inmutel Mania

GitHub - herumi/xbyak: A JIT assembler for x86/x64 architectures supporting MMX, SSE (1-4), AVX (1-2, 512), FPU, APX, and AVX10.2 A JIT assembler for x86/x64 architectures supporting MMX, SSE (1-4), AVX (1-2, 512), FPU, APX, and AVX10.2 - herumi/xbyak

Diamond Rapids用AMX命令群(AMX-{MOVRS,AVX512,FP8,TF32,TRANSPOSE})に対応したXbyak 7.26 release。
github.com/herumi/xbyak

AVX10.2がrev 4.0でYMMレジスタの埋め込み丸めやsae/erなどの仕様が削除されたけどXbyakはまだ残ってた(xed 9.53も残ってる)けど、何かとトラブルになるのでその機能を削除してv7.27をリリースした。
github.com/herumi/xbyak...

XbyakにSolaris対応のちょっとしたpull reqが来たのだけど、Solarisってまだがんばってるんだ。知らなかった。大学のときに触ってた以来だなあ。

定数除算最適化再考3 コンパイラを越えろ

定数除算最適化再考3 コンパイラを越えろ
zenn.dev/herumi/artic...
x64/M4でアセンブリ言語レベルでの試行錯誤の結果を書きました。

Why limit it to Intel? :)

Let me add another bit to the rant. I am sick as shit of sites, youtube idiots, and the rest puking back obvious BS without the bare minimum of checking. 3 sites said it so it has to be true! You are being used by people with ulterior motives, don't be a tool. No hope for society....

Is AMD fabbing at Intel Foundry? A few days ago, a financial note went around saying AMD is going to fab chips at Intel Foundry.

Is #AMD really fabbing at #Intel? We got the definitive story. Really, not clickbait like the others, we got to the bottom of it!
www.semiaccurate.com/2025/10/03/i...

With the changes today, I am now FIRMLY back in the "Intel will die" camp. The company is avoiding the problem source and addressing the symptoms. Badly. In a way that will worsen the problem. I plan on writing this up as soon as I get free time, tomorrow is shot though. More soon.

Remember how people laughed when I said, in 2019, that I saw a clear path to #Intel failing? I wasn't joking. Then Pat Gelsinger came in and addressed the root of the problem turning things around.

Those things you cite had a different purpose, basically to distract Wall Street from viewing Intel as not a player in 'hot' markets. Nothing more nothing less. I can name a dozen others too.

Size wasn't the issue, culture was. Pat fixed that mostly, but will those changes stick?

Very interesting to me that it will only be a 6 X increase in DGEMM.

Fugaku is famously under provisioned for low precision flops, I must wonder if this is an over correction?

TT-QuietBox (Blackhole)

Look what just landed in the lab

Here is an early mention of 512b vectors in a 2009 #Intel #Nehalem optimization slide:

Well... AVX512 was known as AVX3 at some point... (and AVX512F was AVX3.1 for KNL guess) .. SKX was supposed to be AVX3.2 (F, CD, BW, DQ, and VL)

It's too bad that this AVX3.1 nomenclature disappeared. I think it is much more seamless, like SSE4.2, SSE5 (original name of AMD XOP) or Armv8.2. AMX2 and AMX3.1 would be also better.

#Intel xAPIC depreciation plan 1.0:
www.intel.com/content/www/...

#Intel refreshed the xAPIC depreciation plan with #NovaLake and #DiamondRapids:

There are a few working #PantherLake B0_2 among #Intel test machines: (CPUID C06C2, 12c/12t (4P+4E+4LPE probably), 3000 MHz, no HTT, no AVX512, Intel 18A)
intel-gfx-ci.01.org/tree/intel-x...
#CougarCove #Darkmont
For comparison, #LunarLake was 3100MHz (8c/8t 4P+4LPE) at similar stage

#Intel microcode refresh 20250812:
github.com/intel/Intel-...
Release Notes:
github.com/intel/Intel-...

#AMD refreshed the "AMD64 Architecture Programmer's Manual, Volumes 1" 24592 pdf to v3.23 with #AVX512
docs.amd.com/v/u/en-US/24...

ARM unveils GPU Neural Unit and Neural Super Sampling Tech Today ARM is announcing it’s Neural Unit and various GPU upscaling technologies that use it.

New story up on #ARM's Neural Super Sampling Tech. Not deep, waiting on hardware details.
www.semiaccurate.com/2025/08/12/a...

Kicking off the EUMaster4HPC FPGA workshop with Intel oneAPI on #MeluXina! Huge thanks to @luxprovide.bsky.social, and @uni.lu for making this happen. On today's agenda: Hands-on learning with @eurohpc-ju.bsky.social and Luxembourg’s national supercomputer!

#HPC #Supercomputing #oneAPI #FPGA

seems intel has since redacted this section...

💥Spack v1.0 is out!💥

This is a huge milestone. We reworked the core to add compiler dependencies, and we're introducing a stable package API.

🚀1.0 also adds concurrent builds, better includes, and much more -- read it all in the release notes!

github.com/spack/spack/...

OK, serious question that I have been thinking about for a while. What is a PC? Is it x86? Windows? Form factor? Screen size? Features? Discuss.
I am seriously interested in what you think. I have asked several industry heavyweights and gotten as many answers as people I asked. Your turn.

#Intel released the 58th edition of the ISA Extensions Reference with fixes and clarifications:
#PantherLake #DiamondRapids #ClearwaterForest
Download:
cdrdv2-public.intel.com/859029/31943...
CPUID.(EAX=23H,ECX=0H):EBX[2] = #RDPMCUserDisable

#Intel released the 88th edition of the Software Developer’s Manuals with #SLSM (Static LockStep Mode) integrity feature:
CPUID.(EAX=07H,ECX=01H):EDX[24]=SLSM
All-in-One:
cdrdv2-public.intel.com/858440/32546...
Changes:
cdrdv2-public.intel.com/858441/25204...

#Intel refreshed the "Flexible Return and Event Delivery" ( #FRED) 346446 pdf to 9.0:
cdrdv2-public.intel.com/819481/34644...