Add x86 intrinsics for bit manipulation (BMI 1.0, BMI 2.0, and TBM). #34412

gnzlbg · 2016-06-22T14:48:37Z

This PR adds the LLVM x86 intrinsics for the bit manipulation instruction sets (BMI 1.0, BMI 2.0, and TBM).

The objective of this pull-request is to allow building a library that implements all the algorithms offered by those instruction sets, using compiler intrinsics for the targets that support them (by means of target_feature).

The target features added are:

bmi: Bit Manipulation Instruction Set 1.0, available in Intel >= Haswell and AMD's >= Jaguar/Piledriver,
bmi2: Bit Manipulation Instruction Set 2.0, available in Intel >= Haswell and AMD's >= Excavator,
tbm: Trailing Bit Manipulation, available only in AMD's Piledriver (won't be available in newer CPUs).

The intrinsics added are:

BMI 1.0:
- bextr: Bit field extract (with register).
BMI 2.0:
- bzhi: Zero high bits starting with specified bit position.
- pdep: Parallel bits deposit.
- pext: Parallel bits extract.
TBM:
- bextri: Bit field extract (with immediate).

…nerator.py

…I 2.0, and TBM.

rust-highfive · 2016-06-22T14:48:52Z

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @nikomatsakis (or someone else) soon.

If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes.

Please see the contribution instructions for more information.

…0, BMI 2.0, and TBM.

gnzlbg · 2016-06-23T08:26:30Z

I don' have access to all the platforms that the SIMD crate supports. Is there a better way to test it than compiling the SIMD crate for all the platforms it supports?

gnzlbg · 2016-06-24T10:06:40Z

@huonw you used to work on the SIMD crate, any ideas about how to test it?

Ping: @ranma42 @brson @alexcrichton

brson · 2016-06-25T01:38:29Z

Might be good to get some more eyes on this still. Previous issue didn't generate much discussion.

Niko's on vacation. r? @nrc

alexcrichton · 2016-06-27T20:10:51Z

Currently the src/etc/platform-intrinsics folder is primarily used for SIMD stuff, and I don't think that we basically expose any other platform intrinsics just yet (afaik). Doing it through that interface seems somewhat reasonable to me, however, especially when there's a wide number of variations in terms of types for one intrinsics to define in LLVM.

nrc · 2016-06-27T21:43:59Z

I like the looks of this PR. I'm not sure about the second commit - I don't understand the issues there. So, r? @alexcrichton

gnzlbg · 2016-06-27T22:41:26Z

@nrc

I'm not sure about the second commit - I don't understand the issues there.

It is what @alexcrichton mentioned above, the src/etc/platform-intrinsics is only used for SIMD instructions, and as a consequence for x86 all the intrinsics had the name x86_mm_... (mm for multi-media).

I could probably have used x86_mm_ for the bit manipulation instructions as well, but I read somewhere (in the docs of the generator.py?) that the intrinsic functions should be "closely named" to the ones exported by clang's C headers.

The second commit allows having differently named instruction sets per platform, e.g., x86_bmi_... along x86_mm_, and handling those in the match statement for the platform (e.g. x86).

Doing it through that interface seems somewhat reasonable to me, however, especially when there's a wide number of variations in terms of types for one intrinsics to define in LLVM.

While generator.py was a bit overkill for these intrinsics (the number of variations is ~2 per intrinsic), it was actually useful to generate both the match statement ("compiler-defs) and the function signatures for the intrinsics automatically.

alexcrichton · 2016-06-28T17:17:35Z

Discussed during the libs triage meeting yesterday, the conclusion was that this is fine to merge, thanks for the PR @gnzlbg! We're in general pretty amenable to adding any support for fancy intrinsics like this, although we'd be much more hesitant about adding bits and pieces to the standard library, but that's not happening here!

@bors: r+ 483bec7

bors · 2016-06-29T11:00:53Z

⌛ Testing commit 483bec7 with merge 849864d...

bors · 2016-06-29T12:42:06Z

💔 Test failed - auto-linux-64-x-android-t

gnzlbg · 2016-06-29T13:11:59Z

It's my first PR so I don't really know what to look for in the build logs yet to figure out what failed, but this looks like some sort of time out while compiling.

alexcrichton · 2016-06-29T16:19:04Z

Ah no worries that just means that the Android build timed out, which is unlikely to be related to this patch. so let's just...

@bors: retry

gnzlbg · 2016-06-30T19:32:36Z

Maybe I did something wrong?

alexcrichton · 2016-06-30T22:34:38Z

Oh nah this PR is fine, we're just having a lot of automation trouble lately apparently so it may take awhile for bors to get around to this PR

bors · 2016-07-01T12:25:20Z

⌛ Testing commit 483bec7 with merge 7e269b2...

bors · 2016-07-01T12:29:53Z

💔 Test failed - auto-win-msvc-64-opt-rustbuild

gnzlbg · 2016-07-01T15:47:52Z

@bors: retry

alexcrichton · 2016-07-01T15:57:05Z

@bors: retry

Cascading failure on bots when interrupted when performing git operations #34595

bors · 2016-07-01T19:26:59Z

⌛ Testing commit 483bec7 with merge f8be993...

bors · 2016-07-01T19:31:56Z

💔 Test failed - auto-win-msvc-64-opt-rustbuild

gnzlbg · 2016-07-01T19:59:47Z

Come on @bors you can do it!

@bors: retry

gnzlbg · 2016-07-05T13:06:29Z

@bors: retry

GuillaumeGomez · 2016-07-05T13:16:59Z

@bors: retry

bors · 2016-07-05T15:25:38Z

⌛ Testing commit 483bec7 with merge ec58d0c...

…r=alexcrichton Add x86 intrinsics for bit manipulation (BMI 1.0, BMI 2.0, and TBM). This PR adds the LLVM x86 intrinsics for the bit manipulation instruction sets (BMI 1.0, BMI 2.0, and TBM). The objective of this pull-request is to allow building a library that implements all the algorithms offered by those instruction sets, using compiler intrinsics for the targets that support them (by means of `target_feature`). The target features added are: - `bmi`: Bit Manipulation Instruction Set 1.0, available in Intel >= Haswell and AMD's >= Jaguar/Piledriver, - `bmi2`: Bit Manipulation Instruction Set 2.0, available in Intel >= Haswell and AMD's >= Excavator, - `tbm`: Trailing Bit Manipulation, available only in AMD's Piledriver (won't be available in newer CPUs). The intrinsics added are: - BMI 1.0: - `bextr`: Bit field extract (with register). - BMI 2.0: - `bzhi`: Zero high bits starting with specified bit position. - `pdep`: Parallel bits deposit. - `pext`: Parallel bits extract. - TBM: - `bextri`: Bit field extract (with immediate).

bors · 2016-07-05T19:17:43Z

gnzlbg added 3 commits June 22, 2016 14:20

Add usage examples to the documentation of etc/platform-intrinsics/ge…

cb4f54d

…nerator.py

Allow different instruction set prefixes within the same architecture

1520554

Add intrinsics for x86 bit manipulation instruction sets: BMI 1.0, BM…

10cbc37

…I 2.0, and TBM.

rust-highfive assigned nikomatsakis Jun 22, 2016

gnzlbg mentioned this pull request Jun 22, 2016

Support detecting all llc's x86 target features #34397

Closed

Add target_features for the bit manipulation instruction sets: BMI 1.…

483bec7

…0, BMI 2.0, and TBM.

gnzlbg force-pushed the document_platform_intrinsics_generate branch from 7ee154e to 483bec7 Compare June 22, 2016 15:11

brson added T-libs-api Relevant to the library API team, which will review and decide on the PR/issue. T-lang Relevant to the language team, which will review and decide on the PR/issue. I-nominated labels Jun 25, 2016

rust-highfive assigned nrc and unassigned nikomatsakis Jun 25, 2016

brson mentioned this pull request Jun 25, 2016

Testing brson/miri#15

Open

rust-highfive assigned alexcrichton and unassigned nrc Jun 27, 2016

alexcrichton removed I-nominated T-lang Relevant to the language team, which will review and decide on the PR/issue. labels Jun 28, 2016

bors merged commit 483bec7 into rust-lang:master Jul 5, 2016

briansmith mentioned this pull request Dec 15, 2016

Tracking issue for cfg_target_feature #29717

Closed

1 task

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add x86 intrinsics for bit manipulation (BMI 1.0, BMI 2.0, and TBM). #34412

Add x86 intrinsics for bit manipulation (BMI 1.0, BMI 2.0, and TBM). #34412

gnzlbg commented Jun 22, 2016 •

edited

Loading

rust-highfive commented Jun 22, 2016

gnzlbg commented Jun 23, 2016

gnzlbg commented Jun 24, 2016

brson commented Jun 25, 2016

alexcrichton commented Jun 27, 2016

nrc commented Jun 27, 2016

gnzlbg commented Jun 27, 2016 •

edited

Loading

alexcrichton commented Jun 28, 2016

bors commented Jun 29, 2016

bors commented Jun 29, 2016

gnzlbg commented Jun 29, 2016

alexcrichton commented Jun 29, 2016

gnzlbg commented Jun 30, 2016

alexcrichton commented Jun 30, 2016

bors commented Jul 1, 2016

bors commented Jul 1, 2016

gnzlbg commented Jul 1, 2016

alexcrichton commented Jul 1, 2016 •

edited

Loading

bors commented Jul 1, 2016

bors commented Jul 1, 2016

gnzlbg commented Jul 1, 2016

gnzlbg commented Jul 5, 2016

GuillaumeGomez commented Jul 5, 2016

bors commented Jul 5, 2016

bors commented Jul 5, 2016

Add x86 intrinsics for bit manipulation (BMI 1.0, BMI 2.0, and TBM). #34412

Add x86 intrinsics for bit manipulation (BMI 1.0, BMI 2.0, and TBM). #34412

Conversation

gnzlbg commented Jun 22, 2016 • edited Loading

rust-highfive commented Jun 22, 2016

gnzlbg commented Jun 23, 2016

gnzlbg commented Jun 24, 2016

brson commented Jun 25, 2016

alexcrichton commented Jun 27, 2016

nrc commented Jun 27, 2016

gnzlbg commented Jun 27, 2016 • edited Loading

alexcrichton commented Jun 28, 2016

bors commented Jun 29, 2016

bors commented Jun 29, 2016

gnzlbg commented Jun 29, 2016

alexcrichton commented Jun 29, 2016

gnzlbg commented Jun 30, 2016

alexcrichton commented Jun 30, 2016

bors commented Jul 1, 2016

bors commented Jul 1, 2016

gnzlbg commented Jul 1, 2016

alexcrichton commented Jul 1, 2016 • edited Loading

bors commented Jul 1, 2016

bors commented Jul 1, 2016

gnzlbg commented Jul 1, 2016

gnzlbg commented Jul 5, 2016

GuillaumeGomez commented Jul 5, 2016

bors commented Jul 5, 2016

bors commented Jul 5, 2016

gnzlbg commented Jun 22, 2016 •

edited

Loading

gnzlbg commented Jun 27, 2016 •

edited

Loading

alexcrichton commented Jul 1, 2016 •

edited

Loading