wmurmann
Reputation: 115
Segmentation Fault when using vmovupd
I am trying to input four flouting point numbers at time into the stack then transfer it into a ymm(avx) register. A friend of mine is working on the same project and our code seems identical but I'm getting a core dump when I call vmovupd ymm0, [rsp]. I narrowed it down to that call with a debug program.
;===== Begin code area ====================================================================================================================================================
%include "debug.inc"
extern printf ;External C++ function for writing to standard output device
extern scanf ;External C++ function for reading from the standard input device
global intrest_calc ;This makes trapezoid_area callable by functions outside of this file.
segment .data ;Place initialized data here
;===== Declare some messages ==============================================================================================================================================
initialmessage db "Welcome to the Bank of Catalina Island", 10, 0
officer db "William Murmann, Cheif Loan Officer: ", 0
rate db "Please enter the current interest rate as a float number: ", 0
months db "Please enter the the time in months for the loan: ", 0
amounts db "Please enter the amounts of the four loans:", 0
xsavenotsupported.notsupportedmessage db "The xsave instruction and the xrstor instruction are not supported in this microprocessor.", 10
db "However, processing will continue without backing up state component data", 10, 0
goodbye db "Have a nice day. Enjoy your trapezoids. ", 10, 0
stringformat db "%s", 0 ;general string format
fourfloatformat db "%lf %lf %lf %lf",0 ;four float format
xsavenotsupported.stringformat db "%s", 0
eight_byte_format db "%lf", 0 ;general 8-byte float format
segment .bss ;Place un-initialized data here.
align 64 ;Insure that the inext data declaration starts on a 64-byte boundar.
backuparea resb 832 ;Create an array for backup storage having 832 bytes.
;===== Begin executable instructions here =================================================================================================================================
segment .text ;Place executable instructions in this segment.
intrest_calc: ;Entry point. Execution begins here.
;=========== Back up all the GPRs whether used in this program or not =====================================================================================================
push rbp ;Save a copy of the stack base pointer
mov rbp, rsp ;We do this in order to be 100% compatible with C and C++.
push rbx ;Back up rbx
push rcx ;Back up rcx
push rdx ;Back up rdx
push rsi ;Back up rsi
push rdi ;Back up rdi
push r8 ;Back up r8
push r9 ;Back up r9
push r10 ;Back up r10
push r11 ;Back up r11
push r12 ;Back up r12
push r13 ;Back up r13
push r14 ;Back up r14
push r15 ;Back up r15
pushf ;Back up rflags
;==========================================================================================================================================================================
;===== Begin State Component Backup =======================================================================================================================================
;==========================================================================================================================================================================
;=========== Before proceeding verify that this computer supports xsave and xrstor ========================================================================================
;Bit #26 of rcx, written rcx[26], must be 1; otherwise xsave and xrstor are not supported by this computer.
;Preconditions: rax holds 1.
mov rax, 1
;Execute the cpuid instruction
cpuid
;Postconditions: If rcx[26]==1 then xsave is supported. If rcx[26]==0 then xsave is not supported.
;=========== Extract bit #26 and test it ==================================================================================================================================
and rcx, 0x0000000004000000 ;The mask 0x0000000004000000 has a 1 in position #26. Now rcx is either all zeros or
;has a single 1 in position #26 and zeros everywhere else.
cmp rcx, 0 ;Is (rcx == 0)?
je xsavenotsupported ;Skip the section that backs up state component data.
;========== Call the function to obtain the bitmap of state components ====================================================================================================
;Preconditions
mov rax, 0x000000000000000d ;Place 13 in rax. This number is provided in the Intel manual
mov rcx, 0 ;0 is parameter for subfunction 0
;Call the function
cpuid ;cpuid is an essential function that returns information about the cpu
;Postconditions (There are 2 of these):
;1. edx:eax is a bit map of state components managed by xsave. At the time this program was written (2014 June) there were exactly 3 state components. Therefore, bits
; numbered 2, 1, and 0 are important for current cpu technology.
;2. ecx holds the number of bytes required to store all the data of enabled state components. [Post condition 2 is not used in this program.]
;This program assumes that under current technology (year 2014) there are at most three state components having a maximum combined data storage requirement of 832 bytes.
;Therefore, the value in ecx will be less than or equal to 832.
;Precaution: As an insurance against a future time when there will be more than 3 state components in a processor of the X86 family the state component bitmap is masked to
;allow only 3 state components maximum.
mov r15, 7 ;7 equals three 1 bits.
and rax, r15 ;Bits 63-3 become zeros.
mov r15, 0 ;0 equals 64 binary zeros.
and rdx, r15 ;Zero out rdx.
;========== Save all the data of all three components except GPRs =========================================================================================================
;The instruction xsave will save those state components with on bits in the bitmap. At this point edx:eax continues to hold the state component bitmap.
;Precondition: edx:eax holds the state component bit map. This condition has been met by the two pops preceding this statement.
xsave [backuparea] ;All the data of state components managed by xsave have been written to backuparea.
push qword -1 ;Set a flag (-1 = true) to indicate that state component data were backed up.
jmp startapplication
;========== Show message xsave is not supported on this platform ==========================================================================================================
xsavenotsupported:
mov rax, 0
mov rdi, .stringformat
mov rsi, .notsupportedmessage ;"The xsave instruction is not suported in this microprocessor.
call printf
push qword 0 ;Set a flag (0 = false) to indicate that state component data were not backed up.
;==========================================================================================================================================================================
;===== End of State Component Backup ======================================================================================================================================
;==========================================================================================================================================================================
;==========================================================================================================================================================================
startapplication: ;===== Begin the application here: Trapezoid area calculation =======================================================================================
;==========================================================================================================================================================================
;=========== Show the initial message =====================================================================================================================================
mov qword rax, 0 ;No data from SSE will be printed
mov rdi, stringformat ;"%s"
mov rsi, initialmessage ;
call printf ;Call a library function to make the output
;=========== Show the officier name =====================================================================================================================================
mov qword rax, 0 ;No data from SSE will be printed
mov rdi, stringformat ;"%s"
mov rsi, officer ;
call printf ;Call a library function to make the output
;=========== Prompt for the rate =============================================================================================================================
mov qword rax, 0 ;No data from SSE will be printed
mov rdi, stringformat ;"%s"
mov rsi, rate ;"Enter the rate: "
call printf ;Call a library function to make the output
;===== Obtain the first base number from the standard input device and store a copy in xmm15 =============================================================================
push qword 0 ;Reserve 8 bytes of storage for the incoming number
mov qword rax, 0 ;SSE is not involved in this scanf operation
mov rdi, eight_byte_format ;"%lf"
mov rsi, rsp ;Give scanf a point to the reserved storage
call scanf ;Call a library function to do the input work
movsd xmm15, [rsp] ;Copy the inputted number to xmm15
pop rax ;Make free the storage that was used by scanf
;=========== Prompt for the amounts =============================================================================================================================
mov qword rax, 0 ;No data from SSE will be printed
mov rdi, stringformat ;"%s"
mov rsi, amounts ;"Enter the other base number: "
call printf ;Call a library function to make the output
;=========== Obtain the amounts =============================================================================================================================
push qword 0
push qword 0
push qword 0
push qword 0
mov rax,0
mov rdi, fourfloatformat
mov rsi, rsp
add rdx, rsp
mov rdx, 8
mov rcx, rsp
add rcx, 16
mov r8, rsp
add r8, 24
call scanf
vmovupd ymm0, [rsp]
pop rax
pop rax
pop rax
pop rax
;=========== Prompt for the amounts =============================================================================================================================
mov qword rax, 0 ;No data from SSE will be printed
mov rdi, stringformat ;"%s"
mov rsi, amounts ;"Enter the other base number: "
call printf ;Call a library function to make the output
;===== Retrieve a copy of the quotient that was backed up earlier =========================================================================================================
pop r14 ;A copy of the quotient is in r14 (temporary storage)
;Now the stack is in the same state as when the application area was entered. It is safe to leave this application area.
;==========================================================================================================================================================================
;===== Begin State Component Restore ======================================================================================================================================
;==========================================================================================================================================================================
;===== Check the flag to determine if state components were really backed up ==============================================================================================
pop rbx ;Obtain a copy of the flag that indicates state component backup or not.
cmp rbx, 0 ;If there was no backup of state components then jump past the restore section.
je setreturnvalue ;Go to set up the return value.
;Continue with restoration of state components;
;Precondition: edx:eax must hold the state component bitmap. Therefore, go get a new copy of that bitmap.
;Preconditions for obtaining the bitmap from the cpuid instruction
mov rax, 0x000000000000000d ;Place 13 in rax. This number is provided in the Intel manual
mov rcx, 0 ;0 is parameter for subfunction 0
;Call the function
cpuid ;cpuid is an essential function that returns information about the cpu
;Postcondition: The bitmap in now in edx:eax
;Future insurance: Make sure the bitmap is limited to a maximum of 3 state components.
mov r15, 7
and rax, r15
mov r15, 0
and rdx, r15
xrstor [backuparea]
;==========================================================================================================================================================================
;===== End State Component Restore ========================================================================================================================================
;==========================================================================================================================================================================
setreturnvalue: ;=========== Set the value to be returned to the caller ===================================================================================================
push r14 ;r14 continues to hold the first computed floating point value.
movsd xmm0, [rsp] ;That first computed floating point value is copied to xmm0[63-0]
pop r14 ;Reverse the push of two lines earlier.
;=========== Restore GPR values and return to the caller ==================================================================================================================
popf ;Restore rflags
pop r15 ;Restore r15
pop r14 ;Restore r14
pop r13 ;Restore r13
pop r12 ;Restore r12
pop r11 ;Restore r11
pop r10 ;Restore r10
pop r9 ;Restore r9
pop r8 ;Restore r8
pop rdi ;Restore rdi
pop rsi ;Restore rsi
pop rdx ;Restore rdx
pop rcx ;Restore rcx
pop rbx ;Restore rbx
pop rbp ;Restore rbp
ret ;No parameter with this instruction. This instruction will pop 8 bytes from
;the integer stack, and jump to the address found on the stack.
;========== End of program
Upvotes: 0
Views: 304
Answers (1)
Jester
Reputation: 58762
You used the debugger wrong. Pay more attention.
The fault is not at the vmovupd, rather it is inside scanf. The reason is that you have 2 lines swapped in your code, hence the pointer argument rdx is messed up. I trust you can see the problem from this hint.
Furthermore, I can't find the counterpart to the Retrieve a copy of the quotient that was backed up earlier, and apparently neither can my cpu. As such, the stack is unbalanced and the return address will be corrupted.
Also, you don't seem to care much about stack alignment, which may bite you with unexpected segfaults too.
Upvotes: 2
Related Questions
- YASM: vmovaps instruction causing segmentation fault
- Why I got a segmentation fault?
- NASM-64bits-segmentation fault calling procedure
- Segmentation fault in mov instruction
- Why does mov rax,cr0 cause a segmentation fault in nasm 64 bit
- Why does vmovapd cause segfault?
- segmentation fault for `vmovaps'
- Segmentation fault movsb nasm in 64 bits linux
- NASM x86 Segfault
- Segmentation Fault when using "mov" in Assembly