From 68b3d680f20d5feb5181c00a9c09210e8e6bb999 Mon Sep 17 00:00:00 2001
From: Yi Kong <Yi.Kong@arm.com>
Date: Tue, 5 Aug 2014 12:46:47 +0000
Subject: [PATCH] AArch64: Add support for instruction prefetch intrinsic

Instruction prefetch is not implemented for AArch64, it is incorrectly
translated into data prefetch instruction.

Differential Revision: http://reviews.llvm.org/D4777


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@214860 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AArch64/AArch64ISelLowering.cpp |  4 +--
 test/CodeGen/AArch64/arm64-prefetch.ll     | 55 ++++++++++++++++++++++++------
 2 files changed, 47 insertions(+), 12 deletions(-)

diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index 4fcc409e071..2f55b4a19f9 100644
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -1334,8 +1334,7 @@ static SDValue LowerPREFETCH(SDValue Op, SelectionDAG &DAG) {
   SDLoc DL(Op);
   unsigned IsWrite = cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue();
   unsigned Locality = cast<ConstantSDNode>(Op.getOperand(3))->getZExtValue();
-  // The data thing is not used.
-  // unsigned isData = cast<ConstantSDNode>(Op.getOperand(4))->getZExtValue();
+  unsigned IsData = cast<ConstantSDNode>(Op.getOperand(4))->getZExtValue();
 
   bool IsStream = !Locality;
   // When the locality number is set
@@ -1350,6 +1349,7 @@ static SDValue LowerPREFETCH(SDValue Op, SelectionDAG &DAG) {
 
   // built the mask value encoding the expected behavior.
   unsigned PrfOp = (IsWrite << 4) |     // Load/Store bit
+                   (!IsData << 3) |     // IsDataCache bit
                    (Locality << 1) |    // Cache level bits
                    (unsigned)IsStream;  // Stream bit
   return DAG.getNode(AArch64ISD::PREFETCH, DL, MVT::Other, Op.getOperand(0),
diff --git a/test/CodeGen/AArch64/arm64-prefetch.ll b/test/CodeGen/AArch64/arm64-prefetch.ll
index b2e06edf931..9dc6301edb2 100644
--- a/test/CodeGen/AArch64/arm64-prefetch.ll
+++ b/test/CodeGen/AArch64/arm64-prefetch.ll
@@ -17,6 +17,15 @@ entry:
   ; CHECK: prfum pldl1keep
   call void @llvm.prefetch(i8* %tmp, i32 0, i32 3, i32 1)
 
+  ; CHECK: prfum plil1strm
+  call void @llvm.prefetch(i8* %tmp, i32 0, i32 0, i32 0)
+  ; CHECK: prfum plil3keep
+  call void @llvm.prefetch(i8* %tmp, i32 0, i32 1, i32 0)
+  ; CHECK: prfum plil2keep
+  call void @llvm.prefetch(i8* %tmp, i32 0, i32 2, i32 0)
+  ; CHECK: prfum plil1keep
+  call void @llvm.prefetch(i8* %tmp, i32 0, i32 3, i32 0)
+
   ; CHECK: prfum pstl1strm
   call void @llvm.prefetch(i8* %tmp, i32 1, i32 0, i32 1)
   ; CHECK: prfum pstl3keep
@@ -57,26 +66,52 @@ entry:
   %arrayidx12 = getelementptr inbounds i32* %tmp10, i64 %idxprom
   %tmp11 = bitcast i32* %arrayidx12 to i8*
 
-  ; CHECK: prfm pstl1strm
-  call void @llvm.prefetch(i8* %tmp11, i32 1, i32 0, i32 1)
+
+  ; CHECK: prfm plil1strm
+  call void @llvm.prefetch(i8* %tmp11, i32 0, i32 0, i32 0)
   %tmp12 = load i32** @a, align 8, !tbaa !3
   %arrayidx15 = getelementptr inbounds i32* %tmp12, i64 %idxprom
-  %tmp13 = bitcast i32* %arrayidx15 to i8*
+  %tmp13 = bitcast i32* %arrayidx3 to i8*
 
-  ; CHECK: prfm pstl3keep
-  call void @llvm.prefetch(i8* %tmp13, i32 1, i32 1, i32 1)
+  ; CHECK: prfm plil3keep
+  call void @llvm.prefetch(i8* %tmp13, i32 0, i32 1, i32 0)
   %tmp14 = load i32** @a, align 8, !tbaa !3
   %arrayidx18 = getelementptr inbounds i32* %tmp14, i64 %idxprom
-  %tmp15 = bitcast i32* %arrayidx18 to i8*
+  %tmp15 = bitcast i32* %arrayidx6 to i8*
 
-  ; CHECK: prfm pstl2keep
-  call void @llvm.prefetch(i8* %tmp15, i32 1, i32 2, i32 1)
+  ; CHECK: prfm plil2keep
+  call void @llvm.prefetch(i8* %tmp15, i32 0, i32 2, i32 0)
   %tmp16 = load i32** @a, align 8, !tbaa !3
   %arrayidx21 = getelementptr inbounds i32* %tmp16, i64 %idxprom
-  %tmp17 = bitcast i32* %arrayidx21 to i8*
+  %tmp17 = bitcast i32* %arrayidx9 to i8*
+
+  ; CHECK: prfm plil1keep
+  call void @llvm.prefetch(i8* %tmp17, i32 0, i32 3, i32 0)
+  %tmp18 = load i32** @a, align 8, !tbaa !3
+  %arrayidx24 = getelementptr inbounds i32* %tmp18, i64 %idxprom
+  %tmp19 = bitcast i32* %arrayidx12 to i8*
+
+
+  ; CHECK: prfm pstl1strm
+  call void @llvm.prefetch(i8* %tmp19, i32 1, i32 0, i32 1)
+  %tmp20 = load i32** @a, align 8, !tbaa !3
+  %arrayidx27 = getelementptr inbounds i32* %tmp20, i64 %idxprom
+  %tmp21 = bitcast i32* %arrayidx15 to i8*
+
+  ; CHECK: prfm pstl3keep
+  call void @llvm.prefetch(i8* %tmp21, i32 1, i32 1, i32 1)
+  %tmp22 = load i32** @a, align 8, !tbaa !3
+  %arrayidx30 = getelementptr inbounds i32* %tmp22, i64 %idxprom
+  %tmp23 = bitcast i32* %arrayidx18 to i8*
+
+  ; CHECK: prfm pstl2keep
+  call void @llvm.prefetch(i8* %tmp23, i32 1, i32 2, i32 1)
+  %tmp24 = load i32** @a, align 8, !tbaa !3
+  %arrayidx33 = getelementptr inbounds i32* %tmp24, i64 %idxprom
+  %tmp25 = bitcast i32* %arrayidx21 to i8*
 
   ; CHECK: prfm pstl1keep
-  call void @llvm.prefetch(i8* %tmp17, i32 1, i32 3, i32 1)
+  call void @llvm.prefetch(i8* %tmp25, i32 1, i32 3, i32 1)
   ret void
 }
 
-- 
2.11.0