From a1840d2f8803714f8f1d004577dff59d9d06f333 Mon Sep 17 00:00:00 2001
From: Daniel Sanders <daniel.sanders@imgtec.com>
Date: Mon, 11 Nov 2013 17:23:41 +0000
Subject: [PATCH] Vector forms of SHL, SRA, and SRL can be constant folded
 using SimplifyVBinOp too

Reviewers: dsanders

Reviewed By: dsanders

CC: llvm-commits, nadav

Differential Revision: http://llvm-reviews.chandlerc.com/D1958

llvm-svn: 194393
---
 llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 18 +++++
 .../test/CodeGen/Mips/msa/shift-dagcombine.ll | 70 +++++++++++++++++++
 2 files changed, 88 insertions(+)
 create mode 100644 llvm/test/CodeGen/Mips/msa/shift-dagcombine.ll

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index de0f6ce26d95..7343236e6a1a 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -3691,6 +3691,12 @@ SDValue DAGCombiner::visitSHL(SDNode *N) {
   EVT VT = N0.getValueType();
   unsigned OpSizeInBits = VT.getScalarType().getSizeInBits();
 
+  // fold vector ops
+  if (VT.isVector()) {
+    SDValue FoldedVOp = SimplifyVBinOp(N);
+    if (FoldedVOp.getNode()) return FoldedVOp;
+  }
+
   // fold (shl c1, c2) -> c1<<c2
   if (N0C && N1C)
     return DAG.FoldConstantArithmetic(ISD::SHL, VT, N0C, N1C);
@@ -3842,6 +3848,12 @@ SDValue DAGCombiner::visitSRA(SDNode *N) {
   EVT VT = N0.getValueType();
   unsigned OpSizeInBits = VT.getScalarType().getSizeInBits();
 
+  // fold vector ops
+  if (VT.isVector()) {
+    SDValue FoldedVOp = SimplifyVBinOp(N);
+    if (FoldedVOp.getNode()) return FoldedVOp;
+  }
+
   // fold (sra c1, c2) -> (sra c1, c2)
   if (N0C && N1C)
     return DAG.FoldConstantArithmetic(ISD::SRA, VT, N0C, N1C);
@@ -3987,6 +3999,12 @@ SDValue DAGCombiner::visitSRL(SDNode *N) {
   EVT VT = N0.getValueType();
   unsigned OpSizeInBits = VT.getScalarType().getSizeInBits();
 
+  // fold vector ops
+  if (VT.isVector()) {
+    SDValue FoldedVOp = SimplifyVBinOp(N);
+    if (FoldedVOp.getNode()) return FoldedVOp;
+  }
+
   // fold (srl c1, c2) -> c1 >>u c2
   if (N0C && N1C)
     return DAG.FoldConstantArithmetic(ISD::SRL, VT, N0C, N1C);
diff --git a/llvm/test/CodeGen/Mips/msa/shift-dagcombine.ll b/llvm/test/CodeGen/Mips/msa/shift-dagcombine.ll
new file mode 100644
index 000000000000..0d809fb4fbf1
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/msa/shift-dagcombine.ll
@@ -0,0 +1,70 @@
+; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | FileCheck %s
+
+define void @ashr_v4i32(<4 x i32>* %c) nounwind {
+  ; CHECK-LABEL: ashr_v4i32:
+
+  %1 = ashr <4 x i32> <i32 1, i32 2, i32 4, i32 8>,
+                      <i32 0, i32 1, i32 2, i32 3>
+  ; CHECK-NOT: sra
+  ; CHECK-DAG: ldi.w [[R1:\$w[0-9]+]], 1
+  ; CHECK-NOT: sra
+  store volatile <4 x i32> %1, <4 x i32>* %c
+  ; CHECK-DAG: st.w [[R1]], 0($4)
+
+  %2 = ashr <4 x i32> <i32 -2, i32 -4, i32 -8, i32 -16>,
+                      <i32 0, i32 1, i32 2, i32 3>
+  ; CHECK-NOT: sra
+  ; CHECK-DAG: ldi.w [[R1:\$w[0-9]+]], -2
+  ; CHECK-NOT: sra
+  store volatile <4 x i32> %2, <4 x i32>* %c
+  ; CHECK-DAG: st.w [[R1]], 0($4)
+
+  ret void
+  ; CHECK-LABEL: .size ashr_v4i32
+}
+
+define void @lshr_v4i32(<4 x i32>* %c) nounwind {
+  ; CHECK-LABEL: lshr_v4i32:
+
+  %1 = lshr <4 x i32> <i32 1, i32 2, i32 4, i32 8>,
+                      <i32 0, i32 1, i32 2, i32 3>
+  ; CHECK-NOT: srl
+  ; CHECK-DAG: ldi.w [[R1:\$w[0-9]+]], 1
+  ; CHECK-NOT: srl
+  store volatile <4 x i32> %1, <4 x i32>* %c
+  ; CHECK-DAG: st.w [[R1]], 0($4)
+
+  %2 = lshr <4 x i32> <i32 -2, i32 -4, i32 -8, i32 -16>,
+                      <i32 0, i32 1, i32 2, i32 3>
+  ; CHECK-NOT: srl
+  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], %lo
+  ; CHECK-NOT: srl
+  store volatile <4 x i32> %2, <4 x i32>* %c
+  ; CHECK-DAG: st.w [[R1]], 0($4)
+
+  ret void
+  ; CHECK-LABEL: .size lshr_v4i32
+}
+
+define void @shl_v4i32(<4 x i32>* %c) nounwind {
+  ; CHECK-LABEL: shl_v4i32:
+
+  %1 = shl <4 x i32> <i32 8, i32 4, i32 2, i32 1>,
+                     <i32 0, i32 1, i32 2, i32 3>
+  ; CHECK-NOT: sll
+  ; CHECK-DAG: ldi.w [[R1:\$w[0-9]+]], 8
+  ; CHECK-NOT: sll
+  store volatile <4 x i32> %1, <4 x i32>* %c
+  ; CHECK-DAG: st.w [[R1]], 0($4)
+
+  %2 = shl <4 x i32> <i32 -8, i32 -4, i32 -2, i32 -1>,
+                     <i32 0, i32 1, i32 2, i32 3>
+  ; CHECK-NOT: sll
+  ; CHECK-DAG: ldi.w [[R1:\$w[0-9]+]], -8
+  ; CHECK-NOT: sll
+  store volatile <4 x i32> %2, <4 x i32>* %c
+  ; CHECK-DAG: st.w [[R1]], 0($4)
+
+  ret void
+  ; CHECK-LABEL: .size shl_v4i32
+}