Use switch tables instead of long strings of cmp/je combos

[faissaloo]

Instead of doing stuff like:

os_system_config:
    cmp rdx, 0
    je os_system_config_timecounter
    cmp rdx, 1
    je os_system_config_argc
    cmp rdx, 2
    je os_system_config_argv
    cmp rdx, 3
    je os_system_config_networkcallback_get
    cmp rdx, 4
    je os_system_config_networkcallback_set
    cmp rdx, 5
    je os_system_config_clockcallback_get
    cmp rdx, 6
    je os_system_config_clockcallback_set
    cmp rdx, 20
    je os_system_config_video_base
    cmp rdx, 21
    je os_system_config_video_x
    cmp rdx, 22
    je os_system_config_video_y
    cmp rdx, 23
    je os_system_config_video_bpp
    cmp rdx, 30
    je os_system_config_mac
    ret

You could do:

os_system_functions dq   os_system_config_timecounter, os_system_config_argc, os_system_config_argv, os_system_config_networkcallback_get, os_system_config_networkcallback_set, os_system_config_clockcallback_get, os_system_config_clockcallback_get, os_system_config_clockcallback_set, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_video_x, os_system_config_video_y, os_system_config_video_bpp, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_mac

os_system_config:
    jmp qword [os_system_functions+rdx*4]
    os_system_config_no_function:
    ret

And I believe it would be much faster, I also believe this is what GCC does (or at least, something similar) for switch statements.

[effbiae]

i got curious on this question of multiple cmps vs lookup table.

i found that cmps win for rdx<7 and then lookup win for the rest. funny -
i remember back in school, a lecturer said it's faster to do linear search
on n<=7 - i wonder if it's still true.

here's the gas code i used on linux - you can call lookup() and cmps()

$ cat j.s
.intel_syntax noprefix
.globl cmps,lookup
.data
a:;b:;c:;d:;e:;f:;g:;h:;i:;j:;k:;l:;
ret
cmps:
cmp rdi,0
je a
cmp rdi,1
je b
cmp rdi,2
je c
cmp rdi,3
je d
cmp rdi,4
je e
cmp rdi,5
je f
cmp rdi,6
je g
cmp rdi,7
je h
cmp rdi,22
je i
cmp rdi,23
je j
cmp rdi,24
je k
cmp rdi,31
je l
ret
.data;.align 8
funs:
.quad a,b,c,d,e,f,g,h,z,z,z,z,z,z,z,z,z,z,z,z,z,z,i,j,k,z,z,z,z,z,z,l,z
lookup:
jmp qword [funs+rdi*8]
z:
ret

On 7 October 2016 at 20:44, Faissal Bensefia notifications@github.com
wrote:

Instead of doing stuff like:

os_system_config:
cmp rdx, 0
je os_system_config_timecounter
cmp rdx, 1
je os_system_config_argc
cmp rdx, 2
je os_system_config_argv
cmp rdx, 3
je os_system_config_networkcallback_get
cmp rdx, 4
je os_system_config_networkcallback_set
cmp rdx, 5
je os_system_config_clockcallback_get
cmp rdx, 6
je os_system_config_clockcallback_set
cmp rdx, 20
je os_system_config_video_base
cmp rdx, 21
je os_system_config_video_x
cmp rdx, 22
je os_system_config_video_y
cmp rdx, 23
je os_system_config_video_bpp
cmp rdx, 30
je os_system_config_mac
ret

You could do:

os_system_functions dq os_system_config_timecounter, os_system_config_argc, os_system_config_argv, os_system_config_networkcallback_get, os_system_config_networkcallback_set, os_system_config_clockcallback_get, os_system_config_clockcallback_get, os_system_config_clockcallback_set, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_video_x, os_system_config_video_y, os_system_config_video_bpp, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_no_function, os_system_config_mac

os_system_config:
jmp qword [os_system_functions+rdx*4]
os_system_config_no_function:
ret

And I believe it would be much faster, I also believe this is what GCC
does (or at least, something similar) for switch statements.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
ReturnInfinity/BareMetal-OS#117, or mute the
thread
https://github.com/notifications/unsubscribe-auth/AAn-H3ioEg_aDGf2VmRNt1Oy2zm_YLdBks5qxhSKgaJpZM4KQ2f6
.

[IanSeyler]

Awesome idea!! I'll give it a test on my end.