GCC 10 Introduces A Static Analyzer - Static Analysis On C Code With "-fanalyzer" Option

[pal666]

But its 8 cores are like the 16 cores in my system:

like by what metric, moron? by output of random command? or by cost? or by performance? (hint: nothing of the above has any relevance to number of cores)

Yeah, there's 16 of 'em up there, but come on now. You know as well as I do that's only 8 actual cores.

16 threads of execution. 8 actually running at the same time, 8 sleeping in cache misses. bulldozer runs all cores at the same time

Calling "shared module cores" actual cores is like trying to bullshit someone that a 2x4 Dually pickup is a 4wd because the 4 tires in the rear are driving it forward.
Technically speaking, that is 4wd, but it sure isn't 4wd in the meaning anyone reasonable would expect.

only clueless idiot can compare modules to hyperthreading. sane person would expect scaling from additional cores and bulldozer cores do scale well except on code which consists exclusively of floating point instructions. i.e. on benchmarks for idiots like pi number calculation. sane people need heavy fp for 3d or video acceleration. and guess what? they are doing it on videocards anyway. all real software scales well on bulldozer because most of instructions are not fp. bulldozers are just slower per core than contemporary intel cpu, but that is not a crime and amd never marketed "faster per core" bulldozers. but they are faster per core than contemporary arm cpus, where is a queue of imbeciles suing arm for "calling their single-core cpus multi-core" because "they reasonably expect them to be as fast as i9 with same number of cores"?

[pal666]

No, it had 4 cores. Some of the resources inside those cores were doubled up - just like hyperthreading does too, and nobody tries to call those full cores.

just i said, internet is full of idiots. hyperthreading doesn't double resources, hyperthreading masks cache misses. bulldozer has complete cores, just slightly weaker than competition - that was whole point of it - more weaker cores for same money. hyperthreading can easily scale negatively without any synchronization contention, unlike real cores. what idiots call "doubled up resources" is an optimization - by having two fpus shared by two cores you will have better performance when not every instruction in every thread uses fpu. alternative is to have one fpu per core - same number of transistors, same performance in worst case, worse performance on average.

[ktraglin]

https://www.phoronix.com/scan.php?pa...e-0.1-Released

Wow - I missed that article. Seems we definately need a better education system. It's sad to see that so many people can become so triggered by an acronym that can have so many different meanings. My guess is it's mainly people in the U.S. Makes me want to leave - move to Canada.

[smitty3268]

It had 8 cores with shared smt fpu between every two cores.

False.

It had 8 "integer clusters". An integer cluster is not a core.

It only had 1 instruction decoder per module. 1 dispatch block per module. 1 FPU per module. Etc. Only about 12% of the module was duplicated. That's a bit more than you get from hyperthreading SMT, but not that much more. Hyperthreading results in duplicated silicon as well, just to a slightly lesser degree.

A module = 1 core.

Even AMD ended up basically admitting this, when they settled the false advertising lawsuit.

[Rallos Zek]

No, it had 4 cores. Some of the resources inside those cores were doubled up - just like hyperthreading does too, and nobody tries to call those full cores.

It had 8 cores with shared smt fpu between every two cores.

[skeevy420]

it really has 8 cores, just weak ones(which is not a crime). but internet is full of idiots with strange fantasies

But its 8 cores are like the 16 cores in my system:

inxi -C
CPU:       Topology: 2x Quad Core model: Intel Xeon X5687 bits: 64 type: MT MCP SMP L2 cache: 24.0 MiB
           Speed: 1597 MHz min/max: 1596/3592 MHz Core speeds (MHz): 1: 1596 2: 1596 3: 1596 4: 1596 5: 1596 6: 1597 7: 1596
           8: 1596 9: 1596 10: 1596 11: 1596 12: 1596 13: 1596 14: 1596 15: 1596 16: 1597

Yeah, there's 16 of 'em up there, but come on now. You know as well as I do that's only 8 actual cores.

Calling "shared module cores" actual cores is like trying to bullshit someone that a 2x4 Dually pickup is a 4wd because the 4 tires in the rear are driving it forward. Technically speaking, that is 4wd, but it sure isn't 4wd in the meaning anyone reasonable would expect.

[smitty3268]

it really has 8 cores, just weak ones(which is not a crime). but internet is full of idiots with strange fantasies

No, it had 4 cores. Some of the resources inside those cores were doubled up - just like hyperthreading does too, and nobody tries to call those full cores.

[pal666]

The fuck is bulldozer parenting? Buying your kid an FX-8350 and making them believe it really has 8 cores?

it really has 8 cores, just weak ones(which is not a crime). but internet is full of idiots with strange fantasies

[pal666]

people shouldn't write raw malloc and free calls, so analyzer analyzing malloc and free has zero value for correct by design code.
really valuable analyzer is being designed by competitor http://www.open-std.org/jtc1/sc22/wg...18/p1179r0.pdf

[Michael]

I see. Now I'm curious... have people freaked out regarding the acronym for the GNU Image Manipulation Program?

https://www.phoronix.com/scan.php?page=news_item&px=Glimpse-0.1-Released