[llvm] [AMDGPU] Prevent SCC clobber in frame index lowering in scaling path (PR #223789)

[Pankaj Dwivedi via llvm-commits]

================
@@ -3411,8 +3416,7 @@ bool SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
       // Convert to a swizzled stack address by scaling by the wave size.
       // In an entry function/kernel the offset is already swizzled.
       bool IsSALU = isSGPRClass(TII->getRegClass(MI->getDesc(), FIOperandNum));
-      bool LiveSCC = RS->isRegUsed(AMDGPU::SCC) &&
-                     !MI->definesRegister(AMDGPU::SCC, /*TRI=*/nullptr);
+      bool LiveSCC = isSCCLiveInto(*RS, *MI);
----------------
PankajDwivedi-25 wrote:

Looks orthogonal to this patch, The LiveSCC path can still lose SCC when the scavenge spills?

Ex:

```
# llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -run-pass=prolog-epilog
---
name:            scc_read_by_fi_user_no_free_sgpr
tracksRegLiveness: true
stack:
  - { id: 0, size: 64, alignment: 4 }
machineFunctionInfo:
  isEntryFunction: false
  scratchRSrcReg:  '$sgpr0_sgpr1_sgpr2_sgpr3'
  frameOffsetReg:  '$sgpr33'
  stackPtrOffsetReg: '$sgpr32'
  sgprForEXECCopy: '$sgpr100_sgpr101'
  occupancy:       2
body:             |
  bb.0:
    S_BRANCH %bb.1

  bb.1:
    liveins: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, $sgpr96_sgpr97_sgpr98_sgpr99, $vcc

    S_CMP_EQ_U32 renamable $sgpr96, 0, implicit-def $scc
    renamable $vcc_lo = S_CSELECT_B32 %stack.0, renamable $sgpr97, implicit $scc
    S_NOP 0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, implicit killed $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, implicit killed $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, implicit killed $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, implicit killed $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, implicit killed $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, implicit killed $sgpr96_sgpr97_sgpr98_sgpr99
    S_BRANCH %bb.2

  bb.2:
    liveins: $vcc

    S_ENDPGM 0, implicit $vcc
...
```

https://github.com/llvm/llvm-project/pull/223789