Prevent compiler creating potential unaligned 32 bit access.

kvleonhardt — Tue, 28 Feb 2023 10:23:45 GMT

When using highest optimization level following code for RT1170:

for (nWFixed = obj->isActiveIdx[2]; nWFixed <= obj->mConstrMax; nWFixed++) {
obj->isActiveConstr.data[nWFixed - 1] = false;
}

Compiles to this:

11485 for (nWFixed = obj->isActiveIdx[2]; nWFixed <= obj->mConstrMax; nWFixed++) { 30016b8c: ldr r3, [sp, #152] ; 0x98 30016b8e: ldr.w r5, [r3, #1112] ; 0x458 30016b92: mov r3, r2 30016b94: cmp r5, r2 30016b96: blt.n 0x30016bee <ThrustAllocation_fmincon_m+4342> 30016b98: subs r0, r5, r2 30016b9a: adds r1, r5, #1 30016b9c: cmp r0, #2 30016b9e: sub.w r1, r1, r2 30016ba2: bls.n 0x30016bce <ThrustAllocation_fmincon_m+4310> 30016ba4: movw r0, #12880 ; 0x3250 30016ba8: ldr r3, [pc, #276] ; (0x30016cc0 <ThrustAllocation_fmincon_m+4552>) 11486 obj->isActiveConstr.data[nWFixed - 1] = false; 30016baa: movs r4, #0 30016bac: add r0, sp 30016bae: add r3, r2 30016bb0: ldr r0, [r0, #0] 30016bb2: add r3, r0 30016bb4: bic.w r0, r1, #3 30016bb8: add r0, r3 30016bba: str.w r4, [r3], #4 11485 for (nWFixed = obj->isActiveIdx[2]; nWFixed <= obj->mConstrMax; nWFixed++) { 30016bbe: cmp r3, r0 30016bc0: bne.n 0x30016bba <ThrustAllocation_fmincon_m+4290> 30016bc2: bic.w r0, r1, #3 30016bc6: cmp r0, r1 30016bc8: add.w r3, r2, r0 30016bcc: beq.n 0x30016bee <ThrustAllocation_fmincon_m+4342>

The problematic line is this:

30016bba: str.w r4, [r3], #4

As the operation could result in unaligned access. The array is placed in external SDRAM, which does not support unaligned access. Both arrays are basically of unsigned char.

In other words, the compiler uptimizes the code to clear 4 bytes at a time, instead of clearing 1 by 1 byte.

Is there a way to instruct the compiler to avoid creating potential unaligned 32 bit access.

I cannot rewrite the code, as it is a part of an auto generated code from Matlab/Simulink. An I do not want to use lower uptimization level.

Kasper.

topic Prevent compiler creating potential unaligned 32 bit access. in i.MX Processors

Prevent compiler creating potential unaligned 32 bit access.