Enable unaligned memory access on PPC for scalar types

Unaligned access is supported on PPC for non-vector types, and is generally
more efficient than manually expanding the loads and stores.

A few of the existing test cases were using expanded unaligned loads and stores
to test other features (like load/store with update), and for these test cases,
unaligned access remains disabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177160 91177308-0d34-0410-b5e6-96231b3b80d8

7 files changed, 110 insertions, 4 deletions

diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index 13cb358fc0..a7e9d56fa9 100644
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -57,6 +57,9 @@ cl::desc("disable preincrement load/store generation on PPC"), cl::Hidden);
 static cl::opt<bool> DisableILPPref("disable-ppc-ilp-pref",
 cl::desc("disable setting the node scheduling preference to ILP on PPC"), cl::Hidden);
 
+static cl::opt<bool> DisablePPCUnaligned("disable-ppc-unaligned",
+cl::desc("disable unaligned load/store generation on PPC"), cl::Hidden);
+
 static TargetLoweringObjectFile *CreateTLOF(const PPCTargetMachine &TM) {
   if (TM.getSubtargetImpl()->isDarwin())
     return new TargetLoweringObjectFileMachO();
@@ -6851,6 +6854,32 @@ EVT PPCTargetLowering::getOptimalMemOpType(uint64_t Size,
   }
 }
 
+bool PPCTargetLowering::allowsUnalignedMemoryAccesses(EVT VT,
+                                                      bool *Fast) const {
+  if (DisablePPCUnaligned)
+    return false;
+
+  // PowerPC supports unaligned memory access for simple non-vector types.
+  // Although accessing unaligned addresses is not as efficient as accessing
+  // aligned addresses, it is generally more efficient than manual expansion,
+  // and generally only traps for software emulation when crossing page
+  // boundaries.
+
+  if (!VT.isSimple())
+    return false;
+
+  if (VT.getSimpleVT().isVector())
+    return false;
+
+  if (VT == MVT::ppcf128)
+    return false;
+
+  if (Fast)
+    *Fast = true;
+
+  return true;
+}
+
 /// isFMAFasterThanMulAndAdd - Return true if an FMA operation is faster than
 /// a pair of mul and add instructions. fmuladd intrinsics will be expanded to
 /// FMAs when this method returns true (and FMAs are legal), otherwise fmuladd
diff --git a/lib/Target/PowerPC/PPCISelLowering.h b/lib/Target/PowerPC/PPCISelLowering.h
index 3931384d89..8d44d9ff46 100644
--- a/lib/Target/PowerPC/PPCISelLowering.h
+++ b/lib/Target/PowerPC/PPCISelLowering.h
@@ -449,6 +449,10 @@ namespace llvm {
                         bool IsMemset, bool ZeroMemset, bool MemcpyStrSrc,
                         MachineFunction &MF) const;
 
+    /// Is unaligned memory access allowed for the given type, and is it fast
+    /// relative to software emulation.
+    virtual bool allowsUnalignedMemoryAccesses(EVT VT, bool *Fast = 0) const;
+
     /// isFMAFasterThanMulAndAdd - Return true if an FMA operation is faster than
     /// a pair of mul and add instructions. fmuladd intrinsics will be expanded to
     /// FMAs when this method returns true (and FMAs are legal), otherwise fmuladd
diff --git a/test/CodeGen/PowerPC/lbzux.ll b/test/CodeGen/PowerPC/lbzux.ll
index 12f1d1f130..98951306fd 100644
--- a/test/CodeGen/PowerPC/lbzux.ll
+++ b/test/CodeGen/PowerPC/lbzux.ll
@@ -1,6 +1,6 @@
 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128-n32:64"
 target triple = "powerpc64-unknown-linux-gnu"
-; RUN: llc < %s | FileCheck %s
+; RUN: llc -disable-ppc-unaligned < %s | FileCheck %s
 
 define fastcc void @allocateSpace(i1 %cond1, i1 %cond2) nounwind {
 entry:
diff --git a/test/CodeGen/PowerPC/structsinmem.ll b/test/CodeGen/PowerPC/structsinmem.ll
index 884d3a89d1..59b108a814 100644
--- a/test/CodeGen/PowerPC/structsinmem.ll
+++ b/test/CodeGen/PowerPC/structsinmem.ll
@@ -1,4 +1,4 @@
-; RUN: llc -mcpu=pwr7 -O0 -disable-fp-elim < %s | FileCheck %s
+; RUN: llc -mcpu=pwr7 -O0 -disable-fp-elim -disable-ppc-unaligned < %s | FileCheck %s
 
 ; FIXME: The code generation for packed structs is very poor because the
 ; PowerPC target wrongly rejects all unaligned loads.  This test case will
diff --git a/test/CodeGen/PowerPC/structsinregs.ll b/test/CodeGen/PowerPC/structsinregs.ll
index c7cbfa8ee6..c746039f43 100644
--- a/test/CodeGen/PowerPC/structsinregs.ll
+++ b/test/CodeGen/PowerPC/structsinregs.ll
@@ -1,4 +1,4 @@
-; RUN: llc -mcpu=pwr7 -O0 -disable-fp-elim < %s | FileCheck %s
+; RUN: llc -mcpu=pwr7 -O0 -disable-fp-elim -disable-ppc-unaligned < %s | FileCheck %s
 
 ; FIXME: The code generation for packed structs is very poor because the
 ; PowerPC target wrongly rejects all unaligned loads.  This test case will
diff --git a/test/CodeGen/PowerPC/stwu8.ll b/test/CodeGen/PowerPC/stwu8.ll
index 897bfc6d6c..e0bd043454 100644
--- a/test/CodeGen/PowerPC/stwu8.ll
+++ b/test/CodeGen/PowerPC/stwu8.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s | FileCheck %s
+; RUN: llc -disable-ppc-unaligned < %s | FileCheck %s
 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128-n32:64"
 target triple = "powerpc64-unknown-linux-gnu"
 
diff --git a/test/CodeGen/PowerPC/unaligned.ll b/test/CodeGen/PowerPC/unaligned.ll
new file mode 100644
index 0000000000..d05080338f
--- /dev/null
+++ b/test/CodeGen/PowerPC/unaligned.ll
@@ -0,0 +1,73 @@
+; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 | FileCheck %s
+target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f128:64:128-n32"
+
+define void @foo1(i16* %p, i16* %r) nounwind {
+entry:
+  %v = load i16* %p, align 1
+  store i16 %v, i16* %r, align 1
+  ret void
+
+; CHECK: @foo1
+; CHECK: lhz
+; CHECK: sth
+}
+
+define void @foo2(i32* %p, i32* %r) nounwind {
+entry:
+  %v = load i32* %p, align 1
+  store i32 %v, i32* %r, align 1
+  ret void
+
+; CHECK: @foo2
+; CHECK: lwz
+; CHECK: stw
+}
+
+define void @foo3(i64* %p, i64* %r) nounwind {
+entry:
+  %v = load i64* %p, align 1
+  store i64 %v, i64* %r, align 1
+  ret void
+
+; CHECK: @foo3
+; CHECK: ld
+; CHECK: std
+}
+
+define void @foo4(float* %p, float* %r) nounwind {
+entry:
+  %v = load float* %p, align 1
+  store float %v, float* %r, align 1
+  ret void
+
+; CHECK: @foo4
+; CHECK: lfs
+; CHECK: stfs
+}
+
+define void @foo5(double* %p, double* %r) nounwind {
+entry:
+  %v = load double* %p, align 1
+  store double %v, double* %r, align 1
+  ret void
+
+; CHECK: @foo5
+; CHECK: lfd
+; CHECK: stfd
+}
+
+define void @foo6(<4 x float>* %p, <4 x float>* %r) nounwind {
+entry:
+  %v = load <4 x float>* %p, align 1
+  store <4 x float> %v, <4 x float>* %r, align 1
+  ret void
+
+; These loads and stores are legalized into aligned loads and stores
+; using aligned stack slots.
+; CHECK: @foo6
+; CHECK: ld
+; CHECK: ld
+; CHECK: std
+; CHECK: std
+}
+