[llvm] r222006 - [Reassociate] Canonicalize operands of vector binary operators.

[Chad Rosier]

Author: mcrosier
Date: Fri Nov 14 11:08:15 2014
New Revision: 222006

URL: http://llvm.org/viewvc/llvm-project?rev=222006&view=rev
Log:
[Reassociate] Canonicalize operands of vector binary operators.

Prior to this commit fmul and fadd binary operators were being canonicalized for
both scalar and vector versions.  We now canonicalize add, mul, and, or, and xor
vector instructions.

Modified:
    llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp
    llvm/trunk/test/Transforms/Reassociate/fast-ReassociateVector.ll

Modified: llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp?rev=222006&r1=222005&r2=222006&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp Fri Nov 14 11:08:15 2014
@@ -2083,14 +2083,9 @@ void Reassociate::OptimizeInst(Instructi
   // writing other transformations simpler.
   if (I->getType()->isFloatingPointTy() || I->getType()->isVectorTy()) {
 
-    // FAdd and FMul can be commuted.
-    unsigned Opcode = I->getOpcode();
-    if (Opcode == Instruction::FMul || Opcode == Instruction::FAdd)
+    if (I->isCommutative())
       canonicalizeOperands(I);
 
-    // FIXME: We should commute vector instructions as well.  However, this 
-    // requires further analysis to determine the effect on later passes.
-
     // Don't try to optimize vector instructions or anything that doesn't have
     // unsafe algebra.
     if (I->getType()->isVectorTy() || !I->hasUnsafeAlgebra())

Modified: llvm/trunk/test/Transforms/Reassociate/fast-ReassociateVector.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/Reassociate/fast-ReassociateVector.ll?rev=222006&r1=222005&r2=222006&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/Reassociate/fast-ReassociateVector.ll (original)
+++ llvm/trunk/test/Transforms/Reassociate/fast-ReassociateVector.ll Fri Nov 14 11:08:15 2014
@@ -11,11 +11,11 @@ define <4 x float> @test1() {
   ret <4 x float> %tmp2
 }
 
-; We don't currently commute integer vector operations.
+; Commute integer vector operations.
 define <2 x i32> @test2(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: test2
 ; CHECK-NEXT: %tmp1 = add <2 x i32> %x, %y
-; CHECK-NEXT: %tmp2 = add <2 x i32> %y, %x
+; CHECK-NEXT: %tmp2 = add <2 x i32> %x, %y
 ; CHECK-NEXT: %tmp3 = add <2 x i32> %tmp1, %tmp2
 
   %tmp1 = add <2 x i32> %x, %y
@@ -23,3 +23,51 @@ define <2 x i32> @test2(<2 x i32> %x, <2
   %tmp3 = add <2 x i32> %tmp1, %tmp2
   ret <2 x i32> %tmp3
 }
+
+define <2 x i32> @test3(<2 x i32> %x, <2 x i32> %y) {
+; CHECK-LABEL: test3
+; CHECK-NEXT: %tmp1 = mul <2 x i32> %x, %y
+; CHECK-NEXT: %tmp2 = mul <2 x i32> %x, %y
+; CHECK-NEXT: %tmp3 = mul <2 x i32> %tmp1, %tmp2
+
+  %tmp1 = mul <2 x i32> %x, %y
+  %tmp2 = mul <2 x i32> %y, %x
+  %tmp3 = mul <2 x i32> %tmp1, %tmp2
+  ret <2 x i32> %tmp3
+}
+
+define <2 x i32> @test4(<2 x i32> %x, <2 x i32> %y) {
+; CHECK-LABEL: test4
+; CHECK-NEXT: %tmp1 = and <2 x i32> %x, %y
+; CHECK-NEXT: %tmp2 = and <2 x i32> %x, %y
+; CHECK-NEXT: %tmp3 = and <2 x i32> %tmp1, %tmp2
+
+  %tmp1 = and <2 x i32> %x, %y
+  %tmp2 = and <2 x i32> %y, %x
+  %tmp3 = and <2 x i32> %tmp1, %tmp2
+  ret <2 x i32> %tmp3
+}
+
+define <2 x i32> @test5(<2 x i32> %x, <2 x i32> %y) {
+; CHECK-LABEL: test5
+; CHECK-NEXT: %tmp1 = or <2 x i32> %x, %y
+; CHECK-NEXT: %tmp2 = or <2 x i32> %x, %y
+; CHECK-NEXT: %tmp3 = or <2 x i32> %tmp1, %tmp2
+
+  %tmp1 = or <2 x i32> %x, %y
+  %tmp2 = or <2 x i32> %y, %x
+  %tmp3 = or <2 x i32> %tmp1, %tmp2
+  ret <2 x i32> %tmp3
+}
+
+define <2 x i32> @test6(<2 x i32> %x, <2 x i32> %y) {
+; CHECK-LABEL: test6
+; CHECK-NEXT: %tmp1 = xor <2 x i32> %x, %y
+; CHECK-NEXT: %tmp2 = xor <2 x i32> %x, %y
+; CHECK-NEXT: %tmp3 = xor <2 x i32> %tmp1, %tmp2
+
+  %tmp1 = xor <2 x i32> %x, %y
+  %tmp2 = xor <2 x i32> %y, %x
+  %tmp3 = xor <2 x i32> %tmp1, %tmp2
+  ret <2 x i32> %tmp3
+}