On Fri, Jun 23, 2017 at 05:01:35PM +0200, Matthieu Bouron wrote:
> On Thu, Jun 22, 2017 at 12:53:52PM -0300, James Almer wrote:
> > On 6/22/2017 9:56 AM, Matthieu Bouron wrote:
> > > ---
> > >
> > > The following patchset applies on top of Clément's aacpsdsp patchset.
> > >
> > > ---
> > > tests/checkasm/Makefile | 3 +-
> > > tests/checkasm/checkasm.c | 1 +
> > > tests/checkasm/checkasm.h | 1 +
> > > tests/checkasm/sbrdsp.c | 297
> > > ++++++++++++++++++++++++++++++++++++++++++++++
> > > 4 files changed, 301 insertions(+), 1 deletion(-)
> > > create mode 100644 tests/checkasm/sbrdsp.c
> > >
> > > diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
> > > index 638e811931..60e80ab738 100644
> > > --- a/tests/checkasm/Makefile
> > > +++ b/tests/checkasm/Makefile
> > > @@ -13,7 +13,8 @@ AVCODECOBJS-$(CONFIG_VP8DSP) += vp8dsp.o
> > > AVCODECOBJS-$(CONFIG_VIDEODSP) += videodsp.o
> > >
> > > # decoders/encoders
> > > -AVCODECOBJS-$(CONFIG_AAC_DECODER) += aacpsdsp.o
> > > +AVCODECOBJS-$(CONFIG_AAC_DECODER) += aacpsdsp.o \
> > > + sbrdsp.o
> > > AVCODECOBJS-$(CONFIG_ALAC_DECODER) += alacdsp.o
> > > AVCODECOBJS-$(CONFIG_DCA_DECODER) += synth_filter.o
> > > AVCODECOBJS-$(CONFIG_JPEG2000_DECODER) += jpeg2000dsp.o
> > > diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
> > > index e66744b162..29f201b1b3 100644
> > > --- a/tests/checkasm/checkasm.c
> > > +++ b/tests/checkasm/checkasm.c
> > > @@ -67,6 +67,7 @@ static const struct {
> > > #if CONFIG_AVCODEC
> > > #if CONFIG_AAC_DECODER
> > > { "aacpsdsp", checkasm_check_aacpsdsp },
> > > + { "sbrdsp", checkasm_check_sbrdsp },
> > > #endif
> > > #if CONFIG_ALAC_DECODER
> > > { "alacdsp", checkasm_check_alacdsp },
> > > diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
> > > index dfb0ce561c..fa51e71e4b 100644
> > > --- a/tests/checkasm/checkasm.h
> > > +++ b/tests/checkasm/checkasm.h
> > > @@ -50,6 +50,7 @@ void checkasm_check_hevc_idct(void);
> > > void checkasm_check_jpeg2000dsp(void);
> > > void checkasm_check_llviddsp(void);
> > > void checkasm_check_pixblockdsp(void);
> > > +void checkasm_check_sbrdsp(void);
> > > void checkasm_check_synth_filter(void);
> > > void checkasm_check_v210enc(void);
> > > void checkasm_check_vp8dsp(void);
> > > diff --git a/tests/checkasm/sbrdsp.c b/tests/checkasm/sbrdsp.c
> > > new file mode 100644
> > > index 0000000000..8333510c6b
> > > --- /dev/null
> > > +++ b/tests/checkasm/sbrdsp.c
> > > @@ -0,0 +1,297 @@
> > > +/*
> > > + * This file is part of FFmpeg.
> > > + *
> > > + * FFmpeg is free software; you can redistribute it and/or modify
> > > + * it under the terms of the GNU General Public License as published by
> > > + * the Free Software Foundation; either version 2 of the License, or
> > > + * (at your option) any later version.
> > > + *
> > > + * FFmpeg is distributed in the hope that it will be useful,
> > > + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> > > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > > + * GNU General Public License for more details.
> > > + *
> > > + * You should have received a copy of the GNU General Public License
> > > along
> > > + * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
> > > + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
> > > + */
> > > +
> > > +#include "libavcodec/sbrdsp.h"
> > > +
> > > +#include "checkasm.h"
> > > +
> > > +#define randomize(buf, len) do { \
> > > + int i; \
> > > + for (i = 0; i < len; i++) { \
> > > + const INTFLOAT f = (INTFLOAT)rnd() / UINT_MAX; \
> > > + (buf)[i] = f; \
> > > + } \
> > > +} while (0)
> > > +
> > > +#define EPS 0.0001
> > > +
> > > +static void test_sum64x5(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [64 + 256]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [64 + 256]);
> > > +
> > > + declare_func(void, INTFLOAT *z);
> > > +
> > > + randomize((INTFLOAT *)dst0, 64 + 256);
> > > + memcpy(dst1, dst0, (64 + 256) * sizeof(INTFLOAT));
> > > + call_ref(dst0);
> > > + call_new(dst1);
> > > + if (!float_near_abs_eps_array(dst0, dst1, EPS, 64 + 256))
> > > + fail();
> > > + bench_new(dst1);
> > > +}
> > > +
> > > +static void test_sum_square(void)
> > > +{
> > > + INTFLOAT res0;
> > > + INTFLOAT res1;
> > > + LOCAL_ALIGNED_16(INTFLOAT, src, [256], [2]);
> > > +
> > > + declare_func(INTFLOAT, INTFLOAT (*x)[2], int n);
> > > +
> > > + randomize((INTFLOAT *)src, 256 * 2);
> > > + res0 = call_ref(src, 256);
> > > + res1 = call_new(src, 256);
> >
> > This one is failing on x86 because the second argument has garbage in
> > the upper 32 bits of the grp.
> >
> > The solution is to either change n from int to ptrdiff_t, or zero the
> > upper bits of n in ff_sbr_sum_square_sse(), which can be done implicitly
> > in the first instruction.
> >
> > > + if (!float_near_abs_eps(res0, res1, EPS))
> > > + fail();
> > > + bench_new(src, 256);
> > > +}
> > > +
> > > +static void test_neg_odd_64(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [64]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [64]);
> > > +
> > > + declare_func(void, INTFLOAT *x);
> > > +
> > > + randomize((INTFLOAT *)dst0, 64);
> > > + memcpy(dst1, dst0, (64) * sizeof(INTFLOAT));
> > > + call_ref(dst0);
> > > + call_new(dst1);
> > > + if (!float_near_abs_eps_array(dst0, dst1, EPS, 64))
> > > + fail();
> > > + bench_new(dst1);
> > > +}
> > > +
> > > +static void test_qmf_pre_shuffle(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [128]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [128]);
> > > +
> > > + declare_func(void, INTFLOAT *z);
> > > +
> > > + randomize((INTFLOAT *)dst0, 128);
> > > + memcpy(dst1, dst0, (128) * sizeof(INTFLOAT));
> > > + call_ref(dst0);
> > > + call_new(dst1);
> > > + if (!float_near_abs_eps_array(dst0, dst1, EPS, 128))
> > > + fail();
> > > + bench_new(dst1);
> > > +}
> > > +
> > > +static void test_qmf_post_shuffle(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, src, [64]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [32], [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [32], [2]);
> > > +
> > > + declare_func(void, INTFLOAT W[32][2], const INTFLOAT *z);
> > > +
> > > + randomize((INTFLOAT *)src, 64);
> > > + call_ref(dst0, src);
> > > + call_new(dst1, src);
> > > + if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1,
> > > EPS, 64))
> > > + fail();
> > > + bench_new(dst1, src);
> > > +}
> > > +
> > > +static void test_qmf_deint_neg(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, src, [64]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [64]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [64]);
> > > +
> > > + declare_func(void, INTFLOAT *v, const INTFLOAT *src);
> > > +
> > > + randomize((INTFLOAT *)src, 64);
> > > + call_ref(dst0, src);
> > > + call_new(dst1, src);
> > > + if (!float_near_abs_eps_array(dst0, dst1, EPS, 64))
> > > + fail();
> > > + bench_new(dst1, src);
> > > +}
> > > +
> > > +static void test_qmf_deint_bfly(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, src0, [64]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, src1, [64]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [128]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [128]);
> > > +
> > > + declare_func(void, INTFLOAT *v, const INTFLOAT *src0, const INTFLOAT
> > > *src1);
> > > +
> > > + memset(dst0, 0, 128 * sizeof(INTFLOAT));
> > > + memset(dst1, 0, 128 * sizeof(INTFLOAT));
> > > +
> > > + randomize((INTFLOAT *)src0, 64);
> > > + randomize((INTFLOAT *)src1, 64);
> > > + call_ref(dst0, src0, src1);
> > > + call_new(dst1, src0, src1);
> > > + if (!float_near_abs_eps_array(dst0, dst1, EPS, 128))
> > > + fail();
> > > + bench_new(dst1, src0, src1);
> > > +}
> > > +
> > > +static void test_autocorrelate(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, src, [40], [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [3], [2][2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [3], [2][2]);
> > > +
> > > + declare_func(void, const INTFLOAT x[40][2], INTFLOAT phi[3][2][2]);
> > > +
> > > + memset(dst0, 0, 3 * 2 * 2 * sizeof(INTFLOAT));
> > > + memset(dst1, 0, 3 * 2 * 2 * sizeof(INTFLOAT));
> > > +
> > > + randomize((INTFLOAT *)src, 80);
> > > + call_ref(src, dst0);
> > > + call_new(src, dst1);
> > > + if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1,
> > > EPS, 3 * 2 * 2))
> > > + fail();
> > > + bench_new(src, dst1);
> > > +}
> > > +
> > > +static void test_hf_gen(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, low, [128], [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, alpha0, [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, alpha1, [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
> > > + INTFLOAT bw = (INTFLOAT)rnd() / UINT_MAX;
> > > + int i;
> > > +
> > > + declare_func(void, INTFLOAT (*X_high)[2], const INTFLOAT (*X_low)[2],
> > > + const INTFLOAT alpha0[2], const INTFLOAT
> > > alpha1[2],
> > > + INTFLOAT bw, int start, int end);
> > > +
> > > + randomize((INTFLOAT *)low, 128 * 2);
> > > + randomize((INTFLOAT *)alpha0, 2);
> > > + randomize((INTFLOAT *)alpha1, 2);
> > > + for (i = 2; i < 64; i += 2) {
> > > + memset(dst0, 0, 128 * 2 * sizeof(INTFLOAT));
> > > + memset(dst1, 0, 128 * 2 * sizeof(INTFLOAT));
> > > + call_ref(dst0, low, alpha0, alpha1, 0.0, i, 128);
> > > + call_new(dst1, low, alpha0, alpha1, 0.0, i, 128);
> > > + if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT
> > > *)dst1, EPS, 128 * 2))
> > > + fail();
> > > + bench_new(dst1, low, alpha0, alpha1, bw, i, 128);
> > > + }
> > > +}
> > > +
> > > +static void test_hf_g_filt(void)
> > > +{
> > > + LOCAL_ALIGNED_16(INTFLOAT, high, [128], [40][2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, g_filt, [128]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
> > > +
> > > + declare_func(void, INTFLOAT (*Y)[2], const INTFLOAT (*X_high)[40][2],
> > > + const INTFLOAT *g_filt, int m_max, intptr_t ixh);
> > > +
> > > + randomize((INTFLOAT *)high, 128 * 40 * 2);
> > > + randomize((INTFLOAT *)g_filt, 128);
> > > +
> > > + call_ref(dst0, high, g_filt, 128, 20);
> > > + call_new(dst1, high, g_filt, 128, 20);
> > > + if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1,
> > > EPS, 128 * 2))
> > > + fail();
> > > + bench_new(dst1, high, g_filt, 128, 20);
> > > +}
> > > +
> > > +static void test_hf_apply_noise(void)
> > > +{
> > > + LOCAL_ALIGNED_16(AAC_FLOAT, s_m, [128]);
> > > + LOCAL_ALIGNED_16(AAC_FLOAT, q_filt, [128]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, ref, [128], [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
> > > + LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
> > > + int noise = 0x2a;
> > > + int i;
> > > +
> > > + declare_func(void, INTFLOAT (*Y)[2], const AAC_FLOAT *s_m,
> > > + const AAC_FLOAT *q_filt, int noise,
> > > + int kx, int m_max);
> > > +
> > > + randomize((INTFLOAT *)ref, 128 * 2);
> > > + randomize((INTFLOAT *)s_m, 128);
> > > + randomize((INTFLOAT *)q_filt, 128);
> > > + for (i = 0; i < 2; i++) {
> > > + memcpy(dst0, ref, 128 * 2 * sizeof(INTFLOAT));
> > > + memcpy(dst1, ref, 128 * 2 * sizeof(INTFLOAT));
> > > + call_ref(dst0, s_m, q_filt, noise, i, 128);
> > > + call_new(dst1, s_m, q_filt, noise, i, 128);
> > > + if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT
> > > *)dst1, EPS, 128 * 2))
> > > + fail();
> > > + bench_new(dst1, s_m, q_filt, noise, i, 128);
> > > + }
> > > +}
> > > +
> > > +void checkasm_check_sbrdsp(void)
> > > +{
> > > + int i;
> > > + SBRDSPContext sbrdsp;
> > > +
> > > + ff_sbrdsp_init(&sbrdsp);
> > > +
> > > + if (check_func(sbrdsp.sum64x5, "sum64x5"))
> > > + test_sum64x5();
> > > + report("sum64x5");
> > > +
> > > + if (check_func(sbrdsp.sum_square, "sum_square"))
> > > + test_sum_square();
> > > + report("sum_square");
> > > +
> > > + if (check_func(sbrdsp.neg_odd_64, "neg_odd_64"))
> > > + test_neg_odd_64();
> > > + report("neg_odd_64");
> > > +
> > > + if (check_func(sbrdsp.qmf_pre_shuffle, "qmf_pre_shuffle"))
> > > + test_qmf_pre_shuffle();
> > > + report("qmf_pre_shuffle");
> > > +
> > > + if (check_func(sbrdsp.qmf_post_shuffle, "qmf_post_shuffle"))
> > > + test_qmf_post_shuffle();
> > > + report("qmf_post_shuffle");
> > > +
> > > + if (check_func(sbrdsp.qmf_deint_neg, "qmf_deint_neg"))
> > > + test_qmf_deint_neg();
> > > + report("qmf_deint_neg");
> > > +
> > > + if (check_func(sbrdsp.qmf_deint_bfly, "qmf_deint_bfly"))
> > > + test_qmf_deint_bfly();
> > > + report("qmf_deint_bfly");
> > > +
> > > + if (check_func(sbrdsp.autocorrelate, "autocorrelate"))
> > > + test_autocorrelate();
> > > + report("autocorrelate");
> > > +
> > > + if (check_func(sbrdsp.hf_gen, "hf_gen"))
> > > + test_hf_gen();
> > > + report("hf_gen");
> > > +
> > > + if (check_func(sbrdsp.hf_g_filt, "hf_g_filt"))
> > > + test_hf_g_filt();
> > > + report("hf_g_filt");
> > > +
> > > + for (i = 0; i < 4; i++) {
> >
> > You could instead add this loop inside test_hf_apply_noise(). It would
> > save you a bunch of unnecessary calls to randomize().
>
> Done in the following version of the patch.
I will apply the patchset in two days if there is no objections.
[...]
--
Matthieu B.
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
