From e478df35f529eaa1390aec0f46b39e81ef3cdb90 Mon Sep 17 00:00:00 2001
From: Gabe Black <gblack@eecs.umich.edu>
Date: Wed, 2 Jun 2010 12:58:14 -0500
Subject: ARM: Implement the VFP version of VCMP.

---
 src/arch/arm/isa/formats/fp.isa | 11 +++++-
 src/arch/arm/isa/insts/fp.isa   | 85 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 95 insertions(+), 1 deletion(-)

(limited to 'src/arch')

diff --git a/src/arch/arm/isa/formats/fp.isa b/src/arch/arm/isa/formats/fp.isa
index d15412825..e553b180d 100644
--- a/src/arch/arm/isa/formats/fp.isa
+++ b/src/arch/arm/isa/formats/fp.isa
@@ -640,8 +640,17 @@ let {{
                 // Between half and single precision.
                 return new WarnUnimplemented("vcvtb, vcvtt", machInst);
               case 0x4:
+                if (single) {
+                    return new VcmpS(machInst, vd, vm);
+                } else {
+                    return new VcmpD(machInst, vd, vm);
+                }
               case 0x5:
-                return new WarnUnimplemented("vcmp, vcmpe", machInst);
+                if (single) {
+                    return new VcmpZeroS(machInst, vd, 0);
+                } else {
+                    return new VcmpZeroD(machInst, vd, 0);
+                }
               case 0x7:
                 if (opc3 == 0x3) {
                     if (single) {
diff --git a/src/arch/arm/isa/insts/fp.isa b/src/arch/arm/isa/insts/fp.isa
index e2c7dd79b..0c1ce626b 100644
--- a/src/arch/arm/isa/insts/fp.isa
+++ b/src/arch/arm/isa/insts/fp.isa
@@ -851,4 +851,89 @@ let {{
     header_output += VfpRegRegOpDeclare.subst(vcvtFpDFpSIop);
     decoder_output += VfpRegRegOpConstructor.subst(vcvtFpDFpSIop);
     exec_output += PredOpExecute.subst(vcvtFpDFpSIop);
+
+    vcmpSCode = '''
+        FPSCR fpscr = Fpscr;
+        if (FpDest == FpOp1) {
+            fpscr.n = 0; fpscr.z = 1; fpscr.c = 1; fpscr.v = 0;
+        } else if (FpDest < FpOp1) {
+            fpscr.n = 1; fpscr.z = 0; fpscr.c = 0; fpscr.v = 0;
+        } else if (FpDest > FpOp1) {
+            fpscr.n = 0; fpscr.z = 0; fpscr.c = 1; fpscr.v = 0;
+        } else {
+            fpscr.n = 0; fpscr.z = 0; fpscr.c = 1; fpscr.v = 1;
+        }
+        Fpscr = fpscr;
+    '''
+    vcmpSIop = InstObjParams("vcmps", "VcmpS", "VfpRegRegOp",
+                                     { "code": vcmpSCode,
+                                       "predicate_test": predicateTest }, [])
+    header_output += VfpRegRegOpDeclare.subst(vcmpSIop);
+    decoder_output += VfpRegRegOpConstructor.subst(vcmpSIop);
+    exec_output += PredOpExecute.subst(vcmpSIop);
+
+    vcmpDCode = '''
+        IntDoubleUnion cOp1, cDest;
+        cDest.bits = ((uint64_t)FpDestP0.uw | ((uint64_t)FpDestP1.uw << 32));
+        cOp1.bits = ((uint64_t)FpOp1P0.uw | ((uint64_t)FpOp1P1.uw << 32));
+        FPSCR fpscr = Fpscr;
+        if (cDest.fp == cOp1.fp) {
+            fpscr.n = 0; fpscr.z = 1; fpscr.c = 1; fpscr.v = 0;
+        } else if (cDest.fp < cOp1.fp) {
+            fpscr.n = 1; fpscr.z = 0; fpscr.c = 0; fpscr.v = 0;
+        } else if (cDest.fp > cOp1.fp) {
+            fpscr.n = 0; fpscr.z = 0; fpscr.c = 1; fpscr.v = 0;
+        } else {
+            fpscr.n = 0; fpscr.z = 0; fpscr.c = 1; fpscr.v = 1;
+        }
+        Fpscr = fpscr;
+    '''
+    vcmpDIop = InstObjParams("vcmpd", "VcmpD", "VfpRegRegOp",
+                                     { "code": vcmpDCode,
+                                       "predicate_test": predicateTest }, [])
+    header_output += VfpRegRegOpDeclare.subst(vcmpDIop);
+    decoder_output += VfpRegRegOpConstructor.subst(vcmpDIop);
+    exec_output += PredOpExecute.subst(vcmpDIop);
+
+    vcmpZeroSCode = '''
+        FPSCR fpscr = Fpscr;
+        if (FpDest == imm) {
+            fpscr.n = 0; fpscr.z = 1; fpscr.c = 1; fpscr.v = 0;
+        } else if (FpDest < imm) {
+            fpscr.n = 1; fpscr.z = 0; fpscr.c = 0; fpscr.v = 0;
+        } else if (FpDest > imm) {
+            fpscr.n = 0; fpscr.z = 0; fpscr.c = 1; fpscr.v = 0;
+        } else {
+            fpscr.n = 0; fpscr.z = 0; fpscr.c = 1; fpscr.v = 1;
+        }
+        Fpscr = fpscr;
+    '''
+    vcmpZeroSIop = InstObjParams("vcmpZeros", "VcmpZeroS", "VfpRegImmOp",
+                                     { "code": vcmpZeroSCode,
+                                       "predicate_test": predicateTest }, [])
+    header_output += VfpRegImmOpDeclare.subst(vcmpZeroSIop);
+    decoder_output += VfpRegImmOpConstructor.subst(vcmpZeroSIop);
+    exec_output += PredOpExecute.subst(vcmpZeroSIop);
+
+    vcmpZeroDCode = '''
+        IntDoubleUnion cDest;
+        cDest.bits = ((uint64_t)FpDestP0.uw | ((uint64_t)FpDestP1.uw << 32));
+        FPSCR fpscr = Fpscr;
+        if (cDest.fp == imm) {
+            fpscr.n = 0; fpscr.z = 1; fpscr.c = 1; fpscr.v = 0;
+        } else if (cDest.fp < imm) {
+            fpscr.n = 1; fpscr.z = 0; fpscr.c = 0; fpscr.v = 0;
+        } else if (cDest.fp > imm) {
+            fpscr.n = 0; fpscr.z = 0; fpscr.c = 1; fpscr.v = 0;
+        } else {
+            fpscr.n = 0; fpscr.z = 0; fpscr.c = 1; fpscr.v = 1;
+        }
+        Fpscr = fpscr;
+    '''
+    vcmpZeroDIop = InstObjParams("vcmpZerod", "VcmpZeroD", "VfpRegImmOp",
+                                     { "code": vcmpZeroDCode,
+                                       "predicate_test": predicateTest }, [])
+    header_output += VfpRegImmOpDeclare.subst(vcmpZeroDIop);
+    decoder_output += VfpRegImmOpConstructor.subst(vcmpZeroDIop);
+    exec_output += PredOpExecute.subst(vcmpZeroDIop);
 }};
-- 
cgit v1.2.3