During SelectionDAG building explicitly set a node to constant zero when the

value is zero.
This allows optmizations to kick in more easily.
Fix some test cases so that they remain meaningful (i.e., not completely dead
coded) when optimizations apply.

<rdar://problem/14096009> superfluous multiply by high part of zero-extended
value.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184222 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 49 insertions, 5 deletions

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 608fc275b5a..a45485382ad 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -718,6 +718,14 @@ SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG,
       unsigned NumSignBits = LOI->NumSignBits;
       unsigned NumZeroBits = LOI->KnownZero.countLeadingOnes();
 
+      if (NumZeroBits == RegSize) {
+        // The current value is a zero.
+        // Explicitly express that as it would be easier for
+        // optimizations to kick in.
+        Parts[i] = DAG.getConstant(0, RegisterVT);
+        continue;
+      }
+
       // FIXME: We capture more information than the dag can represent.  For
       // now, just use the tightest assertzext/assertsext possible.
       bool isSExt = true;
diff --git a/test/CodeGen/ARM/mvn.ll b/test/CodeGen/ARM/mvn.ll
index 571c21a833e..2c5ccd7442e 100644
--- a/test/CodeGen/ARM/mvn.ll
+++ b/test/CodeGen/ARM/mvn.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -march=arm | grep mvn | count 8
+; RUN: llc < %s -march=arm | grep mvn | count 9
 
 define i32 @f1() {
 entry:
diff --git a/test/CodeGen/X86/2007-10-12-CoalesceExtSubReg.ll b/test/CodeGen/X86/2007-10-12-CoalesceExtSubReg.ll
index 8091bd1bc1c..d3a47aefb7d 100644
--- a/test/CodeGen/X86/2007-10-12-CoalesceExtSubReg.ll
+++ b/test/CodeGen/X86/2007-10-12-CoalesceExtSubReg.ll
@@ -7,8 +7,10 @@ entry:
 cond_next127:		; preds = %cond_next391, %entry
 	%v.1 = phi i32 [ undef, %entry ], [ %tmp411, %cond_next391 ]		; <i32> [#uses=1]
 	%tmp149 = mul i32 0, %v.1		; <i32> [#uses=0]
-	%tmp254 = and i32 0, 15		; <i32> [#uses=1]
-	%tmp256 = and i32 0, 15		; <i32> [#uses=2]
+	%tmpss = load i32* %ss, align 4		; <i32> [#uses=1]
+	%tmpbp = load i32* %bp, align 4		; <i32> [#uses=2]
+	%tmp254 = and i32 %tmpss, 15		; <i32> [#uses=1]
+	%tmp256 = and i32 %tmpbp, 15		; <i32> [#uses=2]
 	br label %cond_next391
 
 cond_next391:		; preds = %cond_next127
diff --git a/test/CodeGen/X86/2008-01-08-SchedulerCrash.ll b/test/CodeGen/X86/2008-01-08-SchedulerCrash.ll
index 39af9319c8d..9b9b781cfa2 100644
--- a/test/CodeGen/X86/2008-01-08-SchedulerCrash.ll
+++ b/test/CodeGen/X86/2008-01-08-SchedulerCrash.ll
@@ -19,7 +19,7 @@ bb917:		; preds = %entry
 	ret i32 0
 
 bb951:		; preds = %bb986, %entry
-	%tmp955 = sdiv i32 0, 2		; <i32> [#uses=3]
+	%tmp955 = sdiv i32 %offset, 2		; <i32> [#uses=3]
 	%tmp961 = getelementptr %struct.indexentry* null, i32 %tmp955, i32 0		; <i32*> [#uses=1]
 	br i1 %cond, label %bb986, label %bb967
 
diff --git a/test/CodeGen/X86/i128-mul.ll b/test/CodeGen/X86/i128-mul.ll
index e9d30d67019..c0b85dfd211 100644
--- a/test/CodeGen/X86/i128-mul.ll
+++ b/test/CodeGen/X86/i128-mul.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -march=x86-64
+; RUN: llc < %s -march=x86-64 | FileCheck %s
 ; PR1198
 
 define i64 @foo(i64 %x, i64 %y) {
@@ -10,3 +10,37 @@ define i64 @foo(i64 %x, i64 %y) {
         %tmp4 = trunc i128 %tmp3 to i64
         ret i64 %tmp4
 }
+
+; <rdar://problem/14096009> superfluous multiply by high part of
+; zero-extended value.
+; CHECK: @mul1
+; CHECK-NOT: imulq
+; CHECK: mulq
+; CHECK-NOT: imulq
+define i64 @mul1(i64 %n, i64* nocapture %z, i64* nocapture %x, i64 %y) {
+entry:
+  %conv = zext i64 %y to i128
+  %cmp11 = icmp eq i64 %n, 0
+  br i1 %cmp11, label %for.end, label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %carry.013 = phi i64 [ %conv6, %for.body ], [ 0, %entry ]
+  %i.012 = phi i64 [ %inc, %for.body ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i64* %x, i64 %i.012
+  %0 = load i64* %arrayidx, align 8
+  %conv2 = zext i64 %0 to i128
+  %mul = mul i128 %conv2, %conv
+  %conv3 = zext i64 %carry.013 to i128
+  %add = add i128 %mul, %conv3
+  %conv4 = trunc i128 %add to i64
+  %arrayidx5 = getelementptr inbounds i64* %z, i64 %i.012
+  store i64 %conv4, i64* %arrayidx5, align 8
+  %shr = lshr i128 %add, 64
+  %conv6 = trunc i128 %shr to i64
+  %inc = add i64 %i.012, 1
+  %exitcond = icmp eq i64 %inc, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret i64 0
+}