[MIR trans] Optimize trans for biased switches #33566

dotdash · 2016-05-11T19:53:38Z

Currently, all switches in MIR are exhausitive, meaning that we can have
a lot of arms that all go to the same basic block, the extreme case
being an if-let expression which results in just 2 possible cases, be
might end up with hundreds of arms for large enums.

To improve this situation and give LLVM less code to chew on, we can
detect whether there's a pre-dominant target basic block in a switch
and then promote this to be the default target, not translating the
corresponding arms at all.

In combination with #33544 this makes unoptimized MIR trans of
nickel.rs as fast as using old trans and greatly improves the times for
optimized builds, which are only 30-40% slower instead of ~300%.

cc #33111

rust-highfive · 2016-05-11T19:53:46Z

r? @arielb1

(rust_highfive has picked a reviewer for you, use r? to override)

arielb1 · 2016-05-11T20:03:36Z

src/librustc_trans/mir/block.rs

+                    // If a single target basic blocks is predominant, promote that to be the
+                    // default case for the switch instruction to reduce the size of the generated
+                    // code. This is especially helpful in cases like an if-let on a huge enum.
+                    Some((&&bb, &c)) if c as f32 / targets.len() as f32 >= 0.05 => {


c > targets.len()/2

nagisa · 2016-05-11T20:06:49Z

src/librustc_trans/mir/block.rs

-                // instruction so LLVM knows that
-                let unreachable_blk = self.unreachable_block();
-                let switch = bcx.switch(discr, unreachable_blk.llbb, targets.len());
+                let mut bb_hist = FnvHashMap();


You should add a FIXME here to check for the validity of such transformation. While currently the otherwise branch is always unreachable (and contains a panic), this might not be the case after a bunch of optimisation passes are added to the MIR thus making this transformation invalid.

I added a note.

TerminatorKind::Switch can't have an otherwise branch in any case.

There's also the assert_eq!(adt_def.variants.len(), targets.len()); which should in theory make sure we've covered all the cases.

Currently, all switches in MIR are exhausitive, meaning that we can have a lot of arms that all go to the same basic block, the extreme case being an if-let expression which results in just 2 possible cases, be might end up with hundreds of arms for large enums. To improve this situation and give LLVM less code to chew on, we can detect whether there's a pre-dominant target basic block in a switch and then promote this to be the default target, not translating the corresponding arms at all. In combination with rust-lang#33544 this makes unoptimized MIR trans of nickel.rs as fast as using old trans and greatly improves the times for optimized builds, which are only 30-40% slower instead of ~300%. cc rust-lang#33111

nagisa · 2016-05-11T20:25:21Z

I guess this is fine to land. I’d still prefer it to become a proper MIR transformation down the line but this can be done later.

@bors r+

bors · 2016-05-11T20:25:23Z

📌 Commit 49b2cdf has been approved by nagisa

[MIR trans] Optimize trans for biased switches Currently, all switches in MIR are exhausitive, meaning that we can have a lot of arms that all go to the same basic block, the extreme case being an if-let expression which results in just 2 possible cases, be might end up with hundreds of arms for large enums. To improve this situation and give LLVM less code to chew on, we can detect whether there's a pre-dominant target basic block in a switch and then promote this to be the default target, not translating the corresponding arms at all. In combination with rust-lang#33544 this makes unoptimized MIR trans of nickel.rs as fast as using old trans and greatly improves the times for optimized builds, which are only 30-40% slower instead of ~300%. cc rust-lang#33111

Rollup of 23 pull requests - Successful merges: #33282, #33342, #33393, #33450, #33513, #33517, #33531, #33532, #33533, #33534, #33538, #33541, #33544, #33552, #33555, #33560, #33563, #33565, #33566, #33572, #33580, #33590, #33596 - Failed merges: #33578

[MIR trans] Optimize trans for biased switches Currently, all switches in MIR are exhausitive, meaning that we can have a lot of arms that all go to the same basic block, the extreme case being an if-let expression which results in just 2 possible cases, be might end up with hundreds of arms for large enums. To improve this situation and give LLVM less code to chew on, we can detect whether there's a pre-dominant target basic block in a switch and then promote this to be the default target, not translating the corresponding arms at all. In combination with rust-lang#33544 this makes unoptimized MIR trans of nickel.rs as fast as using old trans and greatly improves the times for optimized builds, which are only 30-40% slower instead of ~300%. cc rust-lang#33111

Rollup of 9 pull requests - Successful merges: #33544, #33552, #33554, #33555, #33560, #33566, #33572, #33574, #33576 - Failed merges:

rust-highfive assigned arielb1 May 11, 2016

arielb1 reviewed May 11, 2016
View reviewed changes

dotdash force-pushed the biased_switch branch from 03a1a28 to ff9f1d4 Compare May 11, 2016 20:04

nagisa reviewed May 11, 2016
View reviewed changes

dotdash force-pushed the biased_switch branch from ff9f1d4 to 7a605b4 Compare May 11, 2016 20:07

dotdash force-pushed the biased_switch branch from 7a605b4 to 49b2cdf Compare May 11, 2016 20:18

nagisa mentioned this pull request May 11, 2016

Convert the default-case optimisation to a MIR pass #33567

Closed

Manishearth mentioned this pull request May 12, 2016

Rollup of 11 pull requests #33582

Closed

nagisa mentioned this pull request May 12, 2016

MIR: Don't generate 3-armed boolean switch from match. #33583

Merged

eddyb mentioned this pull request May 12, 2016

Rollup of 15 pull requests #33589

Closed

eddyb mentioned this pull request May 12, 2016

Rollup of 21 pull requests #33595

Closed

eddyb mentioned this pull request May 12, 2016

Rollup of 23 pull requests #33597

Closed

eddyb mentioned this pull request May 13, 2016

Rollup of 29 pull requests #33610

Closed

Manishearth mentioned this pull request May 14, 2016

Rollup of 9 pull requests #33632

Merged

bors added a commit that referenced this pull request May 14, 2016

Auto merge of #33632 - Manishearth:rollup, r=Manishearth

6ba8a1a

Rollup of 9 pull requests - Successful merges: #33544, #33552, #33554, #33555, #33560, #33566, #33572, #33574, #33576 - Failed merges:

bors merged commit 49b2cdf into rust-lang:master May 14, 2016

dotdash deleted the biased_switch branch May 17, 2016 03:43

nagisa mentioned this pull request Jun 1, 2016

generate fewer basic blocks for variant switches #33999

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[MIR trans] Optimize trans for biased switches #33566

[MIR trans] Optimize trans for biased switches #33566

dotdash commented May 11, 2016

rust-highfive commented May 11, 2016

arielb1 May 11, 2016

nagisa May 11, 2016 •

edited

Loading

dotdash May 11, 2016

arielb1 May 12, 2016

luqmana May 13, 2016

nagisa commented May 11, 2016

bors commented May 11, 2016

[MIR trans] Optimize trans for biased switches #33566

[MIR trans] Optimize trans for biased switches #33566

Conversation

dotdash commented May 11, 2016

rust-highfive commented May 11, 2016

arielb1 May 11, 2016

Choose a reason for hiding this comment

nagisa May 11, 2016 • edited Loading

Choose a reason for hiding this comment

dotdash May 11, 2016

Choose a reason for hiding this comment

arielb1 May 12, 2016

Choose a reason for hiding this comment

luqmana May 13, 2016

Choose a reason for hiding this comment

nagisa commented May 11, 2016

bors commented May 11, 2016

nagisa May 11, 2016 •

edited

Loading