Increase vtable layout size #123572

Mark-Simulacrum · 2024-04-06T21:11:49Z

This improves LLVM's codegen by allowing vtable loads to be hoisted out of loops (as just one example). The calculation here is an under-approximation but works for simple trait hierarchies (e.g., FnMut will be improved). We have a runtime assert that the approximation is accurate, so there's no risk of UB as a result of getting this wrong.

#[no_mangle]
pub fn foo(elements: &[u32], callback: &mut dyn Callback) {
    for element in elements.iter() {
        if *element != 0 {
            callback.call(*element);
        }
    }
}

pub trait Callback {
    fn call(&mut self, _: u32);
}

Simplifying a bit (e.g., numbering ends up different):

 ; Function Attrs: nonlazybind uwtable
-define void @foo(ptr noalias noundef nonnull readonly align 4 %elements.0, i64 noundef %elements.1, ptr noundef nonnull align 1 %callback.0, ptr noalias nocapture noundef readonly align 8 dereferenceable(24) %callback.1) unnamed_addr #0 {
+define void @foo(ptr noalias noundef nonnull readonly align 4 %elements.0, i64 noundef %elements.1, ptr noundef nonnull align 1 %callback.0, ptr noalias nocapture noundef readonly align 8 dereferenceable(32) %callback.1) unnamed_addr #0 {
 start:
   %_15 = getelementptr inbounds i32, ptr %elements.0, i64 %elements.1
@@ -13,4 +13,5 @@
 bb4.lr.ph:                                        ; preds = %start
   %1 = getelementptr inbounds i8, ptr %callback.1, i64 24
+  %2 = load ptr, ptr %1, align 8, !nonnull !3
   br label %bb4

 bb6:                                              ; preds = %bb4
-  %4 = load ptr, ptr %1, align 8, !invariant.load !3, !nonnull !3
-  tail call void %4(ptr noundef nonnull align 1 %callback.0, i32 noundef %_9)
+  tail call void %2(ptr noundef nonnull align 1 %callback.0, i32 noundef %_9)
   br label %bb7
 }

rustbot · 2024-04-06T21:11:56Z

r? @BoxyUwU

rustbot has assigned @BoxyUwU.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Mark-Simulacrum · 2024-04-06T21:14:12Z

@bors try @rust-timer queue

Increase vtable layout size This improves LLVM's codegen by allowing vtable loads to be hoisted out of loops (as just one example). Unfortunately it doesn't help with e.g. FnMut (due to FnOnce occupying a slot in the supertraits there) but seems like potentially still a win for simpler cases. ```rust #[no_mangle] pub fn foo(elements: &[u32], callback: &mut dyn Callback) { for element in elements.iter() { if *element != 0 { callback.call(*element); } } } pub trait Callback { fn call(&mut self, _: u32); } ``` Simplifying a bit (e.g., numbering ends up different): ```diff ; Function Attrs: nonlazybind uwtable -define void `@foo(ptr` noalias noundef nonnull readonly align 4 %elements.0, i64 noundef %elements.1, ptr noundef nonnull align 1 %callback.0, ptr noalias nocapture noundef readonly align 8 dereferenceable(24) %callback.1) unnamed_addr #0 { +define void `@foo(ptr` noalias noundef nonnull readonly align 4 %elements.0, i64 noundef %elements.1, ptr noundef nonnull align 1 %callback.0, ptr noalias nocapture noundef readonly align 8 dereferenceable(32) %callback.1) unnamed_addr #0 { start: %_15 = getelementptr inbounds i32, ptr %elements.0, i64 %elements.1 `@@` -13,4 +13,5 `@@` bb4.lr.ph: ; preds = %start %1 = getelementptr inbounds i8, ptr %callback.1, i64 24 + %2 = load ptr, ptr %1, align 8, !nonnull !3 br label %bb4 bb6: ; preds = %bb4 - %4 = load ptr, ptr %1, align 8, !invariant.load !3, !nonnull !3 - tail call void %4(ptr noundef nonnull align 1 %callback.0, i32 noundef %_9) + tail call void %2(ptr noundef nonnull align 1 %callback.0, i32 noundef %_9) br label %bb7 } ```

bors · 2024-04-06T21:15:23Z

⌛ Trying commit 12d2363 with merge dab7be8...

bors · 2024-04-06T22:48:46Z

☀️ Try build successful - checks-actions
Build commit: dab7be8 (dab7be869dcea44b6f6dcb21b20eff3a3f2f75e0)

rust-timer · 2024-04-07T01:55:38Z

Finished benchmarking commit (dab7be8): comparison URL.

Overall result: ❌ regressions - no action needed

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

@bors rollup=never
@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	1.2%	[1.2%, 1.2%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	9.0%	[9.0%, 9.0%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-3.8%	[-4.5%, -3.1%]	2
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.5%	[-4.5%, 9.0%]	3

Cycles

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.2%	[1.1%, 2.7%]	7
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.0%	[0.0%, 0.0%]	22
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.5%	[-0.5%, -0.5%]	3
Improvements ✅ (secondary)	-1.0%	[-1.0%, -1.0%]	1
All ❌✅ (primary)	-0.0%	[-0.5%, 0.0%]	25

Bootstrap: 670.181s -> 668.562s (-0.24%)
Artifact size: 318.20 MiB -> 318.22 MiB (0.00%)

Mark-Simulacrum · 2024-04-08T13:11:08Z

@bors try @rust-timer queue

Pushed what I hope is a fix for the cycle error and an improvement on the accuracy of the approximation that catches FnMut (and other cases) as well. Let's re-run perf since it is a more expensive computation now.

bors · 2024-04-08T13:12:18Z

⌛ Trying commit 26954e7 with merge 3d8cc34...

Increase vtable layout size This improves LLVM's codegen by allowing vtable loads to be hoisted out of loops (as just one example). The calculation here is an under-approximation but works for simple trait hierarchies (e.g., FnMut will be improved). We have a runtime assert that the approximation is accurate, so there's no risk of UB as a result of getting this wrong. ```rust #[no_mangle] pub fn foo(elements: &[u32], callback: &mut dyn Callback) { for element in elements.iter() { if *element != 0 { callback.call(*element); } } } pub trait Callback { fn call(&mut self, _: u32); } ``` Simplifying a bit (e.g., numbering ends up different): ```diff ; Function Attrs: nonlazybind uwtable -define void `@foo(ptr` noalias noundef nonnull readonly align 4 %elements.0, i64 noundef %elements.1, ptr noundef nonnull align 1 %callback.0, ptr noalias nocapture noundef readonly align 8 dereferenceable(24) %callback.1) unnamed_addr #0 { +define void `@foo(ptr` noalias noundef nonnull readonly align 4 %elements.0, i64 noundef %elements.1, ptr noundef nonnull align 1 %callback.0, ptr noalias nocapture noundef readonly align 8 dereferenceable(32) %callback.1) unnamed_addr #0 { start: %_15 = getelementptr inbounds i32, ptr %elements.0, i64 %elements.1 `@@` -13,4 +13,5 `@@` bb4.lr.ph: ; preds = %start %1 = getelementptr inbounds i8, ptr %callback.1, i64 24 + %2 = load ptr, ptr %1, align 8, !nonnull !3 br label %bb4 bb6: ; preds = %bb4 - %4 = load ptr, ptr %1, align 8, !invariant.load !3, !nonnull !3 - tail call void %4(ptr noundef nonnull align 1 %callback.0, i32 noundef %_9) + tail call void %2(ptr noundef nonnull align 1 %callback.0, i32 noundef %_9) br label %bb7 } ```

bors · 2024-04-08T14:44:32Z

☀️ Try build successful - checks-actions
Build commit: 3d8cc34 (3d8cc34f2e1d2a0ba5f98eb7eba9ab89daa48e5e)

rust-timer · 2024-04-08T16:44:43Z

Finished benchmarking commit (3d8cc34): comparison URL.

Overall result: ❌✅ regressions and improvements - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.5%	[0.2%, 1.0%]	7
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.2%	[-0.2%, -0.2%]	3
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

This benchmark run did not return any relevant results for this metric.

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.0%	[0.0%, 0.0%]	22
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.5%	[-0.5%, -0.5%]	3
Improvements ✅ (secondary)	-1.0%	[-1.0%, -1.0%]	1
All ❌✅ (primary)	-0.0%	[-0.5%, 0.0%]	25

Bootstrap: 666.9s -> 666.636s (-0.04%)
Artifact size: 318.48 MiB -> 318.33 MiB (-0.05%)

Mark-Simulacrum · 2024-04-08T22:12:33Z

Regressions look to be just recently introduced bimodality/noise. This should be ready for review (last push is just a typo fix).

BoxyUwU · 2024-04-11T11:19:20Z

r? compiler
this looks cool

oli-obk · 2024-05-23T11:20:14Z

@bors r+

bors · 2024-05-23T11:20:16Z

📌 Commit fbc3dff has been approved by oli-obk

It is now in the queue for this repository.

bors · 2024-05-23T11:54:05Z

⌛ Testing commit fbc3dff with merge 6d9b03e...

…i-obk Increase vtable layout size This improves LLVM's codegen by allowing vtable loads to be hoisted out of loops (as just one example). The calculation here is an under-approximation but works for simple trait hierarchies (e.g., FnMut will be improved). We have a runtime assert that the approximation is accurate, so there's no risk of UB as a result of getting this wrong. ```rust #[no_mangle] pub fn foo(elements: &[u32], callback: &mut dyn Callback) { for element in elements.iter() { if *element != 0 { callback.call(*element); } } } pub trait Callback { fn call(&mut self, _: u32); } ``` Simplifying a bit (e.g., numbering ends up different): ```diff ; Function Attrs: nonlazybind uwtable -define void `@foo(ptr` noalias noundef nonnull readonly align 4 %elements.0, i64 noundef %elements.1, ptr noundef nonnull align 1 %callback.0, ptr noalias nocapture noundef readonly align 8 dereferenceable(24) %callback.1) unnamed_addr #0 { +define void `@foo(ptr` noalias noundef nonnull readonly align 4 %elements.0, i64 noundef %elements.1, ptr noundef nonnull align 1 %callback.0, ptr noalias nocapture noundef readonly align 8 dereferenceable(32) %callback.1) unnamed_addr #0 { start: %_15 = getelementptr inbounds i32, ptr %elements.0, i64 %elements.1 `@@` -13,4 +13,5 `@@` bb4.lr.ph: ; preds = %start %1 = getelementptr inbounds i8, ptr %callback.1, i64 24 + %2 = load ptr, ptr %1, align 8, !nonnull !3 br label %bb4 bb6: ; preds = %bb4 - %4 = load ptr, ptr %1, align 8, !invariant.load !3, !nonnull !3 - tail call void %4(ptr noundef nonnull align 1 %callback.0, i32 noundef %_9) + tail call void %2(ptr noundef nonnull align 1 %callback.0, i32 noundef %_9) br label %bb7 } ```

bors · 2024-05-23T12:43:06Z

💔 Test failed - checks-actions

jieyouxu · 2024-05-26T22:02:55Z

@bors r- (manually r- because the tree looks wonky with some PRs that shouldn't be elligble for rollup)

bors · 2024-05-28T20:49:22Z

☔ The latest upstream changes (presumably #125665) made this pull request unmergeable. Please resolve the merge conflicts.

This improves LLVM's codegen by allowing vtable loads to be hoisted out of loops (as just one example).

Mark-Simulacrum · 2024-06-01T12:57:29Z

@bors r=oli-obk

Rebased and fixed a test (off by one after vtable sizes are now more precise, I think).

bors · 2024-06-01T12:57:32Z

📌 Commit 95e0732 has been approved by oli-obk

It is now in the queue for this repository.

bors · 2024-06-01T14:31:11Z

⌛ Testing commit 95e0732 with merge f2208b3...

bors · 2024-06-01T16:37:17Z

☀️ Test successful - checks-actions
Approved by: oli-obk
Pushing f2208b3 to master...

rust-timer · 2024-06-01T18:59:42Z

Finished benchmarking commit (f2208b3): comparison URL.

Overall result: ✅ improvements - no action needed

@rustbot label: -perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.3%	[-0.6%, -0.2%]	10
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results (primary -1.1%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.1%	[-1.1%, -1.1%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-1.1%	[-1.1%, -1.1%]	1

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

Results (primary -0.1%, secondary -0.2%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.0%	[0.0%, 0.0%]	9
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.1%	[-0.3%, -0.0%]	12
Improvements ✅ (secondary)	-0.2%	[-0.5%, -0.1%]	38
All ❌✅ (primary)	-0.1%	[-0.3%, 0.0%]	21

Bootstrap: 667.575s -> 668.781s (0.18%)
Artifact size: 318.75 MiB -> 318.82 MiB (0.02%)

rustbot assigned BoxyUwU Apr 6, 2024

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Apr 6, 2024