ARM-Cortex M4F 的 R3 寄存器在 ldr 指令后突然改变值

Question

I have this really strange bug where an ldr instruction first loads in a valid value into the R3 register, then when I use the debugger to step over one instruction, it all of a sudden contains the value 0x0.
This assembly is generated from C++ by GCC9 for C++17.

The instructions are:

ldr r3, [r7, #12]    ; after this instruction r3 will have value 0x20009da0
ldr r3, [r3, #8]     ; after this instruction r3 will have value 0x20003e28
ldr r3, [r3]         ; after this instruction r3 will have value 0x00020a2c, the value at this addr is 0000a1a9
ldr r3, [r3]         ; after this instruction r3 will suddenly have value 0x00000000
ldr r2, [r7, #12]        
ldr r2, [r2, #8]         
mov r0, r2       
blx r3               ; at this point r3 still has value 0x00000000 causing the reset handler to be called

Shouldn't r3 have the value 0x0000a1a9 after the last ldr instruction?

To give some more information, I am developing for Microchip's ATSAMD51J20A using Microchip Studio with GCC9 installed. The code in question where this occurs is:

auto& socket = m_device.get_socket();

where get_socket() has as its return value a reference to an interface class and the actual implementation of get_socket() returns a reference to a class which extends this interface.

Answer 1

I remember Microchip compilers having strange (timing?) issues where lines of repeated instructions behave in unexpected ways. I still don't know about the details but sometimes padding NOPs between these lines resolves the problem.