[llvm] r362472 - [DAGCombine] Match a pattern where a wide type scalar value is stored by several narrow stores
QingShan Zhang via llvm-commits
llvm-commits at lists.llvm.org
Tue Jun 4 01:53:53 PDT 2019
Author: qshanz
Date: Tue Jun 4 01:53:53 2019
New Revision: 362472
URL: http://llvm.org/viewvc/llvm-project?rev=362472&view=rev
Log:
[DAGCombine] Match a pattern where a wide type scalar value is stored by several narrow stores
This opportunity is found from spec 2017 557.xz_r. And it is used by the sha encrypt/decrypt. See sha-2/sha512.c
static void store64(u64 x, unsigned char* y)
{
for(int i = 0; i != 8; ++i)
y[i] = (x >> ((7-i) * 8)) & 255;
}
static u64 load64(const unsigned char* y)
{
u64 res = 0;
for(int i = 0; i != 8; ++i)
res |= (u64)(y[i]) << ((7-i) * 8);
return res;
}
The load64 has been implemented by https://reviews.llvm.org/D26149
This patch is trying to implement the store pattern.
Match a pattern where a wide type scalar value is stored by several narrow
stores. Fold it into a single store or a BSWAP and a store if the targets
supports it.
Assuming little endian target:
i8 *p = ...
i32 val = ...
p[0] = (val >> 0) & 0xFF;
p[1] = (val >> 8) & 0xFF;
p[2] = (val >> 16) & 0xFF;
p[3] = (val >> 24) & 0xFF;
>
*((i32)p) = val;
i8 *p = ...
i32 val = ...
p[0] = (val >> 24) & 0xFF;
p[1] = (val >> 16) & 0xFF;
p[2] = (val >> 8) & 0xFF;
p[3] = (val >> 0) & 0xFF;
>
*((i32)p) = BSWAP(val);
Differential Revision: https://reviews.llvm.org/D61843
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/trunk/test/CodeGen/PowerPC/store-combine.ll
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=362472&r1=362471&r2=362472&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Tue Jun 4 01:53:53 2019
@@ -524,6 +524,7 @@ namespace {
const SDLoc &DL);
SDNode *MatchRotate(SDValue LHS, SDValue RHS, const SDLoc &DL);
SDValue MatchLoadCombine(SDNode *N);
+ SDValue MatchStoreCombine(StoreSDNode *N);
SDValue ReduceLoadWidth(SDNode *N);
SDValue ReduceLoadOpStoreWidth(SDNode *N);
SDValue splitMergedValStore(StoreSDNode *ST);
@@ -6262,6 +6263,180 @@ static Optional<bool> isBigEndian(const
return BigEndian;
}
+static SDValue stripTruncAndExt(SDValue Value) {
+ switch (Value.getOpcode()) {
+ case ISD::TRUNCATE:
+ case ISD::ZERO_EXTEND:
+ case ISD::SIGN_EXTEND:
+ case ISD::ANY_EXTEND:
+ return stripTruncAndExt(Value.getOperand(0));
+ }
+ return Value;
+}
+
+/// Match a pattern where a wide type scalar value is stored by several narrow
+/// stores. Fold it into a single store or a BSWAP and a store if the targets
+/// supports it.
+///
+/// Assuming little endian target:
+/// i8 *p = ...
+/// i32 val = ...
+/// p[0] = (val >> 0) & 0xFF;
+/// p[1] = (val >> 8) & 0xFF;
+/// p[2] = (val >> 16) & 0xFF;
+/// p[3] = (val >> 24) & 0xFF;
+/// =>
+/// *((i32)p) = val;
+///
+/// i8 *p = ...
+/// i32 val = ...
+/// p[0] = (val >> 24) & 0xFF;
+/// p[1] = (val >> 16) & 0xFF;
+/// p[2] = (val >> 8) & 0xFF;
+/// p[3] = (val >> 0) & 0xFF;
+/// =>
+/// *((i32)p) = BSWAP(val);
+SDValue DAGCombiner::MatchStoreCombine(StoreSDNode *N) {
+ // Collect all the stores in the chain.
+ SDValue Chain;
+ SmallVector<StoreSDNode *, 8> Stores;
+ for (StoreSDNode *Store = N; Store; Store = dyn_cast<StoreSDNode>(Chain)) {
+ if (Store->getMemoryVT() != MVT::i8 ||
+ Store->isVolatile() || Store->isIndexed())
+ return SDValue();
+ Stores.push_back(Store);
+ Chain = Store->getChain();
+ }
+ // Handle the simple type only.
+ unsigned Width = Stores.size();
+ EVT VT = EVT::getIntegerVT(
+ *DAG.getContext(), Width * N->getMemoryVT().getSizeInBits());
+ if (VT != MVT::i16 && VT != MVT::i32 && VT != MVT::i64)
+ return SDValue();
+
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ if (LegalOperations && !TLI.isOperationLegal(ISD::STORE, VT))
+ return SDValue();
+
+ // Check if all the bytes of the combined value we are looking at are stored
+ // to the same base address. Collect bytes offsets from Base address into
+ // ByteOffsets.
+ SDValue CombinedValue;
+ SmallVector<int64_t, 4> ByteOffsets(Width);
+ int64_t FirstOffset = INT64_MAX;
+ StoreSDNode *FirstStore = nullptr;
+ Optional<BaseIndexOffset> Base;
+ for (auto Store : Stores) {
+ // All the stores store different byte of the CombinedValue. A truncate is
+ // required to get that byte value.
+ SDValue Trunc = Store->getValue();
+ if (Trunc.getOpcode() != ISD::TRUNCATE)
+ return SDValue();
+ // A shift operation is required to get the right byte offset, except the
+ // first byte.
+ int64_t Offset = 0;
+ SDValue Value = Trunc.getOperand(0);
+ if (Value.getOpcode() == ISD::SRL ||
+ Value.getOpcode() == ISD::SRA) {
+ ConstantSDNode *ShiftOffset =
+ dyn_cast<ConstantSDNode>(Value.getOperand(1));
+ // Trying to match the following pattern. The shift offset must be
+ // a constant and a multiple of 8. It is the byte offset in "y".
+ //
+ // x = srl y, offset
+ // i8 z = trunc x
+ // store z, ...
+ if (!ShiftOffset || (ShiftOffset->getSExtValue() % 8))
+ return SDValue();
+
+ Offset = ShiftOffset->getSExtValue()/8;
+ Value = Value.getOperand(0);
+ }
+
+ // Stores must share the same combined value with different offsets.
+ if (!CombinedValue)
+ CombinedValue = Value;
+ else if (stripTruncAndExt(CombinedValue) != stripTruncAndExt(Value))
+ return SDValue();
+
+ // The trunc and all the extend operation should be stripped to get the
+ // real value we are stored.
+ else if (CombinedValue.getValueType() != VT) {
+ if (Value.getValueType() == VT ||
+ Value.getValueSizeInBits() > CombinedValue.getValueSizeInBits())
+ CombinedValue = Value;
+ // Give up if the combined value type is smaller than the store size.
+ if (CombinedValue.getValueSizeInBits() < VT.getSizeInBits())
+ return SDValue();
+ }
+
+ // Stores must share the same base address
+ BaseIndexOffset Ptr = BaseIndexOffset::match(Store, DAG);
+ int64_t ByteOffsetFromBase = 0;
+ if (!Base)
+ Base = Ptr;
+ else if (!Base->equalBaseIndex(Ptr, DAG, ByteOffsetFromBase))
+ return SDValue();
+
+ // Remember the first byte store
+ if (ByteOffsetFromBase < FirstOffset) {
+ FirstStore = Store;
+ FirstOffset = ByteOffsetFromBase;
+ }
+ // Map the offset in the store and the offset in the combined value.
+ if (Offset < 0 || Offset >= Width)
+ return SDValue();
+ ByteOffsets[Offset] = ByteOffsetFromBase;
+ }
+
+ assert(FirstOffset != INT64_MAX && "First byte offset must be set");
+ assert(FirstStore && "First store must be set");
+
+ // Check if the bytes of the combined value we are looking at match with
+ // either big or little endian value store.
+ Optional<bool> IsBigEndian = isBigEndian(ByteOffsets, FirstOffset);
+ if (!IsBigEndian.hasValue())
+ return SDValue();
+
+ // The node we are looking at matches with the pattern, check if we can
+ // replace it with a single bswap if needed and store.
+
+ // If the store needs byte swap check if the target supports it
+ bool NeedsBswap = DAG.getDataLayout().isBigEndian() != *IsBigEndian;
+
+ // Before legalize we can introduce illegal bswaps which will be later
+ // converted to an explicit bswap sequence. This way we end up with a single
+ // store and byte shuffling instead of several stores and byte shuffling.
+ if (NeedsBswap && LegalOperations && !TLI.isOperationLegal(ISD::BSWAP, VT))
+ return SDValue();
+
+ // Check that a store of the wide type is both allowed and fast on the target
+ bool Fast = false;
+ bool Allowed = TLI.allowsMemoryAccess(*DAG.getContext(), DAG.getDataLayout(),
+ VT, FirstStore->getAddressSpace(),
+ FirstStore->getAlignment(), &Fast);
+ if (!Allowed || !Fast)
+ return SDValue();
+
+ if (VT != CombinedValue.getValueType()) {
+ assert(CombinedValue.getValueType().getSizeInBits() > VT.getSizeInBits() &&
+ "Get unexpected store value to combine");
+ CombinedValue = DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT,
+ CombinedValue);
+ }
+
+ if (NeedsBswap)
+ CombinedValue = DAG.getNode(ISD::BSWAP, SDLoc(N), VT, CombinedValue);
+
+ SDValue NewStore =
+ DAG.getStore(Chain, SDLoc(N), CombinedValue, FirstStore->getBasePtr(),
+ FirstStore->getPointerInfo(), FirstStore->getAlignment());
+
+ // Rely on other DAG combine rules to remove the other individual stores.
+ DAG.ReplaceAllUsesWith(N, NewStore.getNode());
+ return NewStore;
+}
+
/// Match a pattern where a wide type scalar value is loaded by several narrow
/// loads and combined by shifts and ors. Fold it into a single load or a load
/// and a BSWAP if the targets supports it.
@@ -15762,6 +15937,10 @@ SDValue DAGCombiner::visitSTORE(SDNode *
if (SDValue NewST = TransformFPLoadStorePair(N))
return NewST;
+ // Try transforming several stores into STORE (BSWAP).
+ if (SDValue Store = MatchStoreCombine(ST))
+ return Store;
+
if (ST->isUnindexed()) {
// Walk up chain skipping non-aliasing memory nodes, on this store and any
// adjacent stores.
Modified: llvm/trunk/test/CodeGen/PowerPC/store-combine.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/store-combine.ll?rev=362472&r1=362471&r2=362472&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/store-combine.ll (original)
+++ llvm/trunk/test/CodeGen/PowerPC/store-combine.ll Tue Jun 4 01:53:53 2019
@@ -10,24 +10,12 @@
define void @store_i32_by_i8(i32 signext %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 3, 0(4)
-; CHECK-PPC64LE-NEXT: stb 5, 1(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: srwi 3, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, 2(4)
-; CHECK-PPC64LE-NEXT: stb 3, 3(4)
+; CHECK-PPC64LE-NEXT: stw 3, 0(4)
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i32_by_i8:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 3, 0(4)
-; CHECK-PPC64-NEXT: stb 5, 1(4)
-; CHECK-PPC64-NEXT: srwi 5, 3, 16
-; CHECK-PPC64-NEXT: srwi 3, 3, 24
-; CHECK-PPC64-NEXT: stb 5, 2(4)
-; CHECK-PPC64-NEXT: stb 3, 3(4)
+; CHECK-PPC64-NEXT: stwbrx 3, 0, 4
; CHECK-PPC64-NEXT: blr
entry:
%conv = trunc i32 %m to i8
@@ -55,24 +43,12 @@ entry:
define void @store_i32_by_i8_bswap(i32 signext %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_bswap:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, 0(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: stb 5, 1(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 5, 2(4)
-; CHECK-PPC64LE-NEXT: stb 3, 3(4)
+; CHECK-PPC64LE-NEXT: stwbrx 3, 0, 4
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i32_by_i8_bswap:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 5, 3, 24
-; CHECK-PPC64-NEXT: srwi 6, 3, 16
-; CHECK-PPC64-NEXT: stb 5, 0(4)
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 6, 1(4)
-; CHECK-PPC64-NEXT: stb 5, 2(4)
-; CHECK-PPC64-NEXT: stb 3, 3(4)
+; CHECK-PPC64-NEXT: stw 3, 0(4)
; CHECK-PPC64-NEXT: blr
entry:
%0 = lshr i32 %m, 24
@@ -104,40 +80,12 @@ entry:
define void @store_i64_by_i8(i64 %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i64_by_i8:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 56, 8
-; CHECK-PPC64LE-NEXT: stb 3, 0(4)
-; CHECK-PPC64LE-NEXT: stb 5, 1(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 48, 16
-; CHECK-PPC64LE-NEXT: stb 5, 2(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 40, 24
-; CHECK-PPC64LE-NEXT: stb 5, 3(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 32, 32
-; CHECK-PPC64LE-NEXT: stb 5, 4(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 24, 40
-; CHECK-PPC64LE-NEXT: stb 5, 5(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 16, 48
-; CHECK-PPC64LE-NEXT: rldicl 3, 3, 8, 56
-; CHECK-PPC64LE-NEXT: stb 5, 6(4)
-; CHECK-PPC64LE-NEXT: stb 3, 7(4)
+; CHECK-PPC64LE-NEXT: stdx 3, 0, 4
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i64_by_i8:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: rldicl 5, 3, 56, 8
-; CHECK-PPC64-NEXT: rldicl 6, 3, 48, 16
-; CHECK-PPC64-NEXT: stb 5, 1(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 40, 24
-; CHECK-PPC64-NEXT: stb 6, 2(4)
-; CHECK-PPC64-NEXT: rldicl 6, 3, 32, 32
-; CHECK-PPC64-NEXT: stb 5, 3(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 24, 40
-; CHECK-PPC64-NEXT: stb 6, 4(4)
-; CHECK-PPC64-NEXT: stb 3, 0(4)
-; CHECK-PPC64-NEXT: stb 5, 5(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 16, 48
-; CHECK-PPC64-NEXT: rldicl 3, 3, 8, 56
-; CHECK-PPC64-NEXT: stb 5, 6(4)
-; CHECK-PPC64-NEXT: stb 3, 7(4)
+; CHECK-PPC64-NEXT: stdbrx 3, 0, 4
; CHECK-PPC64-NEXT: blr
entry:
%conv = trunc i64 %m to i8
@@ -185,40 +133,12 @@ entry:
define void @store_i64_by_i8_bswap(i64 %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i64_by_i8_bswap:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 56, 8
-; CHECK-PPC64LE-NEXT: stb 3, 7(4)
-; CHECK-PPC64LE-NEXT: stb 5, 6(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 48, 16
-; CHECK-PPC64LE-NEXT: stb 5, 5(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 40, 24
-; CHECK-PPC64LE-NEXT: stb 5, 4(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 32, 32
-; CHECK-PPC64LE-NEXT: stb 5, 3(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 24, 40
-; CHECK-PPC64LE-NEXT: stb 5, 2(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 16, 48
-; CHECK-PPC64LE-NEXT: rldicl 3, 3, 8, 56
-; CHECK-PPC64LE-NEXT: stb 5, 1(4)
-; CHECK-PPC64LE-NEXT: stb 3, 0(4)
+; CHECK-PPC64LE-NEXT: stdbrx 3, 0, 4
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i64_by_i8_bswap:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: rldicl 5, 3, 56, 8
-; CHECK-PPC64-NEXT: rldicl 6, 3, 48, 16
-; CHECK-PPC64-NEXT: stb 5, 6(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 40, 24
-; CHECK-PPC64-NEXT: stb 6, 5(4)
-; CHECK-PPC64-NEXT: rldicl 6, 3, 32, 32
-; CHECK-PPC64-NEXT: stb 5, 4(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 24, 40
-; CHECK-PPC64-NEXT: stb 6, 3(4)
-; CHECK-PPC64-NEXT: stb 3, 7(4)
-; CHECK-PPC64-NEXT: stb 5, 2(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 16, 48
-; CHECK-PPC64-NEXT: rldicl 3, 3, 8, 56
-; CHECK-PPC64-NEXT: stb 5, 1(4)
-; CHECK-PPC64-NEXT: stb 3, 0(4)
+; CHECK-PPC64-NEXT: stdx 3, 0, 4
; CHECK-PPC64-NEXT: blr
entry:
%conv = trunc i64 %m to i8
@@ -267,46 +187,18 @@ entry:
define void @store_i64_by_i8_bswap_uses(i32 signext %t, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i64_by_i8_bswap_uses:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: slwi 5, 3, 3
-; CHECK-PPC64LE-NEXT: subf 3, 3, 5
-; CHECK-PPC64LE-NEXT: extsw 3, 3
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 56, 8
-; CHECK-PPC64LE-NEXT: stb 3, 7(4)
-; CHECK-PPC64LE-NEXT: stb 5, 6(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 48, 16
-; CHECK-PPC64LE-NEXT: stb 5, 5(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 40, 24
-; CHECK-PPC64LE-NEXT: stb 5, 4(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 32, 32
-; CHECK-PPC64LE-NEXT: stb 5, 3(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 24, 40
-; CHECK-PPC64LE-NEXT: stb 5, 2(4)
-; CHECK-PPC64LE-NEXT: rldicl 5, 3, 16, 48
-; CHECK-PPC64LE-NEXT: rldicl 3, 3, 8, 56
-; CHECK-PPC64LE-NEXT: stb 5, 1(4)
-; CHECK-PPC64LE-NEXT: stb 3, 0(4)
+; CHECK-PPC64LE-NEXT: slwi [[REG:[0-9]+]], 3, 3
+; CHECK-PPC64LE-NEXT: subf [[REG1:[0-9]+]], 3, [[REG]]
+; CHECK-PPC64LE-NEXT: extsw [[REG2:[0-9]+]], [[REG1]]
+; CHECK-PPC64LE-NEXT: stdbrx [[REG2]], 0, 4
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i64_by_i8_bswap_uses:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: slwi 5, 3, 3
-; CHECK-PPC64-NEXT: subf 3, 3, 5
-; CHECK-PPC64-NEXT: extsw 3, 3
-; CHECK-PPC64-NEXT: rldicl 5, 3, 56, 8
-; CHECK-PPC64-NEXT: rldicl 6, 3, 48, 16
-; CHECK-PPC64-NEXT: stb 5, 6(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 40, 24
-; CHECK-PPC64-NEXT: stb 6, 5(4)
-; CHECK-PPC64-NEXT: rldicl 6, 3, 32, 32
-; CHECK-PPC64-NEXT: stb 5, 4(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 24, 40
-; CHECK-PPC64-NEXT: stb 6, 3(4)
-; CHECK-PPC64-NEXT: stb 3, 7(4)
-; CHECK-PPC64-NEXT: stb 5, 2(4)
-; CHECK-PPC64-NEXT: rldicl 5, 3, 16, 48
-; CHECK-PPC64-NEXT: rldicl 3, 3, 8, 56
-; CHECK-PPC64-NEXT: stb 5, 1(4)
-; CHECK-PPC64-NEXT: stb 3, 0(4)
+; CHECK-PPC64-NEXT: slwi [[REG:[0-9]+]], 3, 3
+; CHECK-PPC64-NEXT: subf [[REG1:[0-9]+]], 3, [[REG]]
+; CHECK-PPC64-NEXT: extsw [[REG2:[0-9]+]], [[REG1]]
+; CHECK-PPC64-NEXT: stdx [[REG2]], 0, 4
; CHECK-PPC64-NEXT: blr
entry:
%mul = mul nsw i32 %t, 7
@@ -356,25 +248,11 @@ entry:
define void @store_i32_by_i8_bswap_volatile(i32 signext %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_bswap_volatile:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 3, 3(4)
-; CHECK-PPC64LE-NEXT: stb 5, 2(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: srwi 3, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, 1(4)
-; CHECK-PPC64LE-NEXT: stb 3, 0(4)
-; CHECK-PPC64LE-NEXT: blr
+; CHECK-PPC64LE-NOT: stwbrx
;
; CHECK-PPC64-LABEL: store_i32_by_i8_bswap_volatile:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 3, 3(4)
-; CHECK-PPC64-NEXT: stb 5, 2(4)
-; CHECK-PPC64-NEXT: srwi 5, 3, 16
-; CHECK-PPC64-NEXT: srwi 3, 3, 24
-; CHECK-PPC64-NEXT: stb 5, 1(4)
-; CHECK-PPC64-NEXT: stb 3, 0(4)
-; CHECK-PPC64-NEXT: blr
+; CHECK-PPC64-NOT: stw
entry:
%conv = trunc i32 %m to i8
%arrayidx = getelementptr inbounds i8, i8* %p, i64 3
@@ -403,29 +281,11 @@ entry:
define void @store_i32_by_i8_bswap_store_in_between(i32 signext %m, i8* %p, i8* %q) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_bswap_store_in_between:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 6, 3, 8
-; CHECK-PPC64LE-NEXT: stb 3, 3(4)
-; CHECK-PPC64LE-NEXT: stb 6, 2(4)
-; CHECK-PPC64LE-NEXT: li 6, 3
-; CHECK-PPC64LE-NEXT: stb 6, 0(5)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: srwi 3, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, 1(4)
-; CHECK-PPC64LE-NEXT: stb 3, 0(4)
-; CHECK-PPC64LE-NEXT: blr
+; CHECK-PPC64LE-NOT: stwbrx
;
; CHECK-PPC64-LABEL: store_i32_by_i8_bswap_store_in_between:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: li 6, 3
-; CHECK-PPC64-NEXT: srwi 7, 3, 8
-; CHECK-PPC64-NEXT: stb 7, 2(4)
-; CHECK-PPC64-NEXT: stb 3, 3(4)
-; CHECK-PPC64-NEXT: stb 6, 0(5)
-; CHECK-PPC64-NEXT: srwi 5, 3, 16
-; CHECK-PPC64-NEXT: srwi 3, 3, 24
-; CHECK-PPC64-NEXT: stb 5, 1(4)
-; CHECK-PPC64-NEXT: stb 3, 0(4)
-; CHECK-PPC64-NEXT: blr
+; CHECK-PPC64-NOT: stw
entry:
%conv = trunc i32 %m to i8
%arrayidx = getelementptr inbounds i8, i8* %p, i64 3
@@ -448,25 +308,11 @@ entry:
define void @store_i32_by_i8_bswap_unrelated_store(i32 signext %m, i8* %p, i8* %q) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_bswap_unrelated_store:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 6, 3, 8
-; CHECK-PPC64LE-NEXT: stb 3, 3(4)
-; CHECK-PPC64LE-NEXT: stb 6, 2(5)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: srwi 3, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, 1(4)
-; CHECK-PPC64LE-NEXT: stb 3, 0(4)
-; CHECK-PPC64LE-NEXT: blr
+; CHECK-PPC64LE-NOT: stwbrx
;
; CHECK-PPC64-LABEL: store_i32_by_i8_bswap_unrelated_store:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 6, 3, 8
-; CHECK-PPC64-NEXT: stb 3, 3(4)
-; CHECK-PPC64-NEXT: stb 6, 2(5)
-; CHECK-PPC64-NEXT: srwi 5, 3, 16
-; CHECK-PPC64-NEXT: srwi 3, 3, 24
-; CHECK-PPC64-NEXT: stb 5, 1(4)
-; CHECK-PPC64-NEXT: stb 3, 0(4)
-; CHECK-PPC64-NEXT: blr
+; CHECK-PPC64-NOT: stw
entry:
%conv = trunc i32 %m to i8
%arrayidx = getelementptr inbounds i8, i8* %p, i64 3
@@ -493,24 +339,13 @@ entry:
define void @store_i32_by_i8_bswap_nonzero_offset(i32 signext %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_bswap_nonzero_offset:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 5, 3(4)
-; CHECK-PPC64LE-NEXT: stb 3, 4(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: srwi 3, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, 2(4)
-; CHECK-PPC64LE-NEXT: stb 3, 1(4)
+; CHECK-PPC64LE-NEXT: addi [[REG1:[0-9]+]], 4, 1
+; CHECK-PPC64LE-NEXT: stwbrx 3, 0, [[REG1]]
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i32_by_i8_bswap_nonzero_offset:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 3, 4(4)
-; CHECK-PPC64-NEXT: stb 5, 3(4)
-; CHECK-PPC64-NEXT: srwi 5, 3, 16
-; CHECK-PPC64-NEXT: srwi 3, 3, 24
-; CHECK-PPC64-NEXT: stb 5, 2(4)
-; CHECK-PPC64-NEXT: stb 3, 1(4)
+; CHECK-PPC64-NEXT: stw 3, 1(4)
; CHECK-PPC64-NEXT: blr
entry:
%0 = lshr i32 %m, 8
@@ -539,24 +374,13 @@ entry:
define void @store_i32_by_i8_neg_offset(i32 signext %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_neg_offset:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 5, -3(4)
-; CHECK-PPC64LE-NEXT: stb 3, -4(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: srwi 3, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, -2(4)
-; CHECK-PPC64LE-NEXT: stb 3, -1(4)
+; CHECK-PPC64LE-NEXT: stw 3, -4(4)
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i32_by_i8_neg_offset:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 3, -4(4)
-; CHECK-PPC64-NEXT: stb 5, -3(4)
-; CHECK-PPC64-NEXT: srwi 5, 3, 16
-; CHECK-PPC64-NEXT: srwi 3, 3, 24
-; CHECK-PPC64-NEXT: stb 5, -2(4)
-; CHECK-PPC64-NEXT: stb 3, -1(4)
+; CHECK-PPC64-NEXT: addi [[REG1:[0-9]+]], 4, -4
+; CHECK-PPC64-NEXT: stwbrx 3, 0, [[REG1]]
; CHECK-PPC64-NEXT: blr
entry:
%0 = lshr i32 %m, 8
@@ -585,24 +409,13 @@ entry:
define void @store_i32_by_i8_bswap_neg_offset(i32 signext %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_bswap_neg_offset:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: stb 5, -3(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, -4(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 5, -2(4)
-; CHECK-PPC64LE-NEXT: stb 3, -1(4)
+; CHECK-PPC64LE-NEXT: addi [[REG1:[0-9]+]], 4, -4
+; CHECK-PPC64LE-NEXT: stwbrx 3, 0, [[REG1]]
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i32_by_i8_bswap_neg_offset:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 5, 3, 16
-; CHECK-PPC64-NEXT: srwi 6, 3, 24
-; CHECK-PPC64-NEXT: stb 5, -3(4)
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 6, -4(4)
-; CHECK-PPC64-NEXT: stb 5, -2(4)
-; CHECK-PPC64-NEXT: stb 3, -1(4)
+; CHECK-PPC64-NEXT: stw 3, -4(4)
; CHECK-PPC64-NEXT: blr
entry:
%0 = lshr i32 %m, 16
@@ -631,28 +444,17 @@ entry:
define void @store_i32_by_i8_bswap_base_index_offset(i32 %m, i32 %i, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_bswap_base_index_offset:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: extsw 4, 4
-; CHECK-PPC64LE-NEXT: srwi 6, 3, 16
-; CHECK-PPC64LE-NEXT: add 4, 5, 4
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 24
-; CHECK-PPC64LE-NEXT: stb 6, -3(4)
-; CHECK-PPC64LE-NEXT: stb 5, -4(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 5, -2(4)
-; CHECK-PPC64LE-NEXT: stb 3, -1(4)
+; CHECK-PPC64LE-NEXT: extsw [[REG1:[0-9]+]], 4
+; CHECK-PPC64LE-NEXT: add [[REG2:[0-9]+]], 5, [[REG1]]
+; CHECK-PPC64LE-NEXT: addi [[REG3:[0-9]+]], [[REG2]], -4
+; CHECK-PPC64LE-NEXT: stwbrx 3, 0, [[REG3]]
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i32_by_i8_bswap_base_index_offset:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: extsw 4, 4
-; CHECK-PPC64-NEXT: srwi 6, 3, 16
-; CHECK-PPC64-NEXT: add 4, 5, 4
-; CHECK-PPC64-NEXT: srwi 5, 3, 24
-; CHECK-PPC64-NEXT: stb 6, -3(4)
-; CHECK-PPC64-NEXT: srwi 6, 3, 8
-; CHECK-PPC64-NEXT: stb 5, -4(4)
-; CHECK-PPC64-NEXT: stb 6, -2(4)
-; CHECK-PPC64-NEXT: stb 3, -1(4)
+; CHECK-PPC64-NEXT: extsw [[REG1:[0-9]+]], 4
+; CHECK-PPC64-NEXT: add [[REG2:[0-9]+]], 5, [[REG1]]
+; CHECK-PPC64-NEXT: stw 3, -4([[REG2]])
; CHECK-PPC64-NEXT: blr
entry:
%0 = lshr i32 %m, 16
@@ -694,28 +496,17 @@ entry:
define void @store_i32_by_i8_bswap_complicated(i32 %m, i32 %i, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i32_by_i8_bswap_complicated:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: extsw 4, 4
-; CHECK-PPC64LE-NEXT: add 4, 5, 4
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 24
-; CHECK-PPC64LE-NEXT: stb 5, 3(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 16
-; CHECK-PPC64LE-NEXT: stb 5, 4(4)
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 5, 5(4)
-; CHECK-PPC64LE-NEXT: stb 3, 6(4)
+; CHECK-PPC64LE-NEXT: extsw [[REG1:[0-9]+]], 4
+; CHECK-PPC64LE-NEXT: add [[REG2:[0-9]+]], 5, [[REG1]]
+; CHECK-PPC64LE-NEXT: addi [[REG3:[0-9]+]], [[REG2]], 3
+; CHECK-PPC64LE-NEXT: stwbrx 3, 0, [[REG3]]
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i32_by_i8_bswap_complicated:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: extsw 4, 4
-; CHECK-PPC64-NEXT: srwi 6, 3, 24
-; CHECK-PPC64-NEXT: add 4, 5, 4
-; CHECK-PPC64-NEXT: srwi 5, 3, 16
-; CHECK-PPC64-NEXT: stb 6, 3(4)
-; CHECK-PPC64-NEXT: stb 5, 4(4)
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 5, 5(4)
-; CHECK-PPC64-NEXT: stb 3, 6(4)
+; CHECK-PPC64-NEXT: extsw [[REG1:[0-9]+]], 4
+; CHECK-PPC64-NEXT: add [[REG2:[0-9]+]], 5, [[REG1]]
+; CHECK-PPC64-NEXT: stw 3, 3([[REG2]])
; CHECK-PPC64-NEXT: blr
entry:
%idx.ext = sext i32 %i to i64
@@ -745,16 +536,12 @@ entry:
define void @store_i16_by_i8_bswap(i16 %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_i16_by_i8_bswap:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: srwi 5, 3, 8
-; CHECK-PPC64LE-NEXT: stb 5, 0(4)
-; CHECK-PPC64LE-NEXT: stb 3, 1(4)
+; CHECK-PPC64LE-NEXT: sthbrx 3, 0, 4
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_i16_by_i8_bswap:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 5, 0(4)
-; CHECK-PPC64-NEXT: stb 3, 1(4)
+; CHECK-PPC64-NEXT: sth 3, 0(4)
; CHECK-PPC64-NEXT: blr
entry:
%0 = lshr i16 %m, 8
@@ -771,16 +558,12 @@ entry:
define void @store_16_by_i8(i16 %m, i8* %p) {
; CHECK-PPC64LE-LABEL: store_16_by_i8:
; CHECK-PPC64LE: # %bb.0: # %entry
-; CHECK-PPC64LE-NEXT: stb 3, 0(4)
-; CHECK-PPC64LE-NEXT: srwi 3, 3, 8
-; CHECK-PPC64LE-NEXT: stb 3, 1(4)
+; CHECK-PPC64LE-NEXT: sth 3, 0(4)
; CHECK-PPC64LE-NEXT: blr
;
; CHECK-PPC64-LABEL: store_16_by_i8:
; CHECK-PPC64: # %bb.0: # %entry
-; CHECK-PPC64-NEXT: srwi 5, 3, 8
-; CHECK-PPC64-NEXT: stb 3, 0(4)
-; CHECK-PPC64-NEXT: stb 5, 1(4)
+; CHECK-PPC64-NEXT: sthbrx 3, 0, 4
; CHECK-PPC64-NEXT: blr
entry:
%conv1 = trunc i16 %m to i8
More information about the llvm-commits
mailing list