[PATCH] D94429: AMDGPU: Move handling of allocation of fixed ABI inputs

[Madhur Amilkanthwar via Phabricator via llvm-commits]

madhur13490 added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:1890
+  } else
+    Arg = allocateSGPR32InputImpl(CCInfo, &AMDGPU::SGPR_32RegClass, 32);
 }
----------------
Why do we HAVE to allocate?


================
Comment at: llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp:67
+    if (UseFixedABI && CC != CallingConv::AMDGPU_Gfx)
+      ArgInfo = AMDGPUArgumentUsageInfo::FixedABIFunctionInfo;
+
----------------
The constructor needs to be improved. `ArgInfo` will now hold allocated registers but the code ahead (line 93-99) is setting bools to true which is not useful and logically a dead code. Ideally, we should just do the necessary allocation in fixedABI and return.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D94429/new/

https://reviews.llvm.org/D94429