Mercurial > hg > graal-compiler

diff src/cpu/x86/vm/macroAssembler_x86.cpp @ 7475:e2e6bf86682c

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
8005544: Use 256bit YMM registers in arraycopy stubs on x86 Summary: Use YMM registers in arraycopy and array_fill stubs. Reviewed-by: roland, twisti
author kvn
date Thu, 03 Jan 2013 16:30:47 -0800
parents 00af3a3a8df4
children ffa87474d7a4
line wrap: on
line diff
--- a/src/cpu/x86/vm/macroAssembler_x86.cpp	Thu Jan 03 15:09:55 2013 -0800
+++ b/src/cpu/x86/vm/macroAssembler_x86.cpp	Thu Jan 03 16:30:47 2013 -0800
@@ -6011,29 +6011,53 @@
     {
       assert( UseSSE >= 2, "supported cpu only" );
       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
-      // Fill 32-byte chunks
       movdl(xtmp, value);
-      pshufd(xtmp, xtmp, 0);
-
-      subl(count, 8 << shift);
-      jcc(Assembler::less, L_check_fill_8_bytes);
-      align(16);
-
-      BIND(L_fill_32_bytes_loop);
-
-      if (UseUnalignedLoadStores) {
-        movdqu(Address(to, 0), xtmp);
-        movdqu(Address(to, 16), xtmp);
+      if (UseAVX >= 2 && UseUnalignedLoadStores) {
+        // Fill 64-byte chunks
+        Label L_fill_64_bytes_loop, L_check_fill_32_bytes;
+        vpbroadcastd(xtmp, xtmp);
+
+        subl(count, 16 << shift);
+        jcc(Assembler::less, L_check_fill_32_bytes);
+        align(16);
+
+        BIND(L_fill_64_bytes_loop);
+        vmovdqu(Address(to, 0), xtmp);
+        vmovdqu(Address(to, 32), xtmp);
+        addptr(to, 64);
+        subl(count, 16 << shift);
+        jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
+
+        BIND(L_check_fill_32_bytes);
+        addl(count, 8 << shift);
+        jccb(Assembler::less, L_check_fill_8_bytes);
+        vmovdqu(Address(to, 0), xtmp);
+        addptr(to, 32);
+        subl(count, 8 << shift);
       } else {
-        movq(Address(to, 0), xtmp);
-        movq(Address(to, 8), xtmp);
-        movq(Address(to, 16), xtmp);
-        movq(Address(to, 24), xtmp);
+        // Fill 32-byte chunks
+        pshufd(xtmp, xtmp, 0);
+
+        subl(count, 8 << shift);
+        jcc(Assembler::less, L_check_fill_8_bytes);
+        align(16);
+
+        BIND(L_fill_32_bytes_loop);
+
+        if (UseUnalignedLoadStores) {
+          movdqu(Address(to, 0), xtmp);
+          movdqu(Address(to, 16), xtmp);
+        } else {
+          movq(Address(to, 0), xtmp);
+          movq(Address(to, 8), xtmp);
+          movq(Address(to, 16), xtmp);
+          movq(Address(to, 24), xtmp);
+        }
+
+        addptr(to, 32);
+        subl(count, 8 << shift);
+        jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
       }
-
-      addptr(to, 32);
-      subl(count, 8 << shift);
-      jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
       BIND(L_check_fill_8_bytes);
       addl(count, 8 << shift);
       jccb(Assembler::zero, L_exit);