Perform mul combine when multiplying wiht negative constants.

Patch by Weiming Zhao!
This fixes PR12212

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@153049 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 459d82cdbe9f700c7cb498758bec010a318bed4c..e26dd22c9294d989714096e8e31b3e6ecdf9ca87 100644 (file)
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -6865,33 +6865,63 @@ static SDValue PerformMULCombine(SDNode *N,
   if (!C)
     return SDValue();
 
-  uint64_t MulAmt = C->getZExtValue();
+  int64_t MulAmt = C->getSExtValue();
   unsigned ShiftAmt = CountTrailingZeros_64(MulAmt);
+
   ShiftAmt = ShiftAmt & (32 - 1);
   SDValue V = N->getOperand(0);
   DebugLoc DL = N->getDebugLoc();
 
   SDValue Res;
   MulAmt >>= ShiftAmt;
-  if (isPowerOf2_32(MulAmt - 1)) {
-    // (mul x, 2^N + 1) => (add (shl x, N), x)
-    Res = DAG.getNode(ISD::ADD, DL, VT,
-                      V, DAG.getNode(ISD::SHL, DL, VT,
-                                     V, DAG.getConstant(Log2_32(MulAmt-1),
-                                                        MVT::i32)));
-  } else if (isPowerOf2_32(MulAmt + 1)) {
-    // (mul x, 2^N - 1) => (sub (shl x, N), x)
-    Res = DAG.getNode(ISD::SUB, DL, VT,
-                      DAG.getNode(ISD::SHL, DL, VT,
-                                  V, DAG.getConstant(Log2_32(MulAmt+1),
-                                                     MVT::i32)),
-                                                     V);
-  } else
-    return SDValue();
+
+  if (MulAmt >= 0) {
+    if (isPowerOf2_32(MulAmt - 1)) {
+      // (mul x, 2^N + 1) => (add (shl x, N), x)
+      Res = DAG.getNode(ISD::ADD, DL, VT,
+                        V,
+                        DAG.getNode(ISD::SHL, DL, VT,
+                                    V,
+                                    DAG.getConstant(Log2_32(MulAmt - 1),
+                                                    MVT::i32)));
+    } else if (isPowerOf2_32(MulAmt + 1)) {
+      // (mul x, 2^N - 1) => (sub (shl x, N), x)
+      Res = DAG.getNode(ISD::SUB, DL, VT,
+                        DAG.getNode(ISD::SHL, DL, VT,
+                                    V,
+                                    DAG.getConstant(Log2_32(MulAmt + 1),
+                                                    MVT::i32)),
+                        V);
+    } else
+      return SDValue();
+  } else {
+    uint64_t MulAmtAbs = -MulAmt;
+    if (isPowerOf2_32(MulAmtAbs + 1)) {
+      // (mul x, -(2^N - 1)) => (sub x, (shl x, N))
+      Res = DAG.getNode(ISD::SUB, DL, VT,
+                        V,
+                        DAG.getNode(ISD::SHL, DL, VT,
+                                    V,
+                                    DAG.getConstant(Log2_32(MulAmtAbs + 1),
+                                                    MVT::i32)));
+    } else if (isPowerOf2_32(MulAmtAbs - 1)) {
+      // (mul x, -(2^N + 1)) => - (add (shl x, N), x)
+      Res = DAG.getNode(ISD::ADD, DL, VT,
+                        V,
+                        DAG.getNode(ISD::SHL, DL, VT,
+                                    V,
+                                    DAG.getConstant(Log2_32(MulAmtAbs-1),
+                                                    MVT::i32)));
+      Res = DAG.getNode(ISD::SUB, DL, VT,
+                        DAG.getConstant(0, MVT::i32),Res);
+
+    } else
+      return SDValue();
+  }
 
   if (ShiftAmt != 0)
-    Res = DAG.getNode(ISD::SHL, DL, VT, Res,
-                      DAG.getConstant(ShiftAmt, MVT::i32));
+    Res = DAG.getNode(ISD::SHL, DL, VT,
+                      Res, DAG.getConstant(ShiftAmt, MVT::i32));
 
   // Do not add new nodes to DAG combiner worklist.
   DCI.CombineTo(N, Res, false);

diff --git a/test/CodeGen/ARM/mul_const.ll b/test/CodeGen/ARM/mul_const.ll
index 3cb8a8e816f68e4c32b4f2fbd0a8ec0f88042fe8..c50a23354678d2e66ef838c0414f9c5f6b655768 100644 (file)
--- a/test/CodeGen/ARM/mul_const.ll
+++ b/test/CodeGen/ARM/mul_const.ll
@@ -41,3 +41,45 @@ entry:
         ret i32 %0
 }
 
+define i32 @tn9(i32 %v) nounwind readnone {
+entry:
+; CHECK: tn9:
+; CHECK: add   r0, r0, r0, lsl #3
+; CHECK: rsb   r0, r0, #0
+        %0 = mul i32 %v, -9
+        ret i32 %0
+}
+
+define i32 @tn7(i32 %v) nounwind readnone {
+entry:
+; CHECK: tn7:
+; CHECK: sub r0, r0, r0, lsl #3
+       %0 = mul i32 %v, -7
+       ret i32 %0
+}
+
+define i32 @tn5(i32 %v) nounwind readnone {
+entry:
+; CHECK: tn5:
+; CHECK: add r0, r0, r0, lsl #2
+; CHECK: rsb r0, r0, #0
+        %0 = mul i32 %v, -5
+        ret i32 %0
+}
+
+define i32 @tn3(i32 %v) nounwind readnone {
+entry:
+; CHECK: tn3:
+; CHECK: sub r0, r0, r0, lsl #2
+        %0 = mul i32 %v, -3
+        ret i32 %0
+}
+
+define i32 @tn12288(i32 %v) nounwind readnone {
+entry:
+; CHECK: tn12288:
+; CHECK: sub r0, r0, r0, lsl #2
+; CHECK: lsl{{.*}}#12
+        %0 = mul i32 %v, -12288
+        ret i32 %0
+}