Fix regression from #4130 (#4158)

That PR reused the same node twice in the output, which fails on the
assertion in BINARYEN_PASS_DEBUG=1 mode.

No new test is needed because the existing test suite fails already
in that mode. That the PR managed to land seems to say that we are
not testing pass-debug mode on our lit tests, which we need to
investigate.

2 files changed, 219 insertions, 3 deletions

diff --git a/src/passes/OptimizeInstructions.cpp b/src/passes/OptimizeInstructions.cpp
index e48eecb72..c9e4d7ff5 100644
--- a/src/passes/OptimizeInstructions.cpp
+++ b/src/passes/OptimizeInstructions.cpp
@@ -2934,6 +2934,10 @@ private:
   }
 
   Expression* optimizeMemoryFill(MemoryFill* memFill) {
+    if (memFill->type == Type::unreachable) {
+      return nullptr;
+    }
+
     if (!memFill->size->is<Const>()) {
       return nullptr;
     }
@@ -3015,19 +3019,21 @@ private:
                                        Type::v128);
             } else {
               // { i64.store(d, C', 0), i64.store(d, C', 8) }
+              auto destType = memFill->dest->type;
+              Index tempLocal = builder.addVar(getFunction(), destType);
               return builder.makeBlock({
                 builder.makeStore(
                   8,
                   offset,
                   align,
-                  memFill->dest,
+                  builder.makeLocalTee(tempLocal, memFill->dest, destType),
                   builder.makeConst<uint64_t>(value * 0x0101010101010101ULL),
                   Type::i64),
                 builder.makeStore(
                   8,
                   offset + 8,
                   align,
-                  memFill->dest,
+                  builder.makeLocalGet(tempLocal, destType),
                   builder.makeConst<uint64_t>(value * 0x0101010101010101ULL),
                   Type::i64),
               });
diff --git a/test/lit/passes/optimize-instructions-bulk-memory.wast b/test/lit/passes/optimize-instructions-bulk-memory.wast
index 5c9efc893..4f30e3e9a 100644
--- a/test/lit/passes/optimize-instructions-bulk-memory.wast
+++ b/test/lit/passes/optimize-instructions-bulk-memory.wast
@@ -1,5 +1,6 @@
 ;; NOTE: Assertions have been generated by update_lit_checks.py and should not be edited.
-;; RUN: wasm-opt %s --all-features --optimize-instructions --enable-bulk-memory -S -o - | filecheck %s
+;; RUN: wasm-opt %s --optimize-instructions --enable-bulk-memory --enable-simd -S -o - | filecheck %s
+;; RUN: wasm-opt %s --optimize-instructions --enable-bulk-memory -S -o - | filecheck %s --check-prefix NOSIMD
 
 (module
   (memory 0)
@@ -77,6 +78,79 @@
   ;; CHECK-NEXT:   )
   ;; CHECK-NEXT:  )
   ;; CHECK-NEXT: )
+  ;; NOSIMD:      (func $optimize-bulk-memory-copy (param $dst i32) (param $src i32) (param $sz i32)
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $sz)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $src)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store8
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.load8_u
+  ;; NOSIMD-NEXT:    (local.get $src)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store16 align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.load16_u align=1
+  ;; NOSIMD-NEXT:    (local.get $src)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $src)
+  ;; NOSIMD-NEXT:   (i32.const 3)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.load align=1
+  ;; NOSIMD-NEXT:    (local.get $src)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $src)
+  ;; NOSIMD-NEXT:   (i32.const 5)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $src)
+  ;; NOSIMD-NEXT:   (i32.const 6)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $src)
+  ;; NOSIMD-NEXT:   (i32.const 7)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i64.store align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i64.load align=1
+  ;; NOSIMD-NEXT:    (local.get $src)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $src)
+  ;; NOSIMD-NEXT:   (i32.const 16)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $src)
+  ;; NOSIMD-NEXT:   (local.get $sz)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.copy
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:   (i32.load
+  ;; NOSIMD-NEXT:    (i32.const 3)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT: )
   (func $optimize-bulk-memory-copy (param $dst i32) (param $src i32) (param $sz i32)
     (memory.copy  ;; skip
       (local.get $dst)
@@ -251,7 +325,137 @@
   ;; CHECK-NEXT:   (i32.const 0)
   ;; CHECK-NEXT:   (i32.const 17)
   ;; CHECK-NEXT:  )
+  ;; CHECK-NEXT:  (memory.fill
+  ;; CHECK-NEXT:   (unreachable)
+  ;; CHECK-NEXT:   (i32.const 1)
+  ;; CHECK-NEXT:   (i32.const 16)
+  ;; CHECK-NEXT:  )
   ;; CHECK-NEXT: )
+  ;; NOSIMD:      (func $optimize-bulk-memory-fill (param $dst i32) (param $val i32) (param $sz i32)
+  ;; NOSIMD-NEXT:  (local $3 i32)
+  ;; NOSIMD-NEXT:  (local $4 i32)
+  ;; NOSIMD-NEXT:  (local $5 i32)
+  ;; NOSIMD-NEXT:  (memory.fill
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store8
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store8
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $val)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store8
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 1)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store8
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 255)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store16 align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store16 align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 257)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store16 align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 65535)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i64.store align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i64.const 0)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (block
+  ;; NOSIMD-NEXT:   (i64.store align=1
+  ;; NOSIMD-NEXT:    (local.tee $3
+  ;; NOSIMD-NEXT:     (local.get $dst)
+  ;; NOSIMD-NEXT:    )
+  ;; NOSIMD-NEXT:    (i64.const 0)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:   (i64.store offset=8 align=1
+  ;; NOSIMD-NEXT:    (local.get $3)
+  ;; NOSIMD-NEXT:    (i64.const 0)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i32.store align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 16843009)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (i64.store align=1
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i64.const 72340172838076673)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (block
+  ;; NOSIMD-NEXT:   (i64.store align=1
+  ;; NOSIMD-NEXT:    (local.tee $4
+  ;; NOSIMD-NEXT:     (local.get $dst)
+  ;; NOSIMD-NEXT:    )
+  ;; NOSIMD-NEXT:    (i64.const 72340172838076673)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:   (i64.store offset=8 align=1
+  ;; NOSIMD-NEXT:    (local.get $4)
+  ;; NOSIMD-NEXT:    (i64.const 72340172838076673)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (block
+  ;; NOSIMD-NEXT:   (i64.store align=1
+  ;; NOSIMD-NEXT:    (local.tee $5
+  ;; NOSIMD-NEXT:     (local.get $dst)
+  ;; NOSIMD-NEXT:    )
+  ;; NOSIMD-NEXT:    (i64.const -1)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:   (i64.store offset=8 align=1
+  ;; NOSIMD-NEXT:    (local.get $5)
+  ;; NOSIMD-NEXT:    (i64.const -1)
+  ;; NOSIMD-NEXT:   )
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.fill
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $val)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.fill
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:   (local.get $sz)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.fill
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (local.get $val)
+  ;; NOSIMD-NEXT:   (local.get $sz)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.fill
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:   (i32.const 3)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.fill
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 1)
+  ;; NOSIMD-NEXT:   (i32.const 3)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.fill
+  ;; NOSIMD-NEXT:   (local.get $dst)
+  ;; NOSIMD-NEXT:   (i32.const 0)
+  ;; NOSIMD-NEXT:   (i32.const 17)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT:  (memory.fill
+  ;; NOSIMD-NEXT:   (unreachable)
+  ;; NOSIMD-NEXT:   (i32.const 1)
+  ;; NOSIMD-NEXT:   (i32.const 16)
+  ;; NOSIMD-NEXT:  )
+  ;; NOSIMD-NEXT: )
   (func $optimize-bulk-memory-fill (param $dst i32) (param $val i32) (param $sz i32)
     (memory.fill
       (local.get $dst)
@@ -378,5 +582,11 @@
       (i32.const 0)
       (i32.const 17)
     )
+
+    (memory.fill  ;; skip
+      (unreachable)
+      (i32.const 1)
+      (i32.const 16)
+    )
   )
 )