AMDGPU: Partially respect nobuiltin in libcall simplifier

There are more contexts where it's not handled correctly but this is
the simplest one.

https://reviews.llvm.org/D156682

Author: arsenm

llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp

@@ -500,7 +500,7 @@ bool AMDGPULibCalls::sincosUseNative(CallInst *aCI, const FuncInfo &FInfo) {
 bool AMDGPULibCalls::useNative(CallInst *aCI) {
   CI = aCI;
   Function *Callee = aCI->getCalledFunction();
-  if (!Callee)
+  if (!Callee || aCI->isNoBuiltin())
     return false;
 
   FuncInfo FInfo;
@@ -593,7 +593,7 @@ bool AMDGPULibCalls::fold(CallInst *CI, AliasAnalysis *AA) {
   this->CI = CI;
   Function *Callee = CI->getCalledFunction();
   // Ignore indirect calls.
-  if (!Callee)
+  if (!Callee || CI->isNoBuiltin())
     return false;
 
   IRBuilder<> B(CI);

llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll

@@ -630,6 +630,18 @@ entry:
   ret void
 }
 
+; GCN-LABEL: {{^}}define amdgpu_kernel void @test_use_native_powr_nobuiltin
+; GCN: %call = tail call fast float @_Z4powrff(float %tmp, float %tmp1)
+define amdgpu_kernel void @test_use_native_powr_nobuiltin(ptr addrspace(1) nocapture %a) {
+entry:
+  %tmp = load float, ptr addrspace(1) %a, align 4
+  %arrayidx1 = getelementptr inbounds float, ptr addrspace(1) %a, i64 1
+  %tmp1 = load float, ptr addrspace(1) %arrayidx1, align 4
+  %call = call fast float @_Z4powrff(float %tmp, float %tmp1) nobuiltin
+  store float %call, ptr addrspace(1) %a, align 4
+  ret void
+}
+
 ; GCN-LABEL: {{^}}define amdgpu_kernel void @test_use_native_sqrt
 ; GCN-NATIVE: call fast float @_Z11native_sqrtf(float %tmp)
 define amdgpu_kernel void @test_use_native_sqrt(ptr addrspace(1) nocapture %a) {