Skip to content

Commit

Permalink
[ARM] Don't break alignment when combining base updates into load/sto…
Browse files Browse the repository at this point in the history 
…res.

r223862/r224203 tried to also combine base-updating load/stores.
There was a mistake there: the alignment was added as is as an operand to
the ARMISD::VLD/VST node.  However, the VLD/VST selection logic doesn't care
about less-than-standard alignment attributes.
For example, no matter the alignment of a v2i64 load (say 1), SelectVLD picks
VLD1q64 (because of the memory type).  But VLD1q64 ("vld1.64 {dXX, dYY}") is
8-aligned, per ARMARMv7a 3.2.1.
For the 1-aligned load, what we really want is VLD1q8.

This commit introduces bitcasts if necessary, and changes the vld/vst type to
one whose standard alignment matches the original load/store alignment.

Differential Revision: http://reviews.llvm.org/D6759

llvm-svn: 224754
  • Loading branch information
@ahmedbougacha
ahmedbougacha committed Dec 23, 2014
1 parent 279663c commit 4553bff
Show file tree
Hide file tree
Showing 4 changed files with 152 additions and 26 deletions.
49 changes: 47 additions & 2 deletions llvm/lib/Target/ARM/ARMISelLowering.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8976,13 +8976,42 @@ static SDValue CombineBaseUpdate(SDNode *N,
continue;
}

EVT AlignedVecTy = VecTy;

// If this is a less-than-standard-aligned load/store, change the type to
// match the standard alignment.
// The alignment is overlooked when selecting _UPD variants; and it's
// easier to introduce bitcasts here than fix that.
// There are 3 ways to get to this base-update combine:
// - intrinsics: they are assumed to be properly aligned (to the standard
// alignment of the memory type), so we don't need to do anything.
// - ARMISD::VLDx nodes: they are only generated from the aforementioned
// intrinsics, so, likewise, there's nothing to do.
// - generic load/store instructions: the alignment is specified as an
// explicit operand, rather than implicitly as the standard alignment
// of the memory type (like the intrisics). We need to change the
// memory type to match the explicit alignment. That way, we don't
// generate non-standard-aligned ARMISD::VLDx nodes.
if (LSBaseSDNode *LSN = dyn_cast<LSBaseSDNode>(N)) {
unsigned Alignment = LSN->getAlignment();
if (Alignment == 0)
Alignment = 1;
if (Alignment < VecTy.getScalarSizeInBits() / 8) {
MVT EltTy = MVT::getIntegerVT(Alignment * 8);
assert(NumVecs == 1 && "Unexpected multi-element generic load/store.");
assert(!isLaneOp && "Unexpected generic load/store lane.");
unsigned NumElts = NumBytes / (EltTy.getSizeInBits() / 8);
AlignedVecTy = MVT::getVectorVT(EltTy, NumElts);
}
}

// Create the new updating load/store node.
// First, create an SDVTList for the new updating node's results.
EVT Tys[6];
unsigned NumResultVecs = (isLoad ? NumVecs : 0);
unsigned n;
for (n = 0; n < NumResultVecs; ++n)
Tys[n] = VecTy;
Tys[n] = AlignedVecTy;
Tys[n++] = MVT::i32;
Tys[n] = MVT::Other;
SDVTList SDTys = DAG.getVTList(makeArrayRef(Tys, NumResultVecs+2));
Expand All @@ -9000,16 +9029,32 @@ static SDValue CombineBaseUpdate(SDNode *N,
for (unsigned i = AddrOpIdx + 1; i < N->getNumOperands(); ++i)
Ops.push_back(N->getOperand(i));
}

// If this is a non-standard-aligned store, the penultimate operand is the
// stored value. Bitcast it to the aligned type.
if (AlignedVecTy != VecTy && N->getOpcode() == ISD::STORE) {
SDValue &StVal = Ops[Ops.size()-2];
StVal = DAG.getNode(ISD::BITCAST, SDLoc(N), AlignedVecTy, StVal);
}

MemSDNode *MemInt = cast<MemSDNode>(N);
SDValue UpdN = DAG.getMemIntrinsicNode(NewOpc, SDLoc(N), SDTys,
Ops, MemInt->getMemoryVT(),
Ops, AlignedVecTy,
MemInt->getMemOperand());

// Update the uses.
std::vector<SDValue> NewResults;
for (unsigned i = 0; i < NumResultVecs; ++i) {
NewResults.push_back(SDValue(UpdN.getNode(), i));
}

// If this is an non-standard-aligned load, the first result is the loaded
// value. Bitcast it to the expected result type.
if (AlignedVecTy != VecTy && N->getOpcode() == ISD::LOAD) {
SDValue &LdVal = NewResults[0];
LdVal = DAG.getNode(ISD::BITCAST, SDLoc(N), VecTy, LdVal);
}

NewResults.push_back(SDValue(UpdN.getNode(), NumResultVecs+1)); // chain
DCI.CombineTo(N, NewResults);
DCI.CombineTo(User, SDValue(UpdN.getNode(), NumResultVecs));
Expand Down
10 changes: 5 additions & 5 deletions llvm/test/CodeGen/ARM/memcpy-inline.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -46,8 +46,8 @@ entry:
; CHECK: movw [[REG2:r[0-9]+]], #16716
; CHECK: movt [[REG2:r[0-9]+]], #72
; CHECK: str [[REG2]], [r0, #32]
; CHECK: vld1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r1]!
; CHECK: vst1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r0]!
; CHECK: vld1.8 {d{{[0-9]+}}, d{{[0-9]+}}}, [r1]!
; CHECK: vst1.8 {d{{[0-9]+}}, d{{[0-9]+}}}, [r0]!
; CHECK: vld1.8 {d{{[0-9]+}}, d{{[0-9]+}}}, [r1]
; CHECK: vst1.8 {d{{[0-9]+}}, d{{[0-9]+}}}, [r0]
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %C, i8* getelementptr inbounds ([36 x i8]* @.str2, i64 0, i64 0), i64 36, i32 1, i1 false)
Expand All @@ -57,8 +57,8 @@ entry:
define void @t3(i8* nocapture %C) nounwind {
entry:
; CHECK-LABEL: t3:
; CHECK: vld1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r1]!
; CHECK: vst1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r0]!
; CHECK: vld1.8 {d{{[0-9]+}}, d{{[0-9]+}}}, [r1]!
; CHECK: vst1.8 {d{{[0-9]+}}, d{{[0-9]+}}}, [r0]!
; CHECK: vld1.8 {d{{[0-9]+}}}, [r1]
; CHECK: vst1.8 {d{{[0-9]+}}}, [r0]
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %C, i8* getelementptr inbounds ([24 x i8]* @.str3, i64 0, i64 0), i64 24, i32 1, i1 false)
Expand All @@ -69,7 +69,7 @@ define void @t4(i8* nocapture %C) nounwind {
entry:
; CHECK-LABEL: t4:
; CHECK: vld1.8 {[[REG3:d[0-9]+]], [[REG4:d[0-9]+]]}, [r1]
; CHECK: vst1.64 {[[REG3]], [[REG4]]}, [r0]!
; CHECK: vst1.8 {[[REG3]], [[REG4]]}, [r0]!
; CHECK: strh [[REG5:r[0-9]+]], [r0]
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %C, i8* getelementptr inbounds ([18 x i8]* @.str4, i64 0, i64 0), i64 18, i32 1, i1 false)
ret void
Expand Down
59 changes: 50 additions & 9 deletions llvm/test/CodeGen/ARM/vector-load.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,7 @@ define <4 x i16> @load_v4i16(<4 x i16>** %ptr) {

define <4 x i16> @load_v4i16_update(<4 x i16>** %ptr) {
;CHECK-LABEL: load_v4i16_update:
;CHECK: vld1.16 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vld1.8 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <4 x i16>** %ptr
%lA = load <4 x i16>* %A, align 1
%inc = getelementptr <4 x i16>* %A, i34 1
Expand All @@ -49,7 +49,7 @@ define <2 x i32> @load_v2i32(<2 x i32>** %ptr) {

define <2 x i32> @load_v2i32_update(<2 x i32>** %ptr) {
;CHECK-LABEL: load_v2i32_update:
;CHECK: vld1.32 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vld1.8 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x i32>** %ptr
%lA = load <2 x i32>* %A, align 1
%inc = getelementptr <2 x i32>* %A, i32 1
Expand All @@ -67,7 +67,7 @@ define <2 x float> @load_v2f32(<2 x float>** %ptr) {

define <2 x float> @load_v2f32_update(<2 x float>** %ptr) {
;CHECK-LABEL: load_v2f32_update:
;CHECK: vld1.32 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vld1.8 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x float>** %ptr
%lA = load <2 x float>* %A, align 1
%inc = getelementptr <2 x float>* %A, i32 1
Expand All @@ -85,7 +85,7 @@ define <1 x i64> @load_v1i64(<1 x i64>** %ptr) {

define <1 x i64> @load_v1i64_update(<1 x i64>** %ptr) {
;CHECK-LABEL: load_v1i64_update:
;CHECK: vld1.64 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vld1.8 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <1 x i64>** %ptr
%lA = load <1 x i64>* %A, align 1
%inc = getelementptr <1 x i64>* %A, i31 1
Expand Down Expand Up @@ -121,7 +121,7 @@ define <8 x i16> @load_v8i16(<8 x i16>** %ptr) {

define <8 x i16> @load_v8i16_update(<8 x i16>** %ptr) {
;CHECK-LABEL: load_v8i16_update:
;CHECK: vld1.16 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vld1.8 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <8 x i16>** %ptr
%lA = load <8 x i16>* %A, align 1
%inc = getelementptr <8 x i16>* %A, i38 1
Expand All @@ -139,7 +139,7 @@ define <4 x i32> @load_v4i32(<4 x i32>** %ptr) {

define <4 x i32> @load_v4i32_update(<4 x i32>** %ptr) {
;CHECK-LABEL: load_v4i32_update:
;CHECK: vld1.32 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vld1.8 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <4 x i32>** %ptr
%lA = load <4 x i32>* %A, align 1
%inc = getelementptr <4 x i32>* %A, i34 1
Expand All @@ -157,7 +157,7 @@ define <4 x float> @load_v4f32(<4 x float>** %ptr) {

define <4 x float> @load_v4f32_update(<4 x float>** %ptr) {
;CHECK-LABEL: load_v4f32_update:
;CHECK: vld1.32 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vld1.8 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <4 x float>** %ptr
%lA = load <4 x float>* %A, align 1
%inc = getelementptr <4 x float>* %A, i34 1
Expand All @@ -175,22 +175,63 @@ define <2 x i64> @load_v2i64(<2 x i64>** %ptr) {

define <2 x i64> @load_v2i64_update(<2 x i64>** %ptr) {
;CHECK-LABEL: load_v2i64_update:
;CHECK: vld1.64 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vld1.8 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x i64>** %ptr
%lA = load <2 x i64>* %A, align 1
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret <2 x i64> %lA
}

; Make sure we change the type to match alignment if necessary.
define <2 x i64> @load_v2i64_update_aligned2(<2 x i64>** %ptr) {
;CHECK-LABEL: load_v2i64_update_aligned2:
;CHECK: vld1.16 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x i64>** %ptr
%lA = load <2 x i64>* %A, align 2
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret <2 x i64> %lA
}

define <2 x i64> @load_v2i64_update_aligned4(<2 x i64>** %ptr) {
;CHECK-LABEL: load_v2i64_update_aligned4:
;CHECK: vld1.32 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x i64>** %ptr
%lA = load <2 x i64>* %A, align 4
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret <2 x i64> %lA
}

define <2 x i64> @load_v2i64_update_aligned8(<2 x i64>** %ptr) {
;CHECK-LABEL: load_v2i64_update_aligned8:
;CHECK: vld1.64 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}:64]!
%A = load <2 x i64>** %ptr
%lA = load <2 x i64>* %A, align 8
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret <2 x i64> %lA
}

define <2 x i64> @load_v2i64_update_aligned16(<2 x i64>** %ptr) {
;CHECK-LABEL: load_v2i64_update_aligned16:
;CHECK: vld1.64 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}:128]!
%A = load <2 x i64>** %ptr
%lA = load <2 x i64>* %A, align 16
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret <2 x i64> %lA
}

; Make sure we don't break smaller-than-dreg extloads.
define <4 x i32> @zextload_v8i8tov8i32(<4 x i8>** %ptr) {
;CHECK-LABEL: zextload_v8i8tov8i32:
;CHECK: vld1.32 {{{d[0-9]+}}[0]}, [{{r[0-9]+}}:32]
;CHECK: vmovl.u8 {{q[0-9]+}}, {{d[0-9]+}}
;CHECK: vmovl.u16 {{q[0-9]+}}, {{d[0-9]+}}
%A = load <4 x i8>** %ptr
%lA = load <4 x i8>* %A, align 1
%lA = load <4 x i8>* %A, align 4
%zlA = zext <4 x i8> %lA to <4 x i32>
ret <4 x i32> %zlA
}
Expand Down
60 changes: 50 additions & 10 deletions llvm/test/CodeGen/ARM/vector-store.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,7 @@ define void @store_v4i16(<4 x i16>** %ptr, <4 x i16> %val) {

define void @store_v4i16_update(<4 x i16>** %ptr, <4 x i16> %val) {
;CHECK-LABEL: store_v4i16_update:
;CHECK: vst1.16 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vst1.8 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <4 x i16>** %ptr
store <4 x i16> %val, <4 x i16>* %A, align 1
%inc = getelementptr <4 x i16>* %A, i34 1
Expand All @@ -49,7 +49,7 @@ define void @store_v2i32(<2 x i32>** %ptr, <2 x i32> %val) {

define void @store_v2i32_update(<2 x i32>** %ptr, <2 x i32> %val) {
;CHECK-LABEL: store_v2i32_update:
;CHECK: vst1.32 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vst1.8 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x i32>** %ptr
store <2 x i32> %val, <2 x i32>* %A, align 1
%inc = getelementptr <2 x i32>* %A, i32 1
Expand All @@ -67,7 +67,7 @@ define void @store_v2f32(<2 x float>** %ptr, <2 x float> %val) {

define void @store_v2f32_update(<2 x float>** %ptr, <2 x float> %val) {
;CHECK-LABEL: store_v2f32_update:
;CHECK: vst1.32 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vst1.8 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x float>** %ptr
store <2 x float> %val, <2 x float>* %A, align 1
%inc = getelementptr <2 x float>* %A, i32 1
Expand All @@ -85,7 +85,7 @@ define void @store_v1i64(<1 x i64>** %ptr, <1 x i64> %val) {

define void @store_v1i64_update(<1 x i64>** %ptr, <1 x i64> %val) {
;CHECK-LABEL: store_v1i64_update:
;CHECK: vst1.64 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vst1.8 {{{d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <1 x i64>** %ptr
store <1 x i64> %val, <1 x i64>* %A, align 1
%inc = getelementptr <1 x i64>* %A, i31 1
Expand Down Expand Up @@ -121,7 +121,7 @@ define void @store_v8i16(<8 x i16>** %ptr, <8 x i16> %val) {

define void @store_v8i16_update(<8 x i16>** %ptr, <8 x i16> %val) {
;CHECK-LABEL: store_v8i16_update:
;CHECK: vst1.16 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vst1.8 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <8 x i16>** %ptr
store <8 x i16> %val, <8 x i16>* %A, align 1
%inc = getelementptr <8 x i16>* %A, i38 1
Expand All @@ -139,7 +139,7 @@ define void @store_v4i32(<4 x i32>** %ptr, <4 x i32> %val) {

define void @store_v4i32_update(<4 x i32>** %ptr, <4 x i32> %val) {
;CHECK-LABEL: store_v4i32_update:
;CHECK: vst1.32 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vst1.8 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <4 x i32>** %ptr
store <4 x i32> %val, <4 x i32>* %A, align 1
%inc = getelementptr <4 x i32>* %A, i34 1
Expand All @@ -157,7 +157,7 @@ define void @store_v4f32(<4 x float>** %ptr, <4 x float> %val) {

define void @store_v4f32_update(<4 x float>** %ptr, <4 x float> %val) {
;CHECK-LABEL: store_v4f32_update:
;CHECK: vst1.32 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vst1.8 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <4 x float>** %ptr
store <4 x float> %val, <4 x float>* %A, align 1
%inc = getelementptr <4 x float>* %A, i34 1
Expand All @@ -175,14 +175,54 @@ define void @store_v2i64(<2 x i64>** %ptr, <2 x i64> %val) {

define void @store_v2i64_update(<2 x i64>** %ptr, <2 x i64> %val) {
;CHECK-LABEL: store_v2i64_update:
;CHECK: vst1.64 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
;CHECK: vst1.8 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x i64>** %ptr
store <2 x i64> %val, <2 x i64>* %A, align 1
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret void
}

define void @store_v2i64_update_aligned2(<2 x i64>** %ptr, <2 x i64> %val) {
;CHECK-LABEL: store_v2i64_update_aligned2:
;CHECK: vst1.16 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x i64>** %ptr
store <2 x i64> %val, <2 x i64>* %A, align 2
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret void
}

define void @store_v2i64_update_aligned4(<2 x i64>** %ptr, <2 x i64> %val) {
;CHECK-LABEL: store_v2i64_update_aligned4:
;CHECK: vst1.32 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}]!
%A = load <2 x i64>** %ptr
store <2 x i64> %val, <2 x i64>* %A, align 4
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret void
}

define void @store_v2i64_update_aligned8(<2 x i64>** %ptr, <2 x i64> %val) {
;CHECK-LABEL: store_v2i64_update_aligned8:
;CHECK: vst1.64 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}:64]!
%A = load <2 x i64>** %ptr
store <2 x i64> %val, <2 x i64>* %A, align 8
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret void
}

define void @store_v2i64_update_aligned16(<2 x i64>** %ptr, <2 x i64> %val) {
;CHECK-LABEL: store_v2i64_update_aligned16:
;CHECK: vst1.64 {{{d[0-9]+, d[0-9]+}}}, [{{r[0-9]+}}:128]!
%A = load <2 x i64>** %ptr
store <2 x i64> %val, <2 x i64>* %A, align 16
%inc = getelementptr <2 x i64>* %A, i32 1
store <2 x i64>* %inc, <2 x i64>** %ptr
ret void
}

define void @truncstore_v4i32tov4i8(<4 x i8>** %ptr, <4 x i32> %val) {
;CHECK-LABEL: truncstore_v4i32tov4i8:
;CHECK: ldr.w r9, [sp]
Expand All @@ -191,10 +231,10 @@ define void @truncstore_v4i32tov4i8(<4 x i8>** %ptr, <4 x i32> %val) {
;CHECK: vmovn.i32 [[VECLO:d[0-9]+]], {{q[0-9]+}}
;CHECK: vuzp.8 [[VECLO]], {{d[0-9]+}}
;CHECK: ldr r[[PTRREG:[0-9]+]], [r0]
;CHECK: vst1.32 {[[VECLO]][0]}, [r[[PTRREG]]]
;CHECK: vst1.32 {[[VECLO]][0]}, [r[[PTRREG]]:32]
%A = load <4 x i8>** %ptr
%trunc = trunc <4 x i32> %val to <4 x i8>
store <4 x i8> %trunc, <4 x i8>* %A, align 1
store <4 x i8> %trunc, <4 x i8>* %A, align 4
ret void
}

Expand Down

0 comments on commit 4553bff

Please sign in to comment.