[RISCV] Expand multiplication by 2^N * 3/5/9 + 1 with SHL_ADD (llvm#166933)

Author: pfusik

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -16878,12 +16878,23 @@ static SDValue expandMulToShlAddShlAdd(SDNode *N, SelectionDAG &DAG,
     break;
   }
 
-  // 2/4/8 * 3/5/9 + 1 -> (shXadd (shYadd X, X), X)
   int ShX;
   if (int ShY = isShifted359(MulAmt - 1, ShX)) {
     assert(ShX != 0 && "MulAmt=4,6,10 handled before");
+    // 2/4/8 * 3/5/9 + 1 -> (shXadd (shYadd X, X), X)
     if (ShX <= 3)
       return getShlAddShlAdd(N, DAG, ShX, ShY, /*AddX=*/true, Shift);
+    // 2^N * 3/5/9 + 1 -> (add (shYadd (shl X, N), (shl X, N)), X)
+    if (Shift == 0) {
+      SDLoc DL(N);
+      EVT VT = N->getValueType(0);
+      SDValue X = N->getOperand(0);
+      SDValue Shl =
+          DAG.getNode(ISD::SHL, DL, VT, X, DAG.getConstant(ShX, DL, VT));
+      SDValue ShlAdd = DAG.getNode(RISCVISD::SHL_ADD, DL, VT, Shl,
+                                   DAG.getTargetConstant(ShY, DL, VT), Shl);
+      return DAG.getNode(ISD::ADD, DL, VT, ShlAdd, X);
+    }
   }
   return SDValue();
 }
@@ -16944,7 +16955,7 @@ static SDValue expandMul(SDNode *N, SelectionDAG &DAG,
                          DAG.getTargetConstant(Shift, DL, VT), Shift1);
     }
 
-    // TODO: 2^(C1>3) * 3,5,9 +/- 1
+    // TODO: 2^(C1>3) * 3/5/9 - 1
 
     // 2^n + 2/4/8 + 1 -> (add (shl X, C1), (shXadd X, X))
     if (MulAmt > 2 && isPowerOf2_64((MulAmt - 1) & (MulAmt - 2))) {

llvm/test/CodeGen/RISCV/rv64zba.ll

@@ -944,6 +944,58 @@ define i64 @addmul146(i64 %a, i64 %b) {
   ret i64 %d
 }
 
+define i64 @mul49(i64 %a) {
+; RV64I-LABEL: mul49:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a1, 49
+; RV64I-NEXT:    mul a0, a0, a1
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: mul49:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    slli a1, a0, 4
+; RV64ZBA-NEXT:    sh1add a1, a1, a1
+; RV64ZBA-NEXT:    add a0, a1, a0
+; RV64ZBA-NEXT:    ret
+;
+; RV64XANDESPERF-LABEL: mul49:
+; RV64XANDESPERF:       # %bb.0:
+; RV64XANDESPERF-NEXT:    slli a1, a0, 4
+; RV64XANDESPERF-NEXT:    nds.lea.h a1, a1, a1
+; RV64XANDESPERF-NEXT:    add a0, a1, a0
+; RV64XANDESPERF-NEXT:    ret
+  %c = mul i64 %a, 49
+  ret i64 %c
+}
+
+define i64 @zext_mul49(i32 signext %a) {
+; RV64I-LABEL: zext_mul49:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a1, 49
+; RV64I-NEXT:    slli a1, a1, 32
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    mulhu a0, a0, a1
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: zext_mul49:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    slli.uw a1, a0, 4
+; RV64ZBA-NEXT:    sh1add a1, a1, a1
+; RV64ZBA-NEXT:    add.uw a0, a0, a1
+; RV64ZBA-NEXT:    ret
+;
+; RV64XANDESPERF-LABEL: zext_mul49:
+; RV64XANDESPERF:       # %bb.0:
+; RV64XANDESPERF-NEXT:    slli a1, a0, 32
+; RV64XANDESPERF-NEXT:    srli a1, a1, 28
+; RV64XANDESPERF-NEXT:    nds.lea.h a1, a1, a1
+; RV64XANDESPERF-NEXT:    nds.lea.b.ze a0, a1, a0
+; RV64XANDESPERF-NEXT:    ret
+  %b = zext i32 %a to i64
+  %c = mul i64 %b, 49
+  ret i64 %c
+}
+
 define i64 @mul50(i64 %a) {
 ; RV64I-LABEL: mul50:
 ; RV64I:       # %bb.0:
@@ -1044,6 +1096,54 @@ define i64 @addmul100(i64 %a, i64 %b) {
   ret i64 %d
 }
 
+define i64 @mul145(i64 %a) {
+; RV64I-LABEL: mul145:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a1, 145
+; RV64I-NEXT:    mul a0, a0, a1
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: mul145:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    slli a1, a0, 4
+; RV64ZBA-NEXT:    sh3add a1, a1, a1
+; RV64ZBA-NEXT:    add a0, a1, a0
+; RV64ZBA-NEXT:    ret
+;
+; RV64XANDESPERF-LABEL: mul145:
+; RV64XANDESPERF:       # %bb.0:
+; RV64XANDESPERF-NEXT:    slli a1, a0, 4
+; RV64XANDESPERF-NEXT:    nds.lea.d a1, a1, a1
+; RV64XANDESPERF-NEXT:    add a0, a1, a0
+; RV64XANDESPERF-NEXT:    ret
+  %c = mul i64 %a, 145
+  ret i64 %c
+}
+
+define i64 @mul161(i64 %a) {
+; RV64I-LABEL: mul161:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a1, 161
+; RV64I-NEXT:    mul a0, a0, a1
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: mul161:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    slli a1, a0, 5
+; RV64ZBA-NEXT:    sh2add a1, a1, a1
+; RV64ZBA-NEXT:    add a0, a1, a0
+; RV64ZBA-NEXT:    ret
+;
+; RV64XANDESPERF-LABEL: mul161:
+; RV64XANDESPERF:       # %bb.0:
+; RV64XANDESPERF-NEXT:    slli a1, a0, 5
+; RV64XANDESPERF-NEXT:    nds.lea.w a1, a1, a1
+; RV64XANDESPERF-NEXT:    add a0, a1, a0
+; RV64XANDESPERF-NEXT:    ret
+  %c = mul i64 %a, 161
+  ret i64 %c
+}
+
 define i64 @mul162(i64 %a) {
 ; RV64I-LABEL: mul162:
 ; RV64I:       # %bb.0: