[llvm] c06f77e - [SelectionDAG] fold 'Op0 - (X * MulC)' to 'Op0 + (X << log2(-MulC))'

[Ben Shi via llvm-commits]

Author: Ping Deng
Date: 2022-06-15T05:50:18Z
New Revision: c06f77ec0d38163a67c26ddf847adc58f82f9672

URL: https://github.com/llvm/llvm-project/commit/c06f77ec0d38163a67c26ddf847adc58f82f9672
DIFF: https://github.com/llvm/llvm-project/commit/c06f77ec0d38163a67c26ddf847adc58f82f9672.diff

LOG: [SelectionDAG] fold 'Op0 - (X * MulC)' to 'Op0 + (X << log2(-MulC))'

Reviewed By: craig.topper, spatel

Differential Revision: https://reviews.llvm.org/D127474

Added: 
    

Modified: 
    llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
    llvm/test/CodeGen/RISCV/mul.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 03fc1728ac44a..e09f7d664ef04 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -2533,24 +2533,27 @@ bool TargetLowering::SimplifyDemandedBits(
       return 0;
     };
 
-    auto foldMul = [&](SDValue X, SDValue Y, unsigned ShlAmt) {
+    auto foldMul = [&](ISD::NodeType NT, SDValue X, SDValue Y, unsigned ShlAmt) {
       EVT ShiftAmtTy = getShiftAmountTy(VT, TLO.DAG.getDataLayout());
       SDValue ShlAmtC = TLO.DAG.getConstant(ShlAmt, dl, ShiftAmtTy);
       SDValue Shl = TLO.DAG.getNode(ISD::SHL, dl, VT, X, ShlAmtC);
-      SDValue Sub = TLO.DAG.getNode(ISD::SUB, dl, VT, Y, Shl);
-      return TLO.CombineTo(Op, Sub);
+      SDValue Res = TLO.DAG.getNode(NT, dl, VT, Y, Shl);
+      return TLO.CombineTo(Op, Res);
     };
 
     if (isOperationLegalOrCustom(ISD::SHL, VT)) {
       if (Op.getOpcode() == ISD::ADD) {
         // (X * MulC) + Op1 --> Op1 - (X << log2(-MulC))
         if (unsigned ShAmt = getShiftLeftAmt(Op0))
-          return foldMul(Op0.getOperand(0), Op1, ShAmt);
+          return foldMul(ISD::SUB, Op0.getOperand(0), Op1, ShAmt);
         // Op0 + (X * MulC) --> Op0 - (X << log2(-MulC))
         if (unsigned ShAmt = getShiftLeftAmt(Op1))
-          return foldMul(Op1.getOperand(0), Op0, ShAmt);
-        // TODO:
+          return foldMul(ISD::SUB, Op1.getOperand(0), Op0, ShAmt);
+      }
+      if (Op.getOpcode() == ISD::SUB) {
         // Op0 - (X * MulC) --> Op0 + (X << log2(-MulC))
+        if (unsigned ShAmt = getShiftLeftAmt(Op1))
+          return foldMul(ISD::ADD, Op1.getOperand(0), Op0, ShAmt);
       }
     }
 

diff  --git a/llvm/test/CodeGen/RISCV/mul.ll b/llvm/test/CodeGen/RISCV/mul.ll
index f72556e7155f3..ba6fcc4122be9 100644
--- a/llvm/test/CodeGen/RISCV/mul.ll
+++ b/llvm/test/CodeGen/RISCV/mul.ll
@@ -1584,47 +1584,29 @@ define i8 @muladd_demand(i8 %x, i8 %y) nounwind {
 define i8 @mulsub_demand(i8 %x, i8 %y) nounwind {
 ; RV32I-LABEL: mulsub_demand:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    addi sp, sp, -16
-; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    mv s0, a1
-; RV32I-NEXT:    li a1, 14
-; RV32I-NEXT:    call __mulsi3 at plt
-; RV32I-NEXT:    sub a0, s0, a0
+; RV32I-NEXT:    slli a0, a0, 1
+; RV32I-NEXT:    add a0, a1, a0
 ; RV32I-NEXT:    andi a0, a0, 15
-; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    addi sp, sp, 16
 ; RV32I-NEXT:    ret
 ;
 ; RV32IM-LABEL: mulsub_demand:
 ; RV32IM:       # %bb.0:
-; RV32IM-NEXT:    li a2, 14
-; RV32IM-NEXT:    mul a0, a0, a2
-; RV32IM-NEXT:    sub a0, a1, a0
+; RV32IM-NEXT:    slli a0, a0, 1
+; RV32IM-NEXT:    add a0, a1, a0
 ; RV32IM-NEXT:    andi a0, a0, 15
 ; RV32IM-NEXT:    ret
 ;
 ; RV64I-LABEL: mulsub_demand:
 ; RV64I:       # %bb.0:
-; RV64I-NEXT:    addi sp, sp, -16
-; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64I-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
-; RV64I-NEXT:    mv s0, a1
-; RV64I-NEXT:    li a1, 14
-; RV64I-NEXT:    call __muldi3 at plt
-; RV64I-NEXT:    subw a0, s0, a0
+; RV64I-NEXT:    slliw a0, a0, 1
+; RV64I-NEXT:    addw a0, a1, a0
 ; RV64I-NEXT:    andi a0, a0, 15
-; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64I-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
-; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: mulsub_demand:
 ; RV64IM:       # %bb.0:
-; RV64IM-NEXT:    li a2, 14
-; RV64IM-NEXT:    mulw a0, a0, a2
-; RV64IM-NEXT:    subw a0, a1, a0
+; RV64IM-NEXT:    slliw a0, a0, 1
+; RV64IM-NEXT:    addw a0, a1, a0
 ; RV64IM-NEXT:    andi a0, a0, 15
 ; RV64IM-NEXT:    ret
   %m = mul i8 %x, 14