cgribexlib update

src/cgribexlib.c

 
-/* Automatically generated by m214003 at 2012-06-08, do not edit */
+/* Automatically generated by m214003 at 2012-06-14, do not edit */
 
 /* CGRIBEXLIB_VERSION="1.5.3" */
 
@@ -890,7 +890,6 @@ void minmax_val(const double *restrict data, long datasize, double *fmin, double
 
 #else
 
-
 #ifdef __SSE2__
 
   sse2_minmax_val(data, datasize, fmin, fmax);
@@ -944,7 +943,8 @@ void minmax_val(const double *restrict data, long datasize, double *fmin, double
 #endif
 
 #undef __UNROLL_DEPTH_1
-#else
+
+#else // original loop
 
 #ifdef _GET_IBM_COUNTER 
   hpmStart(1, "minmax base");
@@ -973,9 +973,9 @@ void minmax_val(const double *restrict data, long datasize, double *fmin, double
   hpmStop(1);
 #endif
 
-#endif
-#endif
-#endif
+#endif // _ARCH_PWR6 && original loop
+#endif // __SSE2__
+#endif // __AVX__
 
 #ifdef _GET_X86_COUNTER 
   end_minmax = _rdtsc();
@@ -1001,7 +1001,136 @@ void minmax_val(const double *restrict data, long datasize, double *fmin, double
 #ifdef _GET_IBM_COUNTER
 #endif
 
- 
+#if   defined(__GNUC__) && (__GNUC__ >= 4)
+#elif defined(__ICC)    && (__ICC >= 1100)
+#else
+#define DISABLE_SIMD
+#endif
+
+#ifdef DISABLE_SIMD
+#ifndef ENABLE_SSE4_1
+#undef __SSE4_1__
+#endif
+#endif
+
+// SSE4.1
+#if 0
+#ifdef __SSE4_1__
+
+static
+void sse41_encode_double_array_2byte(long datasize, 
+				     unsigned char * restrict lGrib,
+				     const double * restrict data, 
+				     double zref, double factor, long * restrict gz) 
+{
+  long i;
+  const double *dval = data;
+  __m128i *sgrib = (__m128i *) lGrib;
+
+  const __m128i swap = _mm_set_epi8(14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1);
+
+  const __m128d c0 = _mm_set1_pd(zref);
+  const __m128d c1 = _mm_set1_pd(factor);
+  const __m128d c2 = _mm_set1_pd(0.5);
+  
+  __m128d d0, d4, d3, d2, d1;
+  __m128i i0, i1, i2, i3, i4;
+  __m128i s0, s1;  
+
+  for (i = 0; i < datasize; i += 16)
+    {
+      (void) _mm_prefetch(dval+8, _MM_HINT_NTA);
+      //_____________________________________________________________________________
+
+      d0 = _mm_load_pd (dval);
+      d0 = _mm_sub_pd (d0, c0);
+      d0 = _mm_mul_pd (d0, c1);
+      d0 = _mm_add_pd (d0, c2);
+      
+      d4 = _mm_load_pd (dval+2);
+      d4 = _mm_sub_pd (d4, c0);
+      d4 = _mm_mul_pd (d4, c1);
+      d4 = _mm_add_pd (d4, c2);
+
+      i0 = _mm_cvttpd_epi32 (d0);
+      i4 = _mm_cvttpd_epi32 (d4);  
+      i0 = _mm_unpacklo_epi64 (i0, i4);
+      
+      //_____________________________________________________________________________
+      
+      d1 = _mm_load_pd (dval+4);
+      d1 = _mm_sub_pd (d1, c0);
+      d1 = _mm_mul_pd (d1, c1);
+      d1 = _mm_add_pd (d1, c2);
+      
+      d4 = _mm_load_pd (dval+6);
+      d4 = _mm_sub_pd (d4, c0);
+      d4 = _mm_mul_pd (d4, c1);
+      d4 = _mm_add_pd (d4, c2);
+      
+      i1 = _mm_cvttpd_epi32 (d1);
+      i4  = _mm_cvttpd_epi32 (d4);  
+      i1 = _mm_unpacklo_epi64 (i1, i4);
+
+      //_____________________________________________________________________________
+
+      s0 = _mm_packus_epi32(i0, i1);
+      s0 = _mm_shuffle_epi8 (s0, swap);
+      (void) _mm_stream_si128 (sgrib, s0);
+
+      //_____________________________________________________________________________
+
+      (void) _mm_prefetch(dval+16, _MM_HINT_NTA);
+
+      //_____________________________________________________________________________
+      
+      d2 = _mm_load_pd (dval+8);
+      d2 = _mm_sub_pd (d2, c0);
+      d2 = _mm_mul_pd (d2, c1);
+      d2 = _mm_add_pd (d2, c2);
+      
+      d4 = _mm_load_pd (dval+10);
+      d4 = _mm_sub_pd (d4, c0);
+      d4 = _mm_mul_pd (d4, c1);
+      d4 = _mm_add_pd (d4, c2);
+      
+      i2 = _mm_cvttpd_epi32 (d2);
+      i4  = _mm_cvttpd_epi32 (d4);  
+      i2 = _mm_unpacklo_epi64 (i2, i4);
+      
+      //_____________________________________________________________________________
+      
+      d3 = _mm_load_pd (dval+12);
+      d3 = _mm_sub_pd (d3, c0);
+      d3 = _mm_mul_pd (d3, c1);
+      d3 = _mm_add_pd (d3, c2);
+      
+      d4 = _mm_load_pd (dval+14);
+      d4 = _mm_sub_pd (d4, c0);
+      d4 = _mm_mul_pd (d4, c1);
+      d4 = _mm_add_pd (d4, c2);
+      
+      i3 = _mm_cvttpd_epi32 (d3);
+      i4 = _mm_cvttpd_epi32 (d4);  
+      i3 = _mm_unpacklo_epi64 (i3, i4);
+
+      //_____________________________________________________________________________
+
+      s1 = _mm_packus_epi32(i2, i3);
+      s1 = _mm_shuffle_epi8 (s1, swap);
+      (void) _mm_stream_si128 (sgrib+1, s1);
+
+      //_____________________________________________________________________________
+           
+      dval += 16;
+      sgrib += 2;
+    }
+
+  return;
+}
+#endif // SSE4.1
+#endif
+
 static
 void encode_double_array_common(int numBits, long packStart, long datasize, GRIBPACK *lGrib,
 				const double *data, double zref, double factor, long *gz)
@@ -1093,6 +1222,13 @@ void encode_double_array_byte(int numBits, long packStart, long datasize,
 #ifdef _GET_IBM_COUNTER 
       hpmStart(3, "pack 16 bit base");
 #endif
+
+#if 0
+
+      sse41_encode_double_array_2byte(datasize, lGrib, data, zref, factor, &z);
+
+#else
+
 #if   defined (CRAY)
 #pragma _CRI ivdep
 #elif defined (SX)
@@ -1108,6 +1244,8 @@ void encode_double_array_byte(int numBits, long packStart, long datasize,
           lGrib[z+1] = ival;
           z += 2;
       	}
+#endif
+
 #ifdef _GET_IBM_COUNTER 
       hpmStop(3);
 #endif
@@ -9842,7 +9980,7 @@ int  gribUnzip(unsigned char *dbuf, long dbufsize, unsigned char *sbuf, long sbu
 
   return (gribLen);
 }
-static const char grb_libvers[] = "1.5.3" " of ""Jun  8 2012"" ""10:20:01";
+static const char grb_libvers[] = "1.5.3" " of ""Jun 14 2012"" ""13:22:45";
 const char *
 cgribexLibraryVersion(void)
 {