Fix stack-passed arguments on Darwin (AArch64)

[flylai]

Summary

Fixes #4892.

On AArch64, Darwin uses a slightly different ABI from Linux for stack-passed arguments. In particular, Darwin sizes stack slots according to the actual argument size instead of always allocating 8-byte slots.

According to LLVM:
https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/AArch64/AArch64CallingConvention.td#L365-L368

// Darwin uses a calling convention which differs in only two ways
// from the standard one at this level:
//     + i128s (i.e. split i64s) don't need even registers.
//     + Stack slots are sized as needed rather than being at least 64-bit.

Linux ABI

When the number of function arguments exceeds the available argument registers, the remaining arguments are passed on the stack. Each stack-passed argument occupies an 8-byte stack slot, even for 32-bit values:

+-----------------------------------+ <--- sp
|   s0 (i32) [occupies 4 bytes]     | \
|   Padding  [4 bytes]              |  }-- 8-byte stack slot
+-----------------------------------+ <--- sp + 8
|   s1 (i32) [occupies 4 bytes]     | \
|   Padding  [4 bytes]              |  }-- 8-byte stack slot
+-----------------------------------+ <--- sp + 16
|   s2 (i64) [occupies 8 bytes]     | \
|                                   |  }-- 8-byte stack slot
+-----------------------------------+ <--- sp + 24

Darwin ABI

Darwin sizes stack slots according to the actual argument size, allowing adjacent 32-bit arguments to be packed without unnecessary padding:

+-----------------+-----------------+ <--- sp
|  s0 (i32) [4B]  |  s1 (i32) [4B]  |
+-----------------+-----------------+ <--- sp + 8
|  s2 (i64) [8B]                    |
|                                   |
+-----------------------------------+ <--- sp + 16

Test Results

before fix:

============> run test-invoke-native
=========================================
Starting Native Invoke Tests...
=========================================
Running run_test_i32_overflow...
  FAIL: line 374: expected 108, got 0
  FAIL: line 374: expected 109, got 108
  FAIL: line 374: expected 110, got 0
RESULT: FAIL

Running run_test_i64_overflow...
RESULT: PASS

Running run_test_i32_i32_i64_on_stack...
  FAIL: line 406: expected 308, got 0
  FAIL: line 408: expected 1234605616436508552, got 308
RESULT: FAIL

Running run_test_f32_overflow...
  FAIL: line 423: expected 409.000000, got 0.000000
  FAIL: line 423: expected 410.000000, got 409.000000
  FAIL: line 423: expected 411.000000, got 0.000000
RESULT: FAIL

Running run_test_f64_overflow...
RESULT: PASS

Running run_test_f32_f32_f64_on_stack...
  FAIL: line 457: expected 609.000000, got 0.000000
  FAIL: line 458: expected 999.888000, got 0.000000
RESULT: FAIL

Running run_test_mixed_overflow...
  FAIL: line 484: expected 808.000000, got 0.000000
  FAIL: line 485: expected 6148933456521300104, got 1145700352
  FAIL: line 486: expected 12345.678900, got 11982634940256485731267794306136873969953117937862608357662199969694965589666549254097237871668169277440.000000
RESULT: FAIL

Running run_test_i32_i64_alignment...
RESULT: PASS

Running run_test_unaligned_total_stack_size...
  FAIL: line 510: expected 71, got 0
  FAIL: line 511: expected 72, got 71
RESULT: FAIL

Running run_test_float_double_alignment...
RESULT: PASS

Running run_test_return_values...
RESULT: PASS

=========================================
Tests Summary: Passed: 5, Failed: 6
=========================================

after fix:

============> run test-invoke-native
=========================================
Starting Native Invoke Tests...
=========================================
Running run_test_i32_overflow...
RESULT: PASS

Running run_test_i64_overflow...
RESULT: PASS

Running run_test_i32_i32_i64_on_stack...
RESULT: PASS

Running run_test_f32_overflow...
RESULT: PASS

Running run_test_f64_overflow...
RESULT: PASS

Running run_test_f32_f32_f64_on_stack...
RESULT: PASS

Running run_test_mixed_overflow...
RESULT: PASS

Running run_test_i32_i64_alignment...
RESULT: PASS

Running run_test_unaligned_total_stack_size...
RESULT: PASS

Running run_test_float_double_alignment...
RESULT: PASS

Running run_test_return_values...
RESULT: PASS

=========================================
Tests Summary: Passed: 11, Failed: 0
=========================================