[llvm] [AMDGPU] Fix bad removal of s_delay_alu (PR #145728)

[Jay Foad via llvm-commits]

================
@@ -50,8 +50,12 @@ class AMDGPUInsertDelayAlu {
   static bool instructionWaitsForSGPRWrites(const MachineInstr &MI) {
     // These instruction types wait for VA_SDST==0 before issuing.
     const uint64_t VA_SDST_0 = SIInstrFlags::SALU | SIInstrFlags::SMRD;
-
-    return MI.getDesc().TSFlags & VA_SDST_0;
+    if (MI.getDesc().TSFlags & VA_SDST_0) {
+      for (auto &Op : MI.uses())
----------------
jayfoad wrote:

Also don't scan the operands of SMEM instructions. That's just a waste of time.

https://github.com/llvm/llvm-project/pull/145728