X86: Implement a locking version of BTR.

diff --git a/src/arch/x86/isa/insts/general_purpose/compare_and_test/bit_test.py b/src/arch/x86/isa/insts/general_purpose/compare_and_test/bit_test.py
index bc222300d5061e9d797b564080e4cbf1a295d177..4dca52010d327cadad992cd9cbf21470dc902c86 100644 (file)
--- a/src/arch/x86/isa/insts/general_purpose/compare_and_test/bit_test.py
+++ b/src/arch/x86/isa/insts/general_purpose/compare_and_test/bit_test.py
@@ -214,6 +214,27 @@ def macroop BTR_P_I {
     st t1, seg, riprel, disp
 };
 
+def macroop BTR_LOCKED_M_I {
+    limm t1, imm, dataSize=asz
+    limm t4, "(uint64_t(-(2ULL)))"
+    roli t4, t4, imm
+    ldstl t1, seg, sib, disp
+    sexti t0, t1, imm, flags=(CF,)
+    and t1, t1, t4
+    stul t1, seg, sib, disp
+};
+
+def macroop BTR_LOCKED_P_I {
+    rdip t7, dataSize=asz
+    limm t1, imm, dataSize=asz
+    limm t4, "(uint64_t(-(2ULL)))"
+    roli t4, t4, imm
+    ldstl t1, seg, riprel, disp
+    sexti t0, t1, imm, flags=(CF,)
+    and t1, t1, t4
+    stul t1, seg, riprel, disp
+};
+
 def macroop BTR_R_R {
     sext t0, reg, regm, flags=(CF,)
     limm t1, "(uint64_t(-(2ULL)))"
@@ -246,6 +267,31 @@ def macroop BTR_P_R {
     st t1, seg, [1, t3, t7], disp
 };
 
+def macroop BTR_LOCKED_M_R {
+    srai t2, reg, 3, dataSize=asz
+    srai t3, t2, ldsz, dataSize=asz
+    lea t3, flatseg, [dsz, t3, base], dataSize=asz
+    limm t4, "(uint64_t(-(2ULL)))"
+    rol t4, t4, reg
+    ldstl t1, seg, [scale, index, t3], disp
+    sext t0, t1, reg, flags=(CF,)
+    and t1, t1, t4
+    stul t1, seg, [scale, index, t3], disp
+};
+
+def macroop BTR_LOCKED_P_R {
+    rdip t7, dataSize=asz
+    srai t2, reg, 3, dataSize=asz
+    srai t3, t2, ldsz, dataSize=asz
+    lea t3, flatseg, [dsz, t3, base], dataSize=asz
+    limm t4, "(uint64_t(-(2ULL)))"
+    rol t4, t4, reg
+    ldstl t1, seg, [1, t3, t7], disp
+    sext t0, t1, reg, flags=(CF,)
+    and t1, t1, t4
+    stul t1, seg, [1, t3, t7], disp
+};
+
 def macroop BTS_R_I {
     sexti t0, reg, imm, flags=(CF,)
     limm t1, 1