下面是一个多线程(4核)NASM 64程序。这不是最小的,但是是完整的。我发布了完整的代码,因为一个最小的示例可能无法揭示此代码中的问题。
这是我在NASM中的第一个多线程多核程序,昨天下午我很高兴看到它正常工作(所有四个核)并返回了我期望的值。我昨天成功运行了几次,今天早上又成功运行了几次。
突然之间,在我上次运行大约10分钟后,未对代码进行任何更改,未重新组装dll,我再次运行它,这次线程是随机执行的-并非所有线程都执行,并且配置执行的线程因运行而异-有时仅线程1;有时是线程2-4,但不是线程1,有时只有线程2和4,等等。
我的代码没有ExitThread或CloseHandle调用,所以我将它们添加到了(请参阅label_899 :),重新启动并再次运行它。全新重新引导后的第一次运行仍然显示随机执行线程。
我认为问题可能是CreateThread失败,所以我也返回了线程句柄。每次都创建了所有四个线程句柄,即使它们都没有执行,事实并非如此。
我已经做了很多研究,但是我根本没有看到讨论这个问题。即使这是NASM,同一件事也可能适用于C或C ++多线程。我使用WaitForMultipleObjects等待所有线程完成。
这是一个dll,入口点是Main_Entry_fn。 dll是使用ctypes从Python调用的,但是我怀疑这有什么区别。我经常使用ctypes,从来没有遇到过问题。
感谢任何想法。
P.S。忽略realloc代码,因为缓冲区足够大,以至于永不调用realloc。
更新:这是我构造程序流的方式:入口点是程序列表底部的Main_Entry_fn(这是导出操作,因为必须为dll调用而导出)。 Main_Entry_fn从调用函数中获取输入数据,保留寄存器(因为整个程序都经过寄存器优化),并分配输出缓冲区。然后Main_Entry_fn调用Init_Cores_fn(在程序清单的顶部),在其中我们创建四个带有数组的数组,这些数组包含每个内核入口处寄存器的数据(请记住所有var存储在寄存器中,请参见对DupThreadInfo的调用),并在一个循环中创建四个线程。所有线程都调用Prime_Number_fn,并将线程设置为在创建后立即启动。
注册分配:
r15 list_of_results_ptr(指向list_of_results的指针)
r14 list_of_results_ctr(针对list_of_results的计数器)
r13 list_of_results_length
r12numbers_ptr(指向“数字”输入数组的指针)
r11数字_ctr
r10numbers_length
r9 num
r8 i
xmm15 num_float
xmm14:结果
; Header Section
[BITS 64]
[default rel]
extern malloc, calloc, realloc, free
global Main_Entry_fn
export Main_Entry_fn
global FreeMem_fn
export FreeMem_fn
extern CreateThread, CloseHandle, ExitThread
extern WaitForMultipleObjects, GetCurrentThread
section .data align=16
Return_Pointer_Array: dq 0, 0, 0
Input_Length_Array: dq 0, 0,
list_of_results_ptr: dq 0
list_of_results_ctr: dq 0
list_of_results_length: dq 0
data_master_ptr: dq 0
initial_dynamic_length: dq 0
internal_dynamic_length: dq 5000
XMM_Stack: dq 0, 0, 0
numbers_ptr: dq 0
numbers_length: dq 0
numbers: dq 0
list_of_results: dq 0
num_float: dq 0.0
loop_counter_401: dq 0
numbers_ctr: dq 0
num: dq 0
result: dq 0
const_1: dq 1
i: dq 0
range_loop_start_i: dq 0
range_loop_end_i: dq 0
const_0: dq 0
; New vars for threads:
ThreadCount: times 4 dq 0
ThreadInfo: times 10 dq 0
ThreadInfo2: times 10 dq 0
ThreadInfo3: times 10 dq 0
ThreadInfo4: times 10 dq 0
TestInfo: times 4 dq 0
ThreadHandles: times 4 dq 0
Return_Data_Array: times 8 dq 0 ; ptr,ptr,ptr,ptr,length,length,length,length
StartByte: dq 0
stride: dq 8
output_buffer_pointers: times 4 dq 0 ; for malloc
Division_Size: dq 0
Division_Start: dq 0
section .text
; ______________________________________
Init_Cores_fn:
; Calculate the data divisions
mov rax,r10
mov rbx,4 ;cores
xor rdx,rdx
div rbx
mov [Division_Size],rax
mov [Division_Start],rax
; Populate the ThreadInfo array with vars to pass
; ThreadInfo: length, startbyte, stride, vars into registers on entry to each core
mov rdi,ThreadInfo
mov rax,0 ;ThreadInfoLength
mov [rdi],rax ; length (number of vars into registers plus 3 elements)
mov rax,[stride]
mov [rdi+16],rax ; 8 x number of cores (32 in this example)
; Vars (these registers are populated on main entry)
mov rbp,output_buffer_pointers
mov rax,[rbp]
mov [rdi+24],rax ;for r15
mov [rdi+32],r14
mov [rdi+40],r13
mov [rdi+48],r12
mov rax,[Division_Start]
mov [rdi+56],rax
mov rax,0
mov [rdi+64],rax
call DupThreadInfo
mov rbp,rsp ; preserve caller's stack frame
sub rsp,56 ; Shadow space (was 32)
; _____
label_0:
mov rax,[StartByte]
cmp rax,0
jne sb2
mov rdi,ThreadInfo
jmp sb5
sb2:cmp rax,8
jne sb3
mov rdi,ThreadInfo2
jmp sb5
sb3:cmp rax,16
jne sb4
mov rdi,ThreadInfo3
jmp sb5
sb4:cmp rax,24
jne sb5
mov rdi,ThreadInfo4
sb5:
; _____
; Create Threads
mov rcx,0 ; lpThreadAttributes (Security Attributes)
mov rdx,0 ; dwStackSize
mov r8,PrimeNumber_fn ; lpStartAddress (function pointer)
mov r9,rdi ; lpParameter (array of data passed to each core)
mov rax,0
mov [rsp+32],rax ; use default creation flags
mov rdi,ThreadCount
mov [rsp+40],rdi ; ThreadID
call CreateThread
; Move the handle into ThreadHandles array (returned in rax)
mov rdi,ThreadHandles
mov rcx,[StartByte]
mov [rdi+rcx],rax
mov rax,[StartByte]
add rax,8
mov [StartByte],rax
mov rbx,32 ; Four cores
cmp rax,rbx
jl label_0
; _____
; Wait
mov rcx,4 ;rax ; number of handles
mov rdx,ThreadHandles ; pointer to handles array
mov r8,0 ; wait for all threads to complete
mov r9,5000 ; milliseconds to wait
call WaitForMultipleObjects
; _____
mov rsp,rbp
jmp label_900
; ______________________________________
PrimeNumber_fn:
; Populate registers
;(note: rcx is the return value for ThreadProc)
mov rdi,rcx
mov rax,[rdi]
mov r15,[rdi+24]
mov r13,[rdi+40]
mov r12,[rdi+48]
mov r10,[rdi+56]
xor r11,r11
xor r9,r9
mov r8,[rdi+8] ; start byte
pxor xmm15,xmm15
pxor xmm15,xmm14
pxor xmm15,xmm13
; Get the ThreadID based on startbyte
mov rax,[rdi+72] ; 0, 8, 16, 24 for 4-core
push rax
;______
label_401:
mov rdi,r12 ; Pointer
cmp r8,r10
jge label_899
movsd xmm0,qword[rdi+r8]
movsd xmm15,xmm0
add r8,8
;______
cvttsd2si rax,xmm15
mov r9,rax
;______
label_8010:
label_801:
cmp r9,[const_1]
jle label_401
;______
label_12010:
mov rax,2
sub rax,1
movq xmm13,rax
label_1201:
movq rcx,xmm13
movq xmm1,[const_1]
addsd xmm13,xmm1
movq rcx,xmm13
mov rdx,r9
cmp rcx,rdx
jge label_12020
;______
label_16010:
label_1601:
mov rax,r9
movq rbx,xmm13
xor rdx,rdx
div rbx
mov rax,rdx
mov rdx,0
cmp rax,rdx
jne label_16020
;______
movq xmm14,[const_0]
;______
movq rax,xmm14
cvtsi2sd xmm0,rax
movsd [r15+r14],xmm14
add r14,8
mov rax,r13
cmp r14,rax
jl next_17
;[Irrelevant code omitted]
next_17:
;______
jmp label_401
;______
label_1602:
label_16020:
;______
mov rax,r9
cvtsi2sd xmm14,rax
;______
movq rax,xmm14
cvtsi2sd xmm0,rax
movsd [r15+r14],xmm14
add r14,8
mov rax,r13
cmp r14,rax
jl next_21
;[Irrelevant code omitted]
next_21:
jmp label_1201
;______
label_1202:
label_12020:
;______
movq xmm14,[const_0]
;______
movq rax,xmm14
cvtsi2sd xmm0,rax
movsd [r15+r14],xmm14
add r14,8
mov rax,r13
cmp r14,rax
jl next_25
;[Irrelevant code omitted]
next_25:
jmp label_401
;______
label_899:
pop rax
mov rdi,Return_Data_Array
mov [rdi+rax],r15
mov [rdi+rax+32],r14 ; 32 = four cores
call ExitThread
call GetCurrentThread
mov rcx,rax
call CloseHandle
ret
; __________
label_900:
exit_label_for_PrimeNumber_fn:
mov rdi,Return_Data_Array ; Final return to the calling process
mov rax,rdi
ret
;__________
;Free the memory
FreeMem_fn:
sub rsp,40
call free
add rsp,40
ret
; __________
; Main Entry
Main_Entry_fn:
push rdi
push rbp
push rbx
push r15
xor r15,r15
push r14
xor r14,r14
push r13
xor r13,r13
push r12
xor r12,r12
push r11
xor r11,r11
push r10
xor r10,r10
push r9
xor r9,r9
movsd [XMM_Stack+0],xmm15
movsd [XMM_Stack+8],xmm14
movsd [XMM_Stack+16],xmm13
push r8
xor r8,r8
mov [numbers_ptr],rcx
mov [data_master_ptr],rdx
; Now assign lengths
lea rdi,[data_master_ptr]
mov rbp,[rdi]
xor rcx,rcx
movsd xmm0,qword[rbp+rcx]
cvttsd2si rax,xmm0
mov [numbers_length],rax
add rcx,8
; __________
; malloc for dynamic arrays
lea rdi,[data_master_ptr]
mov rbp,[rdi]
movsd xmm0,qword[rbp]
cvttsd2si rax,xmm0
mov r8,rax
mov rdx,10
mul rdx
mov rdx,10000000
cmp rax,rdx
jl malloc_next
mov rax,r8
malloc_next:
mov rax,50000000
mov [initial_dynamic_length],rax
;__________
; Output buffer #1
mov rcx,qword[initial_dynamic_length] ; Initial size
xor rax,rax
sub rsp,40
call malloc
mov rdi,output_buffer_pointers
mov [rdi],rax
add rsp,40
mov rax,qword[initial_dynamic_length]
mov [list_of_results_length],rax
;__________
; Output buffer #2
mov rcx,qword[initial_dynamic_length] ; Initial size
xor rax,rax
sub rsp,40
call malloc
mov rdi,output_buffer_pointers
mov [rdi+8],rax
add rsp,40
mov rax,qword[initial_dynamic_length]
mov [list_of_results_length],rax
;__________
; Output buffer #3
mov rcx,qword[initial_dynamic_length] ; Initial size
xor rax,rax
sub rsp,40
call malloc
mov rdi,output_buffer_pointers
mov [rdi+16],rax
add rsp,40
mov rax,qword[initial_dynamic_length]
mov [list_of_results_length],rax
;__________
; Output buffer #4
mov rcx,qword[initial_dynamic_length] ; Initial size
xor rax,rax
sub rsp,40
call malloc
mov rdi,output_buffer_pointers
mov [rdi+24],rax
add rsp,40
mov rax,qword[initial_dynamic_length]
mov [list_of_results_length],rax
; __________
; Write variables to assigned registers
lea rdi,[rel list_of_results_ptr]
mov r15,qword[rdi]
mov r14,[list_of_results_ctr]
mov r13,[list_of_results_length]
lea rdi,[rel numbers_ptr]
mov r12,qword[rdi]
mov r11,[numbers_ctr]
mov r10,[numbers_length]
mov r9,[num]
movsd xmm15,[num_float]
movsd xmm14,[result]
movsd xmm13,[i]
mov r8,[loop_counter_401]
; __________
call Init_Cores_fn
exit_label_for_Main_Entry_fn:
pop r8
movsd xmm13,[XMM_Stack+0]
movsd xmm14,[XMM_Stack+8]
movsd xmm15,[XMM_Stack+16]
pop r9
pop r10
pop r11
pop r12
pop r13
pop r14
pop r15
pop rbx
pop rbp
pop rdi
ret
;_____________
DupThreadInfo:
mov rdi,ThreadInfo2
; StartByte and EndByte
mov rax,[Division_Start]
mov [rdi+8],rax
add rax,[Division_Size]
mov [Division_Start],rax
sub rax,8
mov [rdi+56],rax
mov rax,[stride]
mov [rdi+16],rax ; 8 x number of cores (32 in this example)
; Vars (these registers are populated on main entry)
mov rbp,output_buffer_pointers
mov rax,[rbp+8]
mov [rdi+24],rax
mov rax,8
mov [rdi+32],rax
mov [rdi+40],r13
mov [rdi+48],r12
; 56 is above
mov rax,8
mov [rdi+64],rax
mov rax,8 ; Thread number based on startbyte, for Return_Data_Array
mov [rdi+72],rax
; _____
mov rdi,ThreadInfo3
; StartByte and EndByte
mov rax,[Division_Start]
mov [rdi+8],rax
mov rbx,[Division_Size]
add rax,rbx
mov [Division_Start],rax
sub rax,8
mov [rdi+56],rax
mov rax,[stride]
mov [rdi+16],rax ; 8 x number of cores (32 in this example)
; Vars (these registers are populated on main entry)
mov rbp,output_buffer_pointers
mov rax,[rbp+16]
mov [rdi+24],rax
mov rax,16
mov [rdi+32],rax
mov [rdi+40],r13
mov [rdi+48],r12
; [rdi+56] is above
mov rax,16
mov [rdi+64],rax
mov rax,16 ; Thread number based on startbyte, for Return_Data_Array
mov [rdi+72],rax
mov rdi,ThreadInfo4
; StartByte and EndByte
mov rax,[Division_Start]
mov [rdi+8],rax
mov rax,[numbers_length] ; final segment goes to the end
mov [rdi+56],rax
mov rax,[stride]
mov [rdi+16],rax ; 8 x number of cores (32 in this example)
; Vars (these registers are populated on main entry)
mov rbp,output_buffer_pointers
mov rax,[rbp+24]
mov [rdi+24],rax
mov rax,24
mov [rdi+32],rax
mov [rdi+40],r13
mov [rdi+48],r12
; [rdi+56] is above
mov rax,24
mov [rdi+64],rax
mov rax,24 ; Thread number based on startbyte, for Return_Data_Array
mov [rdi+72],rax
ret