diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -682,11 +682,12 @@
 
   PB.registerPipelineEarlySimplificationEPCallback(
       [this](ModulePassManager &PM, OptimizationLevel Level) {
+        PM.addPass(AMDGPUPrintfRuntimeBindingPass());
+
         if (Level == OptimizationLevel::O0)
           return;
 
         PM.addPass(AMDGPUUnifyMetadataPass());
-        PM.addPass(AMDGPUPrintfRuntimeBindingPass());
 
         if (InternalizeSymbols) {
           PM.addPass(InternalizePass(mustPreserveGV));
diff --git a/llvm/test/CodeGen/AMDGPU/opencl-printf-pipeline.ll b/llvm/test/CodeGen/AMDGPU/opencl-printf-pipeline.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/opencl-printf-pipeline.ll
@@ -0,0 +1,18 @@
+; RUN: opt -mtriple=amdgcn--amdhsa -passes=amdgpu-printf-runtime-binding -S < %s | FileCheck %s
+; RUN: opt -mtriple=amdgcn--amdhsa -O0 -S < %s | FileCheck %s
+; RUN: opt -mtriple=amdgcn--amdhsa -O1 -S < %s | FileCheck %s
+; RUN: opt -mtriple=amdgcn--amdhsa -O2 -S < %s | FileCheck %s
+; RUN: opt -mtriple=amdgcn--amdhsa -O3 -S < %s | FileCheck %s
+
+@.str = private unnamed_addr addrspace(4) constant [7 x i8] c"hello\0A\00", align 1
+
+; Check that the call to printf is removed when AMDGPUPrintfRuntimeBindingPass is executed
+;   And that this pass is executed in the -O0 and -OX pipelines
+; CHECK-LABEL: define void @foo
+; CHECK-NOT: call i32{{.*}}@printf
+define void @foo() {
+  %call = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) noundef @.str)
+  ret void
+}
+
+declare hidden i32 @printf(ptr addrspace(4) noundef, ...)