crosstool-ng/scripts/build/arch/sparc.sh

# Compute sparc-specific values
CT_DoArchTupleValues() {
    # That's the only thing to override
    CT_TARGET_ARCH="sparc${target_bits_64}${CT_ARCH_SUFFIX}"

    if [ "${CT_KERNEL}" = "linux" -a -z "${CT_ARCH_CPU}" ]; then
        if [ "${CT_ARCH_64}" = "y" ]; then
            # By default, sparc64-*-linux is configured with -mcpu=v9. However,
            # according to https://sourceware.org/ml/libc-alpha/2005-12/msg00027.html,
            # "There is no Linux sparc64 port that runs on non-UltraSPARC-I+ ISA CPUs."
            # There is a patch that would change the default to -mcpu=ultrasparc for
            # sparc64-*-linux configuration: https://patchwork.ozlabs.org/patch/409424/
            # but that patch has not been integrated (yet). One concern raised about
            # this patch was that -mcpu=ultrasparc can suboptimally schedule instructions
            # for newer SPARC CPUs. So, override to -mcpu=ultrasparc and warn the user.
            CT_DoLog WARN "Setting CPU to UltraSPARC-I for sparc64-linux. Set CT_ARCH_CPU if a different CPU is desired."
            CT_ARCH_WITH_CPU="--with-cpu=ultrasparc"
            if [ -n "${CT_ARCH_SUPPORTS_WITH_32_64}" -a -n "${CT_MULTILIB}" ]; then
                CT_ARCH_WITH_CPU_32="--with-cpu-32=ultrasparc"
                CT_ARCH_WITH_CPU_64="--with-cpu-64=ultrasparc"
            fi
        else
            # Similarly, sparc-*-linux defaults to v7. GLIBC 2.31 started to reject
            # pre-v8 configurations with 2.31, and default v8 fails a subsequent test for
            # using atomics (which are presumably, only available on some v8 CPUs).
            # Therefore, default to v9.
            if [ -z "${CT_GLIBC_SPARC_ALLOW_V7}" ]; then
                CT_DoLog WARN "Setting CPU to V9 for sparc-linux. Set CT_ARCH_CPU if a different CPU is desired."
                CT_ARCH_WITH_CPU="--with-cpu=v9"
            fi
        fi
    fi
}

#------------------------------------------------------------------------------
# Get multilib architecture-specific target
# Usage: CT_DoArchMultilibTarget "target variable" "multilib flags"
CT_DoArchMultilibTarget ()
{
    local target_var="${1}"; shift
    local -a multi_flags=( "$@" )
    local target_

    local m32=false
    local m64=false

    for m in "${multi_flags[@]}"; do
        case "$m" in
            -m32)     m32=true ;;
            -m64)     m64=true ;;
        esac
    done

    eval target_=\"\${${target_var}}\"

    # Fix up bitness
    case "${target_}" in
        sparc-*)      $m64 && target_=${target_/#sparc-/sparc64-} ;;
        sparc64-*)    $m32 && target_=${target_/#sparc64-/sparc-} ;;
    esac

    # Set the target variable
    eval ${target_var}=\"${target_}\"
}

# Special tuple adjustment for glibc.
CT_DoArchGlibcAdjustTuple() {
    local target_var="${1}"
    local target_

    eval target_=\"\${${target_var}}\"

    case "${target_}" in
        # SPARC quirk: glibc 2.23 and newer dropped support for SPARCv8 and
        # earlier (corresponding pthread barrier code is missing). Until this
        # support is reintroduced, configure as sparcv9.
        sparc-*)
            if [ "${CT_GLIBC_NO_SPARC_V8}" = y ]; then
                CT_DoLog WARN "GLIBC 2.23 and newer only support SPARCv9"
                target_=${target_/#sparc-/sparcv9-}
            fi
            ;;
    esac

    # Set the target variable
    eval ${target_var}=\"${target_}\"
}

CT_DoArchUClibcConfig() {
    local cfg="${1}"

    CT_DoArchUClibcSelectArch "${cfg}" "${CT_ARCH}"
    # FIXME: select CONFIG_SPARC_V7/V8/V9/V9B according to the CPU selector
}