[PATCH] D119762: AMDGPU: Set up User SGPRs for queue_ptr only when necessary

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp:55
+                    bool HasApertureRegs, bool SupportsGetDoorBellID ) {
+  unsigned CodeObjectVersion = AMDGPU::getAmdhsaCodeObjectVersion();
   switch (ID) {
----------------
This really ought to be something read from the IR


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp:429-430
+      // implicit kernargs.
+      if (AMDGPU::getAmdhsaCodeObjectVersion() == 5)
+        removeAssumedBits(IMPLICIT_ARG_PTR);
+      else
----------------
This should recognize both the intrinsic and load from the specific offset from the implicitarg ptr, similar to the new hostcall handling. We still should be able to infer no queue ptr with it in memory


================
Comment at: llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h:653
+  bool needsQueuePtrUserSGPRs() const {
+    return QueuePtr && AMDGPU::getAmdhsaCodeObjectVersion() < 5;
+  }
----------------
I don't see why this query needs to change, the code object version can be considered when setting QueuePtr initially


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D119762/new/

https://reviews.llvm.org/D119762