Compiler/TMS570LS1227: Compiler v16.9.8.LTS Optimisation Level 3 Unrolling 'for' loop failure

Can you identify one function, or better yet one loop, which computes the wrong result when built with -O3 --opt_for_speed=3?  If so, for the source file which contains that function or loop, please send in a test case as described in the article How to Submit a Compiler Test Case.  While we cannot run the code, we can often identify the problem by inspecting the compiler generated assembly code.

Thanks and regards,

-George

Hi George,

Attached in my source file as requested.

I'm sorry that the scenario functions arenot simpler, but I have been unsuccessful in reducing this remaining complexity without the issue disappearing.

The compiler is CGT v16.9.8.LTS.

Compile options are:

'Building file: ../Tony.c'
'Invoking: ARM Compiler'
"C:/ti/ccsv7/tools/compiler/ti-cgt-arm_16.9.8.LTS/bin/armcl" --cmd_file="C:\work\PD_MV\Firmware\BRANCH\RamBlDev_AJM\Compiler.opt" -mv7R4 --code_state=32 --float_support=VFPv3D16 -O3 --opt_for_speed=2 --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/Release" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/Ancillary" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/PdBootLoader" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/DriverSoftware/CommunicationDrivers" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/DriverSoftware/ComplexDrivers" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/DriverSoftware/InputOutputDrivers" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/DriverSoftware/MemoryDrivers" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/DriverSoftware/MicrocontrollerDrivers" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/ServiceSoftware/CommunicationServices" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/ServiceSoftware/InputOutputServices" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/ServiceSoftware/MemoryServices" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/ServiceSoftware/SystemServices" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/ToolChain/include" --include_path="C:/work/PD_MV/Firmware/BRANCH/RamBlDev_AJM/ToolChain/Rtl" --define=DET_ENABLED --define=__TI_EABI__ -g --plain_char=unsigned --preproc_with_comment --preproc_with_compile --diag_remark=97 --diag_suppress=552 --diag_warning=225 --display_error_number --gen_func_subsections=on --enum_type=packed --abi=eabi --asm_listing --c_src_interlist --gen_opt_info=2 "../Tony.c"
'Finished building: ../Tony.c'

The program Tony_Main() is called as follows:

static unsigned char arrayTony16[16] = {0x00U, 0x11U, 0x22U, 0x33U, 0x44U, 0x55U, 0x66U, 0x77U, 0x88U, 0x99U, 0xAAU, 0xBBU, 0xCCU, 0xDDU, 0xEEU, 0xFFU};
static unsigned char arrayTony48[48] = {0xB8U, 0xEAU, 0x30U, 0xF3U, 0x60U, 0x01U, 0x95U, 0xE5U, 0x8EU, 0x9BU, 0x97U, 0x1AU, 0xC0U, 0x4BU, 0x32U, 0x5FU,
0xBFU, 0x40U, 0xC8U, 0x16U, 0x43U, 0x5BU, 0xF7U, 0xC0U, 0xD9U, 0x08U, 0xE8U, 0xB6U, 0x91U, 0x8AU, 0x0FU, 0x29U,
0x55U, 0xB6U, 0x3AU, 0xF0U, 0xA2U, 0x69U, 0x7EU, 0xA2U, 0x21U, 0x06U, 0xE9U, 0xCEU, 0x7AU, 0x98U, 0x08U, 0xACU};

void main(void)
{
Tony_Main(&arrayTony16[0], &arrayTony48[0]);
}

With -O3 --opt_for_speed=2, resultant arrayTony16 =

{0x45U, 0x03U, 0xDBU, 0xD1U, 0x61U, 0x2FU, 0x86U, 0xC0U, 0x73U, 0x72U, 0X7CU, 0x38U, 0xC1U, 0x65U, 0x21U, 0x7AU}

With -O3 --opt_for_speed=3, resultant arrayTony16 =

{0x9EU, 0xF1U, 0xDBU, 0xD1U, 0x61U, 0x2FU, 0x86U, 0xC0U, 0xA8U, 0x80U, 0X7CU, 0x38U, 0xC1U, 0x65U, 0x21U, 0x7AU}

In view of this issue, I am minded to revert to CGT v5.2.9 which I believe is also long term support.

Regards, Tony

void Tony_Main(unsigned char nDataPtr[], const unsigned char nExtPtr[]);



static void Tony_Munge1(unsigned char nMunge1Ptr[], const unsigned char nPtr[]);
static void Tony_Munge2(unsigned char nMunge2Ptr[]);
static void Tony_Munge3(unsigned char nMunge3Ptr[]);

void Tony_Main(
    unsigned char nDataPtr[],
    const unsigned char nExtPtr[])
{
    signed char nIndex;

    Tony_Munge1(nDataPtr, &nExtPtr[2U * (16U)]);
    Tony_Munge2(nDataPtr);

    for (nIndex = 1; nIndex >= 0; nIndex--)
    {
        Tony_Munge1(nDataPtr, &nExtPtr[((unsigned char)nIndex * (16U))]);
        if (0 != nIndex)
        {
            Tony_Munge3(nDataPtr);
        }
    }
}

static void Tony_Munge1(
    unsigned char nMunge1Ptr[],
    const unsigned char nPtr[])
{
    unsigned char nIndex;

    for (nIndex = 0U; nIndex < (16U); nIndex++)
    {
        (nMunge1Ptr[nIndex]) ^= nPtr[nIndex];
    }
}
static void Tony_Munge2(
    unsigned char nMunge2Ptr[])
{
    unsigned char nTemp;

    nTemp = nMunge2Ptr[0];
    nMunge2Ptr[0] = nMunge2Ptr[4];
    nMunge2Ptr[4] = nMunge2Ptr[8];
    nMunge2Ptr[8] = nMunge2Ptr[12];
    nMunge2Ptr[12] = nTemp;
    nTemp = nMunge2Ptr[1];
    nMunge2Ptr[1] = nMunge2Ptr[5];
    nMunge2Ptr[5] = nMunge2Ptr[9];
    nMunge2Ptr[9] = nMunge2Ptr[13];
    nMunge2Ptr[13] = nTemp;
    nTemp = nMunge2Ptr[2];
    nMunge2Ptr[2] = nMunge2Ptr[6];
    nMunge2Ptr[6] = nMunge2Ptr[10];
    nMunge2Ptr[10] = nMunge2Ptr[14];
    nMunge2Ptr[14] = nTemp;
    nTemp = nMunge2Ptr[3];
    nMunge2Ptr[3] = nMunge2Ptr[7];
    nMunge2Ptr[7] = nMunge2Ptr[11];
    nMunge2Ptr[11] = nMunge2Ptr[15];
    nMunge2Ptr[15] = nTemp;
}

static void Tony_Munge3(
    unsigned char nMunge3Ptr[])
{
    unsigned char nTemp[(16U)];
    unsigned char nIndex;

    nTemp[15] = nMunge3Ptr[0] ^ nMunge3Ptr[1];
    nTemp[14] = nMunge3Ptr[2] ^ nMunge3Ptr[3];
    nTemp[13] = nMunge3Ptr[4] ^ nMunge3Ptr[5];
    nTemp[12] = nMunge3Ptr[6] ^ nMunge3Ptr[7];

    nTemp[11] = nMunge3Ptr[8] ^ nMunge3Ptr[9];
    nTemp[10] = nMunge3Ptr[10] ^ nMunge3Ptr[11];
    nTemp[9]  = nMunge3Ptr[12] ^ nMunge3Ptr[13];
    nTemp[8]  = nMunge3Ptr[14] ^ nMunge3Ptr[15];

    nTemp[7]  = nMunge3Ptr[0] ^ nMunge3Ptr[1];
    nTemp[6]  = nMunge3Ptr[2] ^ nMunge3Ptr[3];
    nTemp[5]  = nMunge3Ptr[4] ^ nMunge3Ptr[5];
    nTemp[4]  = nMunge3Ptr[6] ^ nMunge3Ptr[7];

    nTemp[3]  = nMunge3Ptr[8] ^ nMunge3Ptr[9];
    nTemp[2]  = nMunge3Ptr[10] ^ nMunge3Ptr[11];
    nTemp[1]  = nMunge3Ptr[12] ^ nMunge3Ptr[13];
    nTemp[0]  = nMunge3Ptr[14] ^ nMunge3Ptr[15];

    for (nIndex = 0U; nIndex < (16U); nIndex++)
    {
        nMunge3Ptr[nIndex] = nTemp[nIndex];
    }
}

Please show me the contents of this command file ...

Tony Morrell said:

--cmd_file="C:\work\PD_MV\Firmware\BRANCH\RamBlDev_AJM\Compiler.opt"

Or, if it is easier, attach it to your next post.  It is possible, though not certain, these options could affect things.

Thanks and regards,

-George

Hi George,

I'm afraid the Compiler.opt file only consists of the following two lines:

--check_misra=2.2
--display_error_number

Regards, Tony.

I added the additional options and clarification to the SDOWP record.  Many of the fields in these records are not public, so the entry may appear the same to you.

Thanks and regards,

-George