[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Libffcall] Return small structs in registers for powerpc on openbsd
|
From: |
Josh Elsasser |
|
Subject: |
Re: [Libffcall] Return small structs in registers for powerpc on openbsd |
|
Date: |
Mon, 28 Oct 2019 13:35:14 -0700 |
|
User-agent: |
Mutt/1.5.22 (2013-10-16) |
On Thu, Oct 17, 2019 at 07:16:38PM -0700, Josh Elsasser wrote:
> On Thu, Oct 17, 2019 at 11:29:05PM +0200, Bruno Haible wrote:
> > Hi Josh,
> >
> > Thank you for the patch.
> > It will take me a while to make sure your patch is good, and include it.
>
> Of course, thanks for taking the time to look.
>
> > Maybe you can help me with some infos:
> > - Which OpenBSD version and which qemu-system-ppc or qemu-system-ppc64
> > options do you recommend for running it in a VM?
>
> I've never tried that before, I'll poke around and see if I can find a
> working configuration.
I was unable to run openbsd/macppc under qemu, but had some luck with
netbsd 7.1.1 and qemu 2.8. After building qemu with the ppc-softmmu
target and downloading
https://cdn.netbsd.org/pub/NetBSD/NetBSD-7.1.1/images/NetBSD-7.1.1-macppc.iso
I created the disk image and started qemu:
$ qemu-img create -f qcow2 netbsdmacppc.qcow2 10g
$ qemu-system-ppc -m 512 -M g3beige -hda netbsdmacppc.qcow2 -cdrom
NetBSD-7.1.1-macppc.iso -prom-env boot-device=cd:,ofwboot.xcf -net
nic,model=i82551 -net user
It should boot into the netbsd installer. As per the netbsd install
doc, I chose (S)hell and then partitioned manually: (note C vs. c)
# stty erase '^?'
# pdisk /dev/wd0c
i
C 64 65472 boot Apple_HFS
c 65536 196608 swap b
c 262144 20709376 root a
w
q
# sysinst
Next I followed the menus to install, selecting "Use existing
partitions", installing sets from cdrom, and then configuring the
system. I skipped networking as the installer has no DHCP client.
Then I exited to the shell and finished setting up:
# mount /dev/wd0a /mnt
# chroot /mnt /bin/ksh
# echo dhcp > /etc/ifconfig.fxp0
# dhcpcd -4w
# echo
PKG_PATH=http://cdn.netbsd.org/pub/pkgsrc/packages/NetBSD/powerpc/7.1/All >
/etc/pkg_install.conf
# pkg_add hfsutils
# export PATH="$PATH:/usr/pkg/bin"
# hformat /dev/wd0d
# hcopy /usr/mdec/ofwboot.xcf :
# humount
# mount /dev/cd0a /mnt
# mv /netbsd /netbsd.orig
# gzip -d < /mnt/macppc/binary/kernel/netbsd-GENERIC_MD.gz > /netbsd
# halt -p
Note that I moved the original kernel out of the way and replaced it
with one containing the embedded installer ramdisk. The normal kernel
hung on boot for me and I worked around it by booting the ramdisk
kernel with -a to force the mountroot prompt. You'll have to enter -a
at the Boot: prompt and wd0a at the "root device" prompt. This is
the command I used to boot post-install:
$ qemu-system-ppc -m 512 -M g3beige -hda netbsdmacppc.qcow2 -prom-env
boot-command='boot hd:,ofwboot.xcf -a' -net nic,model=i82551 -net
user,hostfwd=tcp::2222-:22
Good luck, hopefully my success is reproducable.
> > - If gcc 4.1.2 needs patches, to make it compile, maybe a different
> > version is simpler to use. How about using gcc 4.2.4 instead?
>
> I'll give 4.2.4 a try.
>
> > - Do you have a formal documentation of the "secure plt ABI"?
>
> This is the reference I was reading:
>
> https://www.polyomino.org.uk/publications/2011/Power-Arch-32-bit-ABI-supp-1.0-Unified.pdf
>
> I also found this helpful:
>
> http://www.sourceware.org/ml/binutils/2005-05/txt00011.txt
I've dropped the secure plt bits from this version of the diff as
they're not needed for netbsd. I'll send another diff after
openbsd/netbsd small struct return is taken care of.
> > > Replace the unused powerpc small-struct-copying code with hppa's,
> > > which copies structs of all sizes.
> >
> > Oh really? I thought that the hppa code for small structs was so braindead
> > that it would remain the only OS that needs this code. The code for
> > m68k, s390, sparc is simpler.
>
> The existing code doesn't handle 3, 5, 6, or 7 byte structs. It looks
> like riscv uses the hppa code too. I wonder if the compiler pads
> structs to a full word on those other platforms?
>
> > Bruno
>
Here's an updated patch, with the secure plt bits removed, netbsd
added to the ifdefs with openbsd, and with a couple obvious mistakes
fixed. I think I've figured out how to build a proper release tarball
now, and have checked that it builds and passes all tests for powerpc
linux and netbsd.
diff --git avcall/avcall-internal.h avcall/avcall-internal.h
index 0357a77..3a8d318 100644
--- avcall/avcall-internal.h
+++ avcall/avcall-internal.h
@@ -191,7 +191,7 @@ typedef int __av_alist_verify[2*(__AV_ALIST_SIZE_BOUND -
(int)sizeof(__av_alist)
#define __av_start_struct3(LIST) \
((LIST).flags |= __AV_REGISTER_STRUCT_RETURN, 0)
#endif
-#if (defined(__i386__) && !defined(_WIN32)) || defined(__m68k__) ||
(defined(__powerpc__) && !defined(__powerpc64__)) || (defined(__s390__) &&
!defined(__s390x__))
+#if (defined(__i386__) && !defined(_WIN32)) || defined(__m68k__) ||
(defined(__powerpc__) && !defined(__powerpc64__) && !(defined(__OpenBSD__) ||
defined(__NetBSD__))) || (defined(__s390__) && !defined(__s390x__))
#define __av_reg_struct_return(LIST,TYPE_SIZE,TYPE_SPLITTABLE) \
((TYPE_SIZE) == 1 || (TYPE_SIZE) == 2 || (TYPE_SIZE) == 4 \
|| ((TYPE_SIZE) == 8 && (TYPE_SPLITTABLE) \
@@ -247,6 +247,15 @@ typedef int __av_alist_verify[2*(__AV_ALIST_SIZE_BOUND -
(int)sizeof(__av_alist)
#define __av_start_struct3(LIST) \
((LIST).flags |= __AV_REGISTER_STRUCT_RETURN, 0)
#endif
+#if defined(__powerpc__) && !defined(__powerpc64__) && (defined(__OpenBSD__)
|| defined(__NetBSD__))
+#define __av_reg_struct_return(LIST,TYPE_SIZE,TYPE_SPLITTABLE) \
+ ((TYPE_SIZE) <= 8)
+/* Turn on __AV_REGISTER_STRUCT_RETURN if __AV_SMALL_STRUCT_RETURN was set
+ * and the struct will actually be returned in registers.
+ */
+#define __av_start_struct3(LIST) \
+ ((LIST).flags |= __AV_REGISTER_STRUCT_RETURN, 0)
+#endif
#if (defined(__powerpc64__) && !defined(__powerpc64_elfv2__)) ||
defined(__s390x__)
#define __av_reg_struct_return(LIST,TYPE_SIZE,TYPE_SPLITTABLE) \
0
diff --git avcall/avcall-powerpc.c avcall/avcall-powerpc.c
index 5d1b6f8..17cbde8 100644
--- avcall/avcall-powerpc.c
+++ avcall/avcall-powerpc.c
@@ -196,19 +196,96 @@ avcall_call(av_alist* list)
} else
if (l->rtype == __AVstruct) {
if (l->flags & __AV_REGISTER_STRUCT_RETURN) {
- if (l->rsize == sizeof(char)) {
- RETURN(char, i);
- } else
- if (l->rsize == sizeof(short)) {
- RETURN(short, i);
- } else
- if (l->rsize == sizeof(int)) {
- RETURN(int, i);
- } else
- if (l->rsize == 2*sizeof(__avword)) {
+ if (l->rsize > 0 && l->rsize <= 8) {
void* raddr = l->raddr;
- ((__avword*)raddr)[0] = i;
- ((__avword*)raddr)[1] = iret2;
+ #if 0 /* Unoptimized */
+ if (l->rsize == 1) {
+ ((unsigned char *)raddr)[0] = (unsigned char)(i);
+ } else
+ if (l->rsize == 2) {
+ ((unsigned char *)raddr)[0] = (unsigned char)(i>>8);
+ ((unsigned char *)raddr)[1] = (unsigned char)(i);
+ } else
+ if (l->rsize == 3) {
+ ((unsigned char *)raddr)[0] = (unsigned char)(i>>16);
+ ((unsigned char *)raddr)[1] = (unsigned char)(i>>8);
+ ((unsigned char *)raddr)[2] = (unsigned char)(i);
+ } else
+ if (l->rsize == 4) {
+ ((unsigned char *)raddr)[0] = (unsigned char)(i>>24);
+ ((unsigned char *)raddr)[1] = (unsigned char)(i>>16);
+ ((unsigned char *)raddr)[2] = (unsigned char)(i>>8);
+ ((unsigned char *)raddr)[3] = (unsigned char)(i);
+ } else
+ if (l->rsize == 5) {
+ ((unsigned char *)raddr)[0] = (unsigned char)(i);
+ ((unsigned char *)raddr)[1] = (unsigned char)(iret2>>24);
+ ((unsigned char *)raddr)[2] = (unsigned char)(iret2>>16);
+ ((unsigned char *)raddr)[3] = (unsigned char)(iret2>>8);
+ ((unsigned char *)raddr)[4] = (unsigned char)(iret2);
+ } else
+ if (l->rsize == 6) {
+ ((unsigned char *)raddr)[0] = (unsigned char)(i>>8);
+ ((unsigned char *)raddr)[1] = (unsigned char)(i);
+ ((unsigned char *)raddr)[2] = (unsigned char)(iret2>>24);
+ ((unsigned char *)raddr)[3] = (unsigned char)(iret2>>16);
+ ((unsigned char *)raddr)[4] = (unsigned char)(iret2>>8);
+ ((unsigned char *)raddr)[5] = (unsigned char)(iret2);
+ } else
+ if (l->rsize == 7) {
+ ((unsigned char *)raddr)[0] = (unsigned char)(i>>16);
+ ((unsigned char *)raddr)[1] = (unsigned char)(i>>8);
+ ((unsigned char *)raddr)[2] = (unsigned char)(i);
+ ((unsigned char *)raddr)[3] = (unsigned char)(iret2>>24);
+ ((unsigned char *)raddr)[4] = (unsigned char)(iret2>>16);
+ ((unsigned char *)raddr)[5] = (unsigned char)(iret2>>8);
+ ((unsigned char *)raddr)[6] = (unsigned char)(iret2);
+ } else
+ if (l->rsize == 8) {
+ ((unsigned char *)raddr)[0] = (unsigned char)(i>>24);
+ ((unsigned char *)raddr)[1] = (unsigned char)(i>>16);
+ ((unsigned char *)raddr)[2] = (unsigned char)(i>>8);
+ ((unsigned char *)raddr)[3] = (unsigned char)(i);
+ ((unsigned char *)raddr)[4] = (unsigned char)(iret2>>24);
+ ((unsigned char *)raddr)[5] = (unsigned char)(iret2>>16);
+ ((unsigned char *)raddr)[6] = (unsigned char)(iret2>>8);
+ ((unsigned char *)raddr)[7] = (unsigned char)(iret2);
+ }
+ #else /* Optimized: fewer conditional jumps, fewer memory accesses */
+ uintptr_t count = l->rsize; /* > 0, ≤ 2*sizeof(__avword) */
+ __avword* wordaddr = (__avword*)((uintptr_t)raddr &
~(uintptr_t)(sizeof(__avword)-1));
+ uintptr_t start_offset = (uintptr_t)raddr &
(uintptr_t)(sizeof(__avword)-1); /* ≥ 0, < sizeof(__avword) */
+ uintptr_t end_offset = start_offset + count; /* > 0, <
3*sizeof(__avword) */
+ if (count <= sizeof(__avword)) {
+ /* Use iret. */
+ if (end_offset <= sizeof(__avword)) {
+ /* 0 < end_offset ≤ sizeof(__avword) */
+ __avword mask0 = ((__avword)2 <<
(sizeof(__avword)*8-start_offset*8-1)) - ((__avword)1 <<
(sizeof(__avword)*8-end_offset*8));
+ wordaddr[0] ^= (wordaddr[0] ^ (i <<
(sizeof(__avword)*8-end_offset*8))) & mask0;
+ } else {
+ /* sizeof(__avword) < end_offset < 2*sizeof(__avword),
start_offset > 0 */
+ __avword mask0 = ((__avword)2 <<
(sizeof(__avword)*8-start_offset*8-1)) - 1;
+ __avword mask1 = - ((__avword)1 <<
(2*sizeof(__avword)*8-end_offset*8));
+ wordaddr[0] ^= (wordaddr[0] ^ (i >>
(end_offset*8-sizeof(__avword)*8))) & mask0;
+ wordaddr[1] ^= (wordaddr[1] ^ (i <<
(2*sizeof(__avword)*8-end_offset*8))) & mask1;
+ }
+ } else {
+ /* Use iret, iret2. */
+ __avword mask0 = ((__avword)2 <<
(sizeof(__avword)*8-start_offset*8-1)) - 1;
+ if (end_offset <= 2*sizeof(__avword)) {
+ /* sizeof(__avword) < end_offset ≤ 2*sizeof(__avword) */
+ __avword mask1 = - ((__avword)1 <<
(2*sizeof(__avword)*8-end_offset*8));
+ wordaddr[0] ^= (wordaddr[0] ^ ((i <<
(2*sizeof(__avword)*8-end_offset*8)) | (iret2 >>
(end_offset*4-sizeof(__avword)*4) >> (end_offset*4-sizeof(__avword)*4)))) &
mask0;
+ wordaddr[1] ^= (wordaddr[1] ^ (iret2 <<
(2*sizeof(__avword)*8-end_offset*8))) & mask1;
+ } else {
+ /* 2*sizeof(__avword) < end_offset < 3*sizeof(__avword),
start_offset > 0 */
+ __avword mask2 = - ((__avword)1 <<
(3*sizeof(__avword)*8-end_offset*8));
+ wordaddr[0] ^= (wordaddr[0] ^ (i >>
(end_offset*8-2*sizeof(__avword)*8))) & mask0;
+ wordaddr[1] = (i << (3*sizeof(__avword)*8-end_offset*8)) | (iret2
>> (end_offset*8-2*sizeof(__avword)*8));
+ wordaddr[2] ^= (wordaddr[2] ^ (iret2 <<
(3*sizeof(__avword)*8-end_offset*8))) & mask2;
+ }
+ }
+ #endif
}
}
}
diff --git avcall/avcall.h avcall/avcall.h
index f429dc8..cfd6492 100644
--- avcall/avcall.h
+++ avcall/avcall.h
@@ -140,7 +140,7 @@ enum __AV_alist_flags
#if defined(__sparc__) && !defined(__sparc64__) && defined(__sun) &&
(defined(__SUNPRO_C) || defined(__SUNPRO_CC)) /* SUNWspro cc or CC */
__AV_SUNPROCC_STRUCT_RETURN,
#else
-#if (defined(__i386__) && (defined(_WIN32) || defined(__CYGWIN__) ||
(defined(__MACH__) && defined(__APPLE__)) || defined(__FreeBSD__) ||
defined(__DragonFly__) || defined(__OpenBSD__))) || defined(__m68k__) ||
defined(__mipsn32__) || defined(__mips64__) || defined(__sparc64__) ||
defined(__hppa__) || defined(__hppa64__) || defined(__arm__) ||
defined(__armhf__) || defined(__arm64__) || defined(__powerpc64_elfv2__) ||
defined(__ia64__) || defined(__x86_64__) || defined(__riscv32__) ||
defined(__riscv64__)
+#if (defined(__i386__) && (defined(_WIN32) || defined(__CYGWIN__) ||
(defined(__MACH__) && defined(__APPLE__)) || defined(__FreeBSD__) ||
defined(__DragonFly__) || defined(__OpenBSD__))) || defined(__m68k__) ||
defined(__mipsn32__) || defined(__mips64__) || defined(__sparc64__) ||
defined(__hppa__) || defined(__hppa64__) || defined(__arm__) ||
defined(__armhf__) || defined(__arm64__) || (defined(__powerpc__) &&
!defined(__powerpc64__) && (defined(__OpenBSD__) || defined(__NetBSD__))) ||
defined(__powerpc64_elfv2__) || defined(__ia64__) || defined(__x86_64__) ||
defined(__riscv32__) || defined(__riscv64__)
__AV_SMALL_STRUCT_RETURN |
#endif
#if defined(__GNUC__) && !((defined(__mipsn32__) || defined(__mips64__)) &&
((__GNUC__ == 3 && __GNUC_MINOR__ >= 4) || (__GNUC__ > 3)))
diff --git callback/vacall_r/vacall_r.h callback/vacall_r/vacall_r.h
index 0e66c2e..b883029 100644
--- callback/vacall_r/vacall_r.h
+++ callback/vacall_r/vacall_r.h
@@ -167,7 +167,7 @@ enum __VA_alist_flags
#if defined(__sparc__) && !defined(__sparc64__) && defined(__sun) &&
(defined(__SUNPRO_C) || defined(__SUNPRO_CC)) /* SUNWspro cc or CC */
__VA_SUNPROCC_STRUCT_RETURN,
#else
-#if (defined(__i386__) && (defined(_WIN32) || defined(__CYGWIN__) ||
(defined(__MACH__) && defined(__APPLE__)) || defined(__FreeBSD__) ||
defined(__DragonFly__) || defined(__OpenBSD__))) || defined(__m68k__) ||
defined(__mipsn32__) || defined(__mips64__) || defined(__sparc64__) ||
defined(__hppa__) || defined(__hppa64__) || defined(__arm__) ||
defined(__armhf__) || defined(__arm64__) || defined(__powerpc64_elfv2__) ||
defined(__ia64__) || defined(__x86_64__) || defined(__riscv32__) ||
defined(__riscv64__)
+#if (defined(__i386__) && (defined(_WIN32) || defined(__CYGWIN__) ||
(defined(__MACH__) && defined(__APPLE__)) || defined(__FreeBSD__) ||
defined(__DragonFly__) || defined(__OpenBSD__))) || defined(__m68k__) ||
defined(__mipsn32__) || defined(__mips64__) || defined(__sparc64__) ||
defined(__hppa__) || defined(__hppa64__) || defined(__arm__) ||
defined(__armhf__) || defined(__arm64__) || (defined(__powerpc__) &&
!defined(__powerpc64__) && (defined(__OpenBSD__) || defined(__NetBSD__))) ||
defined(__powerpc64_elfv2__) || defined(__ia64__) || defined(__x86_64__) ||
defined(__riscv32__) || defined(__riscv64__)
__VA_SMALL_STRUCT_RETURN |
#endif
#if defined(__GNUC__) && !((defined(__mipsn32__) || defined(__mips64__)) &&
((__GNUC__ == 3 && __GNUC_MINOR__ >= 4) || (__GNUC__ > 3)))
diff --git vacall/vacall-internal.h vacall/vacall-internal.h
index e29a6d4..b4d60d0 100644
--- vacall/vacall-internal.h
+++ vacall/vacall-internal.h
@@ -279,7 +279,7 @@ typedef struct vacall_alist
#define __va_start_struct1(LIST,TYPE_SIZE,TYPE_ALIGN,TYPE_SPLITTABLE) \
((LIST)->flags |= __VA_REGISTER_STRUCT_RETURN, 0)
#endif
-#if (defined(__i386__) && !defined(_WIN32)) || defined(__m68k__) ||
(defined(__powerpc__) && !defined(__powerpc64__)) || (defined(__s390__) &&
!defined(__s390x__))
+#if (defined(__i386__) && !defined(_WIN32)) || defined(__m68k__) ||
(defined(__powerpc__) && !defined(__powerpc64__) && !(defined(__OpenBSD__) ||
defined(__NetBSD__))) || (defined(__s390__) && !defined(__s390x__))
#define __va_reg_struct_return(LIST,TYPE_SIZE,TYPE_SPLITTABLE) \
((TYPE_SIZE) == 1 || (TYPE_SIZE) == 2 || (TYPE_SIZE) == 4 \
|| ((TYPE_SIZE) == 8 && (TYPE_SPLITTABLE) \
@@ -355,6 +355,16 @@ typedef struct vacall_alist
&& ((LIST)->flags |= __VA_REGISTER_DOUBLESTRUCT_RETURN),
\
0)
#endif
+#if defined(__powerpc__) && !defined(__powerpc64__) && (defined(__OpenBSD__)
|| defined(__NetBSD__))
+#define __va_reg_struct_return(LIST,TYPE_SIZE,TYPE_SPLITTABLE) \
+ ((TYPE_SIZE) <= 8)
+/* Turn on __VA_REGISTER_STRUCT_RETURN if __VA_SMALL_STRUCT_RETURN was set
+ * and the struct will actually be returned in registers.
+ */
+#define __va_start_struct1(LIST,TYPE_SIZE,TYPE_ALIGN,TYPE_SPLITTABLE) \
+ ((LIST)->flags |= __VA_REGISTER_STRUCT_RETURN, \
+ 0)
+#endif
#if (defined(__powerpc64__) && !defined(__powerpc64_elfv2__)) ||
defined(__s390x__)
#define __va_reg_struct_return(LIST,TYPE_SIZE,TYPE_SPLITTABLE) \
0
diff --git vacall/vacall-powerpc.c vacall/vacall-powerpc.c
index ea3e208..04a790e 100644
--- vacall/vacall-powerpc.c
+++ vacall/vacall-powerpc.c
@@ -155,18 +155,93 @@ vacall_receiver (__vaword word1, __vaword word2, __vaword
word3, __vaword word4,
} else
if (list.rtype == __VAstruct) {
if (list.flags & __VA_REGISTER_STRUCT_RETURN) {
- if (list.rsize == sizeof(char)) {
- iret = *(unsigned char *) list.raddr;
- } else
- if (list.rsize == sizeof(short)) {
- iret = *(unsigned short *) list.raddr;
- } else
- if (list.rsize == sizeof(int)) {
- iret = *(unsigned int *) list.raddr;
- } else
- if (list.rsize == 2*sizeof(__vaword)) {
- iret = ((__vaword *) list.raddr)[0];
- iret2 = ((__vaword *) list.raddr)[1];
+ if (list.rsize > 0 && list.rsize <= 8) {
+ #if 0 /* Unoptimized */
+ if (list.rsize == 1) {
+ iret = ((unsigned char *) list.raddr)[0];
+ } else
+ if (list.rsize == 2) {
+ iret = (((unsigned char *) list.raddr)[0] << 8)
+ | ((unsigned char *) list.raddr)[1];
+ } else
+ if (list.rsize == 3) {
+ iret = (((unsigned char *) list.raddr)[0] << 16)
+ | (((unsigned char *) list.raddr)[1] << 8)
+ | ((unsigned char *) list.raddr)[2];
+ } else
+ if (list.rsize == 4) {
+ iret = (((unsigned char *) list.raddr)[0] << 24)
+ | (((unsigned char *) list.raddr)[1] << 16)
+ | (((unsigned char *) list.raddr)[2] << 8)
+ | ((unsigned char *) list.raddr)[3];
+ } else
+ if (list.rsize == 5) {
+ iret = ((unsigned char *) list.raddr)[0];
+ iret2 = (((unsigned char *) list.raddr)[1] << 24)
+ | (((unsigned char *) list.raddr)[2] << 16)
+ | (((unsigned char *) list.raddr)[3] << 8)
+ | ((unsigned char *) list.raddr)[4];
+ } else
+ if (list.rsize == 6) {
+ iret = (((unsigned char *) list.raddr)[0] << 8)
+ | ((unsigned char *) list.raddr)[1];
+ iret2 = (((unsigned char *) list.raddr)[2] << 24)
+ | (((unsigned char *) list.raddr)[3] << 16)
+ | (((unsigned char *) list.raddr)[4] << 8)
+ | ((unsigned char *) list.raddr)[5];
+ } else
+ if (list.rsize == 7) {
+ iret = (((unsigned char *) list.raddr)[0] << 16)
+ | (((unsigned char *) list.raddr)[1] << 8)
+ | ((unsigned char *) list.raddr)[2];
+ iret2 = (((unsigned char *) list.raddr)[3] << 24)
+ | (((unsigned char *) list.raddr)[4] << 16)
+ | (((unsigned char *) list.raddr)[5] << 8)
+ | ((unsigned char *) list.raddr)[6];
+ } else
+ if (list.rsize == 8) {
+ iret = (((unsigned char *) list.raddr)[0] << 24)
+ | (((unsigned char *) list.raddr)[1] << 16)
+ | (((unsigned char *) list.raddr)[2] << 8)
+ | ((unsigned char *) list.raddr)[3];
+ iret2 = (((unsigned char *) list.raddr)[4] << 24)
+ | (((unsigned char *) list.raddr)[5] << 16)
+ | (((unsigned char *) list.raddr)[6] << 8)
+ | ((unsigned char *) list.raddr)[7];
+ }
+ #else /* Optimized: fewer conditional jumps, fewer memory accesses */
+ uintptr_t count = list.rsize; /* > 0, ≤ 2*sizeof(__vaword) */
+ __vaword* wordaddr = (__vaword*)((uintptr_t)list.raddr &
~(uintptr_t)(sizeof(__vaword)-1));
+ uintptr_t start_offset = (uintptr_t)list.raddr &
(uintptr_t)(sizeof(__vaword)-1); /* ≥ 0, < sizeof(__vaword) */
+ uintptr_t end_offset = start_offset + count; /* > 0, <
3*sizeof(__vaword) */
+ if (count <= sizeof(__vaword)) {
+ /* Assign iret. */
+ __vaword mask0 = ((__vaword)2 <<
(sizeof(__vaword)*8-start_offset*8-1)) - 1;
+ if (end_offset <= sizeof(__vaword)) {
+ /* 0 < end_offset ≤ sizeof(__vaword) */
+ iret = (wordaddr[0] & mask0) >> (sizeof(__vaword)*8-end_offset*8);
+ } else {
+ /* sizeof(__vaword) < end_offset < 2*sizeof(__vaword),
start_offset > 0 */
+ iret = ((wordaddr[0] & mask0) << (end_offset*8-sizeof(__vaword)*8))
+ | (wordaddr[1] >> (2*sizeof(__vaword)*8-end_offset*8));
+ }
+ } else {
+ /* Assign iret, iret2. */
+ __vaword mask0 = ((__vaword)2 <<
(sizeof(__vaword)*8-start_offset*8-1)) - 1;
+ if (end_offset <= 2*sizeof(__vaword)) {
+ /* sizeof(__vaword) < end_offset ≤ 2*sizeof(__vaword) */
+ iret = (wordaddr[0] & mask0) >>
(2*sizeof(__vaword)*8-end_offset*8);
+ iret2 = ((wordaddr[0] & mask0) <<
(end_offset*4-sizeof(__vaword)*4) << (end_offset*4-sizeof(__vaword)*4))
+ | (wordaddr[1] >> (2*sizeof(__vaword)*8-end_offset*8));
+ } else {
+ /* 2*sizeof(__vaword) < end_offset < 3*sizeof(__vaword),
start_offset > 0 */
+ iret = ((wordaddr[0] & mask0) <<
(end_offset*8-2*sizeof(__vaword)*8))
+ | (wordaddr[1] >> (3*sizeof(__vaword)*8-end_offset*8));
+ iret2 = (wordaddr[1] << (end_offset*8-2*sizeof(__vaword)*8))
+ | (wordaddr[2] >> (3*sizeof(__vaword)*8-end_offset*8));
+ }
+ }
+ #endif
}
}
}
diff --git vacall/vacall.h vacall/vacall.h
index 9485356..4946916 100644
--- vacall/vacall.h
+++ vacall/vacall.h
@@ -128,7 +128,7 @@ enum __VA_alist_flags
#if defined(__sparc__) && !defined(__sparc64__) && defined(__sun) &&
(defined(__SUNPRO_C) || defined(__SUNPRO_CC)) /* SUNWspro cc or CC */
__VA_SUNPROCC_STRUCT_RETURN,
#else
-#if (defined(__i386__) && (defined(_WIN32) || defined(__CYGWIN__) ||
(defined(__MACH__) && defined(__APPLE__)) || defined(__FreeBSD__) ||
defined(__DragonFly__) || defined(__OpenBSD__))) || defined(__m68k__) ||
defined(__mipsn32__) || defined(__mips64__) || defined(__sparc64__) ||
defined(__hppa__) || defined(__hppa64__) || defined(__arm__) ||
defined(__armhf__) || defined(__arm64__) || defined(__powerpc64_elfv2__) ||
defined(__ia64__) || defined(__x86_64__) || defined(__riscv32__) ||
defined(__riscv64__)
+#if (defined(__i386__) && (defined(_WIN32) || defined(__CYGWIN__) ||
(defined(__MACH__) && defined(__APPLE__)) || defined(__FreeBSD__) ||
defined(__DragonFly__) || defined(__OpenBSD__))) || defined(__m68k__) ||
defined(__mipsn32__) || defined(__mips64__) || defined(__sparc64__) ||
defined(__hppa__) || defined(__hppa64__) || defined(__arm__) ||
defined(__armhf__) || defined(__arm64__) || (defined(__powerpc__) &&
!defined(__powerpc64__) && (defined(__OpenBSD__) || defined(__NetBSD__))) ||
defined(__powerpc64_elfv2__) || defined(__ia64__) || defined(__x86_64__) ||
defined(__riscv32__) || defined(__riscv64__)
__VA_SMALL_STRUCT_RETURN |
#endif
#if defined(__GNUC__) && !((defined(__mipsn32__) || defined(__mips64__)) &&
((__GNUC__ == 3 && __GNUC_MINOR__ >= 4) || (__GNUC__ > 3)))