shithub: mc

Info  •  Files  •  Log  •  Branches

Allow CPU feature detection by mbld, and add "sse4" tag

This allows writing graceful degradation of optimizations. In
practice, it justifies less-than-optimal floating point implementations
by allowing a portable upgrade path through AVX, SIMD, &c.

Bootstrap scripts are updated by genboostrap.sh on Linux+amd64, and
by hand on all others.

--- a/lib/math/bld.sub

+++ b/lib/math/bld.sub

@@ -2,7 +2,7 @@

 	fpmath.myr

 	# trunc, floor, ceil

-	fpmath-trunc-impl+posixy-x64.s

+	fpmath-trunc-impl+posixy-x64-sse4.s

 	fpmath-trunc-impl.myr

 	lib ../std:std

--- /dev/null

+++ b/lib/math/fpmath-trunc-impl+posixy-x64-sse4.s

@@ -1,0 +1,41 @@

+.globl math$trunc32

+.globl math$_trunc32

+math$trunc32:

+math$_trunc32:

+	roundps $0x03, %xmm0, %xmm0

+	ret

+

+.globl math$floor32

+.globl math$_floor32

+math$floor32:

+math$_floor32:

+	roundps $0x01, %xmm0, %xmm0

+	ret

+

+.globl math$ceil32

+.globl math$_ceil32

+math$ceil32:

+math$_ceil32:

+	roundps $0x02, %xmm0, %xmm0

+	ret

+

+.globl math$trunc64

+.globl math$_trunc64

+math$trunc64:

+math$_trunc64:

+	roundpd $0x03, %xmm0, %xmm0

+	ret

+

+.globl math$floor64

+.globl math$_floor64

+math$floor64:

+math$_floor64:

+	roundpd $0x01, %xmm0, %xmm0

+	ret

+

+.globl math$ceil64

+.globl math$_ceil64

+math$ceil64:

+math$_ceil64:

+	roundpd $0x02, %xmm0, %xmm0

+	ret

--- a/lib/math/fpmath-trunc-impl+posixy-x64.s

+++ /dev/null

@@ -1,41 +1,0 @@

-.globl math$trunc32

-.globl math$_trunc32

-math$trunc32:

-math$_trunc32:

-	roundps $0x03, %xmm0, %xmm0

-	ret

-

-.globl math$floor32

-.globl math$_floor32

-math$floor32:

-math$_floor32:

-	roundps $0x01, %xmm0, %xmm0

-	ret

-

-.globl math$ceil32

-.globl math$_ceil32

-math$ceil32:

-math$_ceil32:

-	roundps $0x02, %xmm0, %xmm0

-	ret

-

-.globl math$trunc64

-.globl math$_trunc64

-math$trunc64:

-math$_trunc64:

-	roundpd $0x03, %xmm0, %xmm0

-	ret

-

-.globl math$floor64

-.globl math$_floor64

-math$floor64:

-math$_floor64:

-	roundpd $0x01, %xmm0, %xmm0

-	ret

-

-.globl math$ceil64

-.globl math$_ceil64

-math$ceil64:

-math$_ceil64:

-	roundpd $0x02, %xmm0, %xmm0

-	ret

--- a/mbld/bld.sub

+++ b/mbld/bld.sub

@@ -1,5 +1,6 @@

 bin mbld =

 	build.myr

+	cpufeatures+x64.s

 	deps.myr

 	libs.myr

 	install.myr

--- /dev/null

+++ b/mbld/cpufeatures+x64.s

@@ -1,0 +1,10 @@

+.globl bld$cpufeatures

+.globl bld$_cpufeatures

+bld$cpufeatures:

+bld$_cpufeatures:

+	mov	$0x1, %eax

+	cpuid

+	mov	%ecx, %eax

+	rol	$32, %rax

+	shrd	$32, %rdx, %rax

+	ret

--- a/mbld/opts.myr

+++ b/mbld/opts.myr

@@ -7,6 +7,7 @@

 	var opt_arch 		: byte[:]

 	var opt_sys		: byte[:]

 	var opt_sysvers		: (int, int, int)

+	var opt_cpufeatures	: uint64

 	var opt_runtime		: byte[:]

 	var opt_genasm		: bool

 	var opt_incpaths	: byte[:][:]

@@ -30,11 +31,13 @@

 	const initopts	: (-> void)

 	const parseversion	: (v : byte[:] -> (int, int, int))

+	extern const cpufeatures : (-> uint64)

 ;;

 var opt_arch 		= ""

 var opt_sys		= ""

 var opt_binname		= ""

+var opt_cpufeatures	= 0ul

 var opt_libname		= ""

 var opt_runtime		= ""

 var opt_incpaths	/* FIXME: taking a constant slice is a nonconstant initializer */

@@ -74,6 +77,9 @@

 	| "amd64":	opt_arch = "x64"

 	| unknown:	std.fatal("unknown architecture \"{}\"\n", unknown)

 	;;

+

+	/* from cpuid with EAX=1; EDX at top, ECX at bottom */

+	opt_cpufeatures = cpufeatures()

 	opt_maxproc = 2*(thread.ncpu() : std.size)

 	opt_incpaths = [][:]

--- a/mbld/syssel.myr

+++ b/mbld/syssel.myr

@@ -158,6 +158,11 @@

 	| "x64":	tag(b, "x64")

 	| unknown:	std.fatal("unknown architecture {}\n", unknown)

 	;;

+

+	if opt_cpufeatures & 0x180000 == 0x180000

+		tag(b, "sse4")

+	;;

+

 	for t : tags

 		tag(b, t)

 	;;

--- a/mk/bootstrap/bootstrap+Darwin-x86_64.sh

+++ b/mk/bootstrap/bootstrap+Darwin-x86_64.sh

@@ -4,6 +4,7 @@

 pwd=`pwd`

 set -x

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/config.myr

+	as -g -o mbld/cpufeatures.o mbld/cpufeatures+x64.s

 	as -g -o lib/thread/start.o lib/thread/start+osx-x64.s

 	as -g -o lib/thread/atomic-impl.o lib/thread/atomic-impl+x64.s

 	as -g -o lib/std/getbp.o lib/std/getbp+posixy-x64.s

@@ -140,5 +141,5 @@

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/test.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/deps.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/main.myr

-	ld -pagezero_size 0x100000000 -macosx_version_min 10.6 -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys -macosx_version_min 10.6

+	ld -pagezero_size 0x100000000 -macosx_version_min 10.6 -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/cpufeatures.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys -macosx_version_min 10.6

 true

--- a/mk/bootstrap/bootstrap+FreeBSD-amd64.sh

+++ b/mk/bootstrap/bootstrap+FreeBSD-amd64.sh

@@ -4,6 +4,7 @@

 pwd=`pwd`

 set -x

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/config.myr

+	as -g -o mbld/cpufeatures.o mbld/cpufeatures+x64.s

 	as -g -o lib/thread/exit.o lib/thread/exit+freebsd-x64.s

 	as -g -o lib/thread/atomic-impl.o lib/thread/atomic-impl+x64.s

 	as -g -o lib/std/getbp.o lib/std/getbp+posixy-x64.s

@@ -141,5 +142,5 @@

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/test.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/deps.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/main.myr

-	ld --gc-sections -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys

+	ld --gc-sections -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/cpufeatures.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys

 true

--- a/mk/bootstrap/bootstrap+Linux-x86_64.sh

+++ b/mk/bootstrap/bootstrap+Linux-x86_64.sh

@@ -4,6 +4,7 @@

 pwd=`pwd`

 set -x

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/config.myr

+	as -g -o mbld/cpufeatures.o mbld/cpufeatures+x64.s

 	as -g -o lib/thread/exit.o lib/thread/exit+linux-x64.s

 	as -g -o lib/thread/atomic-impl.o lib/thread/atomic-impl+x64.s

 	as -g -o lib/std/getbp.o lib/std/getbp+posixy-x64.s

@@ -141,5 +142,5 @@

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/test.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/deps.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/main.myr

-	ld --gc-sections -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys

+	ld --gc-sections -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/cpufeatures.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys

 true

--- a/mk/bootstrap/bootstrap+NetBSD-amd64.sh

+++ b/mk/bootstrap/bootstrap+NetBSD-amd64.sh

@@ -4,6 +4,7 @@

 pwd=`pwd`

 set -x

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/config.myr

+	as -g -o mbld/cpufeatures.o mbld/cpufeatures+x64.s

 	as -g -o lib/thread/atomic-impl.o lib/thread/atomic-impl+x64.s

 	as -g -o lib/std/getbp.o lib/std/getbp+posixy-x64.s

 	$pwd/6/6m -I . -I lib/sys lib/std/option.myr

@@ -139,5 +140,5 @@

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/test.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/deps.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/main.myr

-	ld -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys

+	ld -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/cpufeatures.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys

 true

--- a/mk/bootstrap/bootstrap+OpenBSD-amd64.sh

+++ b/mk/bootstrap/bootstrap+OpenBSD-amd64.sh

@@ -4,6 +4,7 @@

 pwd=`pwd`

 set -x

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/config.myr

+	as -g -o mbld/cpufeatures.o mbld/cpufeatures+x64.s

 	as -g -o lib/thread/exit.o lib/thread/exit+openbsd-x64.s

 	as -g -o lib/thread/atomic-impl.o lib/thread/atomic-impl+x64.s

 	as -g -o lib/std/getbp.o lib/std/getbp+posixy-x64.s

@@ -141,5 +142,5 @@

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/test.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/deps.myr

 	$pwd/6/6m -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/main.myr

-	ld -nopie --gc-sections -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys

+	ld -nopie --gc-sections -o mbld/mbld $pwd/rt/_myrrt.o mbld/deps.o mbld/main.o mbld/util.o mbld/cpufeatures+x64.s mbld/libs.o mbld/syssel.o mbld/config.o mbld/opts.o mbld/subtest.o mbld/types.o mbld/test.o mbld/install.o mbld/parse.o mbld/build.o -Llib/thread -lthread -Llib/bio -lbio -Llib/regex -lregex -Llib/std -lstd -Llib/sys -lsys

 true

--- a/mk/bootstrap/bootstrap+Plan9-amd64.sh

+++ b/mk/bootstrap/bootstrap+Plan9-amd64.sh

@@ -4,6 +4,7 @@

 pwd=`pwd`

 set -x

 	$pwd/6/6.out -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/config+plan9-x64.myr

+	6a -o mbld/cpufeatures.6 mbld/cpufeatures+x64.s

 	6a -o lib/thread/atomic-impl.6 lib/thread/atomic-impl+plan9-x64.s

 	6a -o lib/std/getbp.6 lib/std/getbp+plan9-x64.s

 	$pwd/6/6.out -I lib/sys lib/std/errno+plan9.myr

@@ -138,5 +139,5 @@

 	$pwd/6/6.out -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/test.myr

 	$pwd/6/6.out -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/deps.myr

 	$pwd/6/6.out -I lib/sys -I lib/std -I lib/bio -I lib/regex -I lib/thread mbld/main.myr

-	6l -l -o mbld/mbld $pwd/rt/_myrrt.6 mbld/deps.6 mbld/main.6 mbld/util.6 mbld/libs.6 mbld/syssel.6 mbld/config.6 mbld/opts.6 mbld/subtest.6 mbld/types.6 mbld/test.6 mbld/install.6 mbld/parse.6 mbld/build.6 lib/thread/libthread.a lib/bio/libbio.a lib/regex/libregex.a lib/std/libstd.a lib/sys/libsys.a

+	6l -l -o mbld/mbld $pwd/rt/_myrrt.6 mbld/deps.6 mbld/main.6 mbld/util.6 mbld/cpufeatures.6 mbld/libs.6 mbld/syssel.6 mbld/config.6 mbld/opts.6 mbld/subtest.6 mbld/types.6 mbld/test.6 mbld/install.6 mbld/parse.6 mbld/build.6 lib/thread/libthread.a lib/bio/libbio.a lib/regex/libregex.a lib/std/libstd.a lib/sys/libsys.a

 true