Fix makeLibCall argument (signed) in SoftenFloatRes_XINT_TO_FP function
authorPetar Jovanovic <petar.jovanovic@imgtec.com>
Tue, 10 Feb 2015 23:30:14 +0000 (23:30 +0000)
committerPetar Jovanovic <petar.jovanovic@imgtec.com>
Tue, 10 Feb 2015 23:30:14 +0000 (23:30 +0000)
The isSigned argument of makeLibCall function was hard-coded to false
(unsigned). This caused zero extension on MIPS64 soft float.
As the result SingleSource/Benchmarks/Stanford/FloatMM test and
SingleSource/UnitTests/2005-07-17-INT-To-FP test failed.
The solution was to use the proper argument.

Patch by Strahinja Petrovic.

Differential Revision: http://reviews.llvm.org/D7292

llvm-svn: 228765

llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
llvm/test/CodeGen/Mips/mips64sinttofpsf.ll [new file with mode: 0644]

index 4591e79..b596715 100644 (file)
@@ -658,7 +658,7 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_XINT_TO_FP(SDNode *N) {
                            NVT, N->getOperand(0));
   return TLI.makeLibCall(DAG, LC,
                          TLI.getTypeToTransformTo(*DAG.getContext(), RVT),
-                         &Op, 1, false, dl).first;
+                         &Op, 1, Signed, dl).first;
 }
 
 
diff --git a/llvm/test/CodeGen/Mips/mips64sinttofpsf.ll b/llvm/test/CodeGen/Mips/mips64sinttofpsf.ll
new file mode 100644 (file)
index 0000000..d3d4603
--- /dev/null
@@ -0,0 +1,15 @@
+; RUN: llc -march=mips64 -mcpu=mips64r2 -soft-float -O0 < %s | FileCheck %s
+
+
+define double @foo() #0 {
+entry:
+  %x = alloca i32, align 4
+  store volatile i32 -32, i32* %x, align 4
+  %0 = load volatile i32* %x, align 4
+  %conv = sitofp i32 %0 to double
+  ret double %conv
+
+; CHECK-NOT:        dsll
+; CHECK-NOT:        dsrl
+
+}