From 1a4e95a9dc594abf0108ed9e97ad9788a7dec65f Mon Sep 17 00:00:00 2001 From: Nikolay Volf Date: Mon, 29 Feb 2016 18:57:46 +0300 Subject: [PATCH 1/2] u256*u256->u512 --- util/benches/bigint.rs | 13 +++ util/src/uint.rs | 257 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 270 insertions(+) diff --git a/util/benches/bigint.rs b/util/benches/bigint.rs index da82084b8..fc41ab628 100644 --- a/util/benches/bigint.rs +++ b/util/benches/bigint.rs @@ -74,6 +74,19 @@ fn u256_mul(b: &mut Bencher) { } +#[bench] +fn u256_full_mul(b: &mut Bencher) { + b.iter(|| { + let n = black_box(10000); + (0..n).fold(U256([rand::random::(), rand::random::(), rand::random::(), rand::random::()]), + |old, new| { + let U512(ref u512words) = old.full_mul(U256([rand::random::(), rand::random::(), rand::random::(), rand::random::()])); + U256([u512words[0], u512words[2], u512words[2], u512words[3]]) + }) + }); +} + + #[bench] fn u128_mul(b: &mut Bencher) { b.iter(|| { diff --git a/util/src/uint.rs b/util/src/uint.rs index 517b7a29f..bbe62e5d4 100644 --- a/util/src/uint.rs +++ b/util/src/uint.rs @@ -1097,6 +1097,157 @@ construct_uint!(U512, 8); construct_uint!(U256, 4); construct_uint!(U128, 2); +impl U256 { + /// Multiplies two 256-bit integers to produce full 512-bit integer + /// No overflow possible + #[cfg(all(x64asm, target_arch="x86_64"))] + pub fn full_mul(self, other: U256) -> U512 { + let self_t: &[u64; 4] = unsafe { &mem::transmute(self) }; + let other_t: &[u64; 4] = unsafe { &mem::transmute(other) }; + let mut result: [u64; 8] = unsafe { mem::uninitialized() }; + unsafe { + asm!(" + mov $8, %rax + mulq $12 + mov %rax, $0 + mov %rdx, $1 + + mov $8, %rax + mulq $13 + add %rax, $1 + adc $$0, %rdx + mov %rdx, $2 + + mov $8, %rax + mulq $14 + add %rax, $2 + adc $$0, %rdx + mov %rdx, $3 + + mov $8, %rax + mulq $15 + add %rax, $3 + adc $$0, %rdx + mov %rdx, $4 + + mov $9, %rax + mulq $12 + add %rax, $1 + adc %rdx, $2 + adc $$0, $3 + adc $$0, $4 + xor $5, $5 + adc $$0, $5 + xor $6, $6 + adc $$0, $6 + xor $7, $7 + adc $$0, $7 + + mov $9, %rax + mulq $13 + add %rax, $2 + adc %rdx, $3 + adc $$0, $4 + adc $$0, $5 + adc $$0, $6 + adc $$0, $7 + + mov $9, %rax + mulq $14 + add %rax, $3 + adc %rdx, $4 + adc $$0, $5 + adc $$0, $6 + adc $$0, $7 + + mov $9, %rax + mulq $15 + add %rax, $4 + adc %rdx, $5 + adc $$0, $6 + adc $$0, $7 + + mov $10, %rax + mulq $12 + add %rax, $2 + adc %rdx, $3 + adc $$0, $4 + adc $$0, $5 + adc $$0, $6 + adc $$0, $7 + + mov $10, %rax + mulq $13 + add %rax, $3 + adc %rdx, $4 + adc $$0, $5 + adc $$0, $6 + adc $$0, $7 + + mov $10, %rax + mulq $14 + add %rax, $4 + adc %rdx, $5 + adc $$0, $6 + adc $$0, $7 + + mov $10, %rax + mulq $15 + add %rax, $5 + adc %rdx, $6 + adc $$0, $7 + + mov $11, %rax + mulq $12 + add %rax, $3 + adc %rdx, $4 + adc $$0, $5 + adc $$0, $6 + adc $$0, $7 + + mov $11, %rax + mulq $13 + add %rax, $4 + adc %rdx, $5 + adc $$0, $6 + adc $$0, $7 + + mov $11, %rax + mulq $14 + add %rax, $5 + adc %rdx, $6 + adc $$0, $7 + + mov $11, %rax + mulq $15 + add %rax, $6 + adc %rdx, $7 + " + : /* $0 */ "={r8}"(result[0]), /* $1 */ "={r9}"(result[1]), /* $2 */ "={r10}"(result[2]), + /* $3 */ "={r11}"(result[3]), /* $4 */ "={r12}"(result[4]), /* $5 */ "={r13}"(result[5]), + /* $6 */ "={r14}"(result[6]), /* $7 */ "={r15}"(result[7]) + + : /* $8 */ "m"(self_t[0]), /* $9 */ "m"(self_t[1]), /* $10 */ "m"(self_t[2]), + /* $11 */ "m"(self_t[3]), /* $12 */ "m"(other_t[0]), /* $13 */ "m"(other_t[1]), + /* $14 */ "m"(other_t[2]), /* $15 */ "m"(other_t[3]) + : "rax", "rdx" + : + ); + } + U512(result) + } + + /// Multiplies two 256-bit integers to produce full 512-bit integer + /// No overflow possible + #[cfg(not(all(x64asm, target_arch="x86_64")))] + pub fn full_mul(self, other: U256) -> U512 { + let self_512 = U512::from(self); + let other_512 = U512::from(other); + let (result, _) = self_512.overflowing_mul(other_512); + result + } +} + impl From for U512 { fn from(value: U256) -> U512 { let U256(ref arr) = value; @@ -1828,5 +1979,111 @@ mod tests { let (_, overflow) = U256([0, 0, 8, 0]).overflowing_mul(U256([0, 0, 7, 0])); assert!(overflow); } + + + #[test] + fn u256_multi_full_mul() { + let result = U256([0, 0, 0, 0]).full_mul(U256([0, 0, 0, 0])); + assert_eq!(U512([0, 0, 0, 0, 0, 0, 0, 0]), result); + + let result = U256([1, 0, 0, 0]).full_mul(U256([1, 0, 0, 0])); + assert_eq!(U512([1, 0, 0, 0, 0, 0, 0, 0]), result); + + let result = U256([5, 0, 0, 0]).full_mul(U256([5, 0, 0, 0])); + assert_eq!(U512([25, 0, 0, 0, 0, 0, 0, 0]), result); + + let result = U256([0, 5, 0, 0]).full_mul(U256([0, 5, 0, 0])); + assert_eq!(U512([0, 0, 25, 0, 0, 0, 0, 0]), result); + + let result = U256([0, 0, 0, 4]).full_mul(U256([4, 0, 0, 0])); + assert_eq!(U512([0, 0, 0, 16, 0, 0, 0, 0]), result); + + let result = U256([0, 0, 0, 5]).full_mul(U256([2, 0, 0, 0])); + assert_eq!(U512([0, 0, 0, 10, 0, 0, 0, 0]), result); + + let result = U256([0, 0, 2, 0]).full_mul(U256([0, 5, 0, 0])); + assert_eq!(U512([0, 0, 0, 10, 0, 0, 0, 0]), result); + + let result = U256([0, 3, 0, 0]).full_mul(U256([0, 0, 3, 0])); + assert_eq!(U512([0, 0, 0, 9, 0, 0, 0, 0]), result); + + let result = U256([0, 0, 8, 0]).full_mul(U256([0, 0, 6, 0])); + assert_eq!(U512([0, 0, 0, 0, 48, 0, 0, 0]), result); + + let result = U256([9, 0, 0, 0]).full_mul(U256([0, 3, 0, 0])); + assert_eq!(U512([0, 27, 0, 0, 0, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, 0, 0, 0]).full_mul(U256([::std::u64::MAX, 0, 0, 0])); + assert_eq!(U512([1, ::std::u64::MAX-1, 0, 0, 0, 0, 0, 0]), result); + + let result = U256([0, ::std::u64::MAX, 0, 0]).full_mul(U256([::std::u64::MAX, 0, 0, 0])); + assert_eq!(U512([0, 1, ::std::u64::MAX-1, 0, 0, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, 0, 0]).full_mul(U256([::std::u64::MAX, 0, 0, 0])); + assert_eq!(U512([1, ::std::u64::MAX, ::std::u64::MAX-1, 0, 0, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, 0, 0, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, 0, 0])); + assert_eq!(U512([1, ::std::u64::MAX, ::std::u64::MAX-1, 0, 0, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, 0, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, 0, 0])); + assert_eq!(U512([1, 0, ::std::u64::MAX-1, ::std::u64::MAX, 0, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, 0, 0, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, 0])); + assert_eq!(U512([1, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX-1, 0, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, 0]).full_mul(U256([::std::u64::MAX, 0, 0, 0])); + assert_eq!(U512([1, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX-1, 0, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, 0, 0, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX])); + assert_eq!(U512([1, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX-1, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX]).full_mul(U256([::std::u64::MAX, 0, 0, 0])); + assert_eq!(U512([1, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX-1, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, 0, 0])); + assert_eq!(U512([1, 0, ::std::u64::MAX, ::std::u64::MAX-1, ::std::u64::MAX, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, 0, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, 0])); + assert_eq!(U512([1, 0, ::std::u64::MAX, ::std::u64::MAX-1, ::std::u64::MAX, 0, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, 0, 0])); + assert_eq!(U512([1, 0, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX-1, ::std::u64::MAX, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, 0, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX])); + assert_eq!(U512([1, 0, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX-1, ::std::u64::MAX, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, 0])); + assert_eq!(U512([1, 0, 0, ::std::u64::MAX-1, ::std::u64::MAX, ::std::u64::MAX, 0, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, 0]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX])); + assert_eq!(U512([1, 0, 0, ::std::u64::MAX, ::std::u64::MAX-1, ::std::u64::MAX, ::std::u64::MAX, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, 0])); + assert_eq!(U512([1, 0, 0, ::std::u64::MAX, ::std::u64::MAX-1, ::std::u64::MAX, ::std::u64::MAX, 0]), result); + + let result = U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX]).full_mul(U256([::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX])); + assert_eq!(U512([1, 0, 0, 0, ::std::u64::MAX-1, ::std::u64::MAX, ::std::u64::MAX, ::std::u64::MAX]), result); + + let result = U256([0, 0, 0, ::std::u64::MAX]).full_mul(U256([0, 0, 0, ::std::u64::MAX])); + assert_eq!(U512([0, 0, 0, 0, 0, 0, 1, ::std::u64::MAX-1]), result); + + let result = U256([1, 0, 0, 0]).full_mul(U256([0, 0, 0, ::std::u64::MAX])); + assert_eq!(U512([0, 0, 0, ::std::u64::MAX, 0, 0, 0, 0]), result); + + let result = U256([1, 2, 3, 4]).full_mul(U256([5, 0, 0, 0])); + assert_eq!(U512([5, 10, 15, 20, 0, 0, 0, 0]), result); + + let result = U256([1, 2, 3, 4]).full_mul(U256([0, 6, 0, 0])); + assert_eq!(U512([0, 6, 12, 18, 24, 0, 0, 0]), result); + + let result = U256([1, 2, 3, 4]).full_mul(U256([0, 0, 7, 0])); + assert_eq!(U512([0, 0, 7, 14, 21, 28, 0, 0]), result); + + let result = U256([1, 2, 3, 4]).full_mul(U256([0, 0, 0, 8])); + assert_eq!(U512([0, 0, 0, 8, 16, 24, 32, 0]), result); + + let result = U256([1, 2, 3, 4]).full_mul(U256([5, 6, 7, 8])); + assert_eq!(U512([5, 16, 34, 60, 61, 52, 32, 0]), result); + } } From b30f066651d480c0e7a5dbd90ed85585de5a6d6c Mon Sep 17 00:00:00 2001 From: Nikolay Volf Date: Mon, 29 Feb 2016 19:15:13 +0300 Subject: [PATCH 2/2] tabified --- util/src/uint.rs | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/util/src/uint.rs b/util/src/uint.rs index bbe62e5d4..88256d5f2 100644 --- a/util/src/uint.rs +++ b/util/src/uint.rs @@ -1105,8 +1105,8 @@ impl U256 { let self_t: &[u64; 4] = unsafe { &mem::transmute(self) }; let other_t: &[u64; 4] = unsafe { &mem::transmute(other) }; let mut result: [u64; 8] = unsafe { mem::uninitialized() }; - unsafe { - asm!(" + unsafe { + asm!(" mov $8, %rax mulq $12 mov %rax, $0 @@ -1230,11 +1230,11 @@ impl U256 { : /* $8 */ "m"(self_t[0]), /* $9 */ "m"(self_t[1]), /* $10 */ "m"(self_t[2]), /* $11 */ "m"(self_t[3]), /* $12 */ "m"(other_t[0]), /* $13 */ "m"(other_t[1]), /* $14 */ "m"(other_t[2]), /* $15 */ "m"(other_t[3]) - : "rax", "rdx" - : - ); - } - U512(result) + : "rax", "rdx" + : + ); + } + U512(result) } /// Multiplies two 256-bit integers to produce full 512-bit integer