From 449db8a37074fedb039e3b8fc6c5e09fdb167d82 Mon Sep 17 00:00:00 2001
From: Gabe Black <gblack@eecs.umich.edu>
Date: Mon, 17 Aug 2009 18:34:19 -0700
Subject: [PATCH] X86: Implement the media sqrt instructions.

---
 src/arch/x86/isa/decoder/two_byte_opcodes.isa |  8 +--
 .../floating_point/arithmetic/square_root.py  | 63 +++++++++++++++++--
 2 files changed, 61 insertions(+), 10 deletions(-)

diff --git a/src/arch/x86/isa/decoder/two_byte_opcodes.isa b/src/arch/x86/isa/decoder/two_byte_opcodes.isa
index 4c608264c..faf9389a4 100644
--- a/src/arch/x86/isa/decoder/two_byte_opcodes.isa
+++ b/src/arch/x86/isa/decoder/two_byte_opcodes.isa
@@ -465,7 +465,7 @@
                 // no prefix
                 0x0: decode OPCODE_OP_BOTTOM3 {
                     0x0: movmskps_Gd_VRo();
-                    0x1: sqrtps_Vo_Wo();
+                    0x1: Inst::SQRTPS(Vo,Wo);
                     0x2: rqsrtps_Vo_Wo();
                     0x3: rcpps_Vo_Wo();
                     0x4: Inst::ANDPS(Vo,Wo);
@@ -475,7 +475,7 @@
                 }
                 // repe (0xF3)
                 0x4: decode OPCODE_OP_BOTTOM3 {
-                    0x1: sqrtss_Vd_Wd();
+                    0x1: Inst::SQRTSS(Vd,Wd);
                     0x2: rsqrtss_Vd_Wd();
                     0x3: rcpss_Vd_Wd();
                     default: Inst::UD2();
@@ -483,7 +483,7 @@
                 // operand size (0x66)
                 0x1: decode OPCODE_OP_BOTTOM3 {
                     0x0: movmskpd_Gd_VRo();
-                    0x1: sqrtpd_Vo_Wo();
+                    0x1: Inst::SQRTPD(Vo,Wo);
                     0x4: Inst::ANDPD(Vo,Wo);
                     0x5: Inst::ANDNPD(Vo,Wo);
                     0x6: Inst::ORPD(Vo,Wo);
@@ -495,7 +495,7 @@
                 format Inst {
                     // repne (0xF2)
                     0x8: decode OPCODE_OP_BOTTOM3 {
-                        0x1: SQRTSD(Vq,Wq);
+                        0x1: Inst::SQRTSD(Vq,Wq);
                         default: UD2();
                     }
                     default: UD2();
diff --git a/src/arch/x86/isa/insts/simd128/floating_point/arithmetic/square_root.py b/src/arch/x86/isa/insts/simd128/floating_point/arithmetic/square_root.py
index 1591eee8e..fdeb30ddc 100644
--- a/src/arch/x86/isa/insts/simd128/floating_point/arithmetic/square_root.py
+++ b/src/arch/x86/isa/insts/simd128/floating_point/arithmetic/square_root.py
@@ -54,22 +54,73 @@
 # Authors: Gabe Black
 
 microcode = '''
-# SQRTPS
-# SQRTPD
-# SQRTSS
+def macroop SQRTSS_XMM_XMM {
+    msqrt xmml, xmmlm, size=4, ext=1
+};
+
+def macroop SQRTSS_XMM_M {
+    ldfp ufp1, seg, sib, disp, dataSize=8
+    msqrt xmml, ufp1, size=4, ext=1
+};
+
+def macroop SQRTSS_XMM_P {
+    rdip t7
+    ldfp ufp1, seg, riprel, disp, dataSize=8
+    msqrt xmml, ufp1, size=4, ext=1
+};
+
+def macroop SQRTPS_XMM_XMM {
+    msqrt xmml, xmmlm, size=4, ext=0
+    msqrt xmmh, xmmhm, size=4, ext=0
+};
+
+def macroop SQRTPS_XMM_M {
+    ldfp ufp1, seg, sib, "DISPLACEMENT", dataSize=8
+    ldfp ufp2, seg, sib, "DISPLACEMENT + 8", dataSize=8
+    msqrt xmml, ufp1, size=4, ext=0
+    msqrt xmmh, ufp2, size=4, ext=0
+};
+
+def macroop SQRTPS_XMM_P {
+    rdip t7
+    ldfp ufp1, seg, riprel, "DISPLACEMENT", dataSize=8
+    ldfp ufp2, seg, riprel, "DISPLACEMENT + 8", dataSize=8
+    msqrt xmml, ufp1, size=4, ext=0
+    msqrt xmmh, ufp2, size=4, ext=0
+};
 
 def macroop SQRTSD_XMM_XMM {
-    sqrtfp xmml, xmml, xmmlm
+    msqrt xmml, xmmlm, size=8, ext=1
 };
 
 def macroop SQRTSD_XMM_M {
     ldfp ufp1, seg, sib, disp, dataSize=8
-    sqrtfp xmml, xmml, ufp1
+    msqrt xmml, ufp1, size=8, ext=1
 };
 
 def macroop SQRTSD_XMM_P {
     rdip t7
     ldfp ufp1, seg, riprel, disp, dataSize=8
-    sqrtfp xmml, xmml, ufp1
+    msqrt xmml, ufp1, size=8, ext=1
+};
+
+def macroop SQRTPD_XMM_XMM {
+    msqrt xmml, xmmlm, size=8, ext=0
+    msqrt xmmh, xmmhm, size=8, ext=0
+};
+
+def macroop SQRTPD_XMM_M {
+    ldfp ufp1, seg, sib, "DISPLACEMENT", dataSize=8
+    ldfp ufp2, seg, sib, "DISPLACEMENT + 8", dataSize=8
+    msqrt xmml, ufp1, size=8, ext=0
+    msqrt xmmh, ufp2, size=8, ext=0
+};
+
+def macroop SQRTPD_XMM_P {
+    rdip t7
+    ldfp ufp1, seg, riprel, "DISPLACEMENT", dataSize=8
+    ldfp ufp2, seg, riprel, "DISPLACEMENT + 8", dataSize=8
+    msqrt xmml, ufp1, size=8, ext=0
+    msqrt xmmh, ufp2, size=8, ext=0
 };
 '''