| 162 | + f3_sticky_shift(14 - exp, x); | |
| 158 | 163 | exp = 0; |
| 159 | 164 | } |
| 160 | 165 | |
| 161 | − error = x.x0 & 3; | |
| 162 | − x.x0 = x.x0 >> 2 | x.x1 << 62; | |
| 163 | − x.x1 = x.x1 >> 2; | |
| 166 | + error = x->x0 & 3; | |
| 167 | + x->x0 = x->x0 >> 2 | x->x1 << 62; | |
| 168 | + x->x1 = x->x1 >> 2; | |
| 164 | 169 | |
| 165 | − if (error == 3 || ((error == 2) & (x.x0 & 1))) { | |
| 166 | − if (!++x.x0) { | |
| 167 | − ++x.x1; | |
| 168 | − if (x.x1 == (uint64_t)1 << 48) | |
| 170 | + if (error == 3 || ((error == 2) & (x->x0 & 1))) { | |
| 171 | + if (!++x->x0) { | |
| 172 | + ++x->x1; | |
| 173 | + if (x->x1 == (uint64_t)1 << 48) | |
| 169 | 174 | exp = 1; |
| 170 | − else if (x.x1 == (uint64_t)1 << 49) { | |
| 175 | + else if (x->x1 == (uint64_t)1 << 49) { | |
| 171 | 176 | ++exp; |
| 172 | − x.x0 = x.x0 >> 1 | x.x1 << 63; | |
| 173 | − x.x1 = x.x1 >> 1; | |
| 177 | + x->x0 = x->x0 >> 1 | x->x1 << 63; | |
| 178 | + x->x1 = x->x1 >> 1; | |
| 174 | 179 | } |
| 175 | 180 | } |
| 176 | 181 | } |
| @@ -178,8 +183,8 @@ static long double f3_round(int sgn, int32_t exp, u128_t x) · 1 unmodified line | ||
| 178 | 183 | if (exp >= 32767) |
| 179 | 184 | return f3_infinity(sgn); |
| 180 | 185 | |
| 181 | − x.x1 = x.x1 << 16 >> 16 | (uint64_t)exp << 48 | (uint64_t)sgn << 63; | |
| 182 | − memcpy(&f, &x, 16); | |
| 186 | + x->x1 = x->x1 << 16 >> 16 | (uint64_t)exp << 48 | (uint64_t)sgn << 63; | |
| 187 | + memcpy(&f, x, 16); | |
| 183 | 188 | return f; |
| 184 | 189 | } |
| 185 | 190 | |
| @@ -214,11 +219,11 @@ static long double f3_add(long double fa, long double fb, int neg) · 28 unmodified lines | ||
| 214 | 219 | b.x0 = b.x0 << 3; |
| 215 | 220 | |
| 216 | 221 | if (a_exp <= b_exp) { |
| 217 | − a = f3_sticky_shift(b_exp - a_exp, a); | |
| 222 | + f3_sticky_shift(b_exp - a_exp, &a); | |
| 218 | 223 | a_exp = b_exp; |
| 219 | 224 | } |
| 220 | 225 | else { |
| 221 | − b = f3_sticky_shift(a_exp - b_exp, b); | |
| 226 | + f3_sticky_shift(a_exp - b_exp, &b); | |
| 222 | 227 | b_exp = a_exp; |
| 223 | 228 | } |
| 224 | 229 | |
| @@ -241,9 +246,9 @@ static long double f3_add(long double fa, long double fb, int neg) · 16 unmodified lines | ||
| 241 | 246 | if (!(x.x0 | x.x1)) |
| 242 | 247 | return f3_zero(0); |
| 243 | 248 | |
| 244 | − x = f3_normalise(&x_exp, x); | |
| 249 | + f3_normalise(&x_exp, &x); | |
| 245 | 250 | |
| 246 | − return f3_round(x_sgn, x_exp + 12, x); | |
| 251 | + return f3_round(x_sgn, x_exp + 12, &x); | |
| 247 | 252 | } |
| 248 | 253 | |
| 249 | 254 | long double __addtf3(long double a, long double b) |
| @@ -278,8 +283,8 @@ long double __multf3(long double fa, long double fb) · 28 unmodified lines | ||
| 278 | 283 | if (!(a.x0 | a.x1) || !(b.x0 | b.x1)) |
| 279 | 284 | return f3_zero(a_sgn ^ b_sgn); |
| 280 | 285 | |
| 281 | − a = f3_normalise(&a_exp, a); | |
| 282 | − b = f3_normalise(&b_exp, b); | |
| 286 | + f3_normalise(&a_exp, &a); | |
| 287 | + f3_normalise(&b_exp, &b); | |
| 283 | 288 | |
| 284 | 289 | x_sgn = a_sgn ^ b_sgn; |
| 285 | 290 | x_exp = a_exp + b_exp - 16352; |
| @@ -318,7 +323,7 @@ long double __multf3(long double fa, long double fb) · 32 unmodified lines | ||
| 318 | 323 | x.x1 = y1; |
| 319 | 324 | } |
| 320 | 325 | |
| 321 | − return f3_round(x_sgn, x_exp, x); | |
| 326 | + return f3_round(x_sgn, x_exp, &x); | |
| 322 | 327 | } |
| 323 | 328 | |
| 324 | 329 | long double __divtf3(long double fa, long double fb) |
| @@ -343,8 +348,8 @@ long double __divtf3(long double fa, long double fb) · 18 unmodified lines | ||
| 343 | 348 | if (!(a.x0 | a.x1) || b_exp == 32767) |
| 344 | 349 | return f3_zero(a_sgn ^ b_sgn); |
| 345 | 350 | |
| 346 | − a = f3_normalise(&a_exp, a); | |
| 347 | − b = f3_normalise(&b_exp, b); | |
| 351 | + f3_normalise(&a_exp, &a); | |
| 352 | + f3_normalise(&b_exp, &b); | |
| 348 | 353 | |
| 349 | 354 | x_sgn = a_sgn ^ b_sgn; |
| 350 | 355 | x_exp = a_exp - b_exp + 16395; |
| @@ -368,9 +373,9 @@ long double __divtf3(long double fa, long double fb) · 17 unmodified lines | ||
| 368 | 373 | } |
| 369 | 374 | x.x0 |= !!(a.x0 | a.x1); |
| 370 | 375 | |
| 371 | − x = f3_normalise(&x_exp, x); | |
| 376 | + f3_normalise(&x_exp, &x); | |
| 372 | 377 | |
| 373 | − return f3_round(x_sgn, x_exp, x); | |
| 378 | + return f3_round(x_sgn, x_exp, &x); | |
| 374 | 379 | } |
| 375 | 380 | |
| 376 | 381 | long double __negtf2(long double f) |