user57

have you tryed with XP ? it is said to not have this problem, but im not certain either

that killed it : "They are handled by SeaBIOS." then it rely´s on a certain BIOS that has this, this is bad for retro stuff like XP

that they doing it now i find interesting, maybe that would be something what i wanted to say but is that all grafic cards up to today then wont be able to en/decode h.266, because its a hardware print https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new then you might have buyed even the latest grafic card from today, but it simply cant encode h.266 either the other way would be to have opcodes that can be changed on a bios/firmware but that dont differ so much from just having 1 core more and giving it the de/encode codes doing it with pure hardware unit is a lot faster, it dont need to parse out the firmware or the opcodes either it is directly progressed at the hardware unit itself maybe the market lacks of an alternativ, why not a pci-e card the hardware unit dont neccesary have to be a grafic card ? and maybe having both (print and cpu based) but i think avx should be fast enough if they can write it for those X(Z)MM registers for hevc there is a check what opcodes can be used, if its mmx it use mmx, if it is sse it use sse and if its avx it use avx

at some point it will reach a cpu limit question however to say is that this is not the case yet chrome or lets say the browser language, is a scriptish language as everybody can see the browser is getting bigger and bigger and has pharse more and more code the high language question therefore also apears ... it would be at least 20 times of the normal work to write the routines in pure assembly the cpu´s are already doing this: https://www.youtube.com/watch?v=RQWpF2Gb-gU the cpu is skipping commands/opcodes in exactly that way also what is not mentioned in that video, that the cpu unit can see a future command and is actually not near at all it can see what memory it accesses and therefore read it from the cache and make a paraell calculation (it dont has to be directly the next command) (also to mention mmx-avx doing that even better) the cpu that is mentioned is rather going back to a 486 cpu also the most compilers are also doing something like that, modern compilers shorten the code if the compiler can reconize that then there is also the MMX-AVX+ question, these are not made of normal opcodes it are opcodes for speed, those are a lot faster sometimes 150 times as fast i say this because we have often the SSE question (like we have to have to non SSE version) at some point as sayed it will reach the cpu question and SSE is just like a hardware acceleration or a grafic card sse(aka mmx-avx+) can speed up the process a lot so we put it the right way, having a grafic card can speed it up however you might not have the right driver or your card is not supported by the OS, or dont have a driver for that OS 2 other ways would be first not a high language (what is not possible in this case chrome has like 300k pages, or at least the part for the grafics would have the be entire rewritten) faster opcodes (aka mmx-sse-avx ect.) also i want to mention that smartphones special olders ones have google chrome and dont support some functions either a simple way would be to buy a supported grafic card, a faster cpu that also supports xp if you dont want trouble wih the chipset or maybe hardware components then rather i would buy hardware that is supported for xp --- in other direction, from what i remember xp had some hidden functions that are written out in for example win10 - xp has them but dont have the function to read/write call them so very well solutions had, just to make 1 of many example a direct access to the eprocess structure that works very well as it then it the same solution as win10 has ... but for like 7 different operating systems like (vista, xp, 7, 8, 8.1, + 64 bit) thats a lot you need like 14 different reactions ... or you might end up in solution that is not very well , like bugged, less functional , has flaws --- amd even tho they sell hardware, they do not offer us a open source driver ... rather it would give amd some money because they dont make money with that driver also it would definatly bring amd money because then xp users buy their new hardware ... is it a kartell ? because doing it that way give a other company money (like buying win10 instead or new hardward to have win10) it would be illegal to form it that way ... - but thats a other long discussion that dont belong here --- here we already talked about, a hardware is a print but the mentioned VP9 codec is supported by some grafic cards but it dont support for example the h.266 codec (and probaly never will, because a hardware is like a print - once printed there is no upgrade) if it really has the other way (just a cpu and a bios for the opcodes (that can be changed) - its just a next core ... (that would raise questions, also that is a lot slower) as mentioned im very busy - i cant join to make a VP9 codec while im like 100 % busy with a different project - there is no room to jump in just in like 5 other projects - actually i dont have even one i could join thats why i must say it again : im really the only person who could make a h.265 en/decoder ? im not the freshest guy either maybe someone should do it instead also to mention is that if you use a older grafic card and even tho you have win10, the right driver then the grafic card cant do that either to mention dx (direct x video) again , directx, opengl, can be either used a "frame/1picture" engine or as "24frames/aka video" engine actually after directx8 there was not real improvements to clarify this we have to put it the right way: directx and opengl animate a 3d model a 3d modell is just a texture in 3d it like 99 % depents not on the engine - rather on the texture of the 3d model gdi might not have a 3d engine, but it can repeat frames so now we have to talk about that cuda (aka the video/frame/ en/decoder) - or just the grafic card the cuda engine or that grafic card just encodes a frame (not a 3d engine) thats why it works with opengl dx and a normal frame buffer (like gdi) cuda/the grafic card gives the encodes frame to a buffer it can be opengl or directx - if we ask it right certainly it could do that for a normal frame buffer for video there is only 1 differens it is just not 1 frame it useally are 24 frames (the 60/50 frames ect. is a other discussion that dont belong here now) maybe we say that you can have 1 frame shot with 1/60 either or even 1/2000 - that is possible - you dont need 50/60 frames to do so i hope that clarify the situations for all the others a bit

i dont want to disappoint anybody but thats why i always said the focus must be on the main operating system of the browser then it was already raining "windows 2000 support" "win10 win8 and win7 must work too" while making a new sp4 iso and 4 other projects meanwhile like firefox too and fixing issues xp for example dont have, like that sandbox discussion it rather gave my old logic saying these things like that to be right i hope he can fix all of that, but its much ... there are alternativ´s for win7 like the chinese version or redfox and maybe some others

maybe it is time that someone look the published codes from microsoft for example the WRK so we could tell why i had left open my fault saying 4 times 512 = 4 k, but that isnt right - it is 8 times however only cixert fixed that one up if FAT32ex on xp can acseed that 2 TB limit and set a sector size, it might be possible if not its not hard to write a loop that actually parse 512 bytes 8 times on the other hand it would raise question to that classical saying "32 bit are limited to 4 gb" according to this logic 32 bit cant address a HDD bigger then 4 gb the overlappend structure just use two dwords (aka 32 bit * 2 = 64 bits) (it has to be mentioned because in windows thats the structure windows use for file offsets) if i continue to talk like this i can only make speculations but lets say it would be able to pass the size of a dword (32 bits) it would not mean that it can pass the 512 sector as we know it can make 512 sectors with 4 gb (aka 2 tb data) but such things you can read out of the microsoft code/either disassembled/debugged or the published code - it is certainly some work - if someone actually know what the problem is it would be faster then just gambling around i actually wrote data to the disc on i/o level but that memory is far to old to get it back, it actually lack like the most part of it, i just remember a few I/O ports like 1f4 and the writes they are either dword (32 bit) word (16 bit) byte (8 bit) this is not a problem as you just give it a loop to write if you have lets say 1024 bits to write you use for the 32 bit writes (1024/32 = downrounded 33 times) 33 * 32 = 1000 then you still have to write 24 bytes , you can either do this with 8 bit writes or one 16 bit write and one 8 bit write 24 / 16 = 1 after that you have the last byte 8 / 8 = 1 = 1024 bits have been written the hardware actually transfer that code into a "next code" this next code dont really care if it was 33 + 1 + 1 writes, it rather finds the data that wants to be progressed (cache should be a a word here) to get this information out you need time it took 2 weeks just to compile chrome up another 2 weeks for getting the things around that that makes at least 4 weeks to dig into this - what time i dont have at the moment sorry maybe thats the right spot ? it definatly has low and high parts https://wiki.osdev.org/ATA_PIO_Mode#Registers quote: ";ATA PI0 33bit singletasking disk read function (up to 64K sectors, using 48bit mode)" quote2: "Note on the "magic bits" sent to port 0x1f6: Bit 6 (value = 0x40) is the LBA bit. This must be set for either LBA28 or LBA48 transfers." quote3: " An example: outb (0x1F2, bytecount/512 = sectorcount) outb (0x1F3, Sector Number -- the S in CHS) outb (0x1F4, Cylinder Low Byte) outb (0x1F5, Cylinder High Byte) " it is written in assembly to me it seems to have 3 words (word = 16 bit) that address a 48 bit offset (aka LBA48 / 16+16+16=48 (it seems low, mid and high/LBAlo, LBAmid, and LBAhi) the logic says it begins with a port that then counts up, it also says this port useally is 0x1F0 (if not its just that "beginning port" + X) (+3 / 1F3) LBAlo - 8-bit / 16-bit (LBA 48 are 16 bit) 8 bit is only for LBA 28 (+4 / 1F4) LBAmid - 8-bit / 16-bit (LBA 48 are 16 bit) 8 bit is only for LBA 28 (+5 / 1F5) LBAhi - 8-bit / 16-bit (LBA 48 are 16 bit) 8 bit is only for LBA 28 that again makes 48 bits, it is not a wire/address or 64 bit question you tell the device that 48 bits in 3 steps ... where you want to write - therefore it dont need a 64 bit address question - together they are 48 bit´s or just LBA48 after that you just write at that spot ... there is no offset - the "offset" has been set before --- if it is like that it isnt hard either you probaly have to set the right settings and address the 48 bits, its different from the paging mechanism (for ram) what actually has 4k pages, 4 megabyte pages, maybe segments, 64 bit PTE/PDE entrys (that can be done on a 32 bit OS, one example is that CMPXCHG8B command in 32 bit mode it can set 64 bit at once (atomic) ) another atmoic way to store 64 bit in 32 bit mode would would be to use the FPU unit the FPU unit can store 64 bits in an offset - to do so you could just put the two 32 bit values to an offset and storing those on the FPU , from the FPU you then store that value to the requied offset (aka where the PTE´s and PDE´s are at), in short talk you can use the FPU as integer if you do it rightm or even "just as memory storage for more then 32 bit" to make it via fpu unit FST / FSTP would be an example the opcodes (these are the ones who have a memory location, could be used) DD /2 (FST), or DD / 3 (FSTP) can write 64 bits to an offset: https://tizee.github.io/x86_ref_book_web/instruction/fst_fstp.html cmpxchg8b: https://www.felixcloutier.com/x86/cmpxchg8b:cmpxchg8b but jumping around from one project to a other just kills not only 1 project it kills both of projects, you guys are on the point

there was DOS then there was win3.11 then there was win95 then there was win98 then there was win98 SE then there was a windows me hard to say but 98 SE was the best if i would rate them then a change was happening a change to a pure 32 bit OS windows nt 3.1 windows nt 4.0 windows nt 5.0 windows 2000 (aka 5.0) windows XP aka - 5.1 windows server 2003 - 5.2 then there was already the 64 bit versions, however they did not provide advantages over 32 bit for that time there already was 32 bit extensions that can hold more then 4 GB ram already ... https://winworldpc.com/product/windows-nt-3x/31 so it wasnt that hard to see the succesor to see from these it was windows xp 32 bit then there was a change however it contained many problems at the beginning (as many first operating system had) vista aka nt 6.0 was born but there was nothing it really provided over XP, so xp somehow survived https://youtu.be/29qnXTw0qr0?t=199 it was a XP vs Vista battle, what xp actually won https://youtu.be/29qnXTw0qr0?t=231 to fix that problem ms came up with a bug fixed vista or overworked nt6.0 now the name changes nt 6.1 - windows 7 windows 7 was a successor what is still very capble even today so now we might think that we need a new version the next version was not a successor but it was the so called windows 8.0 (aka nt6.2) what later on got a problem fixed version aka 8.1 (here i still say it would be NT 6.2 not 6.3 because when a few upgrades kicked in they didnt call it windows 5.3 either) 8.1 wasnt that bad, however it lacked to bring new things that microsoft used to add the spyware called win10 that was because a next operating system had to yet to come - if they choosen 8.1 instead of 10 they could not been a chance to add the spyware components ... so thats the real story then lets remember what microsoft said: there will no be next windows there will just be win10 why would there be a windows 11 ? its just a win10 with a few upgrades - maybe a new GUI

since i dont know this guy i would say you are doing this, you really should

"vista" , supermium was for XP i have to point out again its 15 times as much to go backwards to XP as to go back to from win7 to vista the most liked just the idea what supermium also works on the other operating systems

this is to much getting compared with a nt6.x engine i once told dibya that i dont have interests in creating a vista version, still he pushed it up so dibya came up to make a redfox to "vista" one but it was over 150 internal functions to be changed for xp, and like 10 changes from win7 to vista from what i remember dibya then made some code and the vista version was already working vista has a problem vs 7, 7 is just a upgraded vista so 7 is in advantage - sure you can make it backwards to vista the method dibya used was to change the code itself but, not the redirections in a project with 150-300 functions thats a lot a better method would be to overrule some redirection or different linking in the c-runtime or a different c-runtime, in llvm and maybe others what is easier then in vs2019 vs2019 has a hidden c-runtime.obj file - you can make some changes but hmm i dont know ... it almost sounds like a chinese student made a better version but no - you cant compare nt6.x with xp , in our comparison a compare would be 15 times as much to have a "vista vs xp" version, what the chinese guy dont have he might have a 7 version - that certainly can be changed to vista somehow, there are already codes that can can that the sandbox is also a such example it dont have a real use, the sandbox is rather a mitigation question and mitigations are OS based - also they not function based - the code is going fine without it (and no bugs) reality is that the sandbox in win10 have a lot more then vista and 7 actually have - so having some flags popping up that you have a sandbox wont do much either when the chinese want to challenge supermium he has to challenge it on nt5.1 (aka xp) not 7 vs xp 1 reason more i dont want to make a vista/7 version is exactly this discussion - once done they say they got a better solution - what is not true a other reason you dont make vista/7 stuff is that you lose focus for your normal common OS - dont get me wrong i have nothing against people who like vista or 7 but they have it simplier then we have if not the chinese guy has to come up with a XP version and proof us wrong i would not be unhappen to proofen wrong in this case what i also have to say when i added that redfox code (that also works for vista) i saw redfox works with vista - so the code gone there im glad that he has the code ... but then you get something like this "the chinese guy made a better vista version" - sounds weird to me

this might be a good moment to mention the "engine problem" again first the one-core-api is giving a nice support for some win6+ apis since we talk about phyton stopping the xp support we can point out the engine question again a engine often use functions of a certain OS, is written for that certain OS elder programming languages useally never had a such point neither if it was c, c++, assembly, basic, delphi because that are a programming language ... that dont need a certain "windows, linux, mac" function today that is changing the new c++ styles often get tranlated into a different code (which then use a OS function) -> and then you have it your nt 6+ is involved a such example would be the c-runtime - even tho you written a normal c++ code the c++ code still now involves that c-runtime and that c-runtime use nt 6+ functions for c++ mutex would be a such example https://en.cppreference.com/w/cpp/thread/mutex however there is not only a nt 6+ interpration for this (aka srw locks) you also could use a thread based atom style to solve this problem there are some more, keyed_events, mutex windows functions as as createmutexa, creatthread styles, or criticalsection styles when i saw a new project i saw the following problem it uses DX11 it uses phyton it uses cmake it needs VS2019 (aka new c++ styles + the c-runtime) the project itself already where written with windows 10 functions often you dont have insight into the things these use (i often call them engines) lets say phyton break - then you cant compile it up because phyton decided to longer like xp if cmake use nt6 then you also cant compile up if visual studio wants a newer version you cant compile up if directx wants a newer version of directx you also cant compile up that makes it a lot to go through before you even can do anything the new trends doing exactly so even ffmpeg is going into that direction (for example ffmpegs cuda engine) in this discussion it seems to be bond to phyton a possible solution would be a code translation from phyton to c++ (normal styles ones) a good thing with c is that you can always have a c interpration in comparison to a other language without having a hard time with a lot of math like in assembly assembly for example can represent any language - its because all languages create a assembly code in the end c++ made a good compromise (but new c++ styles going into a direction to be something like a java script) im trying to point out that all of these try not to be just a programming language, they going into a different direction to like a script and engines so if phyton is not possible anymore, i would suggest a translation to c++

"Windows 10 was offered for free from its release on July 29, 2015, until July 29, 2016, for users of eligible previous versions of Windows. However, Microsoft continued to allow free upgrades for several years after that, officially ending the offer on September 20, 2023. Q: Is the upgrade really free? Do I need to purchase Windows 10 after 1 year? A: With Windows 10, we will offer a free upgrade to Windows 10 for qualified Windows 7 and Windows 8.1 devices that upgrade in the first year. "

looks simple to me, there was a reason win10 was free, and win11 also the article rather writes about security reasons the security how they define it is from a person the plan is probaly to get rid of the person who owns the computer so he can only do what they want the TPM chip was also a such direction, you getting "trusted" for microsoft being secure from you we know they had to be something wrong with it when it was offered for free after the establishment they might want money, but before only spreading is important

people like what people like, so i cant blame anyone for using vista for me the vista question was kinda different, i saw the sellings and xp somehow passed vista so i stayed on xp then windows 7 took over the place but windows 7 is just a problem fixed vista (thats why the versions nt 6.0 for vista, nt 6.1 for 7) 8 is also a vista but its real name should be nt 6.2 thats what win8 actually is the versions in between are rather like service packs to me but then microsoft made a big change with 10, as we all know it even was there for free so there had to be a big downside they dont want to tell us - today we know what it having the updates vista was ok, but vista had a bad start today its hard to say, but win7 8 and 10 have a lot of new "nt 6.x area" functions that vista dont have so vista might have a replacement to 7, but xp dont have a replacement

i readed me a bit into this by far not done yet first it´s impressiv much and a very big collection, good job the problem that you need sp1 to sp2 to sp3 or such things is solveable what these installer scripts use is that /.msu/.msi./.inf files for a installer script (some convert all files together like reg to inf ect.) the problem with that installer script however is that is is kinda slow, and it cant install in 1 step there is a thing before that installer script its the "install maker" ? such as InstallShield or Visual Studio Installer (those are very common to create a installer script) those are rather "build-together-ish" and make those scripts there 3 ways to collect that informations 1: from the raw/source file (for example InstallShield, Visual Studio Installer) 2: extracting the information from the .msi/msu or .inf files 3: debugging the installer and the next (non script) functions after the script installer (such as createfile, regopenkey ect.) one way for the files to do it in 1 step would be to use the MoveFileEx function with PendingFileRenameOperations the registry can be changed with the common registry functions then everything dont depent to go through the installer script i remember i once did that with a installer script in like 1998 and it speeded up the installer from like 10 min to 10 seconds a own installer then just dont need .net 1.0 .net 4.0, it simple can install what it wants those installer scripts have a next deeper set of functions those are the registry and file functions - kinda reminds me the discussions about engines and engines for a engine

1 way to solve this problem would be the firmware doing that on the harddrive you often do not have greater files then 4 GB each chuck but xp can handle file sizes greater then 4 GB / per file the overlappend structure is solving that 32 bit problem it use 32 bit dwords combined as "low" and "high" part https://learn.microsoft.com/en-us/windows/win32/api/minwinbase/ns-minwinbase-overlapped a other thing is the rounding number if you fill 4 kb checks with a file of like 2 mb . something you can fill these 4 kb chucks with 512 kb checks (4 times) only the the very end of the file it can be that you only need 1-3 times the 512 chuck then the last chuck of the 4 kb chuck is just not used that actually dont fall into weight because its only a small piece what would be thinkable that it can exactly handle 2 TB instead of 4 GB is that a DWORD (32 bit) is involved, a idea would be the counter but figuring that out somebody can do ... there would be ways to get this out of a code microsoft published some of their codes - as ms did with their dosbox or their wrk a problem with a big project that dont have a project file is rather that it are many of files if there is no project file where you can scroll around and find the right code in this case it looks to me that the files have to be searched for their names then the structures and functions can be isolated depending how much they are splittered up but the ntoskrnl or windows is a big piece, to pull together all the functions, structs, defs and others parts of code is some work ... google has 300 000 pages for example but then you could actually not only look into the functions you also could compile the outdated ntoskrnl for example from server 2003 the code itself might actually be simple but my editor could actually not just open 300 k files, rather you need a program like visual studio, but vs wants a project file i dont think it just can open all the files at once (all tabs opened) then it probaly has to make a search function what search all the files via filecontrol (and guess what vs has a such thing)

its one thing to print a RGB buffer to a screen, that is what directx video is doing what i wanted to point out is first that CUDA is doing the de/encoding (so basicly everything 99.99 % happens in CUDA, not in directx) then second is some parameters are different in directx versions, the example code already needed d3d9ex not d3d9 without the ex so it also could be that some parameters are just not supported in dx9, hard to say exactly thats why you can give that "RGB" buffer to opengl instead to directx that strongly speaks for windows xp´s directx9, however cuda if you want to do that with a grafic card is doing the important thing the RGB buffer we also could draw with a GDI engine - and thats why i said directx might not be the right question another point is that cuda can be done software wise with mmx-avx (so called XMM registers), those are very fast and by far enough for the purpose either en or decoding rather cuda is a hardware interpretation of a software .... thats it if it has to be a card, why not someone invent a PCI-E card doing the en/decoding ? we do not neccesary need a grafic card to do so - maybe a new hardware invention ? that with the high level functions is different to ask if you have paint - paint isnt making the drawing - it use GDI and GDI use NTGDI and NTGDI use internal functions what in the end go into the grafic card driver using a next engine would be a script that controls paint - that you see quite often that "pre CUDA" is a DLL , it probaly has a pre code for the real control to the cuda part in the grafic card - while cuda is being coded in a CUDA SDK - what is also some "scriptish code" to control the CUDA engine (and is like totally bond to dx10/11 and windows10/11) , the phyton version also only control cuda so hard to say what to call this ? a half-script language to control cuda ? a engine code to control cuda ? a simplified programming to control cuda ? i dont know what i would name it, maybe the others have some corrections for me and im happy to hear them why we have to be so indirectly ? why cant we just use cuda rather directly ?

these timers are almost the same just with different names

well maybe there should be a inofficial SP4 (including all of these upgrades also the posready ones) for all languages in a older past there was a ugly solution someone collected all these upgrades. and then run them via createprocess() function , that took 4 hours or something a better way would be to set the files and registry entrys manually via registry and file functions that 4 hours method just runs each update.exe (for every kb upgrade) over and over - until it has the 190 upgrades or something

if the life-cycles are meant, it would be better when the firmware to choose these that raise a certain question tho, if it splits the sectors with into less and more used sectors , then not just a few die on and on, rather the disc would then later die at once

greetings, i would make a better description like screenshots (smartphone pictures are also good). error messages, when the problem apears, do it work in part, KB or upgrade number, your windows xp version

at first it was a good fork for many codecs!, but at some time the engine question kicked in rather of being independent the "newer codecs" are done via engine and ffmpeg is no longer independent and those engines/apis/interfaces (whatever we call them) sit in windows 10, so they keep telling you to install win10 somewhat, somehow neither if its CUDA (what later on calls the high level functions in the grafic card driver(and is called something like "NVIDIA display driver: 551.76"), dx12, LAV engine that also goes for the other like vs2022(what used something called the c-runtime (that uses win10 functions, or the SDT from vs2020), python 3.10 + in future for the compilers today they trigger to use a engine code - that goes into direction of scripts rather then having a c++ code that is really doing the job manual (such as memory managment, file control) this script then is written with win10 functions ... those one-core-api.dll´s are also a such thing that is also the case for example for the nvidia SDK, this one is also to see as some kind of engine -> what then use cuda and d3d12 the HEIC image encoder also was a such thing, it was deeply bound into win10 engines and functions - but it was unbound to me this seems to be a method to get rid of the older operating systems instead of offering the real code - they offer script to do so (and those scripts use win10) - simple said it is not like there would be other ways ... there certainly would be for example if we would have the high level code of CUDA, or the programming control is not going with scripts, rather they are manually controled written - they it would be doable CUDA(and following up the driver) is just a engine that represents hardware video card encoding, such as VP9 or HEVC since CUDA is rather made for win10 its hard to use a code that use CUDA on XP - but the codecs VP9, HEVC are open source - CUDA is not a necessary requirement (and that even tho all the programs out there use this engine ffmpeg being one of them) if put the right way not having a video card is ok if you use a software based en/decoder (and mmx-avx makes such codes by far fast enough - its not like a software mode in a video game where everything would run just superslow) chrome have a such software based decoder for VP9, someone tryed it with 1 core and 2 cores - 2 cores being enough without a grafic card driver being installed (while when he had a grafic card driver installed even 1 core was enough) https://www.youtube.com/watch?v=wsSMmdwh89Y

one way would be that instead the counter 4096 sector size is not repeated if it is a "virtual 4k sector, that has 512 sectors" = it would have to repeat that 512 * 4 if the counter then just access the next 4 k sector instead of sector 513-4096 then it would mis-align them same goes for the oposite "using 4 k step writes on 512 sectors" = it would try to write 4 k on a 512 sector - that would interfere with the next sectors, and leave out 3 512 sectors a buffer overrun oder underrun is also possible seems a matter to understand the entire OS file routines but something else is to see, xp actually handles over 4 GB size here * 512 = 2 TB so hard to say but it might use a 32 bit offsetting while that overlappend structure (what is used on like the entire file functions on xp) has a 64 bit offsetting it tends to say "4 gb * 512 sector size" maybe the issue is small, sure it could be more of problems hard to say to write a such routine (either 4 times 512 or 4k) would be very simple but

the problem with directx is that we dont have dx10/11 or 12 for xp directx10/11/12 is a closed source so we cant see into that engine, so it might end up in a stuckpoint (if you put it the right way "it ask for dx10/11/12") but maybe directx or opengl is even the wrong answer from what i seen in the code its not opengl or directx doing the job its the CUDA engine that gives the RGB buffer to the opengl or directx engine so we might put it the right way you dont need a grafic card to encode / or decode an image or video (the HEIC en/decoder from msfn is proof of that) CUDA is a engine for a grafic card, it makes control/input/output as engine directx/open - uncertain for now - but from what i think its just a engine that gets the data from CUDA if we might think that "software mode" is to slow, no not this time because mmx-avx are from the same nature they 10-500 times faster then normal opcodes thats more then enough to make it fast enough this time so what does CUDA do ? cuda controls a grafic card unit that can make en/decoding of common video formats that sounds good so far - because if you want have something done you want a hardware unit to do the job but nowdays we have many cores - what also can be seen as a hardware unit (that is programmable) while a pure hardware unit is like a print - once printed it cant be changed but the formats are being upgraded - for example CUDA cant en/decode the h.266 codec the grafic card doing the encoding is kinda new, its not like a video game what hardware accelation is from like 1995 or something the video card doing that for a video en/decoder job is not to long ago, and many modes are not supported either the CUDA engine itself is not well supported for windows xp so going that direction might just lead to windows 10 then there you have the right grafic card driver, the right cuda engine, and dx12 so you dont have to deal with any other questions then again, i pointed this out in a other post for a nativ solution we would have insight to the nvidia driver what actually controls the grafic card then we would have a OS-independent solution (high level functions) i dont see this at the moment like at all. when the problems already where spinning around in a LAV engine that makes 3 engines - and no grafic card control was touched at all so at the moment its spinning around in these 3 engines - not where the needs actually would rely a strongpoint could actually be that directx9 dont look that much different to dx10/11/12 so it might could be given a RGB buffer from the CUDA engine - but the CUDA engine for XP is not being upgraded by nvidia https://youtu.be/W3zfb8lLDH0 https://youtu.be/p8387-gu37s https://www.youtube.com/watch?v=mZ-0XBqRxuc the problem i see is that already a next engine came but here is the problem : DX11->CUDA->GRAFIC-CARD-High-FLevel-Functions OPENGL->CUDA->GRAFIC-CARD-High-Level-Functions this time: MPV engine->CUDA->GRAFIC-CARD-High-Level-Functions so everybody can see the problem the high level functions are not touched (neither is CUDA) (MPV calls up cuda ? no ...) we had to sort out a similiar question, at first it was like "you need a grafic card doing this job" it took a while until everybody had an agreement that this is not the case - also not for speed (software mode) basicly this only leaves the software option, in avx ect. its by far enough of speed this is not done by a engine like directx, cuda or video card high levels functions its like the RAW control of the RGB buffer for a video player doing the decoding maybe encoding if wanted

that NdisSetCoalescableTimer seems a timer function for me, it might work without that specific flag be set it has the same KTIMER structure as in KeSetTimerEx and xp has this one https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/wdm/nf-wdm-kesettimer https://www.geoffchappell.com/studies/windows/km/ntoskrnl/inc/ntos/ntosdef_x/dispatcher_header/timercontrolflags.htm if that DPC parameter is in use xp also has the KeInitializeDpc, the same goes for KeInitializeTimerEx https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/wdm/nf-wdm-kesetcoalescabletimer