summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorHal Finkel <hfinkel@anl.gov>2014-03-26 19:39:09 +0000
committerHal Finkel <hfinkel@anl.gov>2014-03-26 19:39:09 +0000
commit6e28e6aaaf68dd75fcadaf6abe48c10cb20a43e4 (patch)
tree5f050af214b586ef0b5db45b758febdd4978dbb0
parent8108f38437f3daca9bd11080e4232980e8cd0d94 (diff)
downloadbcm5719-llvm-6e28e6aaaf68dd75fcadaf6abe48c10cb20a43e4.tar.gz
bcm5719-llvm-6e28e6aaaf68dd75fcadaf6abe48c10cb20a43e4.zip
[PowerPC] VSX loads and stores support unaligned access
I've not yet updated PPCTTI because I'm not sure what the actual relative cost is compared to the aligned uses. llvm-svn: 204848
-rw-r--r--llvm/lib/Target/PowerPC/PPCISelLowering.cpp11
-rw-r--r--llvm/lib/Target/PowerPC/PPCTargetTransformInfo.cpp2
-rw-r--r--llvm/test/CodeGen/PowerPC/vsx.ll18
3 files changed, 28 insertions, 3 deletions
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index e09725d59d2..87639247e21 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -7960,7 +7960,6 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
unsigned ABIAlignment = getDataLayout()->getABITypeAlignment(Ty);
if (ISD::isNON_EXTLoad(N) && VT.isVector() &&
TM.getSubtarget<PPCSubtarget>().hasAltivec() &&
- // FIXME: Update this for VSX!
(VT == MVT::v16i8 || VT == MVT::v8i16 ||
VT == MVT::v4i32 || VT == MVT::v4f32) &&
LD->getAlignment() < ABIAlignment) {
@@ -8716,8 +8715,14 @@ bool PPCTargetLowering::allowsUnalignedMemoryAccesses(EVT VT,
if (!VT.isSimple())
return false;
- if (VT.getSimpleVT().isVector())
- return false;
+ if (VT.getSimpleVT().isVector()) {
+ if (PPCSubTarget.hasVSX()) {
+ if (VT != MVT::v2f64 && VT != MVT::v2i64)
+ return false;
+ } else {
+ return false;
+ }
+ }
if (VT == MVT::ppcf128)
return false;
diff --git a/llvm/lib/Target/PowerPC/PPCTargetTransformInfo.cpp b/llvm/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
index 2491a9e5392..a915b04fa50 100644
--- a/llvm/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
+++ b/llvm/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
@@ -244,6 +244,8 @@ unsigned PPCTTI::getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
// Each load/store unit costs 1.
unsigned Cost = LT.first * 1;
+ // FIXME: Update this for VSX loads/stores that support unaligned access.
+
// PPC in general does not support unaligned loads and stores. They'll need
// to be decomposed based on the alignment factor.
unsigned SrcBytes = LT.second.getStoreSize();
diff --git a/llvm/test/CodeGen/PowerPC/vsx.ll b/llvm/test/CodeGen/PowerPC/vsx.ll
index d2f8bb8ee55..36848427a55 100644
--- a/llvm/test/CodeGen/PowerPC/vsx.ll
+++ b/llvm/test/CodeGen/PowerPC/vsx.ll
@@ -314,6 +314,24 @@ define void @test29(<2 x double>* %a, <2 x double> %b) {
; CHECK: blr
}
+define <2 x double> @test28u(<2 x double>* %a) {
+ %v = load <2 x double>* %a, align 8
+ ret <2 x double> %v
+
+; CHECK-LABEL: @test28u
+; CHECK: lxvd2x 34, 0, 3
+; CHECK: blr
+}
+
+define void @test29u(<2 x double>* %a, <2 x double> %b) {
+ store <2 x double> %b, <2 x double>* %a, align 8
+ ret void
+
+; CHECK-LABEL: @test29u
+; CHECK: stxvd2x 34, 0, 3
+; CHECK: blr
+}
+
define <2 x i64> @test30(<2 x i64>* %a) {
%v = load <2 x i64>* %a, align 16
ret <2 x i64> %v
OpenPOWER on IntegriCloud