A New Testbed for MASM32 Forum

[dedndave]

first, i use a patched version of 6.15
i got the original and the patch from Iczelion's site
the patch adds some text to the command line that i did not like
so, i re-worked the patch a bit
the 6.15 version by MS reports version 6.15.8803
my patched version reports version 6.15.8804
the patched version is available here
ML Version 6.15.8804 (patched)

as for the 10 - no - i used the one you included with your attachment as the Test10.exe

[Antariy]

as for the 10 - no - i used the one you included with your attachment as the Test10.exe

The size of ML10+LINK10 executable is bigger, because LINK put a relocations into EXE, even if it is based on 400000h. Needed to specify /FIXED switch. And LINK.EXE v.10  without patching does not allow to set subsystem version to value less than 5.0.

[dedndave]

actually, the ML10 version is 1 Kb smaller   :bg

[Antariy]

actually, the ML10 version is 1 Kb smaller   :bg

Well, that is not have meaning. Some pages ago I sayed to Frank about this, and he uses POLINK, it seems.

[dedndave]

well - that makes it hard to find the differences - lol

you might be able to use 6.15 and 10 to assemble
use the MS 10 linker
then compare the 2 EXE's

it might make it easy to find out what the problems are
otherwise, i'd have to read every line of code to see what is up   :(

[Antariy]

well - that makes it hard to find the differences - lol

you might be able to use 6.15 and 10 to assemble
use the MS 10 linker
then compare the 2 EXE's

it might make it easy to find out what the problems are

What you mean? If the format - then is possible to compile with ML10+LINK10 not bigger than with POLINK.
If you meant code generation - this is right - ML6.15 and ML10 differently generate code for used "push word ptr [xxx]".
But problem is solved already :lol

[dedndave]

if you are using SSE3 instructions, that is the problem
masm version 6.15 only supports up to SSE2

[Antariy]

otherwise, i'd have to read every line of code to see what is up   :(

No, Dave, no! Not spent your time for this - no need in bothering yourself. Problem was *solved*.
Latest posted by Frank archive at previous page, is the right version.
Current code compilable right with ML6.15

[dedndave]

it runs very slow, now   :(

but, it does assemble and run correctly with 6.15

[Antariy]

it runs very slow, now   :(

What, what is? Which archive do you use? Which tested algo run slower?

[dedndave]

you updated the link, right ?

they all run slow - meaning it takes a long time to run the tests

Code Select Expand

┌────────────────────────────────────────────────────────────────────────────────────────┐
│OS  : Microsoft Windows XP Professional Service Pack 2 (build 2600)                     │
│CPU : Intel(R) Pentium(R) 4 CPU 3.00GHz with MMX, SSE1, SSE2, SSE3                      │
│                                                                                        │
│Test: Conversion of a screen buffer from DOS to Windows CHAR_INFO structure             │
├─────────────────────────────────┬─────────┬──────────┬──────────┬──────────┬──────────┬┤
│        Algorithm notes          │Proc Size│ Test # 1 │ Test # 2 │ Test # 3 │ Test # 4 ││
├─────────────────────────────────┼─────────┼──────────┼──────────┼──────────┼──────────┼┤
│01 Alex / MMX - PUNPCKLBW        │   62    │    9.724 │    9.666 │    9.660 │    9.690 ││
├─────────────────────────────────┼─────────┼──────────┼──────────┼──────────┼──────────┼┤
│02 Frank / 486 - MOV-BSWAP       │   43    │   13.943 │   13.953 │   13.971 │   14.028 ││
├─────────────────────────────────┼─────────┼──────────┼──────────┼──────────┼──────────┼┤
│03 Frank / XMM PUNPCKLBW MOVDQA  │   44    │    4.663 │    5.170 │    5.418 │    4.760 ││

did you change the loop count ?

[Antariy]

you updated the link, right ?

they all run slow - meaning it takes a long time to run the tests

did you change the loop count ?

Ah... This - well, if you have readed previous page - you saw that I suggest to Frank decrease the counter of loops.
I have not posted archives, Dave, all archives posted by Frank. I only posted code pieces which was needed in modifying to be compilable right with ML6.15.

Yes, I have noticed that Frank increase a time of test, too. But you can ask to him about that tomorrow only.

[dedndave]

we usually try to adjust the loop counts so that each run takes 0.5 to 1.0 seconds

[frktons]

The LOOP COUNT was 1.000.000 in the latest version.
I've set it to 100.000 that is fast enough and accurate enough as well.  :P

If you want to rerun the test, just press the [R/r] key.

Frank

[dedndave]

it could make a measurement and calculate a loop count variable
that way, the tests always take ~0.5 seconds