diff options
Diffstat (limited to 'src/integer128.cpp')
| -rw-r--r-- | src/integer128.cpp | 59 |
1 files changed, 56 insertions, 3 deletions
diff --git a/src/integer128.cpp b/src/integer128.cpp index 73cd1c7a0..35023a220 100644 --- a/src/integer128.cpp +++ b/src/integer128.cpp @@ -1,3 +1,10 @@ + +#if defined(GB_COMPILER_MSVC) && defined(GB_ARCH_64_BIT) && defined(GB_CPU_X86) + #define MSVC_AMD64_INTRINSICS + #include <intrin.h> + #pragma intrinsic(_mul128) +#endif + #define BIT128_U64_HIGHBIT 0x8000000000000000ull #define BIT128_U64_BITS62 0x7fffffffffffffffull #define BIT128_U64_ALLBITS 0xffffffffffffffffull @@ -376,7 +383,11 @@ u128 u128_shl(u128 a, u32 n) { if (n >= 128) { return u128_lo_hi(0, 0); } - +#if 0 && defined(MSVC_AMD64_INTRINSICS) + a.hi = __shiftleft128(a.lo, a.hi, n); + a.lo = a.lo << n; + return a; +#else if (n >= 64) { n -= 64; a.hi = a.lo; @@ -391,13 +402,18 @@ u128 u128_shl(u128 a, u32 n) { a.lo <<= n; } return a; +#endif } u128 u128_shr(u128 a, u32 n) { if (n >= 128) { return u128_lo_hi(0, 0); } - +#if 0 && defined(MSVC_AMD64_INTRINSICS) + a.lo = __shiftright128(a.lo, a.hi, n); + a.hi = a.hi >> n; + return a; +#else if (n >= 64) { n -= 64; a.lo = a.hi; @@ -411,6 +427,7 @@ u128 u128_shr(u128 a, u32 n) { a.hi >>= n; } return a; +#endif } @@ -427,6 +444,14 @@ u128 u128_mul(u128 a, u128 b) { return a; } + +#if defined(MSVC_AMD64_INTRINSICS) + if (a.hi == 0 && b.hi == 0) { + a.lo = _umul128(a.lo, b.lo, &a.hi); + return a; + } +#endif + u128 res = {0}; u128 t = b; for (u32 i = 0; i < 128; i++) { @@ -440,6 +465,8 @@ u128 u128_mul(u128 a, u128 b) { return res; } +bool u128_hibit(u128 *d) { return (d->hi & BIT128_U64_HIGHBIT) != 0; } + void u128_divide(u128 num, u128 den, u128 *quo, u128 *rem) { if (u128_eq(den, U128_ZERO)) { if (quo) *quo = u128_from_u64(num.lo/den.lo); @@ -450,7 +477,7 @@ void u128_divide(u128 num, u128 den, u128 *quo, u128 *rem) { u128 x = U128_ONE; u128 r = U128_ZERO; - while (u128_ge(n, d) && ((u128_shr(d, 128-1).lo&1) == 0)) { + while (u128_ge(n, d) && !u128_hibit(&d)) { x = u128_shl(x, 1); d = u128_shl(d, 1); } @@ -471,11 +498,18 @@ void u128_divide(u128 num, u128 den, u128 *quo, u128 *rem) { } u128 u128_quo(u128 a, u128 b) { + if (a.hi == 0 && b.hi == 0) { + return u128_from_u64(a.lo/b.lo); + } + u128 res = {0}; u128_divide(a, b, &res, NULL); return res; } u128 u128_mod(u128 a, u128 b) { + if (a.hi == 0 && b.hi == 0) { + return u128_from_u64(a.lo%b.lo); + } u128 res = {0}; u128_divide(a, b, NULL, &res); return res; @@ -535,6 +569,11 @@ i128 i128_shl(i128 a, u32 n) { return i128_lo_hi(0, 0); } +#if 0 && defined(MSVC_AMD64_INTRINSICS) + a.hi = __shiftleft128(a.lo, a.hi, n); + a.lo = a.lo << n; + return a; +#else if (n >= 64) { n -= 64; a.hi = a.lo; @@ -549,6 +588,7 @@ i128 i128_shl(i128 a, u32 n) { a.lo <<= n; } return a; +#endif } i128 i128_shr(i128 a, u32 n) { @@ -556,6 +596,11 @@ i128 i128_shr(i128 a, u32 n) { return i128_lo_hi(0, 0); } +#if 0 && defined(MSVC_AMD64_INTRINSICS) + a.lo = __shiftright128(a.lo, a.hi, n); + a.hi = a.hi >> n; + return a; +#else if (n >= 64) { n -= 64; a.lo = a.hi; @@ -569,6 +614,7 @@ i128 i128_shr(i128 a, u32 n) { a.hi >>= n; } return a; +#endif } @@ -585,6 +631,13 @@ i128 i128_mul(i128 a, i128 b) { return a; } +#if defined(MSVC_AMD64_INTRINSICS) + if (a.hi == 0 && b.hi == 0) { + a.lo = _mul128(a.lo, b.lo, &a.hi); + return a; + } +#endif + i128 res = {0}; i128 t = b; for (u32 i = 0; i < 128; i++) { |