* Re: Questions and possible optimizations about sysv/linux/[x86_64|i386]/swapcontext.S
2018-05-20 13:15 Questions and possible optimizations about sysv/linux/[x86_64|i386]/swapcontext.S Remus Clearwater
@ 2018-05-20 13:36 ` Remus Clearwater
0 siblings, 0 replies; 2+ messages in thread
From: Remus Clearwater @ 2018-05-20 13:36 UTC (permalink / raw)
To: libc-help
Really sorry for the unexpected auto typesetting by the mail-list server.
----
Hi, I'm reading the sources of ucontext.
According to https://en.wikipedia.org/wiki/X86_calling_conventions#cdecl:
Integer values and memory addresses are returned in the EAX register,
floating point values in the ST0 x87 register. Registers EAX, ECX, and EDX
are caller-saved, and the rest are callee-saved. The x87 floating point
registers ST0 to ST7 must be empty (popped or freed) when calling a new
function, and ST1 to ST7 must be empty on exiting a function. ST0 must also
be empty when not used for returning a value.
And https://en.wikipedia.org/wiki/X86_calling_conventions#System_V_AMD64_ABI
:
If the callee wishes to use registers RBX, RBP, and R12–R15, it must
restore their original values before returning control to the caller. All
other registers must be saved by the caller if it wishes to preserve their
values.
Questions:
1. Since `swapcontext` is a function call, so it maybe not need to
save/restore (fnstenv,stmxcsr/fldenv,ldmxcsr) the floating-point context.
Is this correct? If it's wrong, where I could found the detailed calling
convention about the floating-point context ? Or a bug example after the
(fnstenv,stmxcsr/fldenv,ldmxcsr) part is eliminated?
2. Could the save/restore (fnstenv,stmxcsr/fldenv,ldmxcsr) be eliminated if
all the functions which uses fpu are leaf-node function (function which do
not call any other function) ? If it is true, where I could found the
detailed reference?
3. Since `swapcontext` is a function call, so it is only necessary to
save/restore RBX, RBP, R12–R15, RSP and RIP, the save/restore of RDI, RSI,
RDX, RCX, R8 and R8 is quite useless and can be eliminated which could save
at least (6*8*2 = 96) times memory read/writes in each swapcontext
operation (Same optimization also exists in i386/swapcontext.S). Is this
correct?
4. In i386/swapcontext.S:
54 /* Restore the FS segment register. We don't touch the GS
register
55 since it is used for threads. */
56 movl oFS(%eax), %ecx
57 movw %cx, %fs
But there is no such save/restore about register fs in
x86_64/swapcontext.S. Is this a bug or it's no need to save/restore the
register fs at all ? (The save/restore of register fs in i386/swapcontext.S
could also be elimianted if the later is correct)
Thanks a lot :)
On Sun, May 20, 2018 at 9:15 PM, Remus Clearwater <
remus.clearwater@gmail.com> wrote:
> Hi, I'm reading the sources of ucontext
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/x86_64/swapcontext.S;h=1110c479fad27829f1b1e480cf7520cf77d795f7;hb=HEAD#l46>
> .
>
> According to https://en.wikipedia.org/wiki/X86_calling_conventions#cdecl:
>
> Integer values and memory addresses are returned in the EAX register
>> <https://en.wikipedia.org/wiki/Processor_register>, floating point
>> values in the ST0 x87 <https://en.wikipedia.org/wiki/X87> register.
>> Registers EAX, ECX, and EDX are caller-saved, and the rest are
>> callee-saved. The x87 <https://en.wikipedia.org/wiki/X87> floating point
>> registers ST0 to ST7 must be empty (popped or freed) when calling a new
>> function, and ST1 to ST7 must be empty on exiting a function. ST0 must also
>> be empty when not used for returning a value.
>
>
> And https://en.wikipedia.org/wiki/X86_calling_conventions#Sy
> stem_V_AMD64_ABI:
>
> If the callee wishes to use registers RBX, RBP, and R12–R15, it must
>> restore their original values before returning control to the caller. All
>> other registers must be saved by the caller if it wishes to preserve their
>> values.
>
>
> Questions:
>
> 1. Since `swapcontext` is a function call, so it maybe not need to
> save/restore
> (fnstenv,stmxcsr/fldenv,ldmxcsr) the floating-point context. Is this
> correct?
> If it's wrong, where I could found the detailed calling convention about
> the
> floating-point context ? Or a bug example after the (
> fnstenv,stmxcsr/fldenv,ldmxcsr)
> part is eliminated?
>
> 2. Could the save/restore (fnstenv,stmxcsr/fldenv,ldmxcsr) be eliminated
> if all
> the functions which uses fpu are leaf-node function (function which do not
> call
> any other function) ? If it is true, where I could found the detailed
> reference?
>
> 3. Since `swapcontext` is a function call, so it is only necessary to
> save/restore
> RBX, RBP, R12–R15, RSP and RIP, the save/restore of RDI, RSI, RDX, RCX, R8
> and R8 is quite useless and can be eliminated which could save at least
> (6*8*2 = 96)
> times memory read/writes in each swapcontext operation (Same optimization
> also
> exists in i386/swapcontext.S
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/i386/setcontext.S;h=a604fca0aceaa17f19f536bac46a1fc305f30860;hb=HEAD#l77>).
> Is this correct?
>
> 4. In i386/swapcontext.S
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/i386/setcontext.S;h=a604fca0aceaa17f19f536bac46a1fc305f30860;hb=HEAD#l54>
> :
>
> 54
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/i386/setcontext.S;h=a604fca0aceaa17f19f536bac46a1fc305f30860;hb=HEAD#l54>
> /* Restore the FS segment register. We don't touch
> the GS register
> 55
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/i386/setcontext.S;h=a604fca0aceaa17f19f536bac46a1fc305f30860;hb=HEAD#l55>
> since it is used for threads. */
> 56
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/i386/setcontext.S;h=a604fca0aceaa17f19f536bac46a1fc305f30860;hb=HEAD#l56>
> movl oFS(%eax), %ecx
> 57
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/i386/setcontext.S;h=a604fca0aceaa17f19f536bac46a1fc305f30860;hb=HEAD#l57>
> movw %cx, %fs
>
> But there is no such save/restore about register fs in
> x86_64/swapcontext.S
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/x86_64/swapcontext.S;h=1110c479fad27829f1b1e480cf7520cf77d795f7;hb=HEAD#l46>
> .
> Is this a bug or it's no need to save/restore the register fs at all ?
> (The
> save/restore of register fs in i386/swapcontext.S
> <https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/i386/setcontext.S;h=a604fca0aceaa17f19f536bac46a1fc305f30860;hb=HEAD#l54> could
> also be elimianted
> if the later is correct)
>
> Thanks a lot :)
>
> Remus
>
^ permalink raw reply [flat|nested] 2+ messages in thread