[PATCH] D106139: [AMDGPU] Combine srl of add that intends to get the carry of the add as addcarry

[Abinav Puthan Purayil via Phabricator via llvm-commits]

abinavpp created this revision.
Herald added subscribers: foad, kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, kzhuravl, arsenm.
abinavpp requested review of this revision.
Herald added subscribers: llvm-commits, wdng.
Herald added a project: LLVM.

Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D106139

Files:
  llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
  llvm/test/CodeGen/AMDGPU/combine-srl-add.ll


Index: llvm/test/CodeGen/AMDGPU/combine-srl-add.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/AMDGPU/combine-srl-add.ll
@@ -0,0 +1,13 @@
+; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck %s
+
+; CHECK: v_add_co_u32_e32 v0, vcc, v0, v1
+; CHECK: v_add_co_u32_e32 v0, vcc, vcc_lo, v2
+define i64 @add(i32 %a, i32 %b, i64 %c) {
+entry:
+  %a.zext = zext i32 %a to i64
+  %b.zext = zext i32 %b to i64
+  %add.a.b = add i64 %a.zext, %b.zext
+  %shr = lshr i64 %add.a.b, 32
+  %add.c.shr = add i64 %c, %shr
+  ret i64 %add.c.shr
+}
Index: llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -3246,6 +3246,64 @@
   return SDValue();
 }
 
+// Tries to perform (srl (add (zext a, i2^n,), (zext b, i2^n)), 2^(n-1)) ->
+// (addc a, b).carry where a and b has type i2^(n-1)
+static SDValue combineSrlAdd(SDNode *N, SelectionDAG &DAG) {
+  SDValue N0 = N->getOperand(0);
+  SDValue N1 = N->getOperand(1);
+  ConstantSDNode *N1C = nullptr;
+
+  if (!(N1C = dyn_cast<ConstantSDNode>(N1)))
+    return SDValue();
+
+  if (!N1C->getAPIntValue().isPowerOf2())
+    return SDValue();
+
+  if (N0.getOpcode() != ISD::ADD)
+    return SDValue();
+
+  SDValue N0LHS = N0.getOperand(0);
+  SDValue N0RHS = N0.getOperand(1);
+  unsigned ShiftAmt = N1C->getZExtValue();
+
+  // Make sure that this add was zero-extended from an int type of `ShiftAmt`
+  // width
+  if (N0LHS.getOpcode() != ISD::ZERO_EXTEND ||
+      N0RHS.getOpcode() != ISD::ZERO_EXTEND)
+    return SDValue();
+
+  SDValue N0NoZextLHS = N0LHS.getOperand(0);
+  SDValue N0NoZextRHS = N0RHS.getOperand(0);
+
+  if ((ShiftAmt != N0NoZextLHS.getValueType().getScalarSizeInBits()) ||
+      (ShiftAmt != N0NoZextRHS.getValueType().getScalarSizeInBits()))
+    return SDValue();
+
+  if ((N0.getValueType().getScalarSizeInBits() / 2) != ShiftAmt)
+    return SDValue();
+
+  // Make sure that the users of `N0` is not expecting the zero-extended type
+  for (SDNode *U : N0->uses()) {
+    if (U == N)
+      continue;
+
+    if (U->getOpcode() == ISD::TRUNCATE &&
+        U->getValueType(0).getScalarSizeInBits() == ShiftAmt)
+      continue;
+
+    return SDValue();
+  }
+
+  SDValue Zero = DAG.getConstant(0, SDLoc(N0), N->getValueType(0));
+  SDValue AddC =
+      DAG.getNode(ISD::ADDCARRY, SDLoc(N0),
+                  DAG.getVTList(N0NoZextLHS.getValueType(), N->getValueType(0)),
+                  N0NoZextLHS, N0NoZextRHS, Zero);
+  DAG.ReplaceAllUsesOfValueWith(N0.getValue(0), AddC.getValue(0));
+
+  return AddC.getValue(1);
+}
+
 SDValue AMDGPUTargetLowering::performSrlCombine(SDNode *N,
                                                 DAGCombinerInfo &DCI) const {
   auto *RHS = dyn_cast<ConstantSDNode>(N->getOperand(1));
@@ -3275,6 +3333,9 @@
   if (VT != MVT::i64)
     return SDValue();
 
+  if (SDValue SrlAdd = combineSrlAdd(N, DAG))
+    return SrlAdd;
+
   if (ShiftAmt < 32)
     return SDValue();
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D106139.359273.patch
Type: text/x-patch
Size: 3123 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210716/ce2f4b93/attachment.bin>