[RyuJIT/armel] HFA tests fail

[wateret]

With dotnet/coreclr#13284, most regressions caused by dotnet/coreclr#13023 were fixed. But still there is HFA test regression.

All of these run successfully, but the fails with Unexpected Results. (wrong result value)

• JIT/jit64/hfa/main/testB/hfa_sf2B_d/hfa_sf2B_d.exe
• JIT/jit64/hfa/main/testB/hfa_sf0B_r/hfa_sf0B_r.exe
• JIT/jit64/hfa/main/testB/hfa_sf2B_r/hfa_sf2B_r.exe
• JIT/jit64/hfa/main/testG/hfa_sd0G_r/hfa_sd0G_r.exe
• JIT/jit64/hfa/main/testG/hfa_sd1G_r/hfa_sd1G_r.exe
• JIT/jit64/hfa/main/testG/hfa_sd2G_r/hfa_sd2G_r.exe
• JIT/jit64/hfa/main/testG/hfa_sf2G_r/hfa_sf2G_r.exe
• JIT/jit64/hfa/main/testA/hfa_sd0A_r/hfa_sd0A_r.exe
• JIT/jit64/hfa/main/testA/hfa_sd1A_r/hfa_sd1A_r.exe
• JIT/jit64/hfa/main/testA/hfa_sd2A_d/hfa_sd2A_d.exe
• JIT/jit64/hfa/main/testA/hfa_sd2A_r/hfa_sd2A_r.exe
• JIT/jit64/hfa/main/testA/hfa_sf0A_r/hfa_sf0A_r.exe
• JIT/jit64/hfa/main/testA/hfa_sf1A_r/hfa_sf1A_r.exe
• JIT/jit64/hfa/main/testA/hfa_sf2A_d/hfa_sf2A_d.exe
• JIT/jit64/hfa/main/testA/hfa_sf2A_r/hfa_sf2A_r.exe
• JIT/jit64/hfa/main/testB/hfa_sd0B_d/hfa_sd0B_d.exe
• JIT/jit64/hfa/main/testB/hfa_sd0B_r/hfa_sd0B_r.exe
• JIT/jit64/hfa/main/testB/hfa_sd2B_d/hfa_sd2B_d.exe
• JIT/jit64/hfa/main/testB/hfa_sd2B_r/hfa_sd2B_r.exe
• JIT/jit64/hfa/main/testB/hfa_sf0B_d/hfa_sf0B_d.exe

[wateret]

Result of JIT/jit64/hfa/main/testA/hfa_sd0A_r/hfa_sd0A_r.exe

... more passed tests ...
[Sum11 HFA 01]  PASSED
[Sum11 HFA 02]  FAILED => expected = 3, actual = 80
[Sum11 HFA 03]  FAILED => expected = 6, actual = 83
[Sum11 HFA 05]  PASSED
[Sum11 HFA 08]  PASSED
[Sum11 HFA 11]  PASSED
[Sum11 HFA 19]  PASSED
[Sum19 HFA 01]  PASSED
[Sum19 HFA 02]  PASSED
[Sum19 HFA 03]  PASSED
[Sum19 HFA 05]  PASSED
[Sum19 HFA 08]  PASSED
[Sum19 HFA 11]  PASSED
[Sum19 HFA 19]  PASSED

[wateret]

This is the first part of JitDump gets different. (TC: JIT/jit64/hfa/main/testA/hfa_sd0A_r/hfa_sd0A_r.exe)

Before dotnet/coreclr#13023 merged

  N117.                    PUTARG_STK [+0x00]
  consume=2 produce=0
at start of tree, map contains: { N115. COPY -> (116.N115, 116.N115) }
<RefPosition dotnet/runtime#3959 @117 RefTypeUse <Ivl:312> BB01 regmask=[r0-r9 r12 lr] last>
<RefPosition dotnet/runtime#3960 @117 RefTypeUse <Ivl:313> BB01 regmask=[r0-r9 r12 lr] last>

  N119. t3118           =  CNS_INT(h) 0xD1FFAB1E [ICON_STR_HDL] REG NA
  consume=0 produce=1

After dotnet/coreclr#13023 merged

  N117.                    PUTARG_STK [+0x00]
  consume=2 produce=0
at start of tree, map contains: { N115. COPY -> (116.N115, 116.N115) }
<RefPosition dotnet/runtime#3959 @117 RefTypeFixedReg <Reg:r0 > BB01 regmask=[r0]>
<RefPosition dotnet/runtime#3960 @117 RefTypeUse <Ivl:312> BB01 regmask=[r0] last fixed>
<RefPosition dotnet/runtime#3961 @117 RefTypeFixedReg <Reg:r1 > BB01 regmask=[r1]>
<RefPosition dotnet/coreclr#211 @117 RefTypeUse <Ivl:313> BB01 regmask=[r1] last fixed>

  N119. t3118           =  CNS_INT(h) 0xD1FFAB1E [ICON_STR_HDL] REG NA
  consume=0 produce=1

This is the part that putting argument for a call to public static bool CheckResult(string testName, double actual, double expected), and the argument must be third one since it is the only one that goes to the stack.
It seems like the registers shouldn't be fixed since it is PutArgStk(not PutArgReg).

[wateret]

Closing since fixed by dotnet/coreclr#13332