| Message ID | 20220204201621.20125-1-ffmpeg@tmm1.net |
|---|---|
| State | New |
| Headers | show |
| Series | [FFmpeg-devel] avutil: use getauxval on linux/android for CPU capabilities | expand |
| Context | Check | Description |
|---|---|---|
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
| andriy/make_ppc | success | Make finished |
| andriy/make_fate_ppc | success | Make fate finished |
| andriy/make_aarch64_jetson | success | Make finished |
| andriy/make_fate_aarch64_jetson | success | Make fate finished |
| andriy/make_armv7_RPi4 | success | Make finished |
| andriy/make_fate_armv7_RPi4 | success | Make fate finished |
On Fri, 4 Feb 2022, Aman Karmani wrote: > From: Aman Karmani <aman@tmm1.net> > > fixes #6578 > > Signed-off-by: Aman Karmani <aman@tmm1.net> > --- > libavutil/arm/cpu.c | 17 ++++++++++++++--- > 1 file changed, 14 insertions(+), 3 deletions(-) It would be good if the commit message actually explained the upsides to doing this. It's also important to point out that this function didn't use to exist in all Android versions (hence the /proc/cpuinfo parsing). The point when the function appeared (Android 4.4, API 20, according to the referenced trac issue) is kinda far in the past today, but I'm not sure if strictly all users of the library have stopped supporting older versions still - at least for some use cases I've heard of recently, Android 4.1 was still supported. If coupled with a configure check for the function, so that users explicitly targeting an older version that lack it wouldn't get it, I guess this could be more easily acceptable. // Martin
On Fri, Feb 4, 2022 at 2:20 PM Martin Storsjö <martin@martin.st> wrote: > On Fri, 4 Feb 2022, Aman Karmani wrote: > > > From: Aman Karmani <aman@tmm1.net> > > > > fixes #6578 > > > > Signed-off-by: Aman Karmani <aman@tmm1.net> > > --- > > libavutil/arm/cpu.c | 17 ++++++++++++++--- > > 1 file changed, 14 insertions(+), 3 deletions(-) > > It would be good if the commit message actually explained the upsides to > doing this. > > It's also important to point out that this function didn't use to exist in > all Android versions (hence the /proc/cpuinfo parsing). The point when the > function appeared (Android 4.4, API 20, according to the referenced trac > issue) is kinda far in the past today, but I'm not sure if strictly all > users of the library have stopped supporting older versions still - at > least for some use cases I've heard of recently, Android 4.1 was still > supported. > > If coupled with a configure check for the function, so that users > explicitly targeting an older version that lack it wouldn't get it, I > guess this could be more easily acceptable. Thanks for the review. I discovered some older Linux arm toolchains (< glibc 2.16) are also missing the function, so a configure check seems prudent. Sent v2 patch with the requested changes. > > // Martin > >
diff --git a/libavutil/arm/cpu.c b/libavutil/arm/cpu.c index 81e85e2525..983e5a4d83 100644 --- a/libavutil/arm/cpu.c +++ b/libavutil/arm/cpu.c @@ -36,6 +36,7 @@ #include <stdint.h> #include <stdio.h> #include <string.h> +#include <sys/auxv.h> #include "libavutil/avstring.h" #define AT_HWCAP 16 @@ -48,6 +49,15 @@ #define HWCAP_VFPv3 (1 << 13) #define HWCAP_TLS (1 << 15) +static int get_auxval(uint32_t *hwcap) +{ + unsigned long ret = getauxval(AT_HWCAP); + if (ret == 0) + return -1; + *hwcap = ret; + return 0; +} + static int get_hwcap(uint32_t *hwcap) { struct { uint32_t a_type; uint32_t a_val; } auxv; @@ -106,9 +116,10 @@ int ff_get_cpu_flags_arm(void) int flags = CORE_CPU_FLAGS; uint32_t hwcap; - if (get_hwcap(&hwcap) < 0) - if (get_cpuinfo(&hwcap) < 0) - return flags; + if (get_auxval(&hwcap) < 0) + if (get_hwcap(&hwcap) < 0) + if (get_cpuinfo(&hwcap) < 0) + return flags; #define check_cap(cap, flag) do { \ if (hwcap & HWCAP_ ## cap) \