aom/build/make/rtcd.sh

360 строки
7.0 KiB
Bash
Исходник Обычный вид История

New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
#!/bin/sh
self=$0
usage() {
cat <<EOF >&2
Usage: $self [options] FILE
Reads the Run Time CPU Detections definitions from FILE and generates a
C header file on stdout.
Options:
--arch=ARCH Architecture to generate defs for (required)
--disable-EXT Disable support for EXT extensions
--require-EXT Require support for EXT extensions
--sym=SYMBOL Unique symbol to use for RTCD initialization function
--config=FILE File with CONFIG_FOO=yes lines to parse
EOF
exit 1
}
die() {
echo "$@" >&2
exit 1
}
die_argument_required() {
die "Option $opt requires argument"
}
for opt; do
optval="${opt#*=}"
case "$opt" in
--arch) die_argument_required;;
--arch=*) arch=${optval};;
--disable-*) eval "disable_${opt#--disable-}=true";;
--require-*) REQUIRES="${REQUIRES}${opt#--require-} ";;
--sym) die_argument_required;;
--sym=*) symbol=${optval};;
--config=*) config_file=${optval};;
-h|--help)
usage
;;
-*)
die "Unrecognized option: ${opt%%=*}"
;;
*)
defs_file="$defs_file $opt"
;;
esac
shift
done
for f in $defs_file; do [ -f "$f" ] || usage; done
[ -n "$arch" ] || usage
# Import the configuration
[ -f "$config_file" ] && eval $(grep CONFIG_ "$config_file")
#
# Routines for the RTCD DSL to call
#
prototype() {
local rtyp
case "$1" in
unsigned) rtyp="$1 "; shift;;
esac
rtyp="${rtyp}$1"
local fn="$2"
local args="$3"
eval "${2}_rtyp='$rtyp'"
eval "${2}_args='$3'"
ALL_FUNCS="$ALL_FUNCS $fn"
specialize $fn c
}
specialize() {
local fn="$1"
shift
for opt in "$@"; do
eval "${fn}_${opt}=${fn}_${opt}"
done
}
require() {
for fn in $ALL_FUNCS; do
for opt in "$@"; do
local ofn=$(eval "echo \$${fn}_${opt}")
[ -z "$ofn" ] && continue
# if we already have a default, then we can disable it, as we know
# we can do better.
local best=$(eval "echo \$${fn}_default")
local best_ofn=$(eval "echo \$${best}")
[ -n "$best" ] && [ "$best_ofn" != "$ofn" ] && eval "${best}_link=false"
eval "${fn}_default=${fn}_${opt}"
eval "${fn}_${opt}_link=true"
done
done
}
forward_decls() {
ALL_FORWARD_DECLS="$ALL_FORWARD_DECLS $1"
}
#
# Include the user's directives
#
for f in $defs_file; do
. $f
done
#
# Process the directives according to the command line
#
process_forward_decls() {
for fn in $ALL_FORWARD_DECLS; do
eval $fn
done
}
determine_indirection() {
[ "$CONFIG_RUNTIME_CPU_DETECT" = "yes" ] || require $ALL_ARCHS
for fn in $ALL_FUNCS; do
local n=""
local rtyp="$(eval "echo \$${fn}_rtyp")"
local args="$(eval "echo \"\$${fn}_args\"")"
local dfn="$(eval "echo \$${fn}_default")"
dfn=$(eval "echo \$${dfn}")
for opt in "$@"; do
local ofn=$(eval "echo \$${fn}_${opt}")
[ -z "$ofn" ] && continue
local link=$(eval "echo \$${fn}_${opt}_link")
[ "$link" = "false" ] && continue
n="${n}x"
done
if [ "$n" = "x" ]; then
eval "${fn}_indirect=false"
else
eval "${fn}_indirect=true"
fi
done
}
declare_function_pointers() {
for fn in $ALL_FUNCS; do
local rtyp="$(eval "echo \$${fn}_rtyp")"
local args="$(eval "echo \"\$${fn}_args\"")"
local dfn="$(eval "echo \$${fn}_default")"
dfn=$(eval "echo \$${dfn}")
for opt in "$@"; do
local ofn=$(eval "echo \$${fn}_${opt}")
[ -z "$ofn" ] && continue
echo "$rtyp ${ofn}($args);"
done
if [ "$(eval "echo \$${fn}_indirect")" = "false" ]; then
echo "#define ${fn} ${dfn}"
else
echo "RTCD_EXTERN $rtyp (*${fn})($args);"
fi
echo
done
}
set_function_pointers() {
for fn in $ALL_FUNCS; do
local n=""
local rtyp="$(eval "echo \$${fn}_rtyp")"
local args="$(eval "echo \"\$${fn}_args\"")"
local dfn="$(eval "echo \$${fn}_default")"
dfn=$(eval "echo \$${dfn}")
if $(eval "echo \$${fn}_indirect"); then
echo " $fn = $dfn;"
for opt in "$@"; do
local ofn=$(eval "echo \$${fn}_${opt}")
[ -z "$ofn" ] && continue
[ "$ofn" = "$dfn" ] && continue;
local link=$(eval "echo \$${fn}_${opt}_link")
[ "$link" = "false" ] && continue
local cond="$(eval "echo \$have_${opt}")"
echo " if (${cond}) $fn = $ofn;"
done
fi
echo
done
}
filter() {
local filtered
for opt in "$@"; do
[ -z $(eval "echo \$disable_${opt}") ] && filtered="$filtered $opt"
done
echo $filtered
}
#
# Helper functions for generating the arch specific RTCD files
#
common_top() {
local outfile_basename=$(basename ${symbol:-rtcd.h})
local include_guard=$(echo $outfile_basename | tr '[a-z]' '[A-Z]' | tr -c '[A-Z]' _)
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
cat <<EOF
#ifndef ${include_guard}
#define ${include_guard}
#ifdef RTCD_C
#define RTCD_EXTERN
#else
#define RTCD_EXTERN extern
#endif
$(process_forward_decls)
$(declare_function_pointers c $ALL_ARCHS)
void ${symbol:-rtcd}(void);
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
EOF
}
common_bottom() {
cat <<EOF
#endif
EOF
}
x86() {
determine_indirection c $ALL_ARCHS
# Assign the helper variable for each enabled extension
for opt in $ALL_ARCHS; do
local uc=$(echo $opt | tr '[a-z]' '[A-Z]')
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
eval "have_${opt}=\"flags & HAS_${uc}\""
done
cat <<EOF
$(common_top)
#ifdef RTCD_C
#include "vpx_ports/x86.h"
static void setup_rtcd_internal(void)
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
{
int flags = x86_simd_caps();
(void)flags;
$(set_function_pointers c $ALL_ARCHS)
}
#endif
$(common_bottom)
EOF
}
arm() {
determine_indirection c $ALL_ARCHS
# Assign the helper variable for each enabled extension
for opt in $ALL_ARCHS; do
local uc=$(echo $opt | tr '[a-z]' '[A-Z]')
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
eval "have_${opt}=\"flags & HAS_${uc}\""
done
cat <<EOF
$(common_top)
#include "vpx_config.h"
#ifdef RTCD_C
#include "vpx_ports/arm.h"
static void setup_rtcd_internal(void)
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
{
int flags = arm_cpu_caps();
(void)flags;
$(set_function_pointers c $ALL_ARCHS)
}
#endif
$(common_bottom)
EOF
}
mips() {
determine_indirection c $ALL_ARCHS
cat <<EOF
$(common_top)
#include "vpx_config.h"
void ${symbol:-rtcd}(void);
#ifdef RTCD_C
void ${symbol:-rtcd}(void)
{
$(set_function_pointers c)
#if HAVE_DSPR2
void dsputil_static_init();
dsputil_static_init();
#endif
}
#endif
$(common_bottom)
EOF
}
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
unoptimized() {
determine_indirection c
cat <<EOF
$(common_top)
#include "vpx_config.h"
#ifdef RTCD_C
static void setup_rtcd_internal(void)
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
{
$(set_function_pointers c)
}
#endif
$(common_bottom)
EOF
}
#
# Main Driver
#
require c
case $arch in
x86)
ALL_ARCHS=$(filter mmx sse sse2 sse3 ssse3 sse4_1)
x86
;;
x86_64)
ALL_ARCHS=$(filter mmx sse sse2 sse3 ssse3 sse4_1)
REQUIRES=${REQUIRES:-mmx sse sse2}
require $(filter $REQUIRES)
x86
;;
mips32)
ALL_ARCHS=$(filter mips32)
dspr2=$([ -f "$config_file" ] && eval echo $(grep HAVE_DSPR2 "$config_file"))
HAVE_DSPR2="${dspr2#*=}"
if [ "$HAVE_DSPR2" = "yes" ]; then
ALL_ARCHS=$(filter mips32 dspr2)
fi
mips
;;
New RTCD implementation This is a proof of concept RTCD implementation to replace the current system of nested includes, prototypes, INVOKE macros, etc. Currently only the decoder specific functions are implemented in the new system. Additional functions will be added in subsequent commits. Overview: RTCD "functions" are implemented as either a global function pointer or a macro (when only one eligible specialization available). Functions which have RTCD specializations are listed using a simple DSL identifying the function's base name, its prototype, and the architecture extensions that specializations are available for. Advantages over the old system: - No INVOKE macros. A call to an RTCD function looks like an ordinary function call. - No need to pass vtables around. - If there is only one eligible function to call, the function is called directly, rather than indirecting through a function pointer. - Supports the notion of "required" extensions, so in combination with the above, on x86_64 if the best function available is sse2 or lower it will be called directly, since all x86_64 platforms implement sse2. - Elides all references to functions which will never be called, which could reduce binary size. For example if sse2 is required and there are both mmx and sse2 implementations of a certain function, the code will have no link time references to the mmx code. - Significantly easier to add a new function, just one file to edit. Disadvantages: - Requires global writable data (though this is not a new requirement) - 1 new generated source file. Change-Id: Iae6edab65315f79c168485c96872641c5aa09d55
2011-08-19 22:06:00 +04:00
armv5te)
ALL_ARCHS=$(filter edsp)
arm
;;
armv6)
ALL_ARCHS=$(filter edsp media)
arm
;;
armv7)
ALL_ARCHS=$(filter edsp media neon)
arm
;;
*)
unoptimized
;;
esac