[PATCH] D22199: AMDGPU: Leave WQM earlier before VMEM loads

[Matt Arsenault via llvm-commits]

arsenm added inline comments.

================
Comment at: lib/Target/AMDGPU/SIWholeQuadMode.cpp:355
@@ +354,3 @@
+        // they're loads, to save a bit of memory bandwidth where possible.
+        if (MI.getDesc().TSFlags & SIInstrFlags::VM_CNT)
+          PreferExact.push_back(&MI);
----------------
TII->usesVM_CNT(MI)


https://reviews.llvm.org/D22199