Compiler Optimizations

Morph applies multiple optimization passes to produce efficient code.

---

Optimization Passes

Pass	Description	Example
Constant Folding	Evaluate constant expressions at compile time	2 + 3 → 5
Dead Code Elimination	Remove unreachable or unused code	Unused variables removed
Vectorization	Auto-vectorize loops with SIMD	Element-wise tensor ops
Operation Fusion	Combine compatible operations	a * b + c → FMA
Inlining	Inline small method calls	Reduces call overhead

---

Real Example

From OptTest.mx:

Compute method() {
    x is 2 + 3;       // constant folded to 5
    y is x * 10;       // constant folded to 50
    unused is 99;      // dead code eliminated
    Print(y);
}

After optimization, this becomes equivalent to:

Compute method() {
    Print(50);
}

---

Optimization Levels

Configured in morph.toml:

[build]
optimize = "aggressive"    # none, basic, aggressive

Level	Passes Applied
none	No optimization
basic	Constant folding, dead code elimination
aggressive	All passes including vectorization, fusion, inlining

---

Tensor-Specific Optimizations

• FMA fusion — (a * b) + c becomes a single fused multiply-add
• Kernel caching — Compiled GPU kernels cached to disk
• Auto-tuning — Runtime selection of optimal kernel parameters

---

Viewing Optimized Output

morph compile file.mx    # Show LLVM IR (optimized)