Fixed a bug in the ExecutionDependencyFix pass that caused dependencies to not propagate through implicit defs.

llvm-svn: 165102

2 files changed, 28 insertions, 3 deletions

diff --git a/llvm/lib/CodeGen/ExecutionDepsFix.cpp b/llvm/lib/CodeGen/ExecutionDepsFix.cpp
index fee8e47b832..2311842671a 100644
--- a/llvm/lib/CodeGen/ExecutionDepsFix.cpp
+++ b/llvm/lib/CodeGen/ExecutionDepsFix.cpp
@@ -626,9 +626,12 @@ void ExeDepsFix::visitSoftInstr(MachineInstr *mi, unsigned mask) {
   }
   dv->Instrs.push_back(mi);
 
-  // Finally set all defs and non-collapsed uses to dv.
-  for (unsigned i = 0, e = mi->getDesc().getNumOperands(); i != e; ++i) {
-    MachineOperand &mo = mi->getOperand(i);
+  // Finally set all defs and non-collapsed uses to dv. We must iterate through
+  // all the operators, including imp-def ones.
+  for (MachineInstr::mop_iterator ii = mi->operands_begin(),
+                                  ee = mi->operands_end();
+                                  ii != ee; ++ii) {
+    MachineOperand &mo = *ii;
     if (!mo.isReg()) continue;
     int rx = regIndex(mo.getReg());
     if (rx < 0) continue;
diff --git a/llvm/test/CodeGen/ARM/deps-fix.ll b/llvm/test/CodeGen/ARM/deps-fix.ll
new file mode 100644
index 00000000000..288697a4dc7
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/deps-fix.ll
@@ -0,0 +1,22 @@
+; RUN: llc < %s -march=arm -mcpu=cortex-a9 -mattr=+neon,+neonfp -float-abi=hard -mtriple armv7-linux-gnueabi | FileCheck %s
+
+;; This test checks that the ExecutionDepsFix pass performs the domain changes
+;; even when some dependencies are propagated through implicit definitions.
+
+; CHECK: fun_a
+define <4 x float> @fun_a(<4 x float> %in, <4 x float> %x, float %y) nounwind {
+; CHECK: vext
+; CHECK: vext
+; CHECK: vadd.f32
+  %1 = insertelement <4 x float> %in, float %y, i32 0
+  %2 = fadd <4 x float> %1, %x  
+  ret <4 x float> %2
+}
+; CHECK: fun_b
+define <4 x i32> @fun_b(<4 x i32> %in, <4 x i32> %x, i32 %y) nounwind {
+; CHECK: vmov.32
+; CHECK: vadd.i32
+  %1 = insertelement <4 x i32> %in, i32 %y, i32 0
+  %2 = add <4 x i32> %1, %x  
+  ret <4 x i32> %2
+}