Changed the default method for w=4 to use SSE if it is there.

git-svn-id: svn://mamba.eecs.utk.edu/home/plank/svn/Galois-Library@90 36f187d4-5712-4624-889c-152d48957efa

3 files changed, 9 insertions, 8 deletions

diff --git a/gf.c b/gf.c
index 195bb2f..769caae 100644
--- a/gf.c
+++ b/gf.c
@@ -104,9 +104,9 @@ void gf_alignment_error(char *s, int a)
   exit(1);
 }
 
-/* Lifted this code from Jens Gregor -- thanks, Jens */
+/* Lifted this code from Jens Gregor -- thanks, Jens.  Tests for SSE3 */
 
-int gf_is_sse2()
+int gf_is_sse()
 {
   unsigned int cpeinfo;
   unsigned int cpsse;
@@ -114,7 +114,8 @@ int gf_is_sse2()
                 "cpuid\n\t"
                 "mov %%edx, %0\n\t"
             "mov %%ecx, %1\n" : "=m" (cpeinfo), "=m" (cpsse));
-  if ((cpeinfo >> 26) & 0x1 ) return 1;
+  if ((cpsse       ) & 0x1 ) return 1;
+
   return 0;
 }
 
diff --git a/gf_int.h b/gf_int.h
index fa962b0..593e4d1 100644
--- a/gf_int.h
+++ b/gf_int.h
@@ -14,7 +14,7 @@ extern void     timer_start (double *t);
 extern double   timer_split (const double *t);
 extern void     galois_fill_random (void *buf, int len, unsigned int seed);
 
-extern int      galois_is_sse();
+extern int      gf_is_sse();
 
 typedef struct {
   int mult_type;
diff --git a/gf_w4.c b/gf_w4.c
index 2dced97..ad58fe8 100644
--- a/gf_w4.c
+++ b/gf_w4.c
@@ -385,7 +385,7 @@ int gf_w4_single_table_init(gf_t *gf)
   gf->inverse.w32 = NULL;
   gf->divide.w32 = gf_w4_single_table_divide;
   gf->multiply.w32 = gf_w4_single_table_multiply;
-  if (h->region_type & GF_REGION_SSE) {  
+  if ((h->region_type & GF_REGION_SSE) || (h->mult_type == GF_MULT_DEFAULT && gf_is_sse())) {  
     gf->multiply_region.w32 = gf_w4_single_table_sse_multiply_region;
   } else {
     gf->multiply_region.w32 = gf_w4_single_table_multiply_region;
@@ -1921,12 +1921,13 @@ int gf_w4_scratch_size(int mult_type, int region_type, int divide_type, int arg1
       }
       return sizeof(gf_internal_t) + sizeof(struct gf_bytwo_data);
       break;
+    case GF_MULT_DEFAULT:
     case GF_MULT_TABLE:
       if (arg1 != 0 || arg2 != 0) return -1;
       if (region_type == GF_REGION_CAUCHY || region_type == (GF_REGION_CAUCHY | GF_REGION_SINGLE_TABLE)) {
         return sizeof(gf_internal_t) + sizeof(struct gf_single_table_data) + 64;
       }
-      if (region_type == 0) region_type = GF_REGION_SINGLE_TABLE;
+      if (mult_type == GF_MULT_DEFAULT || region_type == 0) region_type = GF_REGION_SINGLE_TABLE;
       if (region_type & GF_REGION_SINGLE_TABLE) {
         if ((region_type | sss) != sss) return -1;
         if ((region_type & sss) == sss) return -1;
@@ -1944,7 +1945,6 @@ int gf_w4_scratch_size(int mult_type, int region_type, int divide_type, int arg1
       }
       return -1;  
       break;
-    case GF_MULT_DEFAULT:
     case GF_MULT_LOG_TABLE:
       if (arg1 != 0 || arg2 != 0 || (region_type != 0 && region_type != GF_REGION_CAUCHY)) return -1;
       return sizeof(gf_internal_t) + sizeof(struct gf_logtable_data) + 64;
@@ -1977,8 +1977,8 @@ gf_w4_init (gf_t *gf)
     case GF_MULT_BYTWO_p:   
     case GF_MULT_BYTWO_b:   
       if (gf_w4_bytwo_init(gf) == 0) return 0; break;
-    case GF_MULT_DEFAULT:   
     case GF_MULT_LOG_TABLE: if (gf_w4_log_init(gf) == 0) return 0; break;
+    case GF_MULT_DEFAULT:   
     case GF_MULT_TABLE:     if (gf_w4_table_init(gf) == 0) return 0; break;
     default: return 0;
   }