I'm making progress towards the 1.70 milestone - today's 1.69.15 release has added support for several new instruction sets, including AVX and XOP, and therefore the only thing left to implement before 1.70 is long-promised Mach-O support. I hope to finish the 1.69.xx development line around the time of this year's fasmcon.

Note: some of the files in source and examples directories were rearranged, so if you unpack the 1.69.15 package into the same folder as some previous version, you are going to get some redundant files.

For the purpose of handling new 256-bit AVX instructions, the 256-bit size operator was added in this version. The 128-bit and 256-bit size operators can either be specified with fasm-style "dqword" and "qqword", or "xword" and "yword", which reflect the sizes of XMM/YMM registers.

The new instruction sets that got implemented are mainly based on Intels AVX reference, revision 8 (from August 2010) and AMD's XOP/FMA4 reference, revision 3.04 (from November 2009). I hope these specifications are mature enough that I will not have to change too much in the future releases. There may still be some my own bugs there, however, so if you got some spare time, please do a bit of testing/verifying.

I ain't got no bulldozer to clean my sandy bridge


P.S. Who on earth does give them names?!

Whom to chose, Tomasz?
I need no pluggings; add colors; easy float values definition; signed/unsigned division; multiplication; shifting; solve db -1 using different notations; there is no valid encoding for long floats except some cases; why do you stumble with that; are you chasing for a bonus with instruction inclined dev only; are you busy; eh, fasmw renamed to -fasmw.
Is there fasm download count?
Of cource too big list.
“No matter what you do, somebody will call you an idiot.” — Raymond Chen.
That's sayd about me and concerns you too.

It may be just me but I get...


...when unpacking the linux version.

Yeei - the RSS feed works

Sorry about that, the FTP connection must have had broken. Please let me know if it's OK now.

edemko: if you read the older threads about 1.67 and 1.69 development lines you are going to see, that I'm slowly following the established plan, and Mach-O is the only feature left that was promised for 1.70 but is not ready yet.
If there are any other problems with fasm's core that I'm not aware of, report them in the "Compiler Internals" section.

As for the colors and other stuff like that - first, it is not related to core development (that's why reports of fasmw bugs go into "IDE Development", not "Compiler Internals"). Second - the set of fasmw's features (except the built-in debugger, which was planned for 1.x releases, right now fasmw is still at 0.9x) is quite exact fulfillment of my vision, I'm happy with it, and I don't plan to add any new features, but rather keep implementing new IDE interfaces that will provide the same editor environment suited to my very private preferences (with roots in original DOS Navigator's editor) into other operating systems. At least fasmx is in the plans. As for the fasmw plugins project, I still have not decided whether to implement or discard it, as the one problem I have with it is that it does not fit well the vision of the same editor interface for various operating systems.

For the more feature-rich IDE suited to the expectations of community, and not my personal vision, there was a Fresh project, but is now dead (because all of its developers have mysteriously disappeared), and thus when I saw this new "wink", I hoped that some new project like that will emerge. If you have got enough time to participate, I encourage you to go on with such project, as since the death of Fresh fasm really need some new community project like that.

I know.
Ok, enough.
Coming fasmcon, take more photos, please.

When will FASMW understand UNICODE?

It is not going to in any reasonable future. I'm sorry, but you have to turn to some other IDE to edit UTF-8 sources.

Another release today - version 1.69.16 allows .fas file to be generated even in case of error. It then contains just the preprocessed source (at least partial) that can be extracted with PREPSRC tool.

Also, the updated documentation of .fas format is now downloadable from the official Documentation page.

You can make macros for this. :>

Tomasz wrote:

> flat assembler 1.69.15/1.69.16

Thanks

What's new:

- Increased EXE bloat by 12 KiB (from 1.69.13 to 1.69.16)
- Slightly updated and enhanced main manual (AVX not documented)
- Added missing keys and fixed typo in DOS IDE manual
- Split EXPRESSI.INC into EXPRPARS.INC and EXPRCALC.INC
- Added AVX.INC
- Added SYMBDUMP.INC (split from FORMATS.INC)
- TABLES.INC grew from 58 KiB to 100 KiB
- Tiny fixes in portable editor core
- Added DEC ESI instruction in the DOS IDE - should fix this one: http://board.flatassembler.net/topic.php?t=11061 ??? It seems fixed (but (C) still says "2007" ...)

So the DOS IDE might be BUG-free now (no bugs known but no thorough test yet), remaining trivial wishes are about the Search and S&R:

- [F3] fall back to [CTL]-[F] if no text
- Report searched text: Text "assume" not found instead of just Text not found
- S&R summary: Replaced 1'000'000 occurrences of "xchg eax,eax" by "NOP" (see http://board.flatassembler.net/topic.php?p=95886#95886)

PS: today I moved many threads from Main into more appropriate subforums ... just in case someone misses something

Thanks it works now.

You did not notice changes concerning Linux and Windows packages. Otherwise good job.

Also, fasmd was updated once more in the meantime.

@Tomasz:
vinsertf128 with memory operand contradicts Intel AVX documentation
(VEX.NDS.256.66.0F3A.W0 18 /r ib):


I made a compile-time fix:


I don't know how to fix this in code, but I found two interesting variables:
mov [operand_size],0
mov [mmx_size],16
Should operand_size be 1 (according to 128-bit = 0; 256-bit = 1)
or should mmx_size be 32 (where 16 bytes is 128-bit and 32 is 256).

Neither - it's a bit more complex thing.
And the "vinsertf128 ymm0,ymm0,ymm0,0" should not word, in fact, as it should be "vinsertf128 ymm0,ymm0,xmm0,0"
I will upload a fixed version in a few minutes (thank you for help).

Testing method:
1) take as many Intel AVX code samples as you can and wrap into a 64-bit PE executable
2) wrap all testing code into

3) assemble with FASM 1.69.17
4) Run iaca.exe -o out.txt test.exe
5) see if everything dissassembled correctly

1.69.17 has passed this "average AVX code"-test

UPDATE: http://www.anandtech.com/print/3871 AVX is already 'cast' in some CPUs Read and dream...