ARM: Implement the VFP version of vmul.

3 files changed, 56 insertions, 1 deletions

diff --git a/src/arch/arm/insts/static_inst.hh b/src/arch/arm/insts/static_inst.hh
index 33453bec6..3af9ef3b0 100644
--- a/src/arch/arm/insts/static_inst.hh
+++ b/src/arch/arm/insts/static_inst.hh
@@ -50,6 +50,11 @@ namespace ArmISA
 class ArmStaticInst : public StaticInst
 {
   protected:
+    union IntDoubleUnion {
+        uint64_t bits;
+        double fp;
+    };
+
     int32_t shift_rm_imm(uint32_t base, uint32_t shamt,
                          uint32_t type, uint32_t cfval) const;
     int32_t shift_rm_rs(uint32_t base, uint32_t shamt,
diff --git a/src/arch/arm/isa/formats/fp.isa b/src/arch/arm/isa/formats/fp.isa
index 848ce907c..2222b1e62 100644
--- a/src/arch/arm/isa/formats/fp.isa
+++ b/src/arch/arm/isa/formats/fp.isa
@@ -486,7 +486,25 @@ let {{
             return new WarnUnimplemented("vmla, vmls", machInst);
           case 0x2:
             if ((opc3 & 0x1) == 0) {
-                return new WarnUnimplemented("vmul", machInst);
+                uint32_t vd;
+                uint32_t vm;
+                uint32_t vn;
+                if (bits(machInst, 8) == 0) {
+                    vd = bits(machInst, 22) | (bits(machInst, 15, 12) << 1);
+                    vm = bits(machInst, 5) | (bits(machInst, 3, 0) << 1);
+                    vn = bits(machInst, 7) | (bits(machInst, 19, 16) << 1);
+                    return new VmulS(machInst, (IntRegIndex)vd,
+                            (IntRegIndex)vn, (IntRegIndex)vm);
+                } else {
+                    vd = (bits(machInst, 22) << 5) |
+                         (bits(machInst, 15, 12) << 1);
+                    vm = (bits(machInst, 5) << 5) |
+                         (bits(machInst, 3, 0) << 1);
+                    vn = (bits(machInst, 7) << 5) |
+                         (bits(machInst, 19, 16) << 1);
+                    return new VmulD(machInst, (IntRegIndex)vd,
+                            (IntRegIndex)vn, (IntRegIndex)vm);
+                }
             }
           case 0x1:
             return new WarnUnimplemented("vnmla, vnmls, vnmul", machInst);
diff --git a/src/arch/arm/isa/insts/fp.isa b/src/arch/arm/isa/insts/fp.isa
index ab37c3e3b..ef79ea420 100644
--- a/src/arch/arm/isa/insts/fp.isa
+++ b/src/arch/arm/isa/insts/fp.isa
@@ -226,4 +226,36 @@ let {{
     header_output += RegRegRegOpDeclare.subst(vmov2Core2RegIop);
     decoder_output += RegRegRegOpConstructor.subst(vmov2Core2RegIop);
     exec_output += PredOpExecute.subst(vmov2Core2RegIop);
+
+    vmulSCode = '''
+        FpDest = FpOp1 * FpOp2;
+        if ((isinf(FpOp1) && FpOp2 == 0) || (isinf(FpOp2) && FpOp1 == 0)) {
+            FpDest = NAN;
+        }
+    '''
+    vmulSIop = InstObjParams("vmuls", "VmulS", "RegRegRegOp",
+                                     { "code": vmulSCode,
+                                       "predicate_test": predicateTest }, [])
+    header_output += RegRegRegOpDeclare.subst(vmulSIop);
+    decoder_output += RegRegRegOpConstructor.subst(vmulSIop);
+    exec_output += PredOpExecute.subst(vmulSIop);
+
+    vmulDCode = '''
+        IntDoubleUnion cOp1, cOp2, cDest;
+        cOp1.bits = ((uint64_t)FpOp1P0.uw | ((uint64_t)FpOp1P1.uw << 32));
+        cOp2.bits = ((uint64_t)FpOp2P0.uw | ((uint64_t)FpOp2P1.uw << 32));
+        cDest.fp = cOp1.fp * cOp2.fp;
+        if ((isinf(cOp1.fp) && cOp2.fp == 0) ||
+                (isinf(cOp2.fp) && cOp1.fp == 0)) {
+            cDest.fp = NAN;
+        }
+        FpDestP0.uw = cDest.bits;
+        FpDestP1.uw = cDest.bits >> 32;
+    '''
+    vmulDIop = InstObjParams("vmuld", "VmulD", "RegRegRegOp",
+                                     { "code": vmulDCode,
+                                       "predicate_test": predicateTest }, [])
+    header_output += RegRegRegOpDeclare.subst(vmulDIop);
+    decoder_output += RegRegRegOpConstructor.subst(vmulDIop);
+    exec_output += PredOpExecute.subst(vmulDIop);
 }};