[RISCV] Strength reduce mul by 2^n + 2/4/8 + 1

[preames]

With zba, we can expand this to (add (shl X, C1), (shXadd X, X)).

Note that this is our first expansion to a three instruction sequence. I believe this to general be a reasonable tradeoff for most architectures, but we may want to (someday) consider a tuning flag here.

I plan to support 2^n + (2/4/8 + 1) eventually as well, but that comes behind 2^N - 2^M. Both are also three instruction sequences.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

Author: Philip Reames (preames)

Changes

With zba, we can expand this to (add (shl X, C1), (shXadd X, X)).

Note that this is our first expansion to a three instruction sequence. I believe this to general be a reasonable tradeoff for most architectures, but we may want to (someday) consider a tuning flag here.

I plan to support 2^n + (2/4/8 + 1) eventually as well, but that comes behind 2^N - 2^M. Both are also three instruction sequences.

---

Full diff: https://github.com/llvm/llvm-project/pull/88911.diff

2 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+28)
• (modified) llvm/test/CodeGen/RISCV/rv64zba.ll (+36-15)

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 765838aafb58d2..388a67a0575f63 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -13431,6 +13431,34 @@ static SDValue expandMul(SDNode *N, SelectionDAG &DAG,
       return DAG.getNode(ISD::ADD, DL, VT, Shift1, Shift2);
     }
   }
+
+  // (2^(1,2,3) * 3,5,9 + 1 -> (shXadd (shYadd x, x), x)
+  // Matched in tablegen, avoid perturbing patterns.
+  switch (MulAmt) {
+  case 11: case 13: case 19: case 21: case 25: case 27: case 29:
+  case 37: case 41: case 45: case 73: case 91:
+    return SDValue();
+  default:
+    break;
+  }
+
+  // 2^n + 2/4/8 + 1 -> (add (shl X, C1), (shXadd X, X))
+  if (MulAmt > 2 &&
+      isPowerOf2_64((MulAmt - 1) & (MulAmt - 2))) {
+    unsigned ScaleShift = llvm::countr_zero(MulAmt - 1);
+    if (ScaleShift >= 1 && ScaleShift < 4) {
+      unsigned ShiftAmt = Log2_64(((MulAmt - 1) & (MulAmt - 2)));
+      SDLoc DL(N);
+      SDValue Shift1 = DAG.getNode(ISD::SHL, DL, VT, N->getOperand(0),
+                                   DAG.getConstant(ShiftAmt, DL, VT));
+      SDValue Shift2 = DAG.getNode(ISD::SHL, DL, VT, N->getOperand(0),
+                                   DAG.getConstant(ScaleShift, DL, VT));
+      return DAG.getNode(ISD::ADD, DL, VT, Shift1,
+                         DAG.getNode(ISD::ADD, DL, VT, Shift2,
+                                     N->getOperand(0)));
+    }
+  }
+
   return SDValue();
 }
 
diff --git a/llvm/test/CodeGen/RISCV/rv64zba.ll b/llvm/test/CodeGen/RISCV/rv64zba.ll
index a84b9e5e7962f6..d452c9d1b0f0c5 100644
--- a/llvm/test/CodeGen/RISCV/rv64zba.ll
+++ b/llvm/test/CodeGen/RISCV/rv64zba.ll
@@ -598,31 +598,52 @@ define i64 @mul125(i64 %a) {
 }
 
 define i64 @mul131(i64 %a) {
-; CHECK-LABEL: mul131:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a1, 131
-; CHECK-NEXT:    mul a0, a0, a1
-; CHECK-NEXT:    ret
+; RV64I-LABEL: mul131:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a1, 131
+; RV64I-NEXT:    mul a0, a0, a1
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: mul131:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    sh1add a1, a0, a0
+; RV64ZBA-NEXT:    slli a0, a0, 7
+; RV64ZBA-NEXT:    add a0, a0, a1
+; RV64ZBA-NEXT:    ret
   %c = mul i64 %a, 131
   ret i64 %c
 }
 
 define i64 @mul133(i64 %a) {
-; CHECK-LABEL: mul133:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a1, 133
-; CHECK-NEXT:    mul a0, a0, a1
-; CHECK-NEXT:    ret
+; RV64I-LABEL: mul133:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a1, 133
+; RV64I-NEXT:    mul a0, a0, a1
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: mul133:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    sh2add a1, a0, a0
+; RV64ZBA-NEXT:    slli a0, a0, 7
+; RV64ZBA-NEXT:    add a0, a0, a1
+; RV64ZBA-NEXT:    ret
   %c = mul i64 %a, 133
   ret i64 %c
 }
 
 define i64 @mul137(i64 %a) {
-; CHECK-LABEL: mul137:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a1, 137
-; CHECK-NEXT:    mul a0, a0, a1
-; CHECK-NEXT:    ret
+; RV64I-LABEL: mul137:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a1, 137
+; RV64I-NEXT:    mul a0, a0, a1
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: mul137:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    sh3add a1, a0, a0
+; RV64ZBA-NEXT:    slli a0, a0, 7
+; RV64ZBA-NEXT:    add a0, a0, a1
+; RV64ZBA-NEXT:    ret
   %c = mul i64 %a, 137
   ret i64 %c
 }

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[mshockwave]

I guess clang-format did this, is this your desired formatting as well?

[preames]

This is not clang-format's preferred format, I'd hand formatted. I'm going to push the clang-format as this isn't worth the time to bother debating.

[mshockwave]

LGTM