64位平台的256位乘法测试

无心人 · 发表于 2014-11-5 11:26:35

马上注册，结交更多好友，享用更多功能，让你轻松玩转社区。

您需要登录才可以下载或查看，没有账号？欢迎注册

×

先贴上我测试通过的代码

__declspec(naked)
void UInt256x256To512(uint64_t* result, const uint64_t* left, const uint64_t* right)
{
__asm
{
push rbx
mov r9, rdx
mov r10, [r8+0*8]
mov r11, [r9+0*8]
mov r12, [r9+1*8]
mov r13, [r9+2*8]
mov r14, [r9+3*8]
mov rax, r10
mul r11
mov [rcx+0*8], rax
mov r15, rdx
mov rax, r10
mul r12
add rax, r15
adc rdx, 0
mov r15, rdx
movq xmm1, rax
mov rax, r10
mul r13
add rax, r15
adc rdx, 0
mov r15, rdx
movq xmm2, rax
mov rax, r10
mul r14
add rax, r15
adc rdx, 0
mov r15, rdx
movq xmm3, rax
movq xmm4, rdx
mov r10, [r8+1*8]
movq r15, xmm1
mov rax, r10
mul r11
add rax, r15
adc rdx, 0
mov [rcx+1*8], rax
mov rbx, rdx
movq r15, xmm2
mov rax, r10
mul r12
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
movq xmm2, rax
mov rbx, rdx
movq r15, xmm3
mov rax, r10
mul r13
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
movq xmm3, rax
mov rbx, rdx
movq r15, xmm4
mov rax, r10
mul r14
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
movq xmm4, rax
movq xmm5, rdx
mov r10, [r8+2*8]
movq r15, xmm2
mov rax, r10
mul r11
add rax, r15
adc rdx, 0
mov [rcx+2*8], rax
mov rbx, rdx
movq r15, xmm3
mov rax, r10
mul r12
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
movq xmm3, rax
mov rbx, rdx
movq r15, xmm4
mov rax, r10
mul r13
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
movq xmm4, rax
mov rbx, rdx
movq r15, xmm5
mov rax, r10
mul r14
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
movq xmm5, rax
movq xmm6, rdx
mov r10, [r8+3*8]
movq r15, xmm3
mov rax, r10
mul r11
add rax, r15
adc rdx, 0
mov [rcx+3*8], rax
mov rbx, rdx
movq r15, xmm4
mov rax, r10
mul r12
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
mov [rcx+4*8], rax
mov rbx, rdx
movq r15, xmm5
mov rax, r10
mul r13
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
mov [rcx+5*8], rax
mov rbx, rdx
movq r15, xmm6
mov rax, r10
mul r14
add rax, r15
adc rdx, 0
add rax, rbx
adc rdx, 0
mov [rcx+6*8], rax
mov [rcx+7*8], rdx
pop rbx
ret
}
}

复制代码

cn8888 · 发表于 2014-11-5 12:31:15

晕,发这些汇编代码上来干什么?一点注释都没有,谁有兴趣去看呀?

mathe · 发表于 2014-11-5 17:14:07

为什么用xmmi?是寄存器不够用吗？这些寄存器不需要保存原先的初值吗?

gxqcn · 发表于 2014-11-11 08:05:28

为什么不用 YMM 寄存器？

l4m2 · 发表于 2023-5-2 20:50:08

63行汇编，含IO占10个寄存器，读写内存。
llvm-mca表示运行时间43cycle，但mca可能有bug，实际时间没那么长

Timeline view:
0123456789 0123456789
Index 0123456789 0123456789 012
[0,0] DeeeeeER . . . . . . . . movq (%rsi), %rdx
[0,1] DeeeeeeeeeER . . . . . . . mulxq (%rbx), %rax, %r8
[0,2] D========eER . . . . . . . movq %rax, (%rdi)
[0,3] .DeeeeeeeeeER . . . . . . . mulxq 8(%rbx), %rax, %r9
[0,4] .D========eER . . . . . . . addq %rax, %r8
[0,5] . DeeeeeeeeeER . . . . . . . mulxq 16(%rbx), %rax, %r10
[0,6] . D========eER . . . . . . . adcq %rax, %r9
[0,7] . DeeeeeeeeeER. . . . . . . mulxq 24(%rbx), %rax, %r11
[0,8] . D========eER. . . . . . . adcq %rax, %r10
[0,9] . D=========eER . . . . . . adcq $0, %r11
[0,10] . D----------R . . . . . . xorl %ecx, %ecx
[0,11] . DeeeeeE----R . . . . . . movq 8(%rsi), %rdx
[0,12] . DeeeeeeeeeER . . . . . . mulxq (%rbx), %rax, %rcx
[0,13] . D=======eER . . . . . . adoxq %rax, %r8
[0,14] . D========eER . . . . . . adcxq %rcx, %r9
[0,15] . D========eER . . . . . . movq %r8, 8(%rdi)
[0,16] . .DeeeeeeeeeER . . . . . . mulxq 8(%rbx), %rax, %rcx
[0,17] . .D========eER . . . . . . adoxq %rax, %r9
[0,18] . .D=========eER . . . . . . adcxq %rcx, %r10
[0,19] . . DeeeeeeeeeER . . . . . . mulxq 16(%rbx), %rax, %rcx
[0,20] . . D=========eER. . . . . . adoxq %rax, %r10
[0,21] . . D==========eER . . . . . adcxq %rcx, %r11
[0,22] . . DeeeeeeeeeE-R . . . . . mulxq 24(%rbx), %rax, %rcx
[0,23] . . D==========eER . . . . . adoxq %rax, %r11
[0,24] . . DeE----------R . . . . . movl $0, %r8d
[0,25] . . D==========eER . . . . . adoxq %r8, %r8
[0,26] . . D===========eER . . . . . adcxq %rcx, %r8
[0,27] . . DeeeeeE-------R . . . . . movq 16(%rsi), %rdx
[0,28] . . DeeeeeeeeeE--R . . . . . mulxq (%rbx), %rax, %rcx
[0,29] . . D===========eER. . . . . adoxq %rax, %r9
[0,30] . . D============eER . . . . adcxq %rcx, %r10
[0,31] . . .D===========eER . . . . movq %r9, 16(%rdi)
[0,32] . . .DeeeeeeeeeE---R . . . . mulxq 8(%rbx), %rax, %rcx
[0,33] . . .D============eER . . . . adoxq %rax, %r10
[0,34] . . . D============eER . . . . adcxq %rcx, %r11
[0,35] . . . DeeeeeeeeeE----R . . . . mulxq 16(%rbx), %rax, %rcx
[0,36] . . . D=============eER . . . . adoxq %rax, %r11
[0,37] . . . D=============eER. . . . adcxq %rcx, %r8
[0,38] . . . DeeeeeeeeeE-----R. . . . mulxq 24(%rbx), %rax, %rcx
[0,39] . . . D==============eER . . . adoxq %rax, %r8
[0,40] . . . DeE-------------R . . . movl $0, %r9d
[0,41] . . . D==============eER . . . adoxq %r9, %r9
[0,42] . . . D===============eER . . . adcxq %rcx, %r9
[0,43] . . . DeeeeeE-----------R . . . movq 24(%rsi), %rdx
[0,44] . . . DeeeeeeeeeE------R . . . mulxq (%rbx), %rax, %rcx
[0,45] . . . D===============eER . . . adoxq %rax, %r10
[0,46] . . . D================eER. . . adcxq %rcx, %r11
[0,47] . . . .D===============eER. . . movq %r10, 24(%rdi)
[0,48] . . . .DeeeeeeeeeE-------R. . . mulxq 8(%rbx), %rax, %rcx
[0,49] . . . .D================eER . . adoxq %rax, %r11
[0,50] . . . . D================eER . . adcxq %rcx, %r8
[0,51] . . . . D================eER . . movq %r11, 32(%rdi)
[0,52] . . . . DeeeeeeeeeE--------R . . mulxq 16(%rbx), %rax, %rcx
[0,53] . . . . D================eER . . adoxq %rax, %r8
[0,54] . . . . D=================eER . . adcxq %rcx, %r9
[0,55] . . . . D=================eER . . movq %r8, 40(%rdi)
[0,56] . . . . DeeeeeeeeeE--------R . . mulxq 24(%rbx), %rax, %rcx
[0,57] . . . . D=================eER. . adoxq %rax, %r9
[0,58] . . . . DeE-----------------R. . movl $0, %r10d
[0,59] . . . . D=================eER . movq %r9, 48(%rdi)
[0,60] . . . . D=================eER . adoxq %r10, %r10
[0,61] . . . . D==================eER. adcxq %rcx, %r10
[0,62] . . . . D===================eER movq %r10, 56(%rdi)

复制代码

l4m2 · 发表于 2023-5-2 23:42:13

運行崩潰：r8未初始化，在

mov r10, [r8+0*8]

复制代码

.·.·. · 发表于 2023-5-3 12:47:20

（把之前的错误代码删除了）
（进位的时候应该按BASE进位，我进成了DOUBLE，不出错就见鬼了）

use std::ops::Mul;
type BASE=u64;
type DOUBLE=u128;
const _:()=assert!(std::mem::size_of::<BASE>() * 2==std::mem::size_of::<DOUBLE>());
const BASE:u32=8*std::mem::size_of::<BASE>() as u32;
const BASE_LEN:usize=256/BASE as usize;
#[derive(Clone, Copy)]
struct U256([BASE;BASE_LEN]);
// impl Mul for U256 {
// type Output = [BASE;BASE_LEN*2];
// fn mul(self, rhs: Self) -> Self::Output {
// let mut out0=[0 as DOUBLE;BASE_LEN*2];
// for i in 0..BASE_LEN {
// let mut carry=false;
// for j in 0..BASE_LEN {
// (out0[i+j],carry)=out0[i+j].overflowing_add(self.0[i] as DOUBLE*rhs.0[j] as DOUBLE+carry as DOUBLE)
// }
// let mut j=BASE_LEN;
// while carry {
// (out0[i+j],carry)=out0[i+j].overflowing_add(carry as DOUBLE);
// j+=1;
// }
// }
// let mut out=[0 as BASE;BASE_LEN*2];
// let mut carry=0 as DOUBLE;
// for i in 0..BASE_LEN*2 {
// out[i]=(out0[i]+carry) as BASE;
// carry=(carry>>BASE)+(out0[i]>>BASE);
// }
// out
// }
// }
impl Mul for U256 {
type Output = [BASE;BASE_LEN*2];
fn mul(self, rhs: Self) -> Self::Output {
let mut out0=[0 as DOUBLE;BASE_LEN*2];
for i in 0..BASE_LEN {
let mut carry=false;
for j in 0..BASE_LEN {
(out0[i+j],carry)=out0[i+j].overflowing_add(self.0[i] as DOUBLE*rhs.0[j] as DOUBLE+((carry as DOUBLE)<<BASE))
}
let mut j=BASE_LEN;
while carry {
(out0[i+j],carry)=out0[i+j].overflowing_add((carry as DOUBLE)<<BASE);
j+=1;
}
}
let mut out=[0 as BASE;BASE_LEN*2];
let mut carry=0 as DOUBLE;
for i in 0..BASE_LEN*2 {
let (res,of)=out0[i].overflowing_add(carry);
out[i]=res as BASE;
carry=(res>>BASE)+((of as DOUBLE)<<BASE);
}
out
}
}

复制代码

对应汇编

.text
.file "lib.9a6cb7ee-cgu.0"
.section ".text._ZN51_$LT$lib..U256$u20$as$u20$core..ops..arith..Mul$GT$3mul17h5eee3b70e1af20c7E","ax",@progbits
.globl _ZN51_$LT$lib..U256$u20$as$u20$core..ops..arith..Mul$GT$3mul17h5eee3b70e1af20c7E
.p2align 4, 0x90
.type _ZN51_$LT$lib..U256$u20$as$u20$core..ops..arith..Mul$GT$3mul17h5eee3b70e1af20c7E,@function
_ZN51_$LT$lib..U256$u20$as$u20$core..ops..arith..Mul$GT$3mul17h5eee3b70e1af20c7E:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
pushq %r15
.cfi_def_cfa_offset 24
pushq %r14
.cfi_def_cfa_offset 32
pushq %r13
.cfi_def_cfa_offset 40
pushq %r12
.cfi_def_cfa_offset 48
pushq %rbx
.cfi_def_cfa_offset 56
subq $80, %rsp
.cfi_def_cfa_offset 136
.cfi_offset %rbx, -56
.cfi_offset %r12, -48
.cfi_offset %r13, -40
.cfi_offset %r14, -32
.cfi_offset %r15, -24
.cfi_offset %rbp, -16
movq %rdx, %rcx
movq 8(%rdx), %r10
movq (%rsi), %rbx
movq %rbx, 72(%rsp)
movq %r10, %rdx
mulxq %rbx, %r11, %r13
movq 16(%rcx), %r9
movq %r9, %rdx
mulxq %rbx, %r15, %r14
movq %rdi, %rax
movq (%rcx), %r12
movq 24(%rcx), %r8
movq %r8, %rdx
mulxq %rbx, %rdx, %rcx
movq %rdx, 16(%rsp)
movq %rcx, (%rsp)
movq 8(%rsi), %rdi
movq %rsi, 32(%rsp)
movq %r12, %rdx
mulxq %rdi, %rdx, %rcx
addq %r11, %rdx
movq %rdx, 56(%rsp)
adcq %r13, %rcx
movq %rcx, 64(%rsp)
movq %r10, %rdx
mulxq %rdi, %rbp, %r13
adcq $0, %r13
addq %r15, %rbp
adcq %r14, %r13
movq %r9, %rdx
movq %r9, 40(%rsp)
mulxq %rdi, %rbx, %rcx
movq %r8, %r15
movq %r8, 48(%rsp)
movq %r8, %rdx
mulxq %rdi, %rdx, %r14
movq %rdx, 8(%rsp)
adcq $0, %rcx
addq 16(%rsp), %rbx
movq 16(%rsi), %rdi
movq %r12, %rdx
movq %r12, %rsi
movq %r12, 24(%rsp)
mulxq %rdi, %r11, %r8
adcq (%rsp), %rcx
adcq $0, %r14
addq %rbp, %r11
movq %r10, %rdx
mulxq %rdi, %rbp, %r12
adcq %r13, %r8
adcq $0, %r12
addq %rbx, %rbp
movq %r9, %rdx
mulxq %rdi, %rbx, %r9
adcq %rcx, %r12
adcq $0, %r9
addq 8(%rsp), %rbx
movq %r15, %rdx
mulxq %rdi, %rcx, %r15
movq %rcx, (%rsp)
adcq %r14, %r9
adcq $0, %r15
movq 32(%rsp), %rcx
movq 24(%rcx), %rcx
movq %rsi, %rdx
mulxq %rcx, %r14, %r13
addq %rbp, %r14
adcq %r12, %r13
movq %r10, %rdx
mulxq %rcx, %r10, %rsi
adcq $0, %rsi
addq %rbx, %r10
adcq %r9, %rsi
movq 40(%rsp), %rdx
mulxq %rcx, %rbp, %rdi
adcq $0, %rdi
movq 48(%rsp), %rdx
mulxq %rcx, %r12, %r9
movq 24(%rsp), %rdx
mulxq 72(%rsp), %rcx, %rdx
addq (%rsp), %rbp
adcq %r15, %rdi
movq %rcx, (%rax)
adcq $0, %r9
addq 56(%rsp), %rdx
movq %rdx, 8(%rax)
adcq 64(%rsp), %r11
movq %r11, 16(%rax)
adcq $0, %r8
setb %cl
addq %r14, %r8
movzbl %cl, %ecx
adcq %r13, %rcx
movq %r8, 24(%rax)
setb %dl
addq %r10, %rcx
movzbl %dl, %edx
adcq %rsi, %rdx
movq %rcx, 32(%rax)
setb %cl
addq %rbp, %rdx
movzbl %cl, %ecx
adcq %rdi, %rcx
movq %rdx, 40(%rax)
setb %dl
addq %r12, %rcx
movzbl %dl, %edx
adcq %r9, %rdx
movq %rcx, 48(%rax)
movq %rdx, 56(%rax)
addq $80, %rsp
.cfi_def_cfa_offset 56
popq %rbx
.cfi_def_cfa_offset 48
popq %r12
.cfi_def_cfa_offset 40
popq %r13
.cfi_def_cfa_offset 32
popq %r14
.cfi_def_cfa_offset 24
popq %r15
.cfi_def_cfa_offset 16
popq %rbp
.cfi_def_cfa_offset 8
retq
.Lfunc_end0:
.size _ZN51_$LT$lib..U256$u20$as$u20$core..ops..arith..Mul$GT$3mul17h5eee3b70e1af20c7E, .Lfunc_end0-_ZN51_$LT$lib..U256$u20$as$u20$core..ops..arith..Mul$GT$3mul17h5eee3b70e1af20c7E
.cfi_endproc
.section ".note.GNU-stack","",@progbits

复制代码

测试

fn main(){
let r=U256([BASE::MAX;BASE_LEN]);
(r*r).into_iter().for_each(|x|println!("{x:x}"))
}

复制代码

输出

1
0
0
0
fffffffffffffffe
ffffffffffffffff
ffffffffffffffff
ffffffffffffffff

复制代码

l4m2 · 发表于 2023-5-4 00:01:02

本帖最后由 l4m2 于 2023-5-4 00:02 编辑

不使用adx指令，多用幾個寄存器，效率接近

.section .text
.globl main
main:
addq $-128, %rsp
movq %rsp, %rdi
leaq 64(%rsp), %rcx
leaq 128(%rsp), %rsi
xorl %ebx, %ebx
lp:
// mul256 start
movq (%rcx), %rdx
mulxq (%rsi), %r8, %r12
movq %r8, (%rdi)
mulxq 8(%rsi), %r9, %r13
addq %r12, %r9
mulxq 16(%rsi), %r10, %r14
adcq %r13, %r10
mulxq 24(%rsi), %r11, %r8
adcq %r14, %r11
adcq $0, %r8
movq 8(%rcx), %rdx
mulxq (%rsi), %rax, %r12
addq %rax, %r9
movq %r9, 8(%rdi)
mulxq 8(%rsi), %rax, %r13
adcq %rax, %r10
mulxq 16(%rsi), %rax, %r14
adcq %rax, %r11
mulxq 24(%rsi), %rax, %r9
adcq %rax, %r8
adcq $0, %r9
addq %r12, %r10
adcq %r13, %r11
adcq %r14, %r8
adcq $0, %r9
movq 16(%rcx), %rdx
mulxq (%rsi), %rax, %r12
addq %rax, %r10
movq %r10, 16(%rdi)
mulxq 8(%rsi), %rax, %r13
adcq %rax, %r11
mulxq 16(%rsi), %rax, %r14
adcq %rax, %r8
mulxq 24(%rsi), %rax, %r9
adcq %rax, %r9
adcq $0, %r10
addq %r12, %r11
adcq %r13, %r8
adcq %r14, %r9
adcq $0, %r10
movq 24(%rcx), %rdx
mulxq (%rsi), %rax, %r12
addq %rax, %r11
movq %r11, 24(%rdi)
mulxq 8(%rsi), %rax, %r13
adcq %rax, %r8
mulxq 16(%rsi), %rax, %r14
adcq %rax, %r9
mulxq 24(%rsi), %rax, %r9
adcq %rax, %r10
adcq $0, %r10
addq %r12, %r8
movq %r8, 32(%rsi)
adcq %r13, %r9
movq %r9, 40(%rsi)
adcq %r14, %r10
movq %r10, 48(%rsi)
adcq $0, %r11
movq %r11, 56(%rsi)
// mul256 end
subl $1, %ebx
jnz lp
subq $-128, %rsp
ret
// rdi(dst) rsi(A) rdx=>rcx(B)
// r8-r11 current considered 4 bytes
// r12-r14 himul high
// rax(tmp) rdx(mulx reg)
/*
mul256:
movq %rdx, %rcx
movq (%rdx), %rdx
movq %r12, -8(%rsp)
mulxq (%rsi), %r8, %r12
movq %r8, (%rdi)
movq %r13, -16(%rsp)
mulxq 8(%rsi), %r9, %r13
addq %r12, %r9
movq %r14, -24(%rsp)
mulxq 16(%rsi), %r10, %r14
adcq %r13, %r10
mulxq 24(%rsi), %r11, %r8
adcq %r14, %r11
adcq $0, %r8
movq 8(%rcx), %rdx
mulxq (%rsi), %rax, %r12
addq %rax, %r9
movq %r9, 8(%rdi)
mulxq 8(%rsi), %rax, %r13
adcq %rax, %r10
mulxq 16(%rsi), %rax, %r14
adcq %rax, %r11
mulxq 24(%rsi), %rax, %r9
adcq %rax, %r8
adcq $0, %r9
addq %r12, %r10
adcq %r13, %r11
adcq %r14, %r8
adcq $0, %r9
movq 16(%rcx), %rdx
mulxq (%rsi), %rax, %r12
addq %rax, %r10
movq %r10, 16(%rdi)
mulxq 8(%rsi), %rax, %r13
adcq %rax, %r11
mulxq 16(%rsi), %rax, %r14
adcq %rax, %r8
mulxq 24(%rsi), %rax, %r9
adcq %rax, %r9
adcq $0, %r10
addq %r12, %r11
adcq %r13, %r8
adcq %r14, %r9
adcq $0, %r10
movq 24(%rcx), %rdx
mulxq (%rsi), %rax, %r12
addq %rax, %r11
movq %r11, 24(%rdi)
mulxq 8(%rsi), %rax, %r13
adcq %rax, %r8
mulxq 16(%rsi), %rax, %r14
adcq %rax, %r9
mulxq 24(%rsi), %rax, %r9
adcq %rax, %r10
adcq $0, %r10
addq %r12, %r8
movq %r8, 32(%rsi)
adcq %r13, %r9
movq %r9, 40(%rsi)
adcq %r14, %r10
movq %r10, 48(%rsi)
adcq $0, %r11
movq %r11, 56(%rsi)
movq -8(%rsp), %r12
movq -16(%rsp), %r13
movq -24(%rsp), %r14
ret
*/

复制代码

Performance counter stats for './a.out':
25,636.00 msec task-clock # 0.999 CPUs utilized
4,351 context-switches # 169.722 /sec
283 cpu-migrations # 11.039 /sec
47 page-faults # 1.833 /sec
100,564,298,132 cycles # 3.923 GHz (83.32%)
304,933,814 stalled-cycles-frontend # 0.30% frontend cycles idle (83.33%)
77,948,700,897 stalled-cycles-backend # 77.51% backend cycles idle (83.34%)
262,154,531,908 instructions # 2.61 insn per cycle
# 0.30 stalled cycles per insn (83.33%)
4,334,044,774 branches # 169.061 M/sec (83.34%)
1,346,833 branch-misses # 0.03% of all branches (83.34%)
25.669427577 seconds time elapsed
25.625846000 seconds user
0.011995000 seconds sys

复制代码

Timeline view:
0123456789 0123
Index 0123456789 0123456789
[0,0] DeER . . . . . . . movq %rdx, %rcx
[0,1] DeeeeeER . . . . . . movq (%rdx), %rdx
[0,2] DeE----R . . . . . . movq %r12, -8(%rsp)
[0,3] .DeeeeeeeeeER . . . . . mulxq (%rsi), %r8, %r12
[0,4] .D========eER . . . . . movq %r8, (%rdi)
[0,5] .D=========eER . . . . . movq %r13, -16(%rsp)
[0,6] . DeeeeeeeeeER . . . . . mulxq 8(%rsi), %r9, %r13
[0,7] . D========eER . . . . . addq %r12, %r9
[0,8] . D=========eER. . . . . movq %r14, -24(%rsp)
[0,9] . DeeeeeeeeeER. . . . . mulxq 16(%rsi), %r10, %r14
[0,10] . D========eER. . . . . adcq %r13, %r10
[0,11] . DeeeeeeeeeER . . . . mulxq 24(%rsi), %r11, %r8
[0,12] . D========eER . . . . adcq %r14, %r11
[0,13] . D=========eER . . . . adcq $0, %r8
[0,14] . DeeeeeE----R . . . . movq 8(%rcx), %rdx
[0,15] . DeeeeeeeeeER . . . . mulxq (%rsi), %rax, %r12
[0,16] . D========eER . . . . addq %rax, %r9
[0,17] . .D========eER . . . . movq %r9, 8(%rdi)
[0,18] . .DeeeeeeeeeER . . . . mulxq 8(%rsi), %rax, %r13
[0,19] . .D========eER . . . . adcq %rax, %r10
[0,20] . . DeeeeeeeeeER . . . . mulxq 16(%rsi), %rax, %r14
[0,21] . . D========eER . . . . adcq %rax, %r11
[0,22] . . DeeeeeeeeeER. . . . mulxq 24(%rsi), %rax, %r9
[0,23] . . D========eER. . . . adcq %rax, %r8
[0,24] . . D=========eER . . . adcq $0, %r9
[0,25] . . D======eE--R . . . addq %r12, %r10
[0,26] . . D=======eE-R . . . adcq %r13, %r11
[0,27] . . D========eER . . . adcq %r14, %r8
[0,28] . . D=========eER . . . adcq $0, %r9
[0,29] . . DeeeeeE-----R . . . movq 16(%rcx), %rdx
[0,30] . . DeeeeeeeeeER . . . mulxq (%rsi), %rax, %r12
[0,31] . . D========eER . . . addq %rax, %r10
[0,32] . . D=========eER . . . movq %r10, 16(%rdi)
[0,33] . . .DeeeeeeeeeER . . . mulxq 8(%rsi), %rax, %r13
[0,34] . . .D========eER . . . adcq %rax, %r11
[0,35] . . . DeeeeeeeeeER . . . mulxq 16(%rsi), %rax, %r14
[0,36] . . . D========eER . . . adcq %rax, %r8
[0,37] . . . DeeeeeeeeeER. . . mulxq 24(%rsi), %rax, %r9
[0,38] . . . D=========eER . . adcq %rax, %r9
[0,39] . . . D==========eER . . adcq $0, %r10
[0,40] . . . D======eE---R . . addq %r12, %r11
[0,41] . . . D=======eE--R . . adcq %r13, %r8
[0,42] . . . D=========eER . . adcq %r14, %r9
[0,43] . . . D==========eER . . adcq $0, %r10
[0,44] . . . DeeeeeE------R . . movq 24(%rcx), %rdx
[0,45] . . . DeeeeeeeeeE-R . . mulxq (%rsi), %rax, %r12
[0,46] . . . D========eE-R . . addq %rax, %r11
[0,47] . . . D=========eER . . movq %r11, 24(%rdi)
[0,48] . . . .DeeeeeeeeeER . . mulxq 8(%rsi), %rax, %r13
[0,49] . . . .D========eER . . adcq %rax, %r8
[0,50] . . . . DeeeeeeeeeER . . mulxq 16(%rsi), %rax, %r14
[0,51] . . . . D========eER . . adcq %rax, %r9
[0,52] . . . . DeeeeeeeeeER. . mulxq 24(%rsi), %rax, %r9
[0,53] . . . . D========eER. . adcq %rax, %r10
[0,54] . . . . D=========eER . adcq $0, %r10
[0,55] . . . . D======eE--R . addq %r12, %r8
[0,56] . . . . D=======eE-R . movq %r8, 32(%rsi)
[0,57] . . . . D========eER . adcq %r13, %r9
[0,58] . . . . D=========eER . movq %r9, 40(%rsi)
[0,59] . . . . D=========eER . adcq %r14, %r10
[0,60] . . . . D==========eER. movq %r10, 48(%rsi)
[0,61] . . . . D=========eER. adcq $0, %r11
[0,62] . . . . D==========eER movq %r11, 56(%rsi)
[0,63] . . . . DeeeeeE------R movq -8(%rsp), %r12
[0,64] . . . . DeeeeeE------R movq -16(%rsp), %r13
[0,65] . . . . D=eeeeeE-----R movq -24(%rsp), %r14
[0,66] . . . . .DeeeeeeeE---R retq

复制代码

补充内容 (2023-5-5 09:32):
程序修復後測試了正確性及效率 https://pastebin.com/nu1xu5jE

nyy · 发表于 2023-5-4 08:59:41

汇编是怎么写怎么调试的？

l4m2 · 发表于 2023-5-6 02:24:32

https://pastebin.com/gktPUxJY
這個要40s但是我不知道瓶頸在哪。B讀進寄存器也差不多40

账号		自动登录	找回密码
密码			欢迎注册

[讨论] 64位平台的256位乘法测试

马上注册，结交更多好友，享用更多功能，让你轻松玩转社区。

点评

点评

点评

点评

点评

点评

点评