The order of function's argument recovered is inconsistent with that in source code

[CupCupRay]

Version and Platform (required):

• Binary Ninja Version: Both tested in 4.0.4958 and 3.5.4526
• OS: Both tested in Ubuntu Linux and Windows11
• OS Version: 22.04
• CPU Architecture: x64

Bug Description: When I disassemble a function and try to recover its function prototypes, I find the argument information (mainly its order) is inconsistent with the source code. For example, in the function ConvertXYZtoLMS, the definition of argument arg1, 2, 3 in MLIL and pseudo code should be arg4, 5, 6 in source code. Are there any special settings that force BN to exchange arg1, 2, and 3 with arg4, 5, and 6?

Test project: ImageMagick-7.0.10-27 Source file: ImageMagick-7.0.10-27/MagickCore/colorspace.c

Expected Behavior: The recovered arg1 should be indicated by the instruction 0x08b3aa instead of 0x08b3b9. Same as other arguments.

Screenshots:

[xusheng6]

It seems we are putting the xmm registers ahead of the rdi/rsi/rdx registers. The resulting decompilation is not technically wrong, but it would be better if we can change the order to match that in the source code.