For emitEpilogue, emitCSRSpillRestores' argument in FPSaved case to handle chain functions in same way as non-FPSaved case, an offset. Remove prior hardcode to s33 and do not restore FP for chain functions.

Author: jofrn

llvm/lib/Target/AMDGPU/SIFrameLowering.cpp

@@ -1408,11 +1408,12 @@ void SIFrameLowering::emitEpilogue(MachineFunction &MF,
       LiveUnits.addReg(FramePtrRegScratchCopy);
     }
 
-    emitCSRSpillRestores(MF, MBB, MBBI, DL, LiveUnits, FramePtrReg,
+    emitCSRSpillRestores(MF, MBB, MBBI, DL, LiveUnits,
+                         FuncInfo->isChainFunction() ? Register() : FramePtrReg,
                          FramePtrRegScratchCopy);
   }
 
-  if (FPSaved) {
+  if (FPSaved && !FuncInfo->isChainFunction()) {
     // Insert the copy to restore FP.
     Register SrcReg = SGPRForFPSaveRestoreCopy ? SGPRForFPSaveRestoreCopy
                                                : FramePtrRegScratchCopy;

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp

@@ -98,7 +98,6 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const Function &F,
     // set one up. For now, we can use s32 to match what amdgpu_gfx functions
     // would use if called, but this can be revisited.
     // FIXME: Only reserve this if we actually need it.
-    FrameOffsetReg = AMDGPU::SGPR33;
     StackPtrOffsetReg = AMDGPU::SGPR32;
 
     ScratchRSrcReg = AMDGPU::SGPR48_SGPR49_SGPR50_SGPR51;

llvm/test/CodeGen/AMDGPU/amdgpu-cs-chain-frame-pointer.ll

@@ -50,10 +50,15 @@ define amdgpu_cs_chain void @indirect(ptr %callee) {
 ; CHECK-NEXT:    s_mov_b64 s[4:5], 0
 ; CHECK-NEXT:    v_readlane_b32 s3, v41, 0
 ; CHECK-NEXT:    s_xor_saveexec_b64 s[6:7], -1
-; CHECK-NEXT:    scratch_load_dword v40, off, s33 ; 4-byte Folded Reload
-; CHECK-NEXT:    scratch_load_dword v41, off, s33 offset:4 ; 4-byte Folded Reload
+; CHECK-NEXT:    scratch_load_dword v40, off, off ; 4-byte Folded Reload
+; CHECK-NEXT:    scratch_load_dword v41, off, off offset:4 ; 4-byte Folded Reload
 ; CHECK-NEXT:    s_mov_b64 exec, s[6:7]
-; CHECK-NEXT:    s_mov_b32 s33, s3
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    v_readlane_b32 s3, v41, 0
+; CHECK-NEXT:    s_xor_saveexec_b64 s[8:9], -1
+; CHECK-NEXT:    scratch_load_dword v40, off, off ; 4-byte Folded Reload
+; CHECK-NEXT:    scratch_load_dword v41, off, off offset:4 ; 4-byte Folded Reload
+; CHECK-NEXT:    s_mov_b64 exec, s[8:9]
 ; CHECK-NEXT:    s_mov_b64 exec, 0
 ; CHECK-NEXT:    s_setpc_b64 s[4:5]
   call void @indirect(ptr null)