Merge commit 'e71b747e9dc56cb84f8a06ec8214d5f3bd98bb6d'
authorHendrik Leppkes <h.leppkes@gmail.com>
Sat, 2 Jan 2016 11:38:46 +0000 (12:38 +0100)
committerHendrik Leppkes <h.leppkes@gmail.com>
Sat, 2 Jan 2016 11:38:46 +0000 (12:38 +0100)
* commit 'e71b747e9dc56cb84f8a06ec8214d5f3bd98bb6d':
  checkasm: add tests for dcadsp

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>
1  2 
tests/checkasm/Makefile
tests/checkasm/checkasm.c
tests/checkasm/checkasm.h
tests/checkasm/dcadsp.c

@@@ -1,13 -1,10 +1,14 @@@
  # libavcodec tests
 +AVCODECOBJS-$(CONFIG_ALAC_DECODER) += alacdsp.o
  AVCODECOBJS-$(CONFIG_BSWAPDSP) += bswapdsp.o
+ AVCODECOBJS-$(CONFIG_DCA_DECODER) += dcadsp.o
 +AVCODECOBJS-$(CONFIG_FLACDSP)  += flacdsp.o
  AVCODECOBJS-$(CONFIG_H264PRED) += h264pred.o
  AVCODECOBJS-$(CONFIG_H264QPEL) += h264qpel.o
 -AVCODECOBJS-$(CONFIG_HEVC_DECODER) += hevc_mc.o
 +AVCODECOBJS-$(CONFIG_JPEG2000_DECODER) += jpeg2000dsp.o
 +AVCODECOBJS-$(CONFIG_PIXBLOCKDSP) += pixblockdsp.o
  AVCODECOBJS-$(CONFIG_V210_ENCODER) += v210enc.o
 +AVCODECOBJS-$(CONFIG_VP9_DECODER) += vp9dsp.o
  
  CHECKASMOBJS-$(CONFIG_AVCODEC) += $(AVCODECOBJS-yes)
  
@@@ -58,34 -58,23 +58,37 @@@ static const struct 
      const char *name;
      void (*func)(void);
  } tests[] = {
 -#if CONFIG_BSWAPDSP
 -    { "bswapdsp", checkasm_check_bswapdsp },
 -#endif
 -#if CONFIG_DCA_DECODER
 -    { "dcadsp", checkasm_check_dcadsp },
 -#endif
 -#if CONFIG_H264PRED
 -    { "h264pred", checkasm_check_h264pred },
 -#endif
 -#if CONFIG_H264QPEL
 -    { "h264qpel", checkasm_check_h264qpel },
 -#endif
 -#if CONFIG_HEVC_DECODER
 -    { "hevc_mc", checkasm_check_hevc_mc },
 -#endif
 -#if CONFIG_V210_ENCODER
 -    { "v210enc", checkasm_check_v210enc },
 +#if CONFIG_AVCODEC
 +    #if CONFIG_ALAC_DECODER
 +        { "alacdsp", checkasm_check_alacdsp },
 +    #endif
 +    #if CONFIG_BSWAPDSP
 +        { "bswapdsp", checkasm_check_bswapdsp },
 +    #endif
++    #if CONFIG_DCA_DECODER
++        { "dcadsp", checkasm_check_dcadsp },
++    #endif
 +    #if CONFIG_FLACDSP
 +        { "flacdsp", checkasm_check_flacdsp },
 +    #endif
 +    #if CONFIG_H264PRED
 +        { "h264pred", checkasm_check_h264pred },
 +    #endif
 +    #if CONFIG_H264QPEL
 +        { "h264qpel", checkasm_check_h264qpel },
 +    #endif
 +    #if CONFIG_JPEG2000_DECODER
 +        { "jpeg2000dsp", checkasm_check_jpeg2000dsp },
 +    #endif
 +    #if CONFIG_PIXBLOCKDSP
 +        { "pixblockdsp", checkasm_check_pixblockdsp },
 +    #endif
 +    #if CONFIG_V210_ENCODER
 +        { "v210enc", checkasm_check_v210enc },
 +    #endif
 +    #if CONFIG_VP9_DECODER
 +        { "vp9dsp", checkasm_check_vp9dsp },
 +    #endif
  #endif
      { NULL }
  };
  #include "libavutil/lfg.h"
  #include "libavutil/timer.h"
  
 +void checkasm_check_alacdsp(void);
  void checkasm_check_bswapdsp(void);
+ void checkasm_check_dcadsp(void);
 +void checkasm_check_flacdsp(void);
  void checkasm_check_h264pred(void);
  void checkasm_check_h264qpel(void);
 -void checkasm_check_hevc_mc(void);
 +void checkasm_check_jpeg2000dsp(void);
 +void checkasm_check_pixblockdsp(void);
  void checkasm_check_v210enc(void);
 +void checkasm_check_vp9dsp(void);
  
  void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3);
  int checkasm_bench_func(void);
index 0000000,1665cbb..e0a4f4d
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,137 +1,137 @@@
 - * This file is part of Libav.
+ /*
+  * Copyright (c) 2015 Janne Grunau
+  *
 - * Libav is free software; you can redistribute it and/or modify
++ * This file is part of FFmpeg.
+  *
 - * Libav is distributed in the hope that it will be useful,
++ * 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.
+  *
 - * with Libav; if not, write to the Free Software Foundation, Inc.,
++ * 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 <math.h>
+ #include <string.h>
+ #include <stdlib.h>
+ #include "libavutil/internal.h"
+ #include "libavutil/intfloat.h"
+ #include "libavcodec/dca.h"
+ #include "libavcodec/dcadsp.h"
+ #include "libavcodec/dcadata.h"
+ #include "checkasm.h"
+ #define randomize_lfe_fir(size)                                 \
+     do {                                                        \
+         int i;                                                  \
+         for (i = 0; i < size; i++) {                            \
+             float f = (float)rnd() / (UINT_MAX >> 1) - 1.0f;    \
+             in[i] = f;                                          \
+         }                                                       \
+         for (i = 0; i < 256; i++) {                             \
+             float f = (float)rnd() / (UINT_MAX >> 1) - 1.0f;    \
+             coeffs[i] = f;                                      \
+         }                                                       \
+     } while (0)
+ #define check_lfe_fir(decifactor, eps)                                  \
+     do {                                                                \
+         LOCAL_ALIGNED_16(float, in,     [256 / decifactor]);            \
+         LOCAL_ALIGNED_16(float, out0,   [decifactor * 2]);              \
+         LOCAL_ALIGNED_16(float, out1,   [decifactor * 2]);              \
+         LOCAL_ALIGNED_16(float, coeffs, [256]);                         \
+         int i;                                                          \
+         const float * in_ptr = in + (256 / decifactor) - 1;             \
+         declare_func(void, float *out, const float *in, const float *coeffs); \
+         /* repeat the test several times */                             \
+         for (i = 0; i < 32; i++) {                                      \
+             int j;                                                      \
+             memset(out0,    0, sizeof(*out0) * 2 * decifactor);         \
+             memset(out1, 0xFF, sizeof(*out1) * 2 * decifactor);         \
+             randomize_lfe_fir(256 / decifactor);                        \
+             call_ref(out0, in_ptr, coeffs);                             \
+             call_new(out1, in_ptr, coeffs);                             \
+             for (j = 0; j < 2 * decifactor; j++) {                      \
+                 if (!float_near_abs_eps(out0[j], out1[j], eps)) {       \
+                     if (0) {                                            \
+                         union av_intfloat32 x, y; x.f = out0[j]; y.f = out1[j]; \
+                         fprintf(stderr, "%3d: %11g (0x%08x); %11g (0x%08x)\n", \
+                                 j, x.f, x.i, y.f, y.i);                 \
+                     }                                                   \
+                     fail();                                             \
+                     break;                                              \
+                 }                                                       \
+             }                                                           \
+             bench_new(out1, in_ptr, coeffs);                            \
+         }                                                               \
+     } while (0)
+ #define randomize_decode_hf()                                   \
+     do {                                                        \
+         int i;                                                  \
+         for (i = 0; i < DCA_SUBBANDS; i++) {                    \
+             vq_num[i]   = rnd() >> 22;                          \
+             scale[i][0] = rnd() >> 26;                          \
+             scale[i][1] = INT32_MIN;                            \
+         }                                                       \
+     } while (0)
+ void checkasm_check_dcadsp(void)
+ {
+     DCADSPContext c;
+     ff_dcadsp_init(&c);
+     /* values are limited to {-8, 8} so absolute epsilon is good enough */
+     if (check_func(c.lfe_fir[0], "dca_lfe_fir0"))
+         check_lfe_fir(32, 1.0e-6f);
+     if (check_func(c.lfe_fir[1], "dca_lfe_fir1"))
+         check_lfe_fir(64, 1.0e-6f);
+     if (check_func(c.decode_hf,  "dca_decode_hf")) {
+         LOCAL_ALIGNED_16(float,   dst0,   [DCA_SUBBANDS], [8]);
+         LOCAL_ALIGNED_16(float,   dst1,   [DCA_SUBBANDS], [8]);
+         LOCAL_ALIGNED_16(int32_t, scale,  [DCA_SUBBANDS], [2]);
+         LOCAL_ALIGNED_16(int32_t, vq_num, [DCA_SUBBANDS]);
+         intptr_t start, end = 32, offset;
+         declare_func(void, float[DCA_SUBBANDS][8], const int32_t[DCA_SUBBANDS],
+                      const int8_t[1024][DCA_SUBBANDS], intptr_t, int32_t[DCA_SUBBANDS][2],
+                      intptr_t, intptr_t);
+         for (start = 0; start < 32; start++) {
+             for (offset = 0; offset < 32; offset += 8) {
+                 int j;
+                 for (j = 0; j < DCA_SUBBANDS; j++) {
+                     memset(dst0[j], 0, sizeof(*(dst0[j])) * 8);
+                     memset(dst1[j], 0, sizeof(*(dst1[j])) * 8);
+                 }
+                 randomize_decode_hf();
+                 call_ref(dst0, vq_num, ff_dca_high_freq_vq, offset, scale, start, end);
+                 call_new(dst1, vq_num, ff_dca_high_freq_vq, offset, scale, start, end);
+                 for (j = 0; j < 8 * DCA_SUBBANDS; j++) {
+                     if (!float_near_ulp(dst0[j>>3][j&7], dst1[j>>3][j&7], 1)) {
+                         fail();
+                         break;
+                     }
+                 }
+                 bench_new(dst1, vq_num, ff_dca_high_freq_vq, offset, scale, start, end);
+             }
+         }
+     }
+     report("dcadsp");
+ }