# HG changeset patch
# User zmajo
# Date 1448288179 -3600
# Node ID abe570308c14cc980aa5ef426bd51721bf238c47
# Parent  ac3017b4c33691a0bb836e4485ed5c1060128528# Parent  a5f1c458b56e246edad8ba3dbf20d6a640b9ceeb
Merge

diff -r ac3017b4c336 -r abe570308c14 hotspot/src/cpu/x86/vm/macroAssembler_x86.cpp
--- a/hotspot/src/cpu/x86/vm/macroAssembler_x86.cpp	Mon Nov 23 15:09:45 2015 +0100
+++ b/hotspot/src/cpu/x86/vm/macroAssembler_x86.cpp	Mon Nov 23 15:16:19 2015 +0100
@@ -7387,7 +7387,8 @@
                                       XMMRegister vec, Register tmp,
                                       int ae) {
   ShortBranchVerifier sbv(this);
-  assert(UseSSE42Intrinsics, "SSE4.2 is required");
+  assert(UseSSE42Intrinsics, "SSE4.2 intrinsics are required");
+  assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
   assert(ae != StrIntrinsicNode::LU, "Invalid encoding");
 
   // This method uses the pcmpestri instruction with bound registers
@@ -7565,7 +7566,8 @@
                                     XMMRegister vec, Register tmp,
                                     int ae) {
   ShortBranchVerifier sbv(this);
-  assert(UseSSE42Intrinsics, "SSE4.2 is required");
+  assert(UseSSE42Intrinsics, "SSE4.2 intrinsics are required");
+  assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
   assert(ae != StrIntrinsicNode::LU, "Invalid encoding");
 
   //
@@ -7882,7 +7884,8 @@
 void MacroAssembler::string_indexof_char(Register str1, Register cnt1, Register ch, Register result,
                                          XMMRegister vec1, XMMRegister vec2, XMMRegister vec3, Register tmp) {
   ShortBranchVerifier sbv(this);
-  assert(UseSSE42Intrinsics, "SSE4.2 is required");
+  assert(UseSSE42Intrinsics, "SSE4.2 intrinsics are required");
+  assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
 
   int stride = 8;
 
@@ -7919,36 +7922,32 @@
     pshufd(vec1, vec1, 0);
     pxor(vec2, vec2);
   }
-  if (UseAVX >= 2 || UseSSE42Intrinsics) {
-    bind(SCAN_TO_8_CHAR);
-    cmpl(cnt1, stride);
-    if (UseAVX >= 2) {
-      jccb(Assembler::less, SCAN_TO_CHAR);
-    }
-    if (!(UseAVX >= 2)) {
-      jccb(Assembler::less, SCAN_TO_CHAR_LOOP);
-      movdl(vec1, ch);
-      pshuflw(vec1, vec1, 0x00);
-      pshufd(vec1, vec1, 0);
-      pxor(vec2, vec2);
-    }
-    movl(tmp, cnt1);
-    andl(tmp, 0xFFFFFFF8);  //vector count (in chars)
-    andl(cnt1,0x00000007);  //tail count (in chars)
-
-    bind(SCAN_TO_8_CHAR_LOOP);
-    movdqu(vec3, Address(result, 0));
-    pcmpeqw(vec3, vec1);
-    ptest(vec2, vec3);
-    jcc(Assembler::carryClear, FOUND_CHAR);
-    addptr(result, 16);
-    subl(tmp, stride);
-    jccb(Assembler::notZero, SCAN_TO_8_CHAR_LOOP);
-  }
+  bind(SCAN_TO_8_CHAR);
+  cmpl(cnt1, stride);
+  if (UseAVX >= 2) {
+    jccb(Assembler::less, SCAN_TO_CHAR);
+  } else {
+    jccb(Assembler::less, SCAN_TO_CHAR_LOOP);
+    movdl(vec1, ch);
+    pshuflw(vec1, vec1, 0x00);
+    pshufd(vec1, vec1, 0);
+    pxor(vec2, vec2);
+  }
+  movl(tmp, cnt1);
+  andl(tmp, 0xFFFFFFF8);  //vector count (in chars)
+  andl(cnt1,0x00000007);  //tail count (in chars)
+
+  bind(SCAN_TO_8_CHAR_LOOP);
+  movdqu(vec3, Address(result, 0));
+  pcmpeqw(vec3, vec1);
+  ptest(vec2, vec3);
+  jcc(Assembler::carryClear, FOUND_CHAR);
+  addptr(result, 16);
+  subl(tmp, stride);
+  jccb(Assembler::notZero, SCAN_TO_8_CHAR_LOOP);
   bind(SCAN_TO_CHAR);
   testl(cnt1, cnt1);
   jcc(Assembler::zero, RET_NOT_FOUND);
-
   bind(SCAN_TO_CHAR_LOOP);
   load_unsigned_short(tmp, Address(result, 0));
   cmpl(ch, tmp);
@@ -7962,16 +7961,14 @@
   movl(result, -1);
   jmpb(DONE_LABEL);
 
-  if (UseAVX >= 2 || UseSSE42Intrinsics) {
-    bind(FOUND_CHAR);
-    if (UseAVX >= 2) {
-      vpmovmskb(tmp, vec3);
-    } else {
-      pmovmskb(tmp, vec3);
-    }
-    bsfl(ch, tmp);
-    addl(result, ch);
-  }
+  bind(FOUND_CHAR);
+  if (UseAVX >= 2) {
+    vpmovmskb(tmp, vec3);
+  } else {
+    pmovmskb(tmp, vec3);
+  }
+  bsfl(ch, tmp);
+  addl(result, ch);
 
   bind(FOUND_SEQ_CHAR);
   subptr(result, str1);
@@ -8060,6 +8057,7 @@
   }
 
   if (UseAVX >= 2 && UseSSE42Intrinsics) {
+    assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
     Label COMPARE_WIDE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_WIDE_TAIL, COMPARE_SMALL_STR;
     Label COMPARE_WIDE_VECTORS_LOOP, COMPARE_16_CHARS, COMPARE_INDEX_CHAR;
     Label COMPARE_TAIL_LONG;
@@ -8195,6 +8193,7 @@
 
     bind(COMPARE_SMALL_STR);
   } else if (UseSSE42Intrinsics) {
+    assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
     Label COMPARE_WIDE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
     int pcmpmask = 0x19;
     // Setup to compare 8-char (16-byte) vectors,
@@ -8327,7 +8326,7 @@
 
   movl(result, len); // copy
 
-  if (UseAVX >= 2) {
+  if (UseAVX >= 2 && UseSSE >= 2) {
     // With AVX2, use 32-byte vector compare
     Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
 
@@ -8362,6 +8361,7 @@
     movl(len, result);
     // Fallthru to tail compare
   } else if (UseSSE42Intrinsics) {
+    assert(UseSSE >= 4, "SSE4 must be  for SSE4.2 intrinsics to be available");
     // With SSE4.2, use double quad vector compare
     Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
 
@@ -8438,7 +8438,7 @@
 
   // That's it
   bind(DONE);
-  if (UseAVX >= 2) {
+  if (UseAVX >= 2 && UseSSE >= 2) {
     // clean upper bits of YMM registers
     vpxor(vec1, vec1);
     vpxor(vec2, vec2);
@@ -8526,6 +8526,7 @@
     movl(limit, result);
     // Fallthru to tail compare
   } else if (UseSSE42Intrinsics) {
+    assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
     // With SSE4.2, use double quad vector compare
     Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
 
@@ -8875,6 +8876,7 @@
   negptr(len);
 
   if (UseSSE42Intrinsics || UseAVX >= 2) {
+    assert(UseSSE42Intrinsics ? UseSSE >= 4 : true, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
     Label L_chars_8_check, L_copy_8_chars, L_copy_8_chars_exit;
     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 
@@ -10647,6 +10649,7 @@
   push(len);
 
   if (UseSSE42Intrinsics) {
+    assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
     Label copy_32_loop, copy_16, copy_tail;
 
     movl(result, len);
@@ -10746,6 +10749,7 @@
   assert_different_registers(src, dst, len, tmp2);
 
   if (UseSSE42Intrinsics) {
+    assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
     Label copy_8_loop, copy_bytes, copy_tail;
 
     movl(tmp2, len);
diff -r ac3017b4c336 -r abe570308c14 hotspot/src/cpu/x86/vm/vm_version_x86.cpp
--- a/hotspot/src/cpu/x86/vm/vm_version_x86.cpp	Mon Nov 23 15:09:45 2015 +0100
+++ b/hotspot/src/cpu/x86/vm/vm_version_x86.cpp	Mon Nov 23 15:16:19 2015 +0100
@@ -930,10 +930,15 @@
         UseXmmI2D = false;
       }
     }
-    if( FLAG_IS_DEFAULT(UseSSE42Intrinsics) ) {
-      if( supports_sse4_2() && UseSSE >= 4 ) {
-        UseSSE42Intrinsics = true;
+    if (supports_sse4_2() && UseSSE >= 4) {
+      if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
+        FLAG_SET_DEFAULT(UseSSE42Intrinsics, true);
       }
+    } else {
+      if (UseSSE42Intrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
+        warning("SSE4.2 intrinsics require SSE4.2 instructions or higher. Intrinsics will be disabled.");
+      }
+      FLAG_SET_DEFAULT(UseSSE42Intrinsics, false);
     }
 
     // some defaults for AMD family 15h
@@ -1007,8 +1012,13 @@
       }
       if (supports_sse4_2() && UseSSE >= 4) {
         if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
-          UseSSE42Intrinsics = true;
+          FLAG_SET_DEFAULT(UseSSE42Intrinsics, true);
         }
+      } else {
+        if (UseSSE42Intrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
+          warning("SSE4.2 intrinsics require SSE4.2 instructions or higher. Intrinsics will be disabled.");
+        }
+        FLAG_SET_DEFAULT(UseSSE42Intrinsics, false);
       }
     }
     if ((cpu_family() == 0x06) &&