Re: [libav-devel] [PATCH] checkasm: vp9dsp: Benchmark the dc-only version of idct_idct separately
Hi, On Thu, Nov 17, 2016 at 7:37 AM, Martin Storsjöwrote: > On Thu, 17 Nov 2016, Ronald S. Bultje wrote: > > Hi, >> >> On Mon, Nov 14, 2016 at 4:46 PM, Martin Storsjö wrote: >> >> The dc-only mode is already checked to work correctly above, but this >>> allows benchmarking this mode for performance tuning, and allows making >>> sure that it actually is correctly hooked up. >>> --- >>> tests/checkasm/vp9dsp.c | 6 ++ >>> 1 file changed, 6 insertions(+) >>> >>> diff --git a/tests/checkasm/vp9dsp.c b/tests/checkasm/vp9dsp.c >>> index 690e0cf..b9d1c73 100644 >>> --- a/tests/checkasm/vp9dsp.c >>> +++ b/tests/checkasm/vp9dsp.c >>> @@ -297,6 +297,12 @@ static void check_itxfm(void) >>> } >>> bench_new(dst, sz * SIZEOF_PIXEL, coef, sz * sz); >>> } >>> +if (txtp == 0 && tx != 4) { >>> +if (check_func(dsp.itxfm_add[tx][txtp], >>> "vp9_inv_%s_%dx%d_dc_add", >>> + txtp_types[txtp], sz, sz)) { >>> +bench_new(dst, sz * SIZEOF_PIXEL, coef, 1); >>> +} >>> +} >>> } >>> } >>> report("itxfm"); >>> -- >>> 2.7.4 >>> >> >> >> I had a different local modification that allows tuning all the relevant >> sub-IDCTs, basically re-arranging the loops so check_func is inside the >> sub-IDCT loop and we bench each sub-IDCT separately. That's more generic >> and probably more useful. >> > > Right, that's probably more useful. Would you care to finish that > modification to get it upstreamed in either project? Sure, no problem. Ronald ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH] checkasm: vp9dsp: Benchmark the dc-only version of idct_idct separately
On Thu, 17 Nov 2016, Ronald S. Bultje wrote: Hi, On Mon, Nov 14, 2016 at 4:46 PM, Martin Storsjöwrote: The dc-only mode is already checked to work correctly above, but this allows benchmarking this mode for performance tuning, and allows making sure that it actually is correctly hooked up. --- tests/checkasm/vp9dsp.c | 6 ++ 1 file changed, 6 insertions(+) diff --git a/tests/checkasm/vp9dsp.c b/tests/checkasm/vp9dsp.c index 690e0cf..b9d1c73 100644 --- a/tests/checkasm/vp9dsp.c +++ b/tests/checkasm/vp9dsp.c @@ -297,6 +297,12 @@ static void check_itxfm(void) } bench_new(dst, sz * SIZEOF_PIXEL, coef, sz * sz); } +if (txtp == 0 && tx != 4) { +if (check_func(dsp.itxfm_add[tx][txtp], "vp9_inv_%s_%dx%d_dc_add", + txtp_types[txtp], sz, sz)) { +bench_new(dst, sz * SIZEOF_PIXEL, coef, 1); +} +} } } report("itxfm"); -- 2.7.4 I had a different local modification that allows tuning all the relevant sub-IDCTs, basically re-arranging the loops so check_func is inside the sub-IDCT loop and we bench each sub-IDCT separately. That's more generic and probably more useful. Right, that's probably more useful. Would you care to finish that modification to get it upstreamed in either project? // Martin ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH] checkasm: vp9dsp: Benchmark the dc-only version of idct_idct separately
Hi, On Mon, Nov 14, 2016 at 4:46 PM, Martin Storsjöwrote: > The dc-only mode is already checked to work correctly above, but this > allows benchmarking this mode for performance tuning, and allows making > sure that it actually is correctly hooked up. > --- > tests/checkasm/vp9dsp.c | 6 ++ > 1 file changed, 6 insertions(+) > > diff --git a/tests/checkasm/vp9dsp.c b/tests/checkasm/vp9dsp.c > index 690e0cf..b9d1c73 100644 > --- a/tests/checkasm/vp9dsp.c > +++ b/tests/checkasm/vp9dsp.c > @@ -297,6 +297,12 @@ static void check_itxfm(void) > } > bench_new(dst, sz * SIZEOF_PIXEL, coef, sz * sz); > } > +if (txtp == 0 && tx != 4) { > +if (check_func(dsp.itxfm_add[tx][txtp], > "vp9_inv_%s_%dx%d_dc_add", > + txtp_types[txtp], sz, sz)) { > +bench_new(dst, sz * SIZEOF_PIXEL, coef, 1); > +} > +} > } > } > report("itxfm"); > -- > 2.7.4 I had a different local modification that allows tuning all the relevant sub-IDCTs, basically re-arranging the loops so check_func is inside the sub-IDCT loop and we bench each sub-IDCT separately. That's more generic and probably more useful. Ronald ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH] checkasm: vp9dsp: Benchmark the dc-only version of idct_idct separately
On 2016-11-14 23:46:14 +0200, Martin Storsjö wrote: > The dc-only mode is already checked to work correctly above, but this > allows benchmarking this mode for performance tuning, and allows making > sure that it actually is correctly hooked up. > --- > tests/checkasm/vp9dsp.c | 6 ++ > 1 file changed, 6 insertions(+) > > diff --git a/tests/checkasm/vp9dsp.c b/tests/checkasm/vp9dsp.c > index 690e0cf..b9d1c73 100644 > --- a/tests/checkasm/vp9dsp.c > +++ b/tests/checkasm/vp9dsp.c > @@ -297,6 +297,12 @@ static void check_itxfm(void) > } > bench_new(dst, sz * SIZEOF_PIXEL, coef, sz * sz); > } > +if (txtp == 0 && tx != 4) { > +if (check_func(dsp.itxfm_add[tx][txtp], > "vp9_inv_%s_%dx%d_dc_add", > + txtp_types[txtp], sz, sz)) { > +bench_new(dst, sz * SIZEOF_PIXEL, coef, 1); > +} > +} > } > } > report("itxfm"); ok Janne ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH] checkasm: vp9dsp: Benchmark the dc-only version of idct_idct separately
On 14/11/2016 22:46, Martin Storsjö wrote: > The dc-only mode is already checked to work correctly above, but this > allows benchmarking this mode for performance tuning, and allows making > sure that it actually is correctly hooked up. > --- > tests/checkasm/vp9dsp.c | 6 ++ > 1 file changed, 6 insertions(+) > > diff --git a/tests/checkasm/vp9dsp.c b/tests/checkasm/vp9dsp.c > index 690e0cf..b9d1c73 100644 > --- a/tests/checkasm/vp9dsp.c > +++ b/tests/checkasm/vp9dsp.c > @@ -297,6 +297,12 @@ static void check_itxfm(void) > } > bench_new(dst, sz * SIZEOF_PIXEL, coef, sz * sz); > } > +if (txtp == 0 && tx != 4) { > +if (check_func(dsp.itxfm_add[tx][txtp], > "vp9_inv_%s_%dx%d_dc_add", > + txtp_types[txtp], sz, sz)) { > +bench_new(dst, sz * SIZEOF_PIXEL, coef, 1); > +} > +} > } > } > report("itxfm"); > Probably Ok. ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
[libav-devel] [PATCH] checkasm: vp9dsp: Benchmark the dc-only version of idct_idct separately
The dc-only mode is already checked to work correctly above, but this allows benchmarking this mode for performance tuning, and allows making sure that it actually is correctly hooked up. --- tests/checkasm/vp9dsp.c | 6 ++ 1 file changed, 6 insertions(+) diff --git a/tests/checkasm/vp9dsp.c b/tests/checkasm/vp9dsp.c index 690e0cf..b9d1c73 100644 --- a/tests/checkasm/vp9dsp.c +++ b/tests/checkasm/vp9dsp.c @@ -297,6 +297,12 @@ static void check_itxfm(void) } bench_new(dst, sz * SIZEOF_PIXEL, coef, sz * sz); } +if (txtp == 0 && tx != 4) { +if (check_func(dsp.itxfm_add[tx][txtp], "vp9_inv_%s_%dx%d_dc_add", + txtp_types[txtp], sz, sz)) { +bench_new(dst, sz * SIZEOF_PIXEL, coef, 1); +} +} } } report("itxfm"); -- 2.7.4 ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel