summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/tools/wasm-split/split-options.cpp36
-rw-r--r--src/tools/wasm-split/split-options.h4
-rw-r--r--src/tools/wasm-split/wasm-split.cpp84
-rw-r--r--test/lit/help/wasm-split.test23
-rw-r--r--test/lit/wasm-split/multi-split.wast220
-rw-r--r--test/lit/wasm-split/multi-split.wast.manifest8
6 files changed, 369 insertions, 6 deletions
diff --git a/src/tools/wasm-split/split-options.cpp b/src/tools/wasm-split/split-options.cpp
index 825efddd9..e77957f1f 100644
--- a/src/tools/wasm-split/split-options.cpp
+++ b/src/tools/wasm-split/split-options.cpp
@@ -61,6 +61,9 @@ std::ostream& operator<<(std::ostream& o, WasmSplitOptions::Mode& mode) {
case WasmSplitOptions::Mode::Split:
o << "split";
break;
+ case WasmSplitOptions::Mode::MultiSplit:
+ o << "multi-split";
+ break;
case WasmSplitOptions::Mode::Instrument:
o << "instrument";
break;
@@ -91,7 +94,14 @@ WasmSplitOptions::WasmSplitOptions()
"Split an input module into two output modules. The default mode.",
WasmSplitOption,
Options::Arguments::Zero,
- [&](Options* o, const std::string& arugment) { mode = Mode::Split; })
+ [&](Options* o, const std::string& argument) { mode = Mode::Split; })
+ .add(
+ "--multi-split",
+ "",
+ "Split an input module into an arbitrary number of output modules.",
+ WasmSplitOption,
+ Options::Arguments::Zero,
+ [&](Options* o, const std::string& argument) { mode = Mode::MultiSplit; })
.add(
"--instrument",
"",
@@ -151,6 +161,25 @@ WasmSplitOptions::WasmSplitOptions()
[&](Options* o, const std::string& argument) {
splitFuncs = parseNameList(argument);
})
+ .add(
+ "--manifest",
+ "",
+ "File describing the functions to be split into each module. Each "
+ "section separated by a blank line begins with the base name of an "
+ "output module, which is followed by a list of functions to place in "
+ "that module, one per line.",
+ WasmSplitOption,
+ {Mode::MultiSplit},
+ Options::Arguments::One,
+ [&](Options* o, const std::string& argument) { manifestFile = argument; })
+ .add("--out-prefix",
+ "",
+ "Prefix prepended to module names in the manifest file to create "
+ "output file names.",
+ WasmSplitOption,
+ {Mode::MultiSplit},
+ Options::Arguments::One,
+ [&](Options* o, const std::string& argument) { outPrefix = argument; })
.add("--primary-output",
"-o1",
"Output file for the primary module.",
@@ -313,7 +342,7 @@ WasmSplitOptions::WasmSplitOptions()
"-g",
"Emit names section in wasm binary (or full debuginfo in wast)",
WasmSplitOption,
- {Mode::Split, Mode::Instrument},
+ {Mode::Split, Mode::MultiSplit, Mode::Instrument},
Options::Arguments::Zero,
[&](Options* o, const std::string& arguments) {
passOptions.debugInfo = true;
@@ -322,7 +351,7 @@ WasmSplitOptions::WasmSplitOptions()
"-o",
"Output file.",
WasmSplitOption,
- {Mode::Instrument, Mode::MergeProfiles},
+ {Mode::Instrument, Mode::MergeProfiles, Mode::MultiSplit},
Options::Arguments::One,
[&](Options* o, const std::string& argument) { output = argument; })
.add("--unescape",
@@ -407,6 +436,7 @@ bool WasmSplitOptions::validate() {
}
switch (mode) {
case Mode::Split:
+ case Mode::MultiSplit:
case Mode::Instrument:
if (inputFiles.size() > 1) {
fail("Cannot have more than one input file.");
diff --git a/src/tools/wasm-split/split-options.h b/src/tools/wasm-split/split-options.h
index b8129f29b..105c90c80 100644
--- a/src/tools/wasm-split/split-options.h
+++ b/src/tools/wasm-split/split-options.h
@@ -26,6 +26,7 @@ const std::string DEFAULT_PROFILE_EXPORT("__write_profile");
struct WasmSplitOptions : ToolOptions {
enum class Mode : unsigned {
Split,
+ MultiSplit,
Instrument,
MergeProfiles,
PrintProfile,
@@ -68,6 +69,9 @@ struct WasmSplitOptions : ToolOptions {
std::string secondaryMemoryName;
std::string exportPrefix;
+ std::string manifestFile;
+ std::string outPrefix;
+
// A hack to ensure the split and instrumented modules have the same table
// size when using Emscripten's SPLIT_MODULE mode with dynamic linking. TODO:
// Figure out a more elegant solution for that use case and remove this.
diff --git a/src/tools/wasm-split/wasm-split.cpp b/src/tools/wasm-split/wasm-split.cpp
index cb148090d..ea1734b6b 100644
--- a/src/tools/wasm-split/wasm-split.cpp
+++ b/src/tools/wasm-split/wasm-split.cpp
@@ -362,6 +362,87 @@ void splitModule(const WasmSplitOptions& options) {
writeModule(*secondary, options.secondaryOutput, options);
}
+void multiSplitModule(const WasmSplitOptions& options) {
+ if (options.manifestFile.empty()) {
+ Fatal() << "--multi-split requires --manifest";
+ }
+ if (options.output.empty()) {
+ Fatal() << "--multi-split requires --output";
+ }
+
+ std::ifstream manifest(options.manifestFile);
+ if (!manifest.is_open()) {
+ Fatal() << "File not found: " << options.manifestFile;
+ }
+
+ Module wasm;
+ parseInput(wasm, options);
+
+ // Map module names to the functions that should be in the modules.
+ std::map<std::string, std::unordered_set<std::string>> moduleFuncs;
+ // The module for which we are currently parsing a set of functions.
+ std::string currModule;
+ // The set of functions we are currently inserting into.
+ std::unordered_set<std::string>* currFuncs = nullptr;
+ // Map functions to their modules to ensure no function is assigned to
+ // multiple modules.
+ std::unordered_map<std::string, std::string> funcModules;
+
+ std::string line;
+ bool newSection = true;
+ while (std::getline(manifest, line)) {
+ if (line.empty()) {
+ newSection = true;
+ continue;
+ }
+ if (newSection) {
+ currModule = line;
+ currFuncs = &moduleFuncs[line];
+ newSection = false;
+ continue;
+ }
+ assert(currFuncs);
+ currFuncs->insert(line);
+ auto [it, inserted] = funcModules.insert({line, currModule});
+ if (!inserted && it->second != currModule) {
+ Fatal() << "Function " << line << "cannot be assigned to module "
+ << currModule << "; it is already assigned to module "
+ << it->second << '\n';
+ }
+ if (inserted && !options.quiet && !wasm.getFunctionOrNull(line)) {
+ std::cerr << "warning: Function " << line << " does not exist\n";
+ }
+ }
+
+ ModuleSplitting::Config config;
+ config.usePlaceholders = false;
+ config.importNamespace = "";
+ config.minimizeNewExportNames = true;
+ for (auto& func : wasm.functions) {
+ config.primaryFuncs.insert(func->name);
+ }
+ for (auto& [mod, funcs] : moduleFuncs) {
+ if (options.verbose) {
+ std::cerr << "Splitting module " << mod << '\n';
+ }
+ if (!options.quiet && funcs.empty()) {
+ std::cerr << "warning: Module " << mod << " will be empty\n";
+ }
+ for (auto& func : funcs) {
+ config.primaryFuncs.erase(Name(func));
+ }
+ auto splitResults = ModuleSplitting::splitFunctions(wasm, config);
+ // TODO: symbolMap, placeholderMap, emitModuleNames
+ // TODO: Support --emit-text and use .wast in that case.
+ auto moduleName = options.outPrefix + mod + ".wasm";
+ PassRunner runner(&*splitResults.secondary);
+ runner.add("remove-unused-module-elements");
+ runner.run();
+ writeModule(*splitResults.secondary, moduleName, options);
+ }
+ writeModule(wasm, options.output, options);
+}
+
void mergeProfiles(const WasmSplitOptions& options) {
// Read the initial profile. We will merge other profiles into this one.
ProfileData data = readProfile(options.inputFiles[0]);
@@ -503,6 +584,9 @@ int main(int argc, const char* argv[]) {
case WasmSplitOptions::Mode::Split:
splitModule(options);
break;
+ case WasmSplitOptions::Mode::MultiSplit:
+ multiSplitModule(options);
+ break;
case WasmSplitOptions::Mode::Instrument:
instrumentModule(options);
break;
diff --git a/test/lit/help/wasm-split.test b/test/lit/help/wasm-split.test
index 4fa534e43..e5e736562 100644
--- a/test/lit/help/wasm-split.test
+++ b/test/lit/help/wasm-split.test
@@ -15,6 +15,9 @@
;; CHECK-NEXT: --split Split an input module into two output
;; CHECK-NEXT: modules. The default mode.
;; CHECK-NEXT:
+;; CHECK-NEXT: --multi-split Split an input module into an arbitrary
+;; CHECK-NEXT: number of output modules.
+;; CHECK-NEXT:
;; CHECK-NEXT: --instrument Instrument an input module to allow it to
;; CHECK-NEXT: generate a profile that can be used to
;; CHECK-NEXT: guide splitting.
@@ -43,6 +46,18 @@
;; CHECK-NEXT: can also pass a file with one function
;; CHECK-NEXT: per line by passing @filename.
;; CHECK-NEXT:
+;; CHECK-NEXT: --manifest [multi-split] File describing the
+;; CHECK-NEXT: functions to be split into each module.
+;; CHECK-NEXT: Each section separated by a blank line
+;; CHECK-NEXT: begins with the base name of an output
+;; CHECK-NEXT: module, which is followed by a list of
+;; CHECK-NEXT: functions to place in that module, one
+;; CHECK-NEXT: per line.
+;; CHECK-NEXT:
+;; CHECK-NEXT: --out-prefix [multi-split] Prefix prepended to module
+;; CHECK-NEXT: names in the manifest file to create
+;; CHECK-NEXT: output file names.
+;; CHECK-NEXT:
;; CHECK-NEXT: --primary-output,-o1 [split] Output file for the primary
;; CHECK-NEXT: module.
;; CHECK-NEXT:
@@ -125,10 +140,12 @@
;; CHECK-NEXT: --emit-text,-S [split, instrument] Emit text instead of
;; CHECK-NEXT: binary for the output file or files.
;; CHECK-NEXT:
-;; CHECK-NEXT: --debuginfo,-g [split, instrument] Emit names section in
-;; CHECK-NEXT: wasm binary (or full debuginfo in wast)
+;; CHECK-NEXT: --debuginfo,-g [split, multi-split, instrument] Emit
+;; CHECK-NEXT: names section in wasm binary (or full
+;; CHECK-NEXT: debuginfo in wast)
;; CHECK-NEXT:
-;; CHECK-NEXT: --output,-o [instrument, merge-profiles] Output file.
+;; CHECK-NEXT: --output,-o [instrument, merge-profiles, multi-split]
+;; CHECK-NEXT: Output file.
;; CHECK-NEXT:
;; CHECK-NEXT: --unescape,-u Un-escape function names (in
;; CHECK-NEXT: print-profile output)
diff --git a/test/lit/wasm-split/multi-split.wast b/test/lit/wasm-split/multi-split.wast
new file mode 100644
index 000000000..9206e60ee
--- /dev/null
+++ b/test/lit/wasm-split/multi-split.wast
@@ -0,0 +1,220 @@
+;; NOTE: Assertions have been generated by update_lit_checks.py --all-items and should not be edited.
+
+;; RUN: wasm-split -all -g --multi-split %s --manifest %s.manifest --out-prefix=%t -o %t.wasm
+;; RUN: wasm-dis %t.wasm | filecheck %s --check-prefix=PRIMARY
+;; RUN: wasm-dis %t1.wasm | filecheck %s --check-prefix=CHECK-A
+;; RUN: wasm-dis %t2.wasm | filecheck %s --check-prefix=CHECK-B
+;; RUN: wasm-dis %t3.wasm | filecheck %s --check-prefix=CHECK-C
+
+(module
+ (type $ret-i32 (func (result i32)))
+ ;; PRIMARY: (type $ret-i64 (func (result i64)))
+ (type $ret-i64 (func (result i64)))
+ ;; PRIMARY: (type $ret-f32 (func (result f32)))
+ (type $ret-f32 (func (result f32)))
+ ;; CHECK-A: (type $0 (func (result i64)))
+
+ ;; CHECK-A: (type $1 (func (result f32)))
+
+ ;; CHECK-A: (type $2 (func (result i32)))
+
+ ;; CHECK-A: (import "" "table" (table $timport$0 1 funcref))
+
+ ;; CHECK-A: (import "" "a" (func $B (result i64)))
+
+ ;; CHECK-A: (import "" "b" (func $C (result f32)))
+
+ ;; CHECK-A: (elem $0 (i32.const 0) $A)
+
+ ;; CHECK-A: (func $A (result i32)
+ ;; CHECK-A-NEXT: (drop
+ ;; CHECK-A-NEXT: (call_ref $2
+ ;; CHECK-A-NEXT: (ref.func $A)
+ ;; CHECK-A-NEXT: )
+ ;; CHECK-A-NEXT: )
+ ;; CHECK-A-NEXT: (drop
+ ;; CHECK-A-NEXT: (call_ref $0
+ ;; CHECK-A-NEXT: (ref.func $B)
+ ;; CHECK-A-NEXT: )
+ ;; CHECK-A-NEXT: )
+ ;; CHECK-A-NEXT: (drop
+ ;; CHECK-A-NEXT: (call_ref $1
+ ;; CHECK-A-NEXT: (ref.func $C)
+ ;; CHECK-A-NEXT: )
+ ;; CHECK-A-NEXT: )
+ ;; CHECK-A-NEXT: (i32.const 0)
+ ;; CHECK-A-NEXT: )
+ (func $A (type $ret-i32) (result i32)
+ (drop
+ (call_ref $ret-i32
+ (ref.func $A)
+ )
+ )
+ (drop
+ (call_ref $ret-i64
+ (ref.func $B)
+ )
+ )
+ (drop
+ (call_ref $ret-f32
+ (ref.func $C)
+ )
+ )
+ (i32.const 0)
+ )
+ ;; CHECK-B: (type $0 (func (result i32)))
+
+ ;; CHECK-B: (type $1 (func (result f32)))
+
+ ;; CHECK-B: (type $2 (func (result i64)))
+
+ ;; CHECK-B: (import "" "table_3" (table $timport$0 2 funcref))
+
+ ;; CHECK-B: (import "" "table" (table $timport$1 1 funcref))
+
+ ;; CHECK-B: (import "" "b" (func $C (result f32)))
+
+ ;; CHECK-B: (elem $0 (table $timport$0) (i32.const 0) func $B $1)
+
+ ;; CHECK-B: (func $B (result i64)
+ ;; CHECK-B-NEXT: (drop
+ ;; CHECK-B-NEXT: (call_ref $0
+ ;; CHECK-B-NEXT: (ref.func $1)
+ ;; CHECK-B-NEXT: )
+ ;; CHECK-B-NEXT: )
+ ;; CHECK-B-NEXT: (drop
+ ;; CHECK-B-NEXT: (call_ref $2
+ ;; CHECK-B-NEXT: (ref.func $B)
+ ;; CHECK-B-NEXT: )
+ ;; CHECK-B-NEXT: )
+ ;; CHECK-B-NEXT: (drop
+ ;; CHECK-B-NEXT: (call_ref $1
+ ;; CHECK-B-NEXT: (ref.func $C)
+ ;; CHECK-B-NEXT: )
+ ;; CHECK-B-NEXT: )
+ ;; CHECK-B-NEXT: (i64.const 0)
+ ;; CHECK-B-NEXT: )
+ (func $B (type $ret-i64) (result i64)
+ (drop
+ (call_ref $ret-i32
+ (ref.func $A)
+ )
+ )
+ (drop
+ (call_ref $ret-i64
+ (ref.func $B)
+ )
+ )
+ (drop
+ (call_ref $ret-f32
+ (ref.func $C)
+ )
+ )
+ (i64.const 0)
+ )
+ ;; CHECK-C: (type $0 (func (result i64)))
+
+ ;; CHECK-C: (type $1 (func (result i32)))
+
+ ;; CHECK-C: (type $2 (func (result f32)))
+
+ ;; CHECK-C: (import "" "table_4" (table $timport$0 2 funcref))
+
+ ;; CHECK-C: (import "" "table_3" (table $timport$1 2 funcref))
+
+ ;; CHECK-C: (elem $0 (table $timport$0) (i32.const 0) func $0 $C)
+
+ ;; CHECK-C: (func $0 (result i64)
+ ;; CHECK-C-NEXT: (call_indirect (type $0)
+ ;; CHECK-C-NEXT: (i32.const 0)
+ ;; CHECK-C-NEXT: )
+ ;; CHECK-C-NEXT: )
+
+ ;; CHECK-C: (func $C (result f32)
+ ;; CHECK-C-NEXT: (drop
+ ;; CHECK-C-NEXT: (call_ref $1
+ ;; CHECK-C-NEXT: (ref.func $3)
+ ;; CHECK-C-NEXT: )
+ ;; CHECK-C-NEXT: )
+ ;; CHECK-C-NEXT: (drop
+ ;; CHECK-C-NEXT: (call_ref $0
+ ;; CHECK-C-NEXT: (ref.func $2)
+ ;; CHECK-C-NEXT: )
+ ;; CHECK-C-NEXT: )
+ ;; CHECK-C-NEXT: (drop
+ ;; CHECK-C-NEXT: (call_ref $2
+ ;; CHECK-C-NEXT: (ref.func $C)
+ ;; CHECK-C-NEXT: )
+ ;; CHECK-C-NEXT: )
+ ;; CHECK-C-NEXT: (f32.const 0)
+ ;; CHECK-C-NEXT: )
+ (func $C (type $ret-f32) (result f32)
+ (drop
+ (call_ref $ret-i32
+ (ref.func $A)
+ )
+ )
+ (drop
+ (call_ref $ret-i64
+ (ref.func $B)
+ )
+ )
+ (drop
+ (call_ref $ret-f32
+ (ref.func $C)
+ )
+ )
+ (f32.const 0)
+ )
+)
+;; PRIMARY: (table $0 1 funcref)
+
+;; PRIMARY: (table $1 2 funcref)
+
+;; PRIMARY: (table $2 2 funcref)
+
+;; PRIMARY: (elem $0 (table $0) (i32.const 0) funcref (item (ref.null nofunc)))
+
+;; PRIMARY: (elem $1 (table $1) (i32.const 0) funcref (item (ref.null nofunc)) (item (ref.null nofunc)))
+
+;; PRIMARY: (elem $2 (table $2) (i32.const 0) funcref (item (ref.null nofunc)) (item (ref.null nofunc)))
+
+;; PRIMARY: (export "a" (func $0))
+
+;; PRIMARY: (export "b" (func $1))
+
+;; PRIMARY: (export "table" (table $0))
+
+;; PRIMARY: (export "table_3" (table $1))
+
+;; PRIMARY: (export "table_4" (table $2))
+
+;; PRIMARY: (func $0 (result i64)
+;; PRIMARY-NEXT: (call_indirect (type $ret-i64)
+;; PRIMARY-NEXT: (i32.const 0)
+;; PRIMARY-NEXT: )
+;; PRIMARY-NEXT: )
+
+;; PRIMARY: (func $1 (result f32)
+;; PRIMARY-NEXT: (call_indirect (type $ret-f32)
+;; PRIMARY-NEXT: (i32.const 1)
+;; PRIMARY-NEXT: )
+;; PRIMARY-NEXT: )
+
+;; CHECK-B: (func $1 (result i32)
+;; CHECK-B-NEXT: (call_indirect (type $0)
+;; CHECK-B-NEXT: (i32.const 0)
+;; CHECK-B-NEXT: )
+;; CHECK-B-NEXT: )
+
+;; CHECK-C: (func $2 (result i64)
+;; CHECK-C-NEXT: (call_indirect (type $0)
+;; CHECK-C-NEXT: (i32.const 0)
+;; CHECK-C-NEXT: )
+;; CHECK-C-NEXT: )
+
+;; CHECK-C: (func $3 (result i32)
+;; CHECK-C-NEXT: (call_indirect (type $1)
+;; CHECK-C-NEXT: (i32.const 1)
+;; CHECK-C-NEXT: )
+;; CHECK-C-NEXT: )
diff --git a/test/lit/wasm-split/multi-split.wast.manifest b/test/lit/wasm-split/multi-split.wast.manifest
new file mode 100644
index 000000000..f6e710fed
--- /dev/null
+++ b/test/lit/wasm-split/multi-split.wast.manifest
@@ -0,0 +1,8 @@
+1
+A
+
+2
+B
+
+3
+C
generated by cgit v1.2.3 (git 2.39.1) at 2025-08-08 10:57:40 +0000