commit 8d1072e7beddc254f78c4dc15e789fdb7c9631d4
author Matt Wala <wala@chromium.org> Fri Jul 11 15:43:51 2014 -0700
committer Matt Wala <wala@chromium.org> Fri Jul 11 15:43:51 2014 -0700
tree 1f76200da5924d052d06e969753a4daa39751491
parent 78b4c0b8168dc867e0b822a3fec4f31ea1e3e16e

Lower vector floating point arithmetic operations.

This adds lowering code for fadd, fsub, fmul, fdiv, and frem. frem, having no native x86 counterpart, is implemented by making a helper call.

BUG=none
R=jvoung@chromium.org, stichnot@chromium.org

Review URL: https://codereview.chromium.org/389653002

src/IceTargetLoweringX8632.cpp

diff --git a/src/IceTargetLoweringX8632.cpp b/src/IceTargetLoweringX8632.cpp
index 2652e8b..f1b8c25 100644
--- a/src/IceTargetLoweringX8632.cpp
+++ b/src/IceTargetLoweringX8632.cpp
@@ -1136,7 +1136,59 @@
       llvm_unreachable("FP instruction with i64 type");
       break;
     }
-  } else { // Dest->getType() != IceType_i64
+  } else if (isVectorType(Dest->getType())) {
+    switch (Inst->getOp()) {
+    case InstArithmetic::_num:
+      llvm_unreachable("Unknown arithmetic operator");
+      break;
+    case InstArithmetic::Add:
+    case InstArithmetic::And:
+    case InstArithmetic::Or:
+    case InstArithmetic::Xor:
+    case InstArithmetic::Sub:
+    case InstArithmetic::Mul:
+    case InstArithmetic::Shl:
+    case InstArithmetic::Lshr:
+    case InstArithmetic::Ashr:
+    case InstArithmetic::Udiv:
+    case InstArithmetic::Sdiv:
+    case InstArithmetic::Urem:
+    case InstArithmetic::Srem:
+      // TODO(wala): Handle these.
+      Func->setError("Unhandled instruction");
+      break;
+    case InstArithmetic::Fadd: {
+      Variable *T = makeReg(Dest->getType());
+      _movp(T, Src0);
+      _addps(T, Src1);
+      _movp(Dest, T);
+    } break;
+    case InstArithmetic::Fsub: {
+      Variable *T = makeReg(Dest->getType());
+      _movp(T, Src0);
+      _subps(T, Src1);
+      _movp(Dest, T);
+    } break;
+    case InstArithmetic::Fmul: {
+      Variable *T = makeReg(Dest->getType());
+      _movp(T, Src0);
+      _mulps(T, Src1);
+      _movp(Dest, T);
+    } break;
+    case InstArithmetic::Fdiv: {
+      Variable *T = makeReg(Dest->getType());
+      _movp(T, Src0);
+      _divps(T, Src1);
+      _movp(Dest, T);
+    } break;
+    case InstArithmetic::Frem: {
+      const SizeT MaxSrcs = 1;
+      InstCall *Call = makeHelperCall("__frem_v4f32", Dest, MaxSrcs);
+      Call->addArg(Src0);
+      lowerCall(Call);
+    } break;
+    }
+  } else { // Dest->getType() is non-i64 scalar
     Variable *T_edx = NULL;
     Variable *T = NULL;
     switch (Inst->getOp()) {
@@ -1454,7 +1506,7 @@
     Inst *FakeUse = InstFakeUse::create(Func, ReturnReg);
     Context.insert(FakeUse);
   }
-  
+
   if (!Dest)
     return;