feat(compiler)!: Custom Grain object files #2104

ospencer · 2024-05-04T00:32:27Z

Overview

This PR replaces Grain's intermediate .gr.wasm files with Grain object files. This is for a variety of reasons, including:

Debuggability. Our wasm-based linker is unable to reconstruct debugging/sourcemap information from wasm binaries. This approach once again generates proper sourcemaps. There is additional work to be done to make the sourcemaps truly usable, but this is a major step in the right direction.
Optimizations. This change will allow us to do "whole program" optimizations to better assist Binaryen with its optimizations.
Performance. Going from our IR to Binaryen IR and serializing is the bulk of compilation time. Since we no longer are serializing/deserializing many files, there is a performance boost. For a hello world program, cold compile time is down 10% and warm compile time is down 25%. When running in JS, cold compile time is down 15% and warm compile time is down 21%. This also gets subsecond warm builds in JS on my machine.

Technical details

Grain object files use the .gro extension. As no wasm files are specific to Grain, the default output filename from the compiler uses the .wasm extension rather than .gr.wasm.

Object files follow this layout:

magic	version len	version	cmi length	cmi	mash_code
0xF0 0x9F 0x8C 0xBE	5	0.6.3	1024	<cmi>	<mash_code>

We use OCaml's Marshal module as our binary format (and for Whole Grain, we can use our Marshal module). js_of_ocaml can't marshal floats into a format that can safely be read back by native code. I wanted object files to be the same between native and js, so we store floats as int64s in the mashtree to guarantee consistent behavior.

Reviewing this PR

The code delta for this PR is fairly small, but snapshot tests are now sexprs of mashtrees. As such, we have all new snapshots. I'd recommend clicking the first commit to review the code diff, or using the GitHub UI to hide changes to snapshots.

Closes #622
Closes #842
Closes #1754
Closes #1903
Closes #1997

stdlib/package.json

spotandjake · 2024-05-04T00:51:17Z

compiler/test/suites/basic_functionality.re

@@ -377,6 +377,6 @@ describe("basic functionality", ({test, testSkip}) => {
    ~config_fn=smallestFileConfig,
    "smallest_grain_program",
    "",
-    4769,
+    6424,


Why did this go up?

Is it because of names changing and debug info?

I meant to leave a comment about this; thanks for pointing it out. The old linker had a concept of hard/soft dependencies. If a file was a dependency but no values/functions from that module were actually uses (just types or an unused import), then the linker would compile in type information but not link in the actual files.

At first thought this seems like good behavior, but if this is done, side effects from importing those modules don't happen, which is bad. For example, a normal Grain program that doesn't use anything in Pervasives doesn't link in Pervasives, which means the setupExceptions side effects don't happen, and throwing exceptions in those programs prints GrainError instead of the actual error. In practice, users don't run into this because most programs use something from Pervasives.

This new linker links in all dependencies, which is more correct. As I mentioned, in practice this doesn't make a real difference for most users. Side note: the major reason the original implementation did this was because the compiler wasn't great at removing unused code from the resulting binary and this resulted in large module sizes. The compiler has gotten much better and this isn't really an issue anymore.

I also discovered a few more configurations producing smaller programs while working on this, as small as 25 bytes.

Ah okay then this probably closes #1903 then.

compiler/src/codegen/transl_anf.re

compiler/src/codegen/mashtree.re

compiler/src/typed/cmi_format.re

compiler/src/codegen/compcore.re

ospencer · 2024-09-22T14:43:23Z

@alex-snezhko This should be ready for you to look at again.

alex-snezhko

Did not see any obvious problems, LGTM!

alex-snezhko · 2024-09-22T22:10:35Z

compiler/src/codegen/compcore.re

+    Function.set_start(wasm_mod, start);
+  } else {
+    ignore @@
+    Export.add_function_export(wasm_mod, "_start", Comp_utils.grain_start);


Nit: inconsistent use of grain_start constant and "_start" literal (a few other times in the lines above)

spotandjake

Mostly looks great to me, I love the new implementation but I have a few questions.

We should probably also update the contributer docs in this pr so they do not get too far out of date.

compiler/src/codegen/compcore.re

compiler/src/codegen/emitmod.re

compiler/src/codegen/linkedtree.re

compiler/src/typed/cmi_format.re

spotandjake

LGTM, can we also open an issue to update the linking section in the contributer docs, we should probably include the binary format there, and some stuff from the pr description.

spotandjake

This looks great to me, I had one tiny question and it looks like some of the tests are failing currently.

I love that we are using an exceptionMod flag it seems a lot less fragile then file paths.

Before we get this merged could we update the static linking contributor docs I think the pr description covers a lot of the content we would want.

compiler/src/codegen/compcore.re

alex-snezhko · 2024-12-28T00:20:58Z

compiler/src/compile.re

@@ -149,6 +149,7 @@ let next_state = (~is_root_file=false, {cstate_desc, cstate_filename} as cs) =>
      Grain_utils.Config.apply_attribute_flags(
        ~no_pervasives=has_attr("noPervasives"),
        ~runtime_mode=has_attr("runtimeMode"),
+        ~no_exception_mod=has_attr("exceptionMod"),


What about "noExceptionMod" to keep with the pattern of "noPervasives"? I assume you did this because currently this attribute would only be used on the exception module so it is taken to mean "this is the exception module", but "noExceptionMod" more accurately reflects what the flag is actually doing I feel.

I had thought of noExceptions but I thought that was too ambiguous, but I've come around on it.

ospencer · 2024-12-28T23:23:16Z

@spotandjake I updated the contributor docs.

spotandjake

This looks great oscar, im excited to get this into main I had one tiny NIT left but otherwise lgtm.

spotandjake · 2024-12-29T00:31:34Z

stdlib/runtime/exception.gr

@@ -1,5 +1,5 @@
 @noPervasives
-@exceptionMod
+@noExceptions


One tiny NIT, do we just want to add a comment here something like, // Exceptions cannot be thrown from the exception module or we would have a circular dependency.?

marcusroberts

LGTM!

I really like the new approach, and the code read fine - I'll not spot any bugs though but looking forward to this being the new way of generating code.

ospencer self-assigned this May 4, 2024

ospencer requested review from phated, peblair, alex-snezhko, marcusroberts and jozanza as code owners May 4, 2024 00:32

spotandjake reviewed May 4, 2024

View reviewed changes

stdlib/package.json Show resolved Hide resolved

spotandjake reviewed May 4, 2024

View reviewed changes

ospencer force-pushed the oscar/gro-linking branch 2 times, most recently from 92e6423 to 89fcb11 Compare May 4, 2024 16:15

ospencer changed the base branch from main to oscar/digest May 4, 2024 16:19

alex-snezhko reviewed Jul 20, 2024

View reviewed changes

compiler/src/codegen/transl_anf.re Show resolved Hide resolved

compiler/src/codegen/mashtree.re Outdated Show resolved Hide resolved

compiler/src/typed/cmi_format.re Show resolved Hide resolved

compiler/src/codegen/compcore.re Show resolved Hide resolved

Base automatically changed from oscar/digest to main July 28, 2024 22:04

spotandjake mentioned this pull request Aug 2, 2024

Update contributor docs for static linking #622

Open

spotandjake added the compiler label Aug 19, 2024

ospencer force-pushed the oscar/gro-linking branch from 193e5ec to 6ac38da Compare September 22, 2024 14:25

alex-snezhko approved these changes Sep 22, 2024

View reviewed changes

spotandjake reviewed Sep 22, 2024

View reviewed changes

ospencer mentioned this pull request Oct 10, 2024

Update debug location of grain_main #2174

Open

spotandjake approved these changes Oct 10, 2024

View reviewed changes

ospencer force-pushed the oscar/gro-linking branch 3 times, most recently from d6920a8 to e4632f3 Compare December 13, 2024 00:58

spotandjake reviewed Dec 13, 2024

View reviewed changes

compiler/src/codegen/compcore.re Show resolved Hide resolved

alex-snezhko approved these changes Dec 28, 2024

View reviewed changes

spotandjake approved these changes Dec 29, 2024

View reviewed changes

marcusroberts reviewed Jan 2, 2025

View reviewed changes

feat(compiler)!: Custom Grain object files

4d60c05

marcusroberts approved these changes Jan 2, 2025

View reviewed changes

ospencer added 19 commits January 2, 2025 13:58

snapshots

ee7936e

disable locations in snapshots

2dc666b

reset idents on every compile

7c942a2

format

072208d

don't compile warning tests past typed well formed

94452f3

convert nan test to runnable

b47815e

Update smallest program size

06ff105

make mash globals a record

165f255

consistent use of grain_start

5f84500

snapshots

4d32c89

refactor compcore

c797d02

refactor objects to not store signature twice

7bb64fb

document object file layout

2d314af

make small function for internal names in linkedtree

aa4c88c

Merge with main with fixed exception handling

462676b

fixed jsoo issue

209b9e4

update contributor docs

2b205ef

rename exceptionMod -> noExceptions

e4a92c1

Update list snapshot

eef00bd

ospencer force-pushed the oscar/gro-linking branch from 72784ac to eef00bd Compare January 2, 2025 20:10

spotandjake mentioned this pull request Jan 4, 2025

Have compiler not allow Pervasives to depend on Pervasives #1632

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(compiler)!: Custom Grain object files #2104

feat(compiler)!: Custom Grain object files #2104

ospencer commented May 4, 2024 •

edited

Loading

spotandjake May 4, 2024

ospencer May 4, 2024

spotandjake May 4, 2024

ospencer commented Sep 22, 2024

alex-snezhko left a comment

alex-snezhko Sep 22, 2024

spotandjake left a comment •

edited

Loading

spotandjake left a comment

spotandjake left a comment •

edited

Loading

alex-snezhko Dec 28, 2024

ospencer Dec 28, 2024

ospencer Dec 28, 2024

ospencer commented Dec 28, 2024

spotandjake left a comment •

edited

Loading

spotandjake Dec 29, 2024

marcusroberts left a comment

feat(compiler)!: Custom Grain object files #2104

Are you sure you want to change the base?

feat(compiler)!: Custom Grain object files #2104

Conversation

ospencer commented May 4, 2024 • edited Loading

Overview

Technical details

Reviewing this PR

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ospencer commented Sep 22, 2024

alex-snezhko left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

spotandjake left a comment • edited Loading

Choose a reason for hiding this comment

spotandjake left a comment

Choose a reason for hiding this comment

spotandjake left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ospencer commented Dec 28, 2024

spotandjake left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marcusroberts left a comment

Choose a reason for hiding this comment

ospencer commented May 4, 2024 •

edited

Loading

spotandjake left a comment •

edited

Loading

spotandjake left a comment •

edited

Loading

spotandjake left a comment •

edited

Loading