[AMDGPU] Skip stack-arg dbg objects while fixing the dead frame indices

[cdevadas]

Summary

Both SGPR->VGPR and VGPR->AGPR spilling code give a fixup to the
spill frame indices referred in debug instructions so that they
can be entirely removed. We should skip the stack argument debug
objects while looking inside the bitvector with FI as the index
that tracks the spill indices being processed. The stack args will
have negative indices and would crash while accessing the bitvector.

llvm/lib/Target/AMDGPU/SIFrameLowering.cpp

for (MachineBasicBlock &MBB : MF) {
       MBB.sortUniqueLiveIns();
 
       if (!SpillFIs.empty() && SeenDbgInstr) {
         // FIXME: The dead frame indices are replaced with a null register from
         // the debug value instructions. We should instead, update it with the
         // correct register value. But not sure the register value alone is
         for (MachineInstr &MI : MBB) {
           if (MI.isDebugValue() && MI.getOperand(0).isFI() &&
+              !MFI.isFixedObjectIndex(MI.getOperand(0).getIndex()) &&
               SpillFIs[MI.getOperand(0).getIndex()]) {
             MI.getOperand(0).ChangeToRegister(Register(), false /*isDef*/);
           }
         }
       }
     }
   }
 

llvm/lib/Target/AMDGPU/SILowerSGPRSpills.cpp

for (MachineBasicBlock &MBB : MF) {
       MBB.sortUniqueLiveIns();
 
       // FIXME: The dead frame indices are replaced with a null register from
       // the debug value instructions. We should instead, update it with the
       // correct register value. But not sure the register value alone is
       // adequate to lower the DIExpression. It should be worked out later.
       for (MachineInstr &MI : MBB) {
         if (MI.isDebugValue() && MI.getOperand(0).isFI() &&
+            !MFI.isFixedObjectIndex(MI.getOperand(0).getIndex()) &&
             SpillFIs[MI.getOperand(0).getIndex()]) {
           MI.getOperand(0).ChangeToRegister(Register(), false /*isDef*/);
         }
       }
     }
 
     // All those frame indices which are dead by now should be removed from the
     // function frame. Otherwise, there is a side effect such as re-mapping of

llvm/test/CodeGen/AMDGPU/sgpr-spill-fi-skip-processing-stack-arg-dbg-value.mir

+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -amdgpu-spill-sgpr-to-vgpr=true -verify-machineinstrs -run-pass=si-lower-sgpr-spills,prologepilog -o - %s | FileCheck %s
+
+# After handling the SGPR spill to VGPR in SILowerSGPRSpills pass, we replace the dead frame index in the DBG_VALUE instruction with reg 0.
+# Skip looking for frame indices in the debug value instruction for incoming arguments passed via stack. The test would crash otherwise.
+# It is safe to skip the fixed stack objects as they will never become the spill objects.
+
+--- |
+  define amdgpu_kernel void @test() { ret void }
+
+  !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !4, producer: "llvm", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !4, retainedTypes: !4)
+  !1 = !DILocalVariable(name: "a", scope: !2, file: !4, line: 126, type: !6)
+  !2 = distinct !DISubprogram(name: "test", scope: !4, file: !4, line: 1, type: !3, isLocal: false, isDefinition: true, scopeLine: 2, flags: DIFlagPrototyped, isOptimized: true, unit: !0, retainedNodes: !5)
+  !3 = !DISubroutineType(types: !4)
+  !4 = !{null}
+  !5 = !{!1}
+  !6 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !7, size: 64, align: 32)
+  !7 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
+  !8 = !DIExpression()
+  !9 = !DILocation(line: 10, column: 9, scope: !2)
+
+...
+---
+name:            test
+tracksRegLiveness: true
+frameInfo:
+  maxAlignment:    4
+fixedStack:
+  - { id: 0, type: default, offset: 4, size: 4, alignment: 4, stack-id: default }
+stack:
+  - { id: 0, type: spill-slot, size: 4, alignment: 4, stack-id: sgpr-spill }
+machineFunctionInfo:
+  maxKernArgAlign: 4
+  isEntryFunction: true
+  waveLimiter:     true
+  scratchRSrcReg:  '$sgpr96_sgpr97_sgpr98_sgpr99'
+  stackPtrOffsetReg: '$sgpr32'
+  frameOffsetReg: '$sgpr33'
+  hasSpilledSGPRs: true
+  argumentInfo:
+    privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }
+    dispatchPtr:     { reg: '$sgpr4_sgpr5' }
+    kernargSegmentPtr: { reg: '$sgpr6_sgpr7' }
+    workGroupIDX:    { reg: '$sgpr8' }
+    privateSegmentWaveByteOffset: { reg: '$sgpr9' }
+body:             |
+  ; CHECK-LABEL: name: test
+  ; CHECK: bb.0:
+  ; CHECK:   DBG_VALUE $noreg, 0
+  bb.0:
+    renamable $sgpr10 = IMPLICIT_DEF
+    SI_SPILL_S32_SAVE killed $sgpr10, %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
+    DBG_VALUE %fixed-stack.0, 0, !1, !8, debug-location !9
+
+  bb.1:
+    renamable $sgpr10 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
+    S_ENDPGM 0

llvm/test/CodeGen/AMDGPU/vgpr-spill-fi-skip-processing-stack-arg-dbg-value.mir

+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -amdgpu-spill-vgpr-to-agpr=true -verify-machineinstrs -run-pass=prologepilog -o - %s | FileCheck %s
+
+# After handling the VGPR spill to AGPR copy, we replace the dead frame index in the DBG_VALUE instruction with reg 0.
+# Skip looking for frame indices in the debug value instruction for incoming arguments passed via stack. The test would crash otherwise.
+# It is safe to skip the fixed stack objects as they will never become the spill objects.
+
+--- |
+  define amdgpu_kernel void @test() { ret void }
+
+  !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !4, producer: "llvm", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !4, retainedTypes: !4)
+  !1 = !DILocalVariable(name: "a", scope: !2, file: !4, line: 126, type: !6)
+  !2 = distinct !DISubprogram(name: "test", scope: !4, file: !4, line: 1, type: !3, isLocal: false, isDefinition: true, scopeLine: 2, flags: DIFlagPrototyped, isOptimized: true, unit: !0, retainedNodes: !5)
+  !3 = !DISubroutineType(types: !4)
+  !4 = !{null}
+  !5 = !{!1}
+  !6 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !7, size: 64, align: 32)
+  !7 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
+  !8 = !DIExpression()
+  !9 = !DILocation(line: 10, column: 9, scope: !2)
+
+...
+---
+name:            test
+tracksRegLiveness: true
+frameInfo:
+  maxAlignment:    4
+fixedStack:
+  - { id: 0, type: default, offset: 4, size: 4, alignment: 4, stack-id: default }
+stack:
+  - { id: 0, type: spill-slot, size: 4, alignment: 4 }
+machineFunctionInfo:
+  maxKernArgAlign: 4
+  isEntryFunction: true
+  waveLimiter:     true
+  scratchRSrcReg:  '$sgpr96_sgpr97_sgpr98_sgpr99'
+  stackPtrOffsetReg: '$sgpr32'
+  frameOffsetReg: '$sgpr33'
+  hasSpilledVGPRs: true
+  argumentInfo:
+    privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }
+    dispatchPtr:     { reg: '$sgpr4_sgpr5' }
+    kernargSegmentPtr: { reg: '$sgpr6_sgpr7' }
+    workGroupIDX:    { reg: '$sgpr8' }
+    privateSegmentWaveByteOffset: { reg: '$sgpr9' }
+body:             |
+  ; CHECK-LABEL: name: test
+  ; CHECK: bb.0:
+  ; CHECK:   DBG_VALUE $noreg, 0
+  bb.0:
+    $vgpr2 = IMPLICIT_DEF
+    SI_SPILL_V32_SAVE $vgpr2, %stack.0, $sgpr32, 0, implicit $exec :: (store (s32) into %stack.0, align 4, addrspace 5)
+    DBG_VALUE %fixed-stack.0, 0, !1, !8, debug-location !9
+
+  bb.1:
+    renamable $vgpr2 = SI_SPILL_V32_RESTORE %stack.0, $sgpr32, 0, implicit $exec :: (load (s32) from %stack.0, align 4, addrspace 5)
+    S_ENDPGM 0