From 14691148cd9300ed7a23dd889b9c0173124b30eb Mon Sep 17 00:00:00 2001
From: Vince Weaver <vince@csl.cornell.edu>
Date: Wed, 21 Oct 2009 13:40:43 -0400
Subject: [PATCH] Implement X86 sse2 movdqu and movdqa instructions

The movdqa instruction should enforce 16-byte alignment.
This implementation does not do that.

These instructions are needed for most of x86_64 spec2k to run.
---
 src/arch/x86/isa/decoder/two_byte_opcodes.isa |  8 +--
 .../simd128/integer/data_transfer/move.py     | 56 ++++++++++++++++++-
 2 files changed, 58 insertions(+), 6 deletions(-)

diff --git a/src/arch/x86/isa/decoder/two_byte_opcodes.isa b/src/arch/x86/isa/decoder/two_byte_opcodes.isa
index c23eeccab..27aabaccc 100644
--- a/src/arch/x86/isa/decoder/two_byte_opcodes.isa
+++ b/src/arch/x86/isa/decoder/two_byte_opcodes.isa
@@ -604,7 +604,7 @@
                     }
                     // repe (0xF3)
                     0x4: decode OPCODE_OP_BOTTOM3 {
-                        0x7: WarnUnimpl::movdqu_Vo_Wo();
+                        0x7: MOVDQU(Vo,Wo);
                         default: UD2();
                     }
                     // operand size (0x66)
@@ -616,7 +616,7 @@
                         0x4: PUNPCKLQDQ(Vo,Wq);
                         0x5: PUNPCKHQDQ(Vo,Wq);
                         0x6: WarnUnimpl::movd_Vo_Ed();
-                        0x7: WarnUnimpl::movdqa_Vo_Wo();
+                        0x7: MOVDQA(Vo,Wo);
                     }
                     default: UD2();
                 }
@@ -702,7 +702,7 @@
                     // repe (0xF3)
                     0x4: decode OPCODE_OP_BOTTOM3 {
                         0x6: MOVQ(Vq,Wq);
-                        0x7: WarnUnimpl::movdqu_Wo_Vo();
+                        0x7: MOVDQU(Wo,Vo);
                         default: UD2();
                     }
                     // operand size (0x66)
@@ -710,7 +710,7 @@
                         0x4: WarnUnimpl::haddpd_Vo_Wo();
                         0x5: WarnUnimpl::hsubpd_Vo_Wo();
                         0x6: WarnUnimpl::movd_Ed_Vd();
-                        0x7: WarnUnimpl::movdqa_Wo_Vo();
+                        0x7: MOVDQA(Wo,Vo);
                         default: UD2();
                     }
                     // repne (0xF2)
diff --git a/src/arch/x86/isa/insts/simd128/integer/data_transfer/move.py b/src/arch/x86/isa/insts/simd128/integer/data_transfer/move.py
index c34bd42bb..ec80ffe73 100644
--- a/src/arch/x86/isa/insts/simd128/integer/data_transfer/move.py
+++ b/src/arch/x86/isa/insts/simd128/integer/data_transfer/move.py
@@ -87,7 +87,59 @@ def macroop MOVQ2DQ_XMM_MMX {
     movfp xmml, mmxm, dataSize=8
     lfpimm xmmh, 0
 };
+
+def macroop MOVDQA_XMM_XMM {
+    movfp xmml, xmmlm
+    movfp xmmh, xmmhm
+};
+
+def macroop MOVDQA_XMM_M {
+    ldfp xmml, seg, sib, "DISPLACEMENT", dataSize=8
+    ldfp xmmh, seg, sib, "DISPLACEMENT + 8", dataSize=8
+};
+
+def macroop MOVDQA_XMM_P {
+    rdip t7
+    ldfp xmml, seg, riprel, "DISPLACEMENT", dataSize=8
+    ldfp xmmh, seg, riprel, "DISPLACEMENT + 8", dataSize=8
+};
+
+def macroop MOVDQA_M_XMM {
+    stfp xmml, seg, sib, "DISPLACEMENT", dataSize=8
+    stfp xmmh, seg, sib, "DISPLACEMENT + 8", dataSize=8
+};
+
+def macroop MOVDQA_P_XMM {
+    rdip t7
+    stfp xmml, seg, riprel, "DISPLACEMENT", dataSize=8
+    stfp xmmh, seg, riprel, "DISPLACEMENT + 8", dataSize=8
+};
+
+def macroop MOVDQU_XMM_XMM {
+    movfp xmml, xmmlm
+    movfp xmmh, xmmhm
+};
+
+def macroop MOVDQU_XMM_M {
+    ldfp xmml, seg, sib, "DISPLACEMENT", dataSize=8
+    ldfp xmmh, seg, sib, "DISPLACEMENT + 8", dataSize=8
+};
+
+def macroop MOVDQU_XMM_P {
+    rdip t7
+    ldfp xmml, seg, riprel, "DISPLACEMENT", dataSize=8
+    ldfp xmmh, seg, riprel, "DISPLACEMENT + 8", dataSize=8
+};
+
+def macroop MOVDQU_M_XMM {
+    stfp xmml, seg, sib, "DISPLACEMENT", dataSize=8
+    stfp xmmh, seg, sib, "DISPLACEMENT + 8", dataSize=8
+};
+
+def macroop MOVDQU_P_XMM {
+    rdip t7
+    stfp xmml, seg, riprel, "DISPLACEMENT", dataSize=8
+    stfp xmmh, seg, riprel, "DISPLACEMENT + 8", dataSize=8
+};
 '''
-# MOVDQA
-# MOVDQU
 # LDDQU
-- 
2.30.2