#!/bin/sh

# Print the gcc cpu specific options tailored for the current CPU

# License: LGPLv2
# Author:
#    http://www.pixelbeat.org/
# Notes:
#    This script currently supports Linux,FreeBSD,Cygwin
#    This script is x86 (32 bit) specific
#    It should work on any gcc >= 2.95 at least
#    It only returns CPU specific options. You probably also want -03 etc.
# Changes:
#    V0.1, 12 Mar 2003, Initial release
#    V0.2, 01 Jun 2005, Added support for 3.2>=gcc<=4.0
#    V0.3, 03 Jun 2005, Added support for pentium-m
#    V0.4, 03 Jun 2005, Fix silly bugs
#    V0.5, 07 Jun 2005, Clarify/Simplify confusing floating point expr usage
#                       Print warning when CPU only supported on a newer gcc
#    V0.6, 15 Dec 2006, Added support for Intel Core and Core 2 processors
#                       Added support for 4.1>=gcc<=4.3
#                       Added support for gcc -msse3 option
#                       Added support for new gcc -march=native option
#    V0.7, 18 Dec 2006, Changes from Conor McDermottroe
#                         Added support for FreeBSD
#                         Remove bash specific constructs
#                         Better error handling
#    V0.8, 19 Dec 2006, Give warnings and 32 bit -march on 64 bit platforms.
#                       Previously it just gave an invalid blank -march.
#                       Reported and tested by Ewan Oughton.
#    V0.9, 30 Apr 2007, Give error if compiler not present.
#                       Warn about rather than default to -march=native option.
#   V0.92, 08 Nov 2007, Change from Krzysztof Jankowski to support Cygwin.
#                       Added support for AMD family 10 processors.
#                       Add support for gcc -msse4 & -msse5 options.
#                       Use "prescott" rather than "pentium4" for all
#                       models >= 3 in intel family 15, not just model 3.
#   V0.93, 13 Nov 2007, Oops, actually not all intel family 15, model >= 3
#                       are prescotts. Use the sse3 flag to distinguish.
#   V0.94, 31 Dec 2007, Oops, actually all intel family 15, model >= 3
#                       are prescotts. This was indicated by Adam Drzewiecki.
#                       I was confused by a bug in older linux kernels where pni
#                       was not reported for my intel "F41" processor at least.
#   V0.95, 18 Jan 2008, Changes from Conor McDermottroe
#                         Support for Mac OS X
#                         Support for FreeBSD base system
#   V0.96, 17 Dec 2008, Following a report from Alfredo Pons, add support
#                       for newer intel core2 processors (models 23,26,29,...)
#                       by assuming models >= 15 are core2.
#                       Report details of unrecognised CPUs.
#   V0.97, 19 Dec 2008, Use prescott for Intel Atom (model 28).
#                       Ensure all errors are output to stderr.
#   V0.98, 24 Feb 2009, Actually Intel Atom (model 28) is core2 ISA compliant.
#                       It's an in-order core (like i586) so mtune as such.
#   V0.99, 30 Apr 2009, Intel Atom (model 28) is getting a corresponding "atom"
#                       option in GCC 4.5
#   V0.99.3, 11 Aug 2009, Support for AMD Geode LX processor.
#   V0.99.4, 24 Oct 2009, Following a report from Maxime de Roucy, add support
#                         for AMD family 17 (griffin).
#   V0.99.14, 16 Jul 2015
#     http://github.com/pixelb/scripts/commits/master/scripts/gcccpuopt

if test $1 = "--version" {
    echo "0.99.14" && exit
}

# This table shows when -march options were introduced into _official_ gcc releases.
# Note there are vendor deviations that complicate this.
# For e.g. Red Hat introduced the prescott option in 3.3-13.
#   gcc-2.95   = i386, i486, i586,pentium, i686,pentiumpro, k6
#   gcc-3.0   += athlon
#   gcc-3.1   += pentium-mmx, pentium2, pentium3, pentium4, k6-2, k6-3, athlon-{tbird, 4,xp,mp}
#   gcc-3.3   += winchip-c6, winchip2, c3
#   gcc-3.4.0 += k8,opteron,athlon64,athlon-fx, c3-2
#   gcc-3.4.1 += pentium-m, pentium3m, pentium4m, prescott, nocona
#   gcc-4.3   += core2, amdfam10, geode
#   gcc-4.5   += atom
#   gcc-4.6   += corei7, corei7-avx, bdver1, btver1
#   gcc-4.7   += core-avx-i, core-avx2, bdver2
#   gcc-4.8   += btver2
#   gcc-4.9   += b[td]ver[34], broadwell, silvermont
#   gcc-5.1   += knl

test -z $CC && setglobal CC = 'gcc'

proc try_gcc_options {
    $CC $ifsjoin(Argv) -S -o /dev/null -xc /dev/null >/dev/null !2 > !1
}

if ! try_gcc_options {
    echo "Error: Couldn't execute your compiler ($CC)" > !2
    exit 1
}

if try_gcc_options -march=native {
    echo "Warning: Your compiler supports the -march=native option which you may prefer" > !2
    #gcc -march=native --verbose -xc /dev/null 2>&1 | sed -n 's|.*/dev/null \(-march.*-mtune[^ ]*\).*|\1|p' >&2
}

if ! try_gcc_options -march=i386 {
    if ! try_gcc_options -m32 -march=i386 {
        echo "Error: This script only supports 32 bit x86 architectures" > !2
        exit 1
    } else {
        echo "Warning: The optimum *32 bit* architecture is reported" > !2
        setglobal m32 = '"-m32 '"
    }
}

proc try_line {
    setglobal skip = '0'
    for arch in [$1] {
        if try_gcc_options $m32 -march=$arch {
            echo $arch
            return
        } elif test $skip = "0" && test $arch != "native" {
            setglobal skip = '1'
            echo "Warning: Newer versions of GCC better support your CPU with -march=$arch" > !2
        }
    }
    return 1
}

proc read_cpu_data_linux {
    setglobal IFS = '":'"
    while read name value {
        unset IFS
        setglobal name = $[echo $name] #strip spaces
        setglobal value = $[echo $value] #strip spaces
        if test $name = "vendor_id" {
            setglobal value = $[echo $value | sed 's/\([^ ]*\).*/\1/] #take first word
        }
        setglobal IFS = '":'"
        if test $name = "vendor_id" {
            setglobal vendor_id = $value
        } elif test $name = "cpu family" {
            setglobal cpu_family = $value
        } elif test $name = "model" {
            setglobal cpu_model = $value
        } elif test $name = "flags" {
            setglobal flags = $value
            break #flags last so break early
        }
    } < /proc/cpuinfo
    unset IFS
}

proc read_cpu_data_freebsd {
    local _line _cpu_id

    if test ! -r /var/run/dmesg.boot {
        echo "Error: /var/run/dmesg.boot does not exist!" > !2
        exit 1;
    }

    setglobal IFS = '"
'"
    for _line in [$[grep -A2 '^CPU: ' /var/run/dmesg.boot]] {
        if test -n $[echo $_line | grep '^  Origin = ] {
            setglobal vendor_id = $[echo $_line | sed -e 's/^  Origin = .//' -e 's/[^A-Za-z0-9].*$//]
            setglobal _cpu_id = $[echo $_line | sed -e 's/^.*Id = //' -e 's/ .*$//' -e 'y/abcdef/ABCDEF/]
            setglobal cpu_family = $shExpr(' ($_cpu_id & 0xF0F) >> 8 ') #FreeBSD 5.0 sh doesn't support 0x
            setglobal cpu_model = $shExpr(' (($_cpu_id & 0xF0) >> 4) + (($_cpu_id & 0xF0000) >> 12) ')
        }
        if test -n $[echo $_line | grep '^  Features=] {
            setglobal flags = $[echo $_line | sed -e 's/^.*<//' -e 's/>.*//' -e 's/,/ /g' | tr 'A-Z' 'a-z]
        }
    }
    unset IFS
}

proc read_cpu_data_darwin {
    setglobal vendor_id = $[/usr/sbin/sysctl -n machdep.cpu.vendor]
    setglobal cpu_family = $[/usr/sbin/sysctl -n machdep.cpu.family]
    setglobal cpu_model = $[/usr/sbin/sysctl -n machdep.cpu.model]
    setglobal flags = $[/usr/sbin/sysctl -n machdep.cpu.features | tr 'A-Z' 'a-z]
}

proc read_cpu_data {
    # Default values
    setglobal vendor_id = '"NotFound'"
    setglobal cpu_family = '"-1'"
    setglobal cpu_model = '"-1'"
    setglobal flags = ''""
    match $[uname] {
        with Linux|CYGWIN*
            read_cpu_data_linux 
        with FreeBSD
            read_cpu_data_freebsd 
        with Darwin
            read_cpu_data_darwin 
        with *
            echo "Error: $[uname] is not a supported operating system" > !2
            exit 1 
    }
}

read_cpu_data

if test $vendor_id = "AuthenticAMD" {
    if test $cpu_family -eq 4 {
        setglobal _CFLAGS = '"-march=i486'"
    } elif test $cpu_family -eq 5 {
        if test $cpu_model -lt 4 {
            setglobal _CFLAGS = '"-march=pentium'"
        } elif test '(' $cpu_model -eq 6 ')' -o '(' $cpu_model -eq 7 ')' {
            setglobal _CFLAGS = '"-march=k6'"
        } elif test '(' $cpu_model -eq 8 ')' -o '(' $cpu_model -eq 12 ')' {
            setglobal line = '"k6-2 k6'"
        } elif test '(' $cpu_model -eq 9 ')' -o '(' $cpu_model -eq 13 ')' {
            setglobal line = '"k6-3 k6-2 k6'"
        } elif test $cpu_model -eq 10 { #geode LX
            setglobal line = '"geode k6-2 k6'"
            #The LX supports 3dnowext in addition to the k6-2 instructions,
            #however gcc doesn't support explicitly selecting that.
        }
    } elif test $cpu_family -eq 6 {
        if test $cpu_model -le 3 {
            setglobal line = '"athlon k6-3 k6-2 k6'"
        } elif test $cpu_model -eq 4 {
            setglobal line = '"athlon-tbird athlon k6-3 k6-2 k6'"
        } elif test $cpu_model -ge 6 { #athlon-{4,xp,mp} (also geode NX)
            setglobal line = '"athlon-4 athlon k6-3 k6-2 k6'"
        }
    } elif test $cpu_family -eq 15 { #k8,opteron,athlon64,athlon-fx
        setglobal line = '"k8 athlon-4 athlon k6-3 k6-2 k6'"
    } elif test $cpu_family -eq 16 ||    #barcelona,amdfam10
         test $cpu_family -eq 17 ||    #griffin
         test $cpu_family -eq 18 { #llano
        setglobal line = '"amdfam10 k8 athlon-4 athlon k6-3 k6-2 k6'"
    } elif test $cpu_family -eq 20 { #bobcat
        setglobal line = '"btver1 amdfam10 k8 athlon-4 athlon k6-3 k6-2 k6'"
    } elif test $cpu_family -eq 21 { #bulldozer
        if echo $flags | grep -q bmi { #piledriver
            setglobal line = '"bdver2 bdver1 btver1 amdfam10 k8 athlon-4 athlon k6-3 k6-2 k6'"
        } else {
            setglobal line = '"bdver1 btver1 amdfam10 k8 athlon-4 athlon k6-3 k6-2 k6'"
        }
    } elif test $cpu_family -eq 22 { #jaguar
        setglobal line = '"btver2 btver1 amdfam10 k8 athlon-4 athlon k6-3 k6-2 k6'"
    }
} elif test $vendor_id = "CentaurHauls" {
    if test $cpu_family -eq 5 {
        if test $cpu_model -eq 4 {
            setglobal line = '"winchip-c6 pentium'"
        } elif test $cpu_model -eq 8 {
            setglobal line = '"winchip2 winchip-c6 pentium'"
        } elif test $cpu_model -ge 9 {
            setglobal line = '"winchip2 winchip-c6 pentium'" #actually winchip3 but gcc doesn't support this currently
        }
    } elif test $cpu_family -eq 6 {
        if echo $flags | grep -q cmov {
            setglobal fallback = 'pentiumpro'
        } else {
            setglobal fallback = 'pentium' #gcc incorrectly assumes i686 always has cmov
        }
        if test $cpu_model -eq 6 {
            setglobal _CFLAGS = '"-march=pentium'" # ? Cyrix 3 (samuel)
        } elif test $cpu_model -eq 7 || test $cpu_model -eq 8 {
            setglobal line = ""c3 winchip2 winchip-c6 $fallback""
        } elif test $cpu_model -ge 9 {
            setglobal line = ""c3-2 c3 winchip2 winchip-c6 $fallback""
        }
    }
} elif test $vendor_id = "GenuineIntel" {
    if test $cpu_family -eq 3 {
        setglobal _CFLAGS = '"-march=i386'"
    } elif test $cpu_family -eq 4 {
        setglobal _CFLAGS = '"-march=i486'"
    } elif test $cpu_family -eq 5 {
        if test $cpu_model -ne 4 {
            setglobal _CFLAGS = '"-march=pentium'"
        } else {
            setglobal line = '"pentium-mmx pentium'" #No overlap with other vendors
        }
    } elif test $cpu_family -eq 6 {
        if test '(' $cpu_model -eq 0 ')' -o '(' $cpu_model -eq 1 ')' {
            setglobal _CFLAGS = '"-march=pentiumpro'"
        } elif test '(' $cpu_model -ge 2 ')' -a '(' $cpu_model -le 6 ')' { #4=TM5600 at least
            setglobal line = '"pentium2 pentiumpro pentium-mmx pentium i486 i386'"
            if test '(' $cpu_model -eq 2 ')' { #qemu generic
                echo "\
Warning: Unspecified QEMU CPU model.
Please consider upgrading QEMU or configuring it to use a specific model." > !2
            }
        } elif test '(' $cpu_model -eq 9 ')' -o '(' $cpu_model -eq 13 ')' { #centrino
            setglobal line = '"pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test '(' $cpu_model -eq 14 ')' { #Core
            setglobal line = '"prescott pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 28 || # pine
             test $cpu_model -eq 38 || # oak
             test $cpu_model -eq 54 { # cedar
            setglobal line = '"atom core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 55 || # Bay trail
             test $cpu_model -eq 74 || # Merrifield
             test $cpu_model -eq 77 || # Avoton
             test $cpu_model -eq 90 || # Moorefield
             test $cpu_model -eq 93 {
            setglobal line = '"silvermont atom core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 26 ||
             test $cpu_model -eq 30 ||
             test $cpu_model -eq 31 ||
             test $cpu_model -eq 46 ||
             # ^ Nehalem ^
             test $cpu_model -eq 37 ||
             test $cpu_model -eq 44 ||
             test $cpu_model -eq 47 {
             # ^ Westmere ^
            setglobal line = '"corei7 core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 42 || #Sandy Bridge
             test $cpu_model -eq 45 { #Sandy Bridge E
            setglobal line = '"corei7-avx corei7 core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 58 || #Ivy Bridge
             test $cpu_model -eq 62 { #Ivy Bridge E
            setglobal line = '"core-avx-i corei7-avx corei7 core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 60 || #Haswell
             test $cpu_model -eq 63 || #Haswell E
             test $cpu_model -eq 69 || #Haswell U
             test $cpu_model -eq 70 { #Haswell Crystalwell
            setglobal line = '"core-avx2 core-avx-i corei7-avx corei7 core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 61 || #Broadwell
             test $cpu_model -eq 71 ||
             test $cpu_model -eq 79 ||
             test $cpu_model -eq 86 {
            setglobal line = '"broadwell core-avx2 core-avx-i corei7-avx corei7 core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 87 { #Knight's landing (haswell compat)
            setglobal line = '"knl core-avx2 core-avx-i corei7-avx corei7 core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test $cpu_model -eq 15 ||
             # ^ Merom ^
             test $cpu_model -eq 22 ||
             # ^ Conroe-L ^
             test $cpu_model -eq 23 ||
             test $cpu_model -eq 29 {
             # ^ Penryn ^
            setglobal line = '"core2 pentium-m pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        } elif test '(' $cpu_model -ge 7 ')' -a '(' $cpu_model -le 11 ')' {
            setglobal line = '"pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        }
    } elif test $cpu_family -eq 15 {
        setglobal line = '"pentium4 pentium3 pentium2 pentiumpro pentium-mmx pentium i486 i386'"
        if test $cpu_model -ge 3 {
            setglobal line = ""prescott $line""
        }
    }
} elif test $vendor_id = "Geode" { #by NSC
    if test $cpu_family -eq 5 {
        if test '(' $cpu_model -eq 4 ')' -o '(' $cpu_model -eq 9 ')' {
            # Note both models 4 and 9 have cmov.
            # In addition, model 9 has cxmmx.
            # Note also, the "geode" gcc arch is for newer AMD geode cores
            # and is not appropriate for this older core.
            setglobal line = '"pentium-mmx pentium'"
        }
    }
}

if test '(' -z $_CFLAGS ')' -a '(' -z $line ')' {
    echo "\
Unrecognised CPU. Please email the following to: P@draigBrady.com
  Vendor:$vendor_id family:$cpu_family model:$cpu_model
  flags:$flags" > !2
    exit 1
}

test -z $_CFLAGS && setglobal _CFLAGS = ""-march=$[try_line $line]""

#The Atom CPU supports the full core2 instruction set,
#but it's an in-order core, the last one of those being the i586.
#Therefore if gcc hasn't explicit support for the Atom,
#tune it for the i586 architecture.
if ! echo $_CFLAGS | grep -q "atom" { #gcc hasn't specific Atom support
    if echo $line | grep -q "core2" { #atom ISA line
        if test $cpu_model -eq 28 {      #atom
            if echo $_CFLAGS | grep -E -q "(core2|pentium[^ ])" { #gcc chose out of order arch
                setglobal _CFLAGS = ""$_CFLAGS -mtune=pentium"" #tune for last in-order core
            }
        }
    }
}

#SSE is not used for floating point by default in gcc 32 bit
#so turn that on here.
if echo $flags | grep -q "sse" {
    if try_gcc_options "-mfpmath=sse" { #gcc >= 3.1
        setglobal _CFLAGS = ""$_CFLAGS -mfpmath=sse""
    }
}

#The SSE options are mostly selected automatically
#when a particular march option is selected.
#There are a few exceptions unfortunately, which we handle here.
#Note the sse instruction lines are:
#   intel: [sse4.2] [sse4.1] ssse3 sse3 sse2 sse ...
#   amd:   [sse5] sse4a [sse3] sse2 sse ...
# The bracketed ones are only available on some cpus
# in a particular family and so need to be added explicitly.
if echo $_CFLAGS | grep -q "amdfam10" {
    if echo $flags | grep -q "sse5" {
        if try_gcc_options "-msse5" { #gcc >= 4.3
            setglobal _CFLAGS = ""$_CFLAGS -msse5""
        }
    }
} elif echo $_CFLAGS | grep -E -q "(k8|c3-2)" {
    if echo $flags | grep -E -q "(sse3|pni)" {
        if try_gcc_options "-msse3" { #gcc >= 3.3.3
            setglobal _CFLAGS = ""$_CFLAGS -msse3""
        }
    }
} elif echo $_CFLAGS | grep -q "core2" {
    if echo $flags | grep -q "sse4_2" {
        if try_gcc_options "-msse4" { #gcc >= 4.3
            setglobal _CFLAGS = ""$_CFLAGS -msse4""
        }
    } elif echo $flags | grep -q "sse4_1" {
        if try_gcc_options "-msse4.1" { #gcc >= 4.3
            setglobal _CFLAGS = ""$_CFLAGS -msse4.1""
        }
    }
}

echo "$m32$_CFLAGS"