Merge remote-tracking branch 'qatar/master'
authorMichael Niedermayer <michaelni@gmx.at>
Tue, 18 Mar 2014 19:04:18 +0000 (20:04 +0100)
committerMichael Niedermayer <michaelni@gmx.at>
Tue, 18 Mar 2014 19:04:28 +0000 (20:04 +0100)
* qatar/master:
  float_dsp: add test program and use it as fate test

Merged-by: Michael Niedermayer <michaelni@gmx.at>
1  2 
libavutil/Makefile
libavutil/float_dsp.c
tests/fate/libavutil.mak

@@@ -147,33 -120,15 +147,34 @@@ TESTPROGS = adler3
              cpu                                                         \
              crc                                                         \
              des                                                         \
 +            error                                                       \
              eval                                                        \
 +            file                                                        \
              fifo                                                        \
+             float_dsp                                                   \
              hmac                                                        \
              lfg                                                         \
 -            lls                                                         \
 +            lls1                                                        \
 +            lls2                                                        \
              md5                                                         \
 +            murmur3                                                     \
              opt                                                         \
 +            pca                                                         \
              parseutils                                                  \
 +            random_seed                                                 \
 +            rational                                                    \
 +            ripemd                                                      \
              sha                                                         \
 +            sha512                                                      \
              tree                                                        \
 +            utf8                                                        \
              xtea                                                        \
 +
 +TESTPROGS-$(HAVE_LZO1X_999_COMPRESS) += lzo
 +
 +TOOLS = crypto_bench ffhash ffeval ffescape
 +
 +tools/crypto_bench$(EXESUF): ELIBS += $(if $(VERSUS),$(subst +, -l,+$(VERSUS)),)
 +tools/crypto_bench$(EXESUF): CFLAGS += -DUSE_EXT_LIBS=0$(if $(VERSUS),$(subst +,+USE_,+$(VERSUS)),)
 +
 +$(SUBDIR)lzo-test$(EXESUF): ELIBS = -llzo2
@@@ -133,7 -130,288 +133,290 @@@ av_cold void avpriv_float_dsp_init(AVFl
      ff_float_dsp_init_ppc(fdsp, bit_exact);
  #elif ARCH_X86
      ff_float_dsp_init_x86(fdsp);
 +#elif ARCH_MIPS
 +    ff_float_dsp_init_mips(fdsp);
  #endif
  }
+ #ifdef TEST
+ #include <float.h>
+ #include <math.h>
+ #include <stdint.h>
+ #include <stdlib.h>
+ #include <string.h>
+ #include "cpu.h"
+ #include "lfg.h"
+ #include "log.h"
+ #include "mem.h"
+ #include "random_seed.h"
+ #define LEN 240
+ static void fill_float_array(AVLFG *lfg, float *a, int len)
+ {
+     int i;
+     double bmg[2], stddev = 10.0, mean = 0.0;
+     for (i = 0; i < len; i += 2) {
+         av_bmg_get(lfg, bmg);
+         a[i]     = bmg[0] * stddev + mean;
+         a[i + 1] = bmg[1] * stddev + mean;
+     }
+ }
+ static int compare_floats(const float *a, const float *b, int len,
+                           float max_diff)
+ {
+     int i;
+     for (i = 0; i < len; i++) {
+         if (fabsf(a[i] - b[i]) > max_diff) {
+             av_log(NULL, AV_LOG_ERROR, "%d: %- .12f - %- .12f = % .12g\n",
+                    i, a[i], b[i], a[i] - b[i]);
+             return -1;
+         }
+     }
+     return 0;
+ }
+ static void fill_double_array(AVLFG *lfg, double *a, int len)
+ {
+     int i;
+     double bmg[2], stddev = 10.0, mean = 0.0;
+     for (i = 0; i < len; i += 2) {
+         av_bmg_get(lfg, bmg);
+         a[i]     = bmg[0] * stddev + mean;
+         a[i + 1] = bmg[1] * stddev + mean;
+     }
+ }
+ static int compare_doubles(const double *a, const double *b, int len,
+                            double max_diff)
+ {
+     int i;
+     for (i = 0; i < len; i++) {
+         if (fabs(a[i] - b[i]) > max_diff) {
+             av_log(NULL, AV_LOG_ERROR, "%d: %- .12f - %- .12f = % .12g\n",
+                    i, a[i], b[i], a[i] - b[i]);
+             return -1;
+         }
+     }
+     return 0;
+ }
+ static int test_vector_fmul(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                             const float *v1, const float *v2)
+ {
+     DECLARE_ALIGNED(32, float, cdst)[LEN];
+     DECLARE_ALIGNED(32, float, odst)[LEN];
+     int ret;
+     cdsp->vector_fmul(cdst, v1, v2, LEN);
+     fdsp->vector_fmul(odst, v1, v2, LEN);
+     if (ret = compare_floats(cdst, odst, LEN, FLT_EPSILON))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ #define ARBITRARY_FMAC_SCALAR_CONST 0.005
+ static int test_vector_fmac_scalar(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                                    const float *v1, const float *src0, float scale)
+ {
+     DECLARE_ALIGNED(32, float, cdst)[LEN];
+     DECLARE_ALIGNED(32, float, odst)[LEN];
+     int ret;
+     memcpy(cdst, v1, LEN * sizeof(*v1));
+     memcpy(odst, v1, LEN * sizeof(*v1));
+     cdsp->vector_fmac_scalar(cdst, src0, scale, LEN);
+     fdsp->vector_fmac_scalar(odst, src0, scale, LEN);
+     if (ret = compare_floats(cdst, odst, LEN, ARBITRARY_FMAC_SCALAR_CONST))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ static int test_vector_fmul_scalar(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                                    const float *v1, float scale)
+ {
+     DECLARE_ALIGNED(32, float, cdst)[LEN];
+     DECLARE_ALIGNED(32, float, odst)[LEN];
+     int ret;
+     cdsp->vector_fmul_scalar(cdst, v1, scale, LEN);
+     fdsp->vector_fmul_scalar(odst, v1, scale, LEN);
+     if (ret = compare_floats(cdst, odst, LEN, FLT_EPSILON))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ static int test_vector_dmul_scalar(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                                    const double *v1, double scale)
+ {
+     DECLARE_ALIGNED(32, double, cdst)[LEN];
+     DECLARE_ALIGNED(32, double, odst)[LEN];
+     int ret;
+     cdsp->vector_dmul_scalar(cdst, v1, scale, LEN);
+     fdsp->vector_dmul_scalar(odst, v1, scale, LEN);
+     if (ret = compare_doubles(cdst, odst, LEN, DBL_EPSILON))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ #define ARBITRARY_FMUL_WINDOW_CONST 0.008
+ static int test_vector_fmul_window(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                                    const float *v1, const float *v2, const float *v3)
+ {
+     DECLARE_ALIGNED(32, float, cdst)[LEN];
+     DECLARE_ALIGNED(32, float, odst)[LEN];
+     int ret;
+     cdsp->vector_fmul_window(cdst, v1, v2, v3, LEN / 2);
+     fdsp->vector_fmul_window(odst, v1, v2, v3, LEN / 2);
+     if (ret = compare_floats(cdst, odst, LEN, ARBITRARY_FMUL_WINDOW_CONST))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ #define ARBITRARY_FMUL_ADD_CONST 0.005
+ static int test_vector_fmul_add(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                                 const float *v1, const float *v2, const float *v3)
+ {
+     DECLARE_ALIGNED(32, float, cdst)[LEN];
+     DECLARE_ALIGNED(32, float, odst)[LEN];
+     int ret;
+     cdsp->vector_fmul_add(cdst, v1, v2, v3, LEN);
+     fdsp->vector_fmul_add(odst, v1, v2, v3, LEN);
+     if (ret = compare_floats(cdst, odst, LEN, ARBITRARY_FMUL_ADD_CONST))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ static int test_vector_fmul_reverse(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                                     const float *v1, const float *v2)
+ {
+     DECLARE_ALIGNED(32, float, cdst)[LEN];
+     DECLARE_ALIGNED(32, float, odst)[LEN];
+     int ret;
+     cdsp->vector_fmul_reverse(cdst, v1, v2, LEN);
+     fdsp->vector_fmul_reverse(odst, v1, v2, LEN);
+     if (ret = compare_floats(cdst, odst, LEN, FLT_EPSILON))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ static int test_butterflies_float(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                                   const float *v1, const float *v2)
+ {
+     DECLARE_ALIGNED(32, float, cv1)[LEN];
+     DECLARE_ALIGNED(32, float, cv2)[LEN];
+     DECLARE_ALIGNED(32, float, ov1)[LEN];
+     DECLARE_ALIGNED(32, float, ov2)[LEN];
+     int ret;
+     memcpy(cv1, v1, LEN * sizeof(*v1));
+     memcpy(cv2, v2, LEN * sizeof(*v2));
+     memcpy(ov1, v1, LEN * sizeof(*v1));
+     memcpy(ov2, v2, LEN * sizeof(*v2));
+     cdsp->butterflies_float(cv1, cv2, LEN);
+     fdsp->butterflies_float(ov1, ov2, LEN);
+     if ((ret = compare_floats(cv1, ov1, LEN, FLT_EPSILON)) ||
+         (ret = compare_floats(cv2, ov2, LEN, FLT_EPSILON)))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ #define ARBITRARY_SCALARPRODUCT_CONST 0.2
+ static int test_scalarproduct_float(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
+                                     const float *v1, const float *v2)
+ {
+     float cprod, oprod;
+     int ret;
+     cprod = cdsp->scalarproduct_float(v1, v2, LEN);
+     oprod = fdsp->scalarproduct_float(v1, v2, LEN);
+     if (ret = compare_floats(&cprod, &oprod, 1, ARBITRARY_SCALARPRODUCT_CONST))
+         av_log(NULL, AV_LOG_ERROR, "%s failed\n", __func__);
+     return ret;
+ }
+ int main(int argc, char **argv)
+ {
+     int ret = 0;
+     uint32_t seed;
+     AVFloatDSPContext fdsp, cdsp;
+     AVLFG lfg;
+     DECLARE_ALIGNED(32, float, src0)[LEN];
+     DECLARE_ALIGNED(32, float, src1)[LEN];
+     DECLARE_ALIGNED(32, float, src2)[LEN];
+     DECLARE_ALIGNED(32, double, dbl_src0)[LEN];
+     DECLARE_ALIGNED(32, double, dbl_src1)[LEN];
+     if (argc > 2 && !strcmp(argv[1], "-s"))
+         seed = strtoul(argv[2], NULL, 10);
+     else
+         seed = av_get_random_seed();
+     av_log(NULL, AV_LOG_INFO, "float_dsp-test: random seed %u\n", seed);
+     av_lfg_init(&lfg, seed);
+     fill_float_array(&lfg, src0, LEN);
+     fill_float_array(&lfg, src1, LEN);
+     fill_float_array(&lfg, src2, LEN);
+     fill_double_array(&lfg, dbl_src0, LEN);
+     fill_double_array(&lfg, dbl_src1, LEN);
+     avpriv_float_dsp_init(&fdsp, 1);
+     av_set_cpu_flags_mask(0);
+     avpriv_float_dsp_init(&cdsp, 1);
+     if (test_vector_fmul(&fdsp, &cdsp, src0, src1))
+         ret -= 1 << 0;
+     if (test_vector_fmac_scalar(&fdsp, &cdsp, src2, src0, src1[0]))
+         ret -= 1 << 1;
+     if (test_vector_fmul_scalar(&fdsp, &cdsp, src0, src1[0]))
+         ret -= 1 << 2;
+     if (test_vector_fmul_window(&fdsp, &cdsp, src0, src1, src2))
+         ret -= 1 << 3;
+     if (test_vector_fmul_add(&fdsp, &cdsp, src0, src1, src2))
+         ret -= 1 << 4;
+     if (test_vector_fmul_reverse(&fdsp, &cdsp, src0, src1))
+         ret -= 1 << 5;
+     if (test_butterflies_float(&fdsp, &cdsp, src0, src1))
+         ret -= 1 << 6;
+     if (test_scalarproduct_float(&fdsp, &cdsp, src0, src1))
+         ret -= 1 << 7;
+     if (test_vector_dmul_scalar(&fdsp, &cdsp, dbl_src0, dbl_src1[0]))
+         ret -= 1 << 8;
+     return ret;
+ }
+ #endif /* TEST */
Simple merge