src/hotspot/cpu/x86/x86.ad

@@ -1448,10 +1448,17 @@
       break;
     case Op_MulAddVS2VI:
       if (UseSSE < 2)
         ret_value = false;
       break;
+    case Op_MaxD:
+    case Op_MaxF:
+    case Op_MinD:
+    case Op_MinF:
+      if (UseAVX < 1) // enabled for AVX only
+        ret_value = false;
+      break;
   }
 
   return ret_value;  // Per default match rules are supported.
 }
 

@@ -2837,10 +2844,109 @@
     __ fmaf($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister);
   %}
   ins_pipe( pipe_slow );
 %}
 
+// Following pseudo code describes the algorithm for max[FD]: 
+// Min algorithm is on similar lines
+//  btmp = (b < 0) ? a : b
+//  atmp = (b < 0) ? b : a
+//  Tmp  = Max_Float( atmp , btmp)
+//  Res  = (atmp == NaN) ? atmp : Tmp 
+instruct maxF_reg(legRegF dst, legRegF a, legRegF b, legRegF tmp, legRegF atmp, legRegF btmp) %{
+  predicate(UseAVX > 0);
+  match(Set dst (MaxF a b));
+  effect(USE a, USE b, TEMP tmp, TEMP atmp, TEMP btmp);
+  format %{ 
+     "blendvps         $btmp,$b,$a,$b           \n\t"
+     "blendvps         $atmp,$a,$b,$b           \n\t"
+     "vmaxps           $tmp,$atmp,$btmp         \n\t"
+     "cmpps.unordered  $btmp, $atmp, $atmp      \n\t"
+     "blendvps         $dst,$tmp,$atmp,$btmp    \n\t"
+  %}
+  ins_encode %{
+    int vector_len = 0;
+    __ blendvps($btmp$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, vector_len);
+    __ blendvps($atmp$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $b$$XMMRegister, vector_len);
+    __ vmaxps($tmp$$XMMRegister, $atmp$$XMMRegister , $btmp$$XMMRegister, vector_len);
+    __ cmpps($btmp$$XMMRegister, $atmp$$XMMRegister, $atmp$$XMMRegister, 0x3, vector_len);
+    __ blendvps($dst$$XMMRegister, $tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister, vector_len);
+ %}
+  ins_pipe( pipe_slow );
+%}
+
+
+// max = java.lang.Max(double a , double b)
+instruct maxD_reg(legRegD dst, legRegD a, legRegD b, legRegD tmp, legRegD atmp, legRegD btmp) %{
+  predicate(UseAVX > 0);
+  match(Set dst (MaxD a b));
+  effect(USE a, USE b, TEMP atmp, TEMP btmp, TEMP tmp);
+  format %{ 
+     "blendvpd         $btmp,$b,$a,$b            \n\t"
+     "blendvpd         $atmp,$a,$b,$b            \n\t"
+     "vmaxpd           $tmp,$atmp,$btmp          \n\t"
+     "cmppd.unordered  $btmp, $atmp, $atmp       \n\t" 
+     "blendvpd         $dst,$tmp,$atmp,$btmp     \n\t"
+  %}
+  ins_encode %{
+    int vector_len = 0;
+    __ blendvpd($btmp$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, vector_len);
+    __ blendvpd($atmp$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $b$$XMMRegister, vector_len);
+    __ vmaxpd($tmp$$XMMRegister, $atmp$$XMMRegister , $btmp$$XMMRegister, vector_len);
+    __ cmppd($btmp$$XMMRegister, $atmp$$XMMRegister, $atmp$$XMMRegister, 0x3, vector_len);
+    __ blendvpd($dst$$XMMRegister, $tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+
+// min = java.lang.Min(float a , float b)
+instruct minF_reg(legRegF dst, legRegF a, legRegF b, legRegF tmp, legRegF atmp, legRegF btmp) %{
+  predicate(UseAVX > 0);
+  match(Set dst (MinF a b));
+  effect(USE a, USE b, TEMP tmp, TEMP atmp, TEMP btmp);
+  format %{ 
+     "blendvps         $atmp,$a,$b,$a             \n\t"
+     "blendvps         $btmp,$b,$a,$a             \n\t"
+     "vminps           $tmp,$atmp,$btmp           \n\t"
+     "cmpps.unordered  $btmp, $atmp, $atmp        \n\t" 
+     "blendvps         $dst,$tmp,$atmp,$btmp      \n\t"
+  %}
+  ins_encode %{
+    int vector_len = 0;
+    __ blendvps($atmp$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, vector_len);
+    __ blendvps($btmp$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, $a$$XMMRegister, vector_len);
+    __ vminps($tmp$$XMMRegister, $atmp$$XMMRegister , $btmp$$XMMRegister, vector_len);
+    __ cmpps($btmp$$XMMRegister, $atmp$$XMMRegister, $atmp$$XMMRegister, 0x3, vector_len);
+    __ blendvps($dst$$XMMRegister, $tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+// min = java.lang.Min(double a , double b)
+instruct minD_reg(legRegD dst, legRegD a, legRegD b, legRegD tmp, legRegD atmp, legRegD btmp) %{
+  predicate(UseAVX > 0);
+  match(Set dst (MinD a b));
+  effect(USE a, USE b, TEMP tmp, TEMP atmp, TEMP btmp);
+  format %{ 
+     "blendvpd         $atmp,$a,$b,$a           \n\t"
+     "blendvpd         $btmp,$b,$a,$a           \n\t"
+     "vminpd           $tmp,$atmp,$btmp         \n\t"
+     "cmppd.unordered  $btmp, $atmp, $atmp      \n\t" 
+     "blendvpd         $dst,$tmp,$atmp,$btmp    \n\t"
+  %}
+  ins_encode %{
+    int vector_len = 0;
+    __ blendvpd($atmp$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, vector_len);
+    __ blendvpd($btmp$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, $a$$XMMRegister, vector_len);
+    __ vminpd($tmp$$XMMRegister, $atmp$$XMMRegister , $btmp$$XMMRegister, vector_len);
+    __ cmppd($btmp$$XMMRegister, $atmp$$XMMRegister, $atmp$$XMMRegister, 0x3, vector_len);
+    __ blendvpd($dst$$XMMRegister, $tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 // ====================VECTOR INSTRUCTIONS=====================================
 
 
 // Load vectors (4 bytes long)
 instruct loadV4(vecS dst, memory mem) %{