Cranelift: GVN spectre guards and run redundant load elimination twice (#5517)

* Cranelift: Make spectre guards GVN-able

While these instructions have a side effect that is otherwise invisible to the
optimizer, the side effect in question is idempotent, so it can be de-duplicated
by GVN.

* Cranelift: Run redundant load replacement and GVN twice

This allows us to actually replace redundant Wasm loads with dynamic memories.

While this improves our hand-crafted test sequences, it doesn't seem to have any
improvement on sightglass benchmarks run with dynamic memories, however it also
isn't a hit to compilation times, so seems generally good to land anyways:

```
$ cargo run --release -- benchmark -e ~/scratch/once.so -e ~/scratch/twice.so -m insts-retired --processes 20 --iterations-per-process 3 --engine-flags="--static-memory-maximum-size 0" -- benchmarks/default.suite
compilation :: instructions-retired :: benchmarks/spidermonkey/benchmark.wasm

  No difference in performance.

  [683595240 683768610.53 684097577] once.so
  [683597068 700115966.83 1664907164] twice.so

instantiation :: instructions-retired :: benchmarks/spidermonkey/benchmark.wasm

  No difference in performance.

  [44107 60411.07 92785] once.so
  [44138 59552.32 92097] twice.so

compilation :: instructions-retired :: benchmarks/bz2/benchmark.wasm

  No difference in performance.

  [17369916 17404839.78 17471458] once.so
  [17369935 17625713.87 30700150] twice.so

compilation :: instructions-retired :: benchmarks/pulldown-cmark/benchmark.wasm

  No difference in performance.

  [126523640 126566170.80 126648265] once.so
  [126523076 127174580.30 163145149] twice.so

instantiation :: instructions-retired :: benchmarks/pulldown-cmark/benchmark.wasm

  No difference in performance.

  [34569 35686.25 36513] once.so
  [34651 35749.97 36953] twice.so

instantiation :: instructions-retired :: benchmarks/bz2/benchmark.wasm

  No difference in performance.

  [35146 36639.10 37707] once.so
  [34472 36580.82 38431] twice.so

execution :: instructions-retired :: benchmarks/spidermonkey/benchmark.wasm

  No difference in performance.

  [7055720115 7055841324.82 7056180024] once.so
  [7055717681 7055877095.85 7056225217] twice.so

execution :: instructions-retired :: benchmarks/pulldown-cmark/benchmark.wasm

  No difference in performance.

  [46436881 46437081.28 46437691] once.so
  [46436883 46437127.68 46437766] twice.so

execution :: instructions-retired :: benchmarks/bz2/benchmark.wasm

  No difference in performance.

  [653010530 653010533.27 653010539] once.so
  [653010531 653010532.95 653010538] twice.so
```

cranelift/codegen/meta/src/cdsl/instructions.rs

@@ -73,6 +73,8 @@ pub(crate) struct InstructionContent {
     pub can_trap: bool,
     /// Does this instruction have other side effects besides can_* flags?
     pub other_side_effects: bool,
+    /// Despite having other side effects, is this instruction okay to GVN?
+    pub side_effects_okay_for_gvn: bool,
 }
 
 impl InstructionContent {
@@ -133,6 +135,7 @@ pub(crate) struct InstructionBuilder {
     can_store: bool,
     can_trap: bool,
     other_side_effects: bool,
+    side_effects_okay_for_gvn: bool,
 }
 
 impl InstructionBuilder {
@@ -152,6 +155,7 @@ impl InstructionBuilder {
             can_store: false,
             can_trap: false,
             other_side_effects: false,
+            side_effects_okay_for_gvn: false,
         }
     }
 
@@ -211,6 +215,11 @@ impl InstructionBuilder {
         self
     }
 
+    pub fn side_effects_okay_for_gvn(mut self, val: bool) -> Self {
+        self.side_effects_okay_for_gvn = val;
+        self
+    }
+
     fn build(self) -> Instruction {
         let operands_in = self.operands_in.unwrap_or_else(Vec::new);
         let operands_out = self.operands_out.unwrap_or_else(Vec::new);
@@ -259,6 +268,7 @@ impl InstructionBuilder {
             can_store: self.can_store,
             can_trap: self.can_trap,
             other_side_effects: self.other_side_effects,
+            side_effects_okay_for_gvn: self.side_effects_okay_for_gvn,
         })
     }
 }

cranelift/codegen/meta/src/gen_inst.rs

@@ -512,6 +512,13 @@ fn gen_opcodes(all_inst: &AllInstructions, fmt: &mut Formatter) {
             "Does this instruction have other side effects besides can_* flags?",
             fmt,
         );
+        gen_bool_accessor(
+            all_inst,
+            |inst| inst.side_effects_okay_for_gvn,
+            "side_effects_okay_for_gvn",
+            "Despite `self.other_side_effects() == true`, is this instruction okay to GVN?",
+            fmt,
+        )
     });
     fmt.line("}");
     fmt.empty_line();

cranelift/codegen/meta/src/shared/instructions.rs

@@ -1389,7 +1389,10 @@ pub(crate) fn define(
         )
         .operands_in(vec![c, x, y])
         .operands_out(vec![a])
-        .other_side_effects(true),
+        .other_side_effects(true)
+        // We can de-duplicate spectre selects since the side effect is
+        // idempotent.
+        .side_effects_okay_for_gvn(true),
     );
 
     let c = &Operand::new("c", Any).with_doc("Controlling value to test");

cranelift/codegen/src/context.rs

@@ -194,8 +194,10 @@ impl Context {
         if isa.flags().use_egraphs() {
             self.egraph_pass()?;
         } else if opt_level != OptLevel::None && isa.flags().enable_alias_analysis() {
-            self.replace_redundant_loads()?;
+            for _ in 0..2 {
-            self.simple_gvn(isa)?;
+                self.replace_redundant_loads()?;
+                self.simple_gvn(isa)?;
+            }
         }
 
         Ok(())

cranelift/codegen/src/simple_gvn.rs

@@ -16,8 +16,8 @@ fn trivially_unsafe_for_gvn(opcode: Opcode) -> bool {
         || opcode.is_terminator()
         || opcode.is_return()
         || opcode.can_trap()
-        || opcode.other_side_effects()
         || opcode.can_store()
+        || (opcode.other_side_effects() && !opcode.side_effects_okay_for_gvn())
 }
 
 /// Test that, if the specified instruction is a load, it doesn't have the `readonly` memflag.

cranelift/filetests/filetests/wasm/duplicate-loads-dynamic-memory.wat

@@ -65,15 +65,15 @@
 ;; @0057                               v10 = icmp ugt v4, v6
 ;;                                     v19 -> v10
 ;; @0057                               v11 = select_spectre_guard v10, v9, v8  ; v9 = 0
+;;                                     v20 -> v11
 ;; @0057                               v12 = load.i32 little heap v11
 ;;                                     v2 -> v12
-;; @005c                               v20 = select_spectre_guard v10, v9, v8  ; v9 = 0
+;;                                     v21 -> v12
-;; @005c                               v21 = load.i32 little heap v20
 ;;                                     v3 -> v21
 ;; @005f                               jump block1
 ;;
 ;;                                 block1:
-;; @005f                               return v12, v21
+;; @005f                               return v12, v12
 ;; }
 ;;
 ;; function u0:1(i32, i64 vmctx) -> i32, i32 fast {
@@ -103,13 +103,13 @@
 ;; @0064                               v11 = icmp ugt v4, v6
 ;;                                     v21 -> v11
 ;; @0064                               v12 = select_spectre_guard v11, v10, v9  ; v10 = 0
+;;                                     v22 -> v12
 ;; @0064                               v13 = load.i32 little heap v12
 ;;                                     v2 -> v13
-;; @006a                               v22 = select_spectre_guard v11, v10, v9  ; v10 = 0
+;;                                     v23 -> v13
-;; @006a                               v23 = load.i32 little heap v22
 ;;                                     v3 -> v23
 ;; @006e                               jump block1
 ;;
 ;;                                 block1:
-;; @006e                               return v13, v23
+;; @006e                               return v13, v13
-;; }
+;; }