Recently normalization (av_normalize_sf) of output was added to av_add_sf.
This normalization is used for better precision for small values and the
purpose of this (quite simple) test case is to test difference between double
and softfloat.
The values used are tailored to maximally highlighte problem with precison when
normalization is not used.
Signed-off-by: Nedeljko Babic <nedeljko.ba...@imgtec.com>
---
libavutil/softfloat.c | 31 +++++++++++++++++++++++++++++--
1 file changed, 29 insertions(+), 2 deletions(-)
diff --git a/libavutil/softfloat.c b/libavutil/softfloat.c
index bf9cfda..b6e1f35 100644
--- a/libavutil/softfloat.c
+++ b/libavutil/softfloat.c
@@ -26,10 +26,21 @@
#undef printf
+static const SoftFloat FLOAT_0_017776489257 = {0x1234, 12};
+static const SoftFloat FLOAT_1374_40625 = {0xabcd, 25};
+static const SoftFloat FLOAT_0_1249694824218 = {0xFFF, 15};
+
+
+static av_const double av_sf2double(SoftFloat v) {
+ v.exp -= ONE_BITS +1;
+ if(v.exp > 0) return (double)v.mant * (double)(1 << v.exp);
+ else return (double)v.mant / (double)(1 << (-v.exp));
+}
+
int main(void){
SoftFloat one= av_int2sf(1, 0);
- SoftFloat sf1, sf2;
- double d1, d2;
+ SoftFloat sf1, sf2, sf3;
+ double d1, d2, d3;
int i, j;
av_log_set_level(AV_LOG_DEBUG);
@@ -67,5 +78,21 @@ int main(void){
STOP_TIMER("softfloat add mul")
}
printf("test2 sf =%d (%d %d)\n", av_sf2int(sf1, 24), sf1.exp, sf1.mant);
+
+ d1 = 0.0177764893;
+ d2 = 1374.40625;
+ d3 = 0.1249694824;
+ d2 += d1;
+ d3 += d2;
+ printf("test3 double: %.10lf\n", d3);
+
+ sf1 = FLOAT_0_017776489257;
+ sf2 = FLOAT_1374_40625;
+ sf3 = FLOAT_0_1249694824218;
+ sf2 = av_add_sf(sf1, sf2);
+ sf3 = av_add_sf(sf3, sf2);
+ printf("test3 softfloat: %.10lf (0x%08x %d)\n", (double)av_sf2double(sf3),
sf3.mant, sf3.exp);
+
return 0;
+
}
--
1.8.2.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
