Optimize x * -1.0 in non-fastMath case (#3315)

We can still make x * -1.0 cheaper for non-fastMath mode as:

x * -1.0  ->  -0.0 - x

Should at least help baseline compilers.

Also it could enable further optimizations, e.g.:

a + b * -1
a + (-0.0 - b)
(a - 0.0) - b
a - b

2 files changed, 22 insertions, 13 deletions

diff --git a/src/passes/OptimizeInstructions.cpp b/src/passes/OptimizeInstructions.cpp
index 003f204a9..9b47f6a0f 100644
--- a/src/passes/OptimizeInstructions.cpp
+++ b/src/passes/OptimizeInstructions.cpp
@@ -1697,9 +1697,18 @@ private:
       curr->left = left;
       return curr;
     }
-    // x * -1.0   ==>   -x
-    if (fastMath && matches(curr, binary(Mul, any(), fval(-1.0)))) {
-      return builder.makeUnary(Abstract::getUnary(type, Neg), left);
+    // x * -1.0   ==>
+    //       -x,  if fastMath == true
+    // -0.0 - x,  if fastMath == false
+    if (matches(curr, binary(Mul, any(), fval(-1.0)))) {
+      if (fastMath) {
+        return builder.makeUnary(Abstract::getUnary(type, Neg), left);
+      }
+      // x * -1.0   ==>  -0.0 - x
+      curr->op = Abstract::getBinary(type, Sub);
+      right->value = Literal::makeZero(type).neg();
+      std::swap(curr->left, curr->right);
+      return curr;
     }
     if (matches(curr, binary(Mul, any(&left), constant(1))) ||
         matches(curr, binary(DivS, any(&left), constant(1))) ||
diff --git a/test/passes/optimize-instructions_all-features.txt b/test/passes/optimize-instructions_all-features.txt
index 1873b8d3a..b80330ad6 100644
--- a/test/passes/optimize-instructions_all-features.txt
+++ b/test/passes/optimize-instructions_all-features.txt
@@ -3354,15 +3354,15 @@
    )
   )
   (drop
-   (f32.mul
+   (f32.sub
+    (f32.const -0)
     (local.get $y32)
-    (f32.const -1)
    )
   )
   (drop
-   (f64.mul
+   (f64.sub
+    (f64.const -0)
     (local.get $y64)
-    (f64.const -1)
    )
   )
   (drop
@@ -4276,15 +4276,15 @@
    )
   )
   (drop
-   (f32.mul
+   (f32.sub
+    (f32.const -0)
     (local.get $fx)
-    (f32.const -1)
    )
   )
   (drop
-   (f64.mul
+   (f64.sub
+    (f64.const -0)
     (local.get $fy)
-    (f64.const -1)
    )
   )
   (drop
@@ -4362,9 +4362,9 @@
    )
   )
   (drop
-   (f32.mul
+   (f32.sub
+    (f32.const -0)
     (local.get $fx)
-    (f32.const -1)
    )
   )
   (drop