[AMDGPU] Assertion `BI.FirstInstr >= Start' failed.

Reproducer: https://godbolt.org/z/aosqsnz4K
Testcase:
```llvm
target datalayout = "e-m:e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128:128:48-p9:192:256:256:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9"
target triple = "amdgcn-amd-amdhsa"

%struct.S1 = type { i8, %struct.S0 }
%struct.S0 = type { i64 }

define i16 @backsmith_pure_2(ptr addrspace(5) %BS_VAR_0, i1 %tobool.not, i64 %add919, <32 x i8> %shuffle57, <32 x i8> %vecinit827, <16 x i16> %0) #0 {
entry:
%BS_VAR_01 = alloca [8 x <32 x i8>], align 32, addrspace(5)
%call2 = call i8 @backsmith_snippet_44()
br i1 %tobool.not, label %if.end, label %if.then, !prof !0

if.then: ; preds = %entry
br i1 %tobool.not, label %for.cond.cleanup, label %for.body

for.cond.cleanup: ; preds = %for.body, %if.then
%call976 = call <4 x i64> @backsmith_pure_0(<16 x i16> zeroinitializer)
br label %if.end

for.body: ; preds = %for.body, %if.then
store <32 x i8> %shuffle57, ptr addrspace(5) %BS_VAR_0, align 32
%call394 = call <4 x i16> @backsmith_snippet_679()
%call393 = call <4 x i16> @backsmith_snippet_679()
%cmp104 = icmp ne <32 x i8> %vecinit827, zeroinitializer
%sext = sext <32 x i1> %cmp104 to <32 x i8>
%vecext347 = extractelement <32 x i8> %sext, i64 29
%cond893.v = select i1 %tobool.not, i8 0, i8 %vecext347
%vecinit895 = insertelement <32 x i8> %vecinit827, i8 %cond893.v, i64 29
store <32 x i8> %vecinit895, ptr addrspace(5) null, align 32
%exitcond.not = icmp eq i64 0, %add919
br i1 %exitcond.not, label %for.cond.cleanup, label %for.body

if.end: ; preds = %for.cond.cleanup, %entry
%call1066 = call <4 x i64> @backsmith_pure_0(<16 x i16> %0)
ret i16 0
}

declare i8 @backsmith_snippet_44()

declare <4 x i16> @backsmith_snippet_679()

declare <4 x i64> @backsmith_pure_0(<16 x i16>)

define %struct.S1 @func_103(i8 %0, i1 %tobool2.not30) #0 {
entry:
%add = shufflevector <4 x i16> splat (i16 1), <4 x i16> zeroinitializer, <4 x i32> zeroinitializer
br label %for.cond1.preheader

for.cond.loopexit: ; preds = %for.end, %for.cond1.preheader
%tobool.not = icmp eq i8 %0, 0
br i1 %tobool.not, label %for.end25, label %for.cond1.preheader

for.cond1.preheader: ; preds = %for.cond.loopexit, %entry
br i1 %tobool2.not30, label %for.cond.loopexit, label %for.body3

for.body3: ; preds = %for.end, %for.cond1.preheader
br label %for.cond4

for.cond4: ; preds = %for.cond4, %for.body3
%tobool7.not = icmp eq i64 0, 0
br i1 %tobool7.not, label %for.end, label %for.cond4

for.end: ; preds = %for.cond4
%call121 = call i16 @backsmith_pure_2(ptr addrspace(5) null, i1 false, i64 0, <32 x i8> zeroinitializer, <32 x i8> zeroinitializer, <16 x i16> zeroinitializer)
store <4 x i16> %add, ptr addrspace(5) null, align 8
tail call void @backsmith_pure_7(<16 x i64> zeroinitializer, i32 -352674805, <2 x i32> splat (i32 -247168742))
tail call void @backsmith_pure_7(<16 x i64> zeroinitializer, i32 -352674805, <2 x i32> splat (i32 -247168742))
%tobool2.not = icmp eq ptr addrspace(5) null, null
br i1 %tobool2.not, label %for.cond.loopexit, label %for.body3

for.end25: ; preds = %for.cond.loopexit
ret %struct.S1 zeroinitializer
}

declare void @backsmith_pure_7(<16 x i64>, i32, <2 x i32>)

; uselistorder directives
uselistorder ptr @backsmith_snippet_679, { 1, 0 }
uselistorder ptr @backsmith_pure_0, { 1, 0 }
uselistorder ptr @backsmith_pure_7, { 1, 0 }

attributes #0 = { "target-cpu"="gfx1201" }

!0 = !{!"branch_weights", !"expected", i32 2000, i32 1}
```

Dump:
```
llc: /root/llvm-project/llvm/lib/CodeGen/SplitKit.cpp:233: void llvm::SplitAnalysis::calcLiveBlockInfo(): Assertion `BI.FirstInstr >= Start' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace and instructions to reproduce the bug.
Stack dump:
0. Program arguments: /opt/compiler-explorer/clang-assertions-trunk/bin/llc -o /app/output.s -x86-asm-syntax=intel -O1
1. Running pass 'CallGraph Pass Manager' on module ''.
2. Running pass 'Greedy Register Allocator' on function '@func_103'
#0 0x0000000004252608 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x4252608)
#1 0x000000000424f4e4 SignalHandler(int, siginfo_t*, void*) Signals.cpp:0:0
#2 0x000075cf6e242520 (/lib/x86_64-linux-gnu/libc.so.6+0x42520)
#3 0x000075cf6e2969fc pthread_kill (/lib/x86_64-linux-gnu/libc.so.6+0x969fc)
#4 0x000075cf6e242476 gsignal (/lib/x86_64-linux-gnu/libc.so.6+0x42476)
#5 0x000075cf6e2287f3 abort (/lib/x86_64-linux-gnu/libc.so.6+0x287f3)
#6 0x000075cf6e22871b (/lib/x86_64-linux-gnu/libc.so.6+0x2871b)
#7 0x000075cf6e239e96 (/lib/x86_64-linux-gnu/libc.so.6+0x39e96)
#8 0x000000000340f8a0 llvm::SplitAnalysis::analyzeUses() (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x340f8a0)
#9 0x000000000340fab8 llvm::SplitAnalysis::analyzeUses() (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x340fab8)
#10 0x000000000334da8f llvm::RAGreedy::tryAssignCSRFirstTime(llvm::LiveInterval const&, llvm::AllocationOrder&, llvm::MCRegister, unsigned char&, llvm::SmallVectorImpl&) (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x334da8f)
#11 0x00000000033552db llvm::RAGreedy::selectOrSplitImpl(llvm::LiveInterval const&, llvm::SmallVectorImpl&, llvm::SmallSet>&, llvm::SmallVector, 8u>&, unsigned int) (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x33552db)
#12 0x0000000003356017 llvm::RAGreedy::selectOrSplit(llvm::LiveInterval const&, llvm::SmallVectorImpl&) (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x3356017)
#13 0x00000000035bdfd0 llvm::RegAllocBase::allocatePhysRegs() (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x35bdfd0)
#14 0x0000000003351c1b llvm::RAGreedy::run(llvm::MachineFunction&) (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x3351c1b)
#15 0x000000000335286a (anonymous namespace)::RAGreedyLegacy::runOnMachineFunction(llvm::MachineFunction&) RegAllocGreedy.cpp:0:0
#16 0x000000000311d0e9 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (.part.0) MachineFunctionPass.cpp:0:0
#17 0x0000000003773256 llvm::FPPassManager::runOnFunction(llvm::Function&) (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x3773256)
#18 0x0000000002b690c8 (anonymous namespace)::CGPassManager::runOnModule(llvm::Module&) CallGraphSCCPass.cpp:0:0
#19 0x0000000003773e6f llvm::legacy::PassManagerImpl::run(llvm::Module&) (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x3773e6f)
#20 0x00000000008eb34d compileModule(char**, llvm::SmallVectorImpl&, llvm::LLVMContext&, std::__cxx11::basic_string, std::allocator>&) llc.cpp:0:0
#21 0x00000000007c6f8c main (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x7c6f8c)
#22 0x000075cf6e229d90 (/lib/x86_64-linux-gnu/libc.so.6+0x29d90)
#23 0x000075cf6e229e40 __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x29e40)
#24 0x00000000008e03f5 _start (/opt/compiler-explorer/clang-assertions-trunk/bin/llc+0x8e03f5)
Program terminated with signal: SIGSEGV
Compiler returned: 139
```

Reproduce with llc at -O1 using the linked Godbolt testcase and the command shown in the report. Start at llvm/lib/CodeGen/SplitKit.cpp:233 and inspect the Greedy Register Allocator path for @func_103. Done means the AMDGPU testcase no longer triggers the BI.FirstInstr assertion and llc completes successfully.

Written by the indexing model from the issue text.