From 506bf83595d0f941ade9e55a192c1d831fc4c90c Mon Sep 17 00:00:00 2001
From: Gabe Black <gblack@eecs.umich.edu>
Date: Sun, 26 Aug 2007 20:36:46 -0700
Subject: [PATCH] X86: Implement cmps (string compare)

--HG--
extra : convert_revision : 0d6b783b2246b8ad8d91e4c63e407307ee11c651
---
 src/arch/x86/isa/decoder/one_byte_opcodes.isa |  4 +-
 .../x86/isa/insts/string/compare_strings.py   | 70 +++++++++++++++----
 2 files changed, 59 insertions(+), 15 deletions(-)

diff --git a/src/arch/x86/isa/decoder/one_byte_opcodes.isa b/src/arch/x86/isa/decoder/one_byte_opcodes.isa
index ee7fbc683..d8db47063 100644
--- a/src/arch/x86/isa/decoder/one_byte_opcodes.isa
+++ b/src/arch/x86/isa/decoder/one_byte_opcodes.isa
@@ -332,8 +332,8 @@
             0x3: mov_Ov_rAX();
             0x4: movs_Yb_Xb();
             0x5: movs_Yv_Xv();
-            0x6: cmps_Yb_Xb();
-            0x7: cmps_Yv_Xv();
+            0x6: StringInst::CMPS(Yb,Xb);
+            0x7: StringInst::CMPS(Yv,Xv);
         }
         0x15: decode OPCODE_OP_BOTTOM3 {
             0x0: Inst::TEST(rAb,Ib);
diff --git a/src/arch/x86/isa/insts/string/compare_strings.py b/src/arch/x86/isa/insts/string/compare_strings.py
index 1484c4706..71b8511b4 100644
--- a/src/arch/x86/isa/insts/string/compare_strings.py
+++ b/src/arch/x86/isa/insts/string/compare_strings.py
@@ -53,16 +53,60 @@
 #
 # Authors: Gabe Black
 
-microcode = ""
-#let {{
-#    class CMPS(Inst):
-#	"GenFault ${new UnimpInstFault}"
-#    class CMPSB(Inst):
-#	"GenFault ${new UnimpInstFault}"
-#    class CMPSW(Inst):
-#	"GenFault ${new UnimpInstFault}"
-#    class CMPSD(Inst):
-#	"GenFault ${new UnimpInstFault}"
-#    class CMPSQ(Inst):
-#	"GenFault ${new UnimpInstFault}"
-#}};
+microcode = '''
+def macroop CMPS_M_M {
+    # Find the constant we need to either add or subtract from rdi
+    ruflag t0, 10
+    movi t3, t3, dsz, flags=(CEZF,), dataSize=asz
+    subi t4, t0, dsz, dataSize=asz
+    mov t3, t3, t4, flags=(nCEZF,), dataSize=asz
+
+    ld t1, seg, [1, t0, rsi]
+    ld t2, es, [1, t0, rdi]
+    sub t0, t1, t2, flags=(OF, SF, ZF, AF, PF, CF)
+
+    add rdi, rdi, t3, dataSize=asz
+    add rsi, rsi, t3, dataSize=asz
+};
+
+#
+# Versions which have the rep prefix. These could benefit from some loop
+# unrolling.
+#
+
+def macroop CMPS_E_M_M {
+    # Find the constant we need to either add or subtract from rdi
+    ruflag t0, 10
+    movi t3, t3, dsz, flags=(CEZF,), dataSize=asz
+    subi t4, t0, dsz, dataSize=asz
+    mov t3, t3, t4, flags=(nCEZF,), dataSize=asz
+
+    ld t1, seg, [1, t0, rsi]
+    ld t2, es, [1, t0, rdi]
+    sub t0, t1, t2, flags=(OF, SF, ZF, AF, PF, CF)
+
+    subi rcx, rcx, 1, flags=(EZF,), dataSize=asz
+    add rdi, rdi, t3, dataSize=asz
+    add rsi, rsi, t3, dataSize=asz
+    bri t0, 4, flags=(CSTRZnEZF,)
+    fault "NoFault"
+};
+
+def macroop CMPS_N_M_M {
+    # Find the constant we need to either add or subtract from rdi
+    ruflag t0, 10
+    movi t3, t3, dsz, flags=(CEZF,), dataSize=asz
+    subi t4, t0, dsz, dataSize=asz
+    mov t3, t3, t4, flags=(nCEZF,), dataSize=asz
+
+    ld t1, seg, [1, t0, rsi]
+    ld t2, es, [1, t0, rdi]
+    sub t0, t1, t2, flags=(OF, SF, ZF, AF, PF, CF)
+
+    subi rcx, rcx, 1, flags=(EZF,), dataSize=asz
+    add rdi, rdi, t3, dataSize=asz
+    add rsi, rsi, t3, dataSize=asz
+    bri t0, 4, flags=(CSTRnZnEZF,)
+    fault "NoFault"
+};
+'''