[Beignet] [PATCH] backend: add double support to floor

[rander]

	do bit operations according to IEEE754 spec

Signed-off-by: rander <rander.wang at intel.com>
---
 backend/src/libocl/tmpl/ocl_math.tmpl.cl    | 33 +++++++++++++++++++++++++++++
 backend/src/libocl/tmpl/ocl_math.tmpl.h     |  2 +-
 backend/src/libocl/tmpl/ocl_math_20.tmpl.cl | 33 +++++++++++++++++++++++++++++
 backend/src/libocl/tmpl/ocl_math_20.tmpl.h  |  3 +--
 4 files changed, 68 insertions(+), 3 deletions(-)

diff --git a/backend/src/libocl/tmpl/ocl_math.tmpl.cl b/backend/src/libocl/tmpl/ocl_math.tmpl.cl
index 83af1c6..bb9bc4d 100644
--- a/backend/src/libocl/tmpl/ocl_math.tmpl.cl
+++ b/backend/src/libocl/tmpl/ocl_math.tmpl.cl
@@ -3931,3 +3931,36 @@ OVERLOADABLE double fabs(double x)
     return as_double(qw);
 }
 
+OVERLOADABLE double floor(double x)
+{
+    ulong lval = as_ulong(x);
+    int exp = ((lval >> 52) & 0x7FF) - 1023;
+    int sign = (lval >> 63) & 0x1;
+    if(exp < 0)
+    {
+        if(sign)
+	{
+		if(lval & DF_ABS_MASK)
+			return -1L;
+		else
+			return 0.0;
+	}
+        else return 0.0;
+    }
+    else
+    {
+        if(exp >= 52)
+            return x;
+         long i = (1L << (52 - exp));
+         i = 0x10000000000000 - i;
+         unsigned long uv = 0xFFF0000000000000 + i;
+         ulong vv = lval & uv;
+         double  dp = as_double(vv);
+         if(vv != lval)
+            dp -= sign;
+
+         return dp;
+    }
+}
+
+
diff --git a/backend/src/libocl/tmpl/ocl_math.tmpl.h b/backend/src/libocl/tmpl/ocl_math.tmpl.h
index dab45ae..f699ab2 100644
--- a/backend/src/libocl/tmpl/ocl_math.tmpl.h
+++ b/backend/src/libocl/tmpl/ocl_math.tmpl.h
@@ -235,4 +235,4 @@ OVERLOADABLE float half_tan(float x);
 //------- double -----------
 OVERLOADABLE double ceil(double x);
 OVERLOADABLE double fabs(double x);
-
+OVERLOADABLE double floor(double x);
diff --git a/backend/src/libocl/tmpl/ocl_math_20.tmpl.cl b/backend/src/libocl/tmpl/ocl_math_20.tmpl.cl
index 83c889a..a032f59 100644
--- a/backend/src/libocl/tmpl/ocl_math_20.tmpl.cl
+++ b/backend/src/libocl/tmpl/ocl_math_20.tmpl.cl
@@ -3832,3 +3832,36 @@ OVERLOADABLE double fabs(double x)
     return as_double(qw);
 }
 
+OVERLOADABLE double floor(double x)
+{
+    ulong lval = as_ulong(x);
+    int exp = ((lval >> 52) & 0x7FF) - 1023;
+    int sign = (lval >> 63) & 0x1;
+    if(exp < 0)
+    {
+        if(sign)
+	{
+		if(lval & DF_ABS_MASK)
+			return -1L;
+		else
+			return 0.0;
+	}
+        else return 0.0;
+    }
+    else
+    {
+        if(exp >= 52)
+            return x;
+         long i = (1L << (52 - exp));
+         i = 0x10000000000000 - i;
+         unsigned long uv = 0xFFF0000000000000 + i;
+         ulong vv = lval & uv;
+         double  dp = as_double(vv);
+         if(vv != lval)
+            dp -= sign;
+
+         return dp;
+    }
+}
+
+
diff --git a/backend/src/libocl/tmpl/ocl_math_20.tmpl.h b/backend/src/libocl/tmpl/ocl_math_20.tmpl.h
index 76778dd..bf7ac13 100644
--- a/backend/src/libocl/tmpl/ocl_math_20.tmpl.h
+++ b/backend/src/libocl/tmpl/ocl_math_20.tmpl.h
@@ -212,5 +212,4 @@ OVERLOADABLE float half_tan(float x);
 //------- double -----------
 OVERLOADABLE double ceil(double x);
 OVERLOADABLE double fabs(double x);
-
-
+OVERLOADABLE double floor(double x);
-- 
2.7.4