Set new LLVM atomic optimizer strategy option

See: https://reviews.llvm.org/D147408#4394537
The LLVM AMDGPU atomic optimizer pass has a new command line option
-amdgpu-atomic-optimizer-strategy= which currently defaults to DPP, but
the plan is to change the default to Iterative. Set it to DPP explicitly
so that LLPC will not be affected by that change.

lgc/state/LgcContext.cpp

@@ -149,6 +149,12 @@ void LgcContext::initialize() {
   setOptionDefault("simplifycfg-sink-common", "0");
   setOptionDefault("amdgpu-vgpr-index-mode", "1"); // force VGPR indexing on GFX8
   setOptionDefault("amdgpu-atomic-optimizations", "1");
+#if LLVM_MAIN_REVISION && LLVM_MAIN_REVISION < 463788
+  // Old version of the code
+#else
+// New version of the code (also handles unknown version, which we treat as latest)
+  setOptionDefault("amdgpu-atomic-optimizer-strategy", "DPP");
+#endif
 #if LLVM_MAIN_REVISION && LLVM_MAIN_REVISION < 458033
   // Old version of the code
   setOptionDefault("use-gpu-divergence-analysis", "1");