Subzero: Add fabs intrinsic support.

src/IceInstX8632.h

Index: src/IceInstX8632.h
diff --git a/src/IceInstX8632.h b/src/IceInstX8632.h
index c86128d0f2deb831acb72cd036927ef66a7be32a..eb927418981cc2c3ce521b29fa0c95f95f639e4a 100644
--- a/src/IceInstX8632.h
+++ b/src/IceInstX8632.h
@@ -234,6 +234,7 @@ public:
     Pshufd,
     Psll,
     Psra,
+    Psrl,
     Psub,
     Push,
     Pxor,
@@ -786,7 +787,7 @@ void emitIASXmmShift(const Cfg *Func, Type Ty, const Variable *Var,
                      const Operand *Src,
                      const x86::AssemblerX86::XmmEmitterShiftOp &Emitter);
 
-template <InstX8632::InstKindX8632 K>
+template <InstX8632::InstKindX8632 K, bool AllowAllTypes = false>
 class InstX8632BinopXmmShift : public InstX8632 {
   InstX8632BinopXmmShift() = delete;
   InstX8632BinopXmmShift(const InstX8632BinopXmmShift &) = delete;
@@ -807,8 +808,8 @@ public:
   }
   void emitIAS(const Cfg *Func) const override {
     Type Ty = getDest()->getType();
-    assert(Ty == IceType_v8i16 || Ty == IceType_v8i1 || Ty == IceType_v4i32 ||
-           Ty == IceType_v4i1);
+    assert(AllowAllTypes || Ty == IceType_v8i16 || Ty == IceType_v8i1 ||

[jvoung (off chromium), 2015/03/19 16:37:53]

We might be able to get away with just removing the assert at this layer, and
the lower level assembler code will know what types are handled:

E.g., for psra:


  if (Ty == IceType_i16) {
    EmitUint8(0xE1);
  } else {
    assert(Ty == IceType_i32);
  //...
  }

but what's missing is the check that isVectorType(Ty)

[Jim Stichnoth, 2015/03/19 18:35:15]

The problem here is that psra's Ty argument is actually the element type (see 2
lines below this).  If Ty==IceType_i32, then ElementTy is also IceType_i32 which
psra will allow.

Maybe we could simplify this as a test for vector type, then let psll/psra/psrl
do the element-type check?

Except that psrl is explicitly used for fabs of scalar float and double so it
shouldn't be subject to an isVectorType check.

I made the isVectorType() simplification, ptal.

[jvoung (off chromium), 2015/03/19 18:50:33]

Ah, ok that's what I missed.
Okay this is a bit simpler / less specific to psra and psll.

+           Ty == IceType_v4i32 || Ty == IceType_v4i1);
     Type ElementTy = typeElementType(Ty);
     assert(getSrcSize() == 2);
     emitIASXmmShift(Func, ElementTy, getDest(), getSrc(1), Emitter);
@@ -1013,6 +1014,7 @@ typedef InstX8632BinopXmm<InstX8632::Divss, false> InstX8632Divss;
 typedef InstX8632BinopGPRShift<InstX8632::Rol> InstX8632Rol;
 typedef InstX8632BinopGPRShift<InstX8632::Shl> InstX8632Shl;
 typedef InstX8632BinopXmmShift<InstX8632::Psll> InstX8632Psll;
+typedef InstX8632BinopXmmShift<InstX8632::Psrl, true> InstX8632Psrl;
 typedef InstX8632BinopGPRShift<InstX8632::Shr> InstX8632Shr;
 typedef InstX8632BinopGPRShift<InstX8632::Sar> InstX8632Sar;
 typedef InstX8632BinopXmmShift<InstX8632::Psra> InstX8632Psra;
@@ -1632,6 +1634,7 @@ template <> void InstX8632Pmull::emit(const Cfg *Func) const;
 template <> void InstX8632Pmuludq::emit(const Cfg *Func) const;
 template <> void InstX8632Psll::emit(const Cfg *Func) const;
 template <> void InstX8632Psra::emit(const Cfg *Func) const;
+template <> void InstX8632Psrl::emit(const Cfg *Func) const;
 template <> void InstX8632Psub::emit(const Cfg *Func) const;
 template <> void InstX8632Sqrtss::emit(const Cfg *Func) const;
 template <> void InstX8632Subss::emit(const Cfg *Func) const;