[PATCH] D46417: wasm: Add a flag to control merging data segments

[Nick Fitzgerald via Phabricator via llvm-commits]

fitzgen added a comment.

I don't know how to write a test for `lld` -- would appreciate some guidance doing that!

I do have a little test case I have been using:

This is `a.c`:

  char A_HELLO[] = "hello";
  char A_GOODBYE[] = "goodbye";
  char A_WHATEVER[] = "whatever";
  int N_ITERATIONS = 42;

This is `b.c`:

  // b.c
  #define WASM_EXPORT __attribute__((visibility("default")))
  
  extern char* A_HELLO;
  extern char* A_GOODBYE;
  extern char* A_WHATEVER;
  extern int N_ITERATIONS;
  
  extern void print(char*);
  
  WASM_EXPORT
  void printem() {
      for (int i = 0; i < N_ITERATIONS; i++) {
          print(A_HELLO);
          print(A_GOODBYE);
          print(A_WHATEVER);
      }
  }

To build and test:

  ~/llvm/obj/bin/clang --target=wasm32-unknown-unknown-wasm -c a.c
  ~/llvm/obj/bin/clang --target=wasm32-unknown-unknown-wasm -c b.c
  
  ~/llvm/obj/bin/wasm-ld --allow-undefined -o merged-data-segments.wasm a.o b.o
  ~/llvm/obj/bin/wasm-ld --allow-undefined --no-merge-data-segments -o no-merged-data-segments.wasm a.o b.o
  
  wasm-objdump -x merged-data-segments.wasm
  wasm-objdump -x no-merged-data-segments.wasm

This is the dump for `merged-data-segments.wasm`:

  merged-data-segments.wasm:	file format wasm 0x1
  
  Section Details:
  
  Type:
   - type[0] () -> nil
   - type[1] (i32) -> nil
  Import:
   - func[0] sig=0 <_start> <- env._start
   - func[1] sig=1 <print> <- env.print
  Function:
   - func[2] sig=0 <__wasm_call_ctors>
   - func[3] sig=0 <printem>
  Table:
   - table[0] type=anyfunc initial=1 max=1
  Memory:
   - memory[0] pages: initial=2
  Global:
   - global[0] i32 mutable=1 - init i32=66592
   - global[1] i32 mutable=0 - init i32=66592
   - global[2] i32 mutable=0 - init i32=1052
  Export:
   - memory[0] -> "memory"
   - global[1] -> "__heap_base"
   - global[2] -> "__data_end"
   - func[3] <printem> -> "printem"
  Data:
   - segment[0] size=28 - init i32=1024
    - 0000400: 6865 6c6c 6f00 676f 6f64 6279 6500 7768  hello.goodbye.wh
    - 0000410: 6174 6576 6572 0000 2a00 0000            atever..*...
  Custom:
   - name: "name"
   - func[0] _start
   - func[1] print
   - func[2] __wasm_call_ctors
   - func[3] printem

And this is the dump for `no-merged-data-segments.wasm`:

  no-merged-data-segments.wasm:	file format wasm 0x1
  
  Section Details:
  
  Type:
   - type[0] () -> nil
   - type[1] (i32) -> nil
  Import:
   - func[0] sig=0 <_start> <- env._start
   - func[1] sig=1 <print> <- env.print
  Function:
   - func[2] sig=0 <__wasm_call_ctors>
   - func[3] sig=0 <printem>
  Table:
   - table[0] type=anyfunc initial=1 max=1
  Memory:
   - memory[0] pages: initial=2
  Global:
   - global[0] i32 mutable=1 - init i32=66592
   - global[1] i32 mutable=0 - init i32=66592
   - global[2] i32 mutable=0 - init i32=1052
  Export:
   - memory[0] -> "memory"
   - global[1] -> "__heap_base"
   - global[2] -> "__data_end"
   - func[3] <printem> -> "printem"
  Data:
   - segment[0] size=6 - init i32=1024
    - 0000400: 6865 6c6c 6f00                           hello.
   - segment[1] size=8 - init i32=1030
    - 0000406: 676f 6f64 6279 6500                      goodbye.
   - segment[2] size=9 - init i32=1038
    - 000040e: 7768 6174 6576 6572 00                   whatever.
   - segment[3] size=4 - init i32=1048
    - 0000418: 2a00 0000                                *...
  Custom:
   - name: "name"
   - func[0] _start
   - func[1] print
   - func[2] __wasm_call_ctors
   - func[3] printem


Repository:
  rLLD LLVM Linker

https://reviews.llvm.org/D46417