Compiler generates wrong code for assigning struct fields

[thefallenidealist]

Zig Version

0.11.0-dev.4000+3022c525e

Steps to Reproduce and Observed Behavior

Hi
Firstly, thanks for the interesting new language (and tutorials & videos)!
I am playing with Zig for the past few months with STM32 MCU board (armv7 thumb2)

Compiler started to generate wrong machine code for this part of my STM32 code:

Systick.CSR = .{
	.ENABLE = 1,
	.TICKINT = 1,
	.CLKSOURCE = 1,
};

I would expect that after it runs value in Systick.CSR register will be 0x7 but it is 0x6. GDB:

p/x *0xe000e010
0x6

watch *0xe000e010
CSR value: 0 -> 1 -> 3 -> 7 -> 6

I would also expect that example above will write 0x7 in the single pass but I'll leave that for the future investigation.

Systick is struct:

const struct_Systick = packed struct {
	CSR:     CSR,
	// ...
	const CSR = packed struct {
		ENABLE:    u1 = 0,
		TICKINT:   u1 = 0,
		CLKSOURCE: u1 = 0,
		_res1:    u13 = 0,
		COUNTFLAG: u1 = 0,
		_res2:    u15 = 0,
	};
};

Noticed on 2023.07.01. with (I think) zig version 0.11.0-dev.3892.
Before it was working but I am not sure with which version.
Tested today with 0.11.0-dev.4000+3022c525e and problem still persist.
Platform is FreeBSD 14.0-CURRENT with LLVM16 from ports.

One of the workaround is to try to write additional bit to the Systick.CSR (which will be ignored):

Systick.CSR = .{
	.ENABLE = 1,
	.TICKINT = 1,
	.CLKSOURCE = 1,
	.COUNTFLAG = 1,	// never written to CSR register, but it will help to write .ENABLE bit to CSR register
};

Now GDB will show values for Systick.CSR: 0 -> 1 -> 3 -> 7

I'll try to create minimal example of this later.
Possible related to bug report #16615

Expected Behavior

After this code runs

Systick.CSR = .{
	.ENABLE = 1,
	.TICKINT = 1,
	.CLKSOURCE = 1,
};

Systick.CSR value should be 7