[1] https://github.com/cloudflare/linux/blob/master/patches/0023...
517 karma · joined April 28, 2012
Netflix. ex-Google. ex-Intel. Gentoo Developer.
[1] https://github.com/cloudflare/linux/blob/master/patches/0023...
Interesting project, if it is still alive.
I sit beside the author of the Iris driver and work on the same team at Intel. The reasoning for only supporting BDW+ has absolutely nothing to do with spectre/meltdown or trying to force people to upgrade their hardware. Such a statement really indicates that you're not familiar with the people on Intel's Mesa team.
FYI the reasoning for only supporting BDW+ is that BDW+ supports 48-bit addressing which allows the CPU and GPU to share addresses trivially. This means no relocations are required and that's a great driver simplification and performance improvement.
BDW+ also supports executing all shader stages in the "scalar" (vs the vec4) mode, which allows much better opportunity for shader compiler optimization, which is the area I spend most of my time.
That is hilariously false. I work on the drivers. They're open source. Go take a look.
I wouldn't normally comment on something like this, but I noticed this mistake twice in the linked post as well. ia64 refers to Itanium, while amd64/x86-64/x64 refer to the 64-bit version of x86.
[ 0.000000] microcode: microcode updated early to revision 0xc2, date = 2017-11-16
[ 0.000000] Linux version 4.14.14+ (mattst88@p50-ethernet) (gcc version 6.4.0 (Gentoo 6.4.0-r1 p1.3)) #19 SMP Sun Feb 4 00:20:02 PST 2018(Disclaimer: I work for Intel, but not on anything related to this and have no knowledge on the topic outside of this PDF)
Rather than assuming individuals are just fickle, I think it makes much more sense that individuals are often only vocal about things they disagree with. We as readers don't know those individuals and lump them into our "hive mind" concept, as mistakenly assume that people's minds have changed.
For one, those people working on Wayland made enormous contributions to the X Server long before Wayland existed. The creator of Wayland, Kristian Høgsberg did AIGLX, just to name one example.
I agree that this has been confusing in the past (as evidenced by the mix of Atom, Celeron, Pentium, i3, i5, i7, Xeon in your link), but I think at least in VT-* it's been greatly simplified.
I'm not an expert on Intel CPU market segmentation, but I do work for Intel on the i965 Mesa driver. I don't think your statement is true.
See http://ark.intel.com/search/advanced?s=t&VTX=true
That's a list of all Intel CPUs with VT-x. It's really easy to narrow down the selection from there. FWIW, I see lots of Skylake i3, i5, and i7s with VT-x.
I sincerely hope that you test our Linux drivers (I work for Intel on them). We'd be excited to learn about anything you find. Please feel free to file bugs here (https://bugs.freedesktop.org/enter_bug.cgi?product=Mesa&comp...) and ping us on #intel-gfx on Freenode.
I could be wrong for doing so, but I read that as somewhat derogatory, in case you didn't mean it that way.
Mentioning Thurmond is simply misleading and detracts from the main point that Sessions was rejected from a federal judgeship because of concerns that he was a racist.
The fact of the matter is that Thurmond voted for Sessions.
> On June 5, 1986, the Committee voted 10–8 against recommending the nomination to the Senate floor, with Republican Senators Charles Mathias of Maryland and Arlen Specter of Pennsylvania voting with the Democrats.
http://koin.com/2014/04/01/1-hurt-car-max-crash-portland/ http://www.oregonlive.com/portland/index.ssf/2013/06/max_col... http://www.oregonlive.com/portland/index.ssf/2015/05/max_blu... https://www.reddit.com/r/Portland/comments/3d5i9g/saw_an_ora...
There may be more to it in Portland than "at grade", but that certainly seems to be a large component of the problem.
We've had some cargo culted patches to call _mesa_error_no_memory() when malloc fails, but it was recently noted that this can't possibly work because it internally calls fprintf, and there's no guarantee that fprintf doesn't call malloc! In fact, glibc's implementation does.
Suggestions for best practices welcome.
We (the team that writes the Intel driver in Mesa) would love to be involved in things like that and would be happy to answer questions where possible -- #intel-gfx on Freenode or mesa-dev@lists.freedesktop.org.
The documentation doesn't contain much if anything in the way of performance data like instruction latency, so we've implemented similar methods to what the author used to measure the number of cycles a piece of code takes (environment variable INTEL_DEBUG=shader_time). We'd likely even learn something about the hardware ourselves by being involved in experiments like these.
SIMD8/SIMD16 refers to the number of fragments processed per thread invocation. There is some overhead to spawning a thread, and so processing 16 fragments at a time is typically faster even though the shader itself is doing more work.
The driver provides both versions because it's the GPU that decides which version is use and where, even using both versions to shade the same primitive. For instance, on a triangle boundary maybe only 4 or 8 fragments are "lit", so the hardware spawns a SIMD8 thread and saves itself a little work.
SIMD16 shaders typically use twice the number of registers as SIMD8, and if they require registers to be spilled to memory it's likely the SIMD8 shader is faster even with the additional thread-spawning overhead. Lots of compiler optimization revolves around trying to squeak in under the register limit to get the program compiled as SIMD16 without spilling. :)
I've never used VTune (but it looks very cool) and some of the notation is different from what I'm used to.
If you're using an Intel GPU made in the last 10 years on Linux, the environment variable INTEL_DEBUG will allow you to see the disassembled shaders for a given program. Try "INTEL_DEBUG=fs,vs glxgears" to see the fragment and vertex shaders' assembly and the dumps of some intermediate representations along the way.
The Gen (or i965, as we call it) instruction set is really powerful, but does take a while to understand fully.
I've actually been trying to finish up an article about some tricks we do in the i965_dri.so driver (think bit twiddling hacks, but each using some interesting features of the instruction set). I'm curious if there's interest in such a thing. I'd probably be more motivated to finish it. :)
#include <stdint.h>
uint64_t inc(uint64_t x) { return x + 1; }
With -O0 gives me: 48 83 c0 01 add $0x1,%rax
and with -O2: 48 8d 47 01 lea 0x1(%rdi),%rax
That is, both are four bytes. array[i++] = this;
array[i++] = that;
array[i++] = the_other;
which also extends nicely when some of those elements are optional.