When promoting the result of fp_to_uint/fp_to_sint,

inform the optimizers that the result must be zero/
sign extended from the smaller type.  For example,
if a fp to unsigned i16 is promoted to fp to i32,
then we are allowed to assume that the extra 16 bits
are zero (because the result of fp to i16 is undefined
if the result does not fit in an i16).  This is
quite aggressive, but should help the optimizers
produce better code.  This requires correcting a
test which thought that fp_to_uint is some kind
of truncation, which it is not: in the testcase
(which does fp to i1), either the fp value converts
to 0 or 1 or the result is undefined, which is
quite different to truncation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58991 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 9bd8bb521b7d89f733c5380d800574ea15179ce8..a4be03a844110b6fcbaad5a7bc595c8f8b2a5c9d 100644 (file)
--- a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -267,7 +267,7 @@ SDValue DAGTypeLegalizer::PromoteIntRes_Constant(SDNode *N) {
   // matter in theory which one we pick, but this tends to give better code?
   unsigned Opc = VT.isByteSized() ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
   SDValue Result = DAG.getNode(Opc, TLI.getTypeToTransformTo(VT),
-                                 SDValue(N, 0));
+                               SDValue(N, 0));
   assert(isa<ConstantSDNode>(Result) && "Didn't constant fold ext?");
   return Result;
 }
@@ -350,22 +350,27 @@ SDValue DAGTypeLegalizer::PromoteIntRes_EXTRACT_VECTOR_ELT(SDNode *N) {
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_FP_TO_XINT(SDNode *N) {
-  unsigned NewOpc = N->getOpcode();
   MVT NVT = TLI.getTypeToTransformTo(N->getValueType(0));
+  unsigned NewOpc = N->getOpcode();
 
   // If we're promoting a UINT to a larger size, check to see if the new node
   // will be legal.  If it isn't, check to see if FP_TO_SINT is legal, since
   // we can use that instead.  This allows us to generate better code for
   // FP_TO_UINT for small destination sizes on targets where FP_TO_UINT is not
   // legal, such as PowerPC.
-  if (N->getOpcode() == ISD::FP_TO_UINT) {
-    if (!TLI.isOperationLegal(ISD::FP_TO_UINT, NVT) &&
-        (TLI.isOperationLegal(ISD::FP_TO_SINT, NVT) ||
-         TLI.getOperationAction(ISD::FP_TO_SINT, NVT)==TargetLowering::Custom))
-      NewOpc = ISD::FP_TO_SINT;
-  }
-
-  return DAG.getNode(NewOpc, NVT, N->getOperand(0));
+  if (N->getOpcode() == ISD::FP_TO_UINT &&
+      !TLI.isOperationLegal(ISD::FP_TO_UINT, NVT) &&
+      TLI.isOperationLegal(ISD::FP_TO_SINT, NVT))
+    NewOpc = ISD::FP_TO_SINT;
+
+  SDValue Res = DAG.getNode(NewOpc, NVT, N->getOperand(0));
+
+  // Assert that the converted value fits in the original type.  If it doesn't
+  // (eg: because the value being converted is too big), then the result of the
+  // original operation was undefined anyway, so the assert is still correct.
+  return DAG.getNode(N->getOpcode() == ISD::FP_TO_UINT ?
+                     ISD::AssertZext : ISD::AssertSext,
+                     NVT, Res, DAG.getValueType(N->getValueType(0)));
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_INT_EXTEND(SDNode *N) {

diff --git a/test/CodeGen/X86/trunc-to-bool.ll b/test/CodeGen/X86/trunc-to-bool.ll
index 9dd14252ace8dfc5f81ab5d3fc899878e4dc60fa..25a1191d8f14f75823f682ca33804244e96e5050 100644 (file)
--- a/test/CodeGen/X86/trunc-to-bool.ll
+++ b/test/CodeGen/X86/trunc-to-bool.ll
@@ -2,7 +2,7 @@
 ; sure only the LSBit survives. Test that this is the case both for a returned
 ; value and as the operand of a branch.
 ; RUN: llvm-as < %s | llc -march=x86 | grep {\\(and\\)\\|\\(test.*\\\$1\\)} | \
-; RUN:   count 6
+; RUN:   count 5
 
 define i1 @test1(i32 %X) zeroext {
     %Y = trunc i32 %X to i1
@@ -40,15 +40,6 @@ cond_false:
     ret i32 42
 }
 
-define i32 @test5(float %f) {
-    %tmp = fptoui float %f to i1
-    br i1 %tmp, label %cond_true, label %cond_false
-cond_true:
-    ret i32 21
-cond_false:
-    ret i32 42
-}
-
 define i32 @test6(double %d) {
     %tmp = fptosi double %d to i1
     br i1 %tmp, label %cond_true, label %cond_false