[PATCH] D125060: [amdgpu] Implement lds kernel id intrinsic

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUArgumentUsageInfo.h:106
     FLAT_SCRATCH_INIT   =  5,
+    LDS_KERNEL_ID       =  6,
     WORKGROUP_ID_X      = 10,
----------------
Needs a comment explaining this isn't a real user SGPR defined in the ABI


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.cpp:106
+ConstantInt *AMDGPUMachineFunction::getLDSKernelIdMetadata(const Function &F) {
+  auto MD = F.getMetadata("llvm.amdgcn.lds_kernel_id");
+  if (MD && MD->getNumOperands() == 1) {
----------------
arsenm wrote:
> arsenm wrote:
> > Why '.' and '_'?
> What is this metadata for? Where does it come from?
probably should have gone the other way, all . instead of _


================
Comment at: llvm/lib/Target/AMDGPU/SIFrameLowering.cpp:530-536
+  if (MFI->hasLDSKernelId()) {
+    ConstantInt *Id = AMDGPUMachineFunction::getLDSKernelIdMetadata(F);
+    Register Reg = MFI->getPreloadedReg(AMDGPUFunctionArgInfo::LDS_KERNEL_ID);      
+    if (Id && Reg) {
+      BuildMI(MBB, I, DL, TII->get(AMDGPU::S_MOV_B32), Reg)
+        .addImm(Id->getZExtValue());
+    }
----------------
This shouldn't get initialized in frame lowering. This should have been emitted as an outgoing call argument only, this isn't a reserved register


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D125060/new/

https://reviews.llvm.org/D125060