SelectionDAG: Fix incorrect condition checks in some cases of folding FADD/FMUL combinations; also improve accuracy of comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183993 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 89 insertions, 20 deletions

diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 046dd416f3..570ea4c0cb 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -5919,7 +5919,7 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
       ConstantFPSDNode *CFP00 = dyn_cast<ConstantFPSDNode>(N0.getOperand(0));
       ConstantFPSDNode *CFP01 = dyn_cast<ConstantFPSDNode>(N0.getOperand(1));
 
-      // (fadd (fmul c, x), x) -> (fmul c+1, x)
+      // (fadd (fmul c, x), x) -> (fmul x, c+1)
       if (CFP00 && !CFP01 && N0.getOperand(1) == N1) {
         SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
                                      SDValue(CFP00, 0),
@@ -5928,7 +5928,7 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
                            N1, NewCFP);
       }
 
-      // (fadd (fmul x, c), x) -> (fmul c+1, x)
+      // (fadd (fmul x, c), x) -> (fmul x, c+1)
       if (CFP01 && !CFP00 && N0.getOperand(0) == N1) {
         SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
                                      SDValue(CFP01, 0),
@@ -5937,7 +5937,7 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
                            N1, NewCFP);
       }
 
-      // (fadd (fmul c, x), (fadd x, x)) -> (fmul c+2, x)
+      // (fadd (fmul c, x), (fadd x, x)) -> (fmul x, c+2)
       if (CFP00 && !CFP01 && N1.getOpcode() == ISD::FADD &&
           N1.getOperand(0) == N1.getOperand(1) &&
           N0.getOperand(1) == N1.getOperand(0)) {
@@ -5948,7 +5948,7 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
                            N0.getOperand(1), NewCFP);
       }
 
-      // (fadd (fmul x, c), (fadd x, x)) -> (fmul c+2, x)
+      // (fadd (fmul x, c), (fadd x, x)) -> (fmul x, c+2)
       if (CFP01 && !CFP00 && N1.getOpcode() == ISD::FADD &&
           N1.getOperand(0) == N1.getOperand(1) &&
           N0.getOperand(0) == N1.getOperand(0)) {
@@ -5964,7 +5964,7 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
       ConstantFPSDNode *CFP10 = dyn_cast<ConstantFPSDNode>(N1.getOperand(0));
       ConstantFPSDNode *CFP11 = dyn_cast<ConstantFPSDNode>(N1.getOperand(1));
 
-      // (fadd x, (fmul c, x)) -> (fmul c+1, x)
+      // (fadd x, (fmul c, x)) -> (fmul x, c+1)
       if (CFP10 && !CFP11 && N1.getOperand(1) == N0) {
         SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
                                      SDValue(CFP10, 0),
@@ -5973,7 +5973,7 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
                            N0, NewCFP);
       }
 
-      // (fadd x, (fmul x, c)) -> (fmul c+1, x)
+      // (fadd x, (fmul x, c)) -> (fmul x, c+1)
       if (CFP11 && !CFP10 && N1.getOperand(0) == N0) {
         SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
                                      SDValue(CFP11, 0),
@@ -5983,26 +5983,26 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
       }
 
 
-      // (fadd (fadd x, x), (fmul c, x)) -> (fmul c+2, x)
-      if (CFP10 && !CFP11 && N1.getOpcode() == ISD::FADD &&
-          N1.getOperand(0) == N1.getOperand(1) &&
-          N0.getOperand(1) == N1.getOperand(0)) {
+      // (fadd (fadd x, x), (fmul c, x)) -> (fmul x, c+2)
+      if (CFP10 && !CFP11 && N0.getOpcode() == ISD::FADD &&
+          N0.getOperand(0) == N0.getOperand(1) &&
+          N1.getOperand(1) == N0.getOperand(0)) {
         SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
                                      SDValue(CFP10, 0),
                                      DAG.getConstantFP(2.0, VT));
         return DAG.getNode(ISD::FMUL, SDLoc(N), VT,
-                           N0.getOperand(1), NewCFP);
+                           N1.getOperand(1), NewCFP);
       }
 
-      // (fadd (fadd x, x), (fmul x, c)) -> (fmul c+2, x)
-      if (CFP11 && !CFP10 && N1.getOpcode() == ISD::FADD &&
-          N1.getOperand(0) == N1.getOperand(1) &&
-          N0.getOperand(0) == N1.getOperand(0)) {
+      // (fadd (fadd x, x), (fmul x, c)) -> (fmul x, c+2)
+      if (CFP11 && !CFP10 && N0.getOpcode() == ISD::FADD &&
+          N0.getOperand(0) == N0.getOperand(1) &&
+          N1.getOperand(0) == N0.getOperand(0)) {
         SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
                                      SDValue(CFP11, 0),
                                      DAG.getConstantFP(2.0, VT));
         return DAG.getNode(ISD::FMUL, SDLoc(N), VT,
-                           N0.getOperand(0), NewCFP);
+                           N1.getOperand(0), NewCFP);
       }
     }
 
diff --git a/test/CodeGen/X86/fp-fast.ll b/test/CodeGen/X86/fp-fast.ll
index 287504801d..7589ac4bdc 100644
--- a/test/CodeGen/X86/fp-fast.ll
+++ b/test/CodeGen/X86/fp-fast.ll
@@ -26,6 +26,42 @@ define float @test2(float %a) {
 ; CHECK: test3
 define float @test3(float %a) {
 ; CHECK-NOT: addss
+; CHECK: mulss
+; CHECK-NOT: addss
+; CHECK: ret
+  %t1 = fmul float %a, 4.0
+  %t2 = fadd float %a, %a
+  %r = fadd float %t1, %t2
+  ret float %r
+}
+
+; CHECK: test4
+define float @test4(float %a) {
+; CHECK-NOT: addss
+; CHECK: mulss
+; CHECK-NOT: addss
+; CHECK: ret
+  %t1 = fadd float %a, %a
+  %t2 = fmul float 4.0, %a
+  %r = fadd float %t1, %t2
+  ret float %r
+}
+
+; CHECK: test5
+define float @test5(float %a) {
+; CHECK-NOT: addss
+; CHECK: mulss
+; CHECK-NOT: addss
+; CHECK: ret
+  %t1 = fadd float %a, %a
+  %t2 = fmul float %a, 4.0
+  %r = fadd float %t1, %t2
+  ret float %r
+}
+
+; CHECK: test6
+define float @test6(float %a) {
+; CHECK-NOT: addss
 ; CHECK: xorps
 ; CHECK-NOT: addss
 ; CHECK: ret
@@ -35,8 +71,20 @@ define float @test3(float %a) {
   ret float %r
 }
 
-; CHECK: test4
-define float @test4(float %a) {
+; CHECK: test7
+define float @test7(float %a) {
+; CHECK-NOT: addss
+; CHECK: xorps
+; CHECK-NOT: addss
+; CHECK: ret
+  %t1 = fmul float %a, 2.0
+  %t2 = fadd float %a, %a
+  %r = fsub float %t1, %t2
+  ret float %r
+}
+
+; CHECK: test8
+define float @test8(float %a) {
 ; CHECK-NOT: fma
 ; CHECK-NOT: mul
 ; CHECK-NOT: add
@@ -46,8 +94,29 @@ define float @test4(float %a) {
   ret float %t2
 }
 
-; CHECK: test5
-define float @test5(float %a) {
+; CHECK: test9
+define float @test9(float %a) {
+; CHECK-NOT: fma
+; CHECK-NOT: mul
+; CHECK-NOT: add
+; CHECK: ret
+  %t1 = fmul float 0.0, %a
+  %t2 = fadd float %t1, %a
+  ret float %t2
+}
+
+; CHECK: test10
+define float @test10(float %a) {
+; CHECK-NOT: add
+; CHECK: vxorps
+; CHECK: ret
+  %t1 = fsub float -0.0, %a
+  %t2 = fadd float %a, %t1
+  ret float %t2
+}
+
+; CHECK: test11
+define float @test11(float %a) {
 ; CHECK-NOT: add
 ; CHECK: vxorps
 ; CHECK: ret