X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Futil%2Frounding.h;h=e329d438244579ab1596b48ba7d5bd713cc8662b;hb=76a1fb3b42c245a425fde55b97a393d3e1984112;hp=c8be45057b6f21e93ed5df5ff1ded8d6eb8c0f91;hpb=dd0d3a2c0fb388745519c8a3be800720541eccfe;p=mesa.git

diff --git a/src/util/rounding.h b/src/util/rounding.h
index c8be45057b6..e329d438244 100644
--- a/src/util/rounding.h
+++ b/src/util/rounding.h
@@ -21,7 +21,22 @@
  * IN THE SOFTWARE.
  */
 
-#include <math.h>
+#ifndef _ROUNDING_H
+#define _ROUNDING_H
+
+#include "c99_math.h"
+
+#include <limits.h>
+#include <stdint.h>
+
+#if defined(__SSE__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 1)) || defined(_M_X64)
+#include <xmmintrin.h>
+#include <emmintrin.h>
+#endif
+
+#ifdef __SSE4_1__
+#include <smmintrin.h>
+#endif
 
 /* The C standard library has functions round()/rint()/nearbyint() that round
  * their arguments according to the rounding mode set in the floating-point
@@ -45,7 +60,15 @@
 static inline float
 _mesa_roundevenf(float x)
 {
+#ifdef __SSE4_1__
+   float ret;
+   __m128 m = _mm_load_ss(&x);
+   m = _mm_round_ss(m, m, _MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC);
+   _mm_store_ss(&ret, m);
+   return ret;
+#else
    return rintf(x);
+#endif
 }
 
 /**
@@ -54,5 +77,72 @@ _mesa_roundevenf(float x)
 static inline double
 _mesa_roundeven(double x)
 {
+#ifdef __SSE4_1__
+   double ret;
+   __m128d m = _mm_load_sd(&x);
+   m = _mm_round_sd(m, m, _MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC);
+   _mm_store_sd(&ret, m);
+   return ret;
+#else
    return rint(x);
+#endif
 }
+
+/**
+ * \brief Rounds \c x to the nearest integer, with ties to the even integer,
+ * and returns the value as a long int.
+ */
+static inline long
+_mesa_lroundevenf(float x)
+{
+#if defined(__SSE__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 1)) || defined(_M_X64)
+#if LONG_MAX == INT64_MAX
+   return _mm_cvtss_si64(_mm_load_ss(&x));
+#elif LONG_MAX == INT32_MAX
+   return _mm_cvtss_si32(_mm_load_ss(&x));
+#else
+#error "Unsupported long size"
+#endif
+#else
+   return lrintf(x);
+#endif
+}
+
+
+/**
+ * \brief Rounds \c x to the nearest integer, with ties to the even integer,
+ * and returns the value as a long int.
+ */
+static inline long
+_mesa_lroundeven(double x)
+{
+#if defined(__SSE2__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(_M_X64)
+#if LONG_MAX == INT64_MAX
+   return _mm_cvtsd_si64(_mm_load_sd(&x));
+#elif LONG_MAX == INT32_MAX
+   return _mm_cvtsd_si32(_mm_load_sd(&x));
+#else
+#error "Unsupported long size"
+#endif
+#else
+   return lrint(x);
+#endif
+}
+
+/**
+ * \brief Rounds \c x to the nearest integer, with ties to the even integer,
+ * and returns the value as an int64_t.
+ */
+static inline int64_t
+_mesa_i64roundevenf(float x)
+{
+#if LONG_MAX == INT64_MAX
+   return _mesa_lroundevenf(x);
+#elif LONG_MAX == INT32_MAX
+   return llrintf(x);
+#else
+#error "Unsupported long size"
+#endif
+}
+
+#endif