diff --git a/src/Pipeline/SpirvShader.cpp b/src/Pipeline/SpirvShader.cpp
index df4c8b8..e92c8b1 100644
--- a/src/Pipeline/SpirvShader.cpp
+++ b/src/Pipeline/SpirvShader.cpp
@@ -3404,7 +3404,12 @@
 		}
 		case GLSLstd450Pow:
 		{
-			UNIMPLEMENTED("GLSLstd450Pow");
+			auto x = GenericValue(this, routine, insn.word(5));
+			auto y = GenericValue(this, routine, insn.word(6));
+			for (auto i = 0u; i < type.sizeInComponents; i++)
+			{
+				dst.move(i, Pow(x.Float(i), y.Float(i)));
+			}
 			break;
 		}
 		case GLSLstd450Exp:
diff --git a/src/Reactor/LLVMReactor.cpp b/src/Reactor/LLVMReactor.cpp
index 46abd51..c04b300 100644
--- a/src/Reactor/LLVMReactor.cpp
+++ b/src/Reactor/LLVMReactor.cpp
@@ -566,6 +566,7 @@
 			func_.emplace("acoshf", reinterpret_cast<void*>(acoshf));
 			func_.emplace("atanhf", reinterpret_cast<void*>(atanhf));
 			func_.emplace("atan2f", reinterpret_cast<void*>(atan2f));
+			func_.emplace("powf", reinterpret_cast<void*>(powf));
 
 #ifdef __APPLE__
 			// LLVM uses this function on macOS for tan.
@@ -3174,6 +3175,13 @@
 		return RValue<Float4>(V(out));
 	}
 
+	RValue<Float4> Pow(RValue<Float4> x, RValue<Float4> y)
+	{
+		auto func = llvm::Intrinsic::getDeclaration(::module, llvm::Intrinsic::pow,
+			{ T(Float4::getType()), T(Float4::getType()) } );
+		return RValue<Float4>(V(::builder->CreateCall(func, { V(x.value), V(y.value) })));
+	}
+
 	Type *Float4::getType()
 	{
 		return T(llvm::VectorType::get(T(Float::getType()), 4));
diff --git a/src/Reactor/Reactor.hpp b/src/Reactor/Reactor.hpp
index a55f8c0..26b06fe 100644
--- a/src/Reactor/Reactor.hpp
+++ b/src/Reactor/Reactor.hpp
@@ -2222,6 +2222,10 @@
 	RValue<Float4> Atanh(RValue<Float4> x);
 	RValue<Float4> Atan2(RValue<Float4> x, RValue<Float4> y);
 
+	// Exponential functions
+	// TODO: Currentlhy unimplemented for Subzero.
+	RValue<Float4> Pow(RValue<Float4> x, RValue<Float4> y);
+
 	template<class T>
 	class Pointer : public LValue<Pointer<T>>
 	{