Passing by address results in compiler miscalculation?

When I use a local struct variable of type RCC_ClocksTypeDef (part of the ST peripheral library) and I call the RCC_GetClocksFreq() function passing the local struct as the parameter, for some reason the compiler increments the address by 4 so the data is all misaligned. However, if I use a global variable of the same type, the compiler handles this ok.

The issue only occurs with compiler optimization level of 0 or 1.

I'm using Keil MDK-Arm 4.13a and targeting the STM32F103. Using the simulator for now. I have the latest version of the ST peripheral library. My code snip in 'C' looks like:

.....

RCC_ClocksTypeDef  grcc_clocks;

int main(void)
{
   RCC_ClocksTypeDef  rcc_clocks;
   RCC_GetClocksFreq(&rcc_clocks);
         RCC_GetClocksFreq(&grcc_clocks);
}

So with the code optimization set to Level 0 or 1, where the code does not result in the expected result, here is the asm:

   106:    RCC_ClocksTypeDef  rcc_clocks;
0x0800357C B086      SUB      sp,sp,#0x18
   107:    RCC_GetClocksFreq(&rcc_clocks);
0x0800357E A801      ADD      r0,sp,#0x04
0x08003580 F7FEFCBC  BL.W     RCC_GetClocksFreq (0x08001EFC)
   108:          RCC_GetClocksFreq(&grcc_clocks);
   109:
0x08003584 4842      LDR      r0,[pc,#264]  ; @0x08003690
0x08003586 F7FEFCB9  BL.W     RCC_GetClocksFreq (0x08001EFC)

So now with the code optimized to 2 or 3, THIS code works properly:

   106:    RCC_ClocksTypeDef  rcc_clocks;
0x0800332C B086      SUB      sp,sp,#0x18
   107:    RCC_GetClocksFreq(&rcc_clocks);
0x0800332E 4668      MOV      r0,sp
0x08003330 F7FEFD88  BL.W     RCC_GetClocksFreq (0x08001E44)
   108:          RCC_GetClocksFreq(&grcc_clocks);
   109:
0x08003334 4842      LDR      r0,[pc,#264]  ; @0x08003440
0x08003336 F7FEFD85  BL.W     RCC_GetClocksFreq (0x08001E44)

So the issue seems to be this line:

0x0800357E A801      ADD      r0,sp,#0x04

Is this a compiler error?