Fix sdisel memcpy, memset, memmove lowering:
1. Makes it possible to lower with floating point loads and stores.
2. Avoid unaligned loads / stores unless it's fast.
3. Fix some memcpy lowering logic bug related to when to optimize a
load from constant string into a constant.
4. Adjust x86 memcpy lowering threshold to make it more sane.
5. Fix x86 target hook so it uses vector and floating point memory
ops more effectively.
rdar://7774704
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100090 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index aa283ad..a336e0a 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -5022,18 +5022,6 @@
SDValue Chain = LD->getChain();
SDValue Ptr = LD->getBasePtr();
- // Try to infer better alignment information than the load already has.
- if (OptLevel != CodeGenOpt::None && LD->isUnindexed()) {
- if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
- if (Align > LD->getAlignment())
- return DAG.getExtLoad(LD->getExtensionType(), N->getDebugLoc(),
- LD->getValueType(0),
- Chain, Ptr, LD->getSrcValue(),
- LD->getSrcValueOffset(), LD->getMemoryVT(),
- LD->isVolatile(), LD->isNonTemporal(), Align);
- }
- }
-
// If load is not volatile and there are no uses of the loaded value (and
// the updated indexed value in case of indexed loads), change uses of the
// chain value into uses of the chain input (i.e. delete the dead load).
@@ -5099,6 +5087,18 @@
}
}
+ // Try to infer better alignment information than the load already has.
+ if (OptLevel != CodeGenOpt::None && LD->isUnindexed()) {
+ if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
+ if (Align > LD->getAlignment())
+ return DAG.getExtLoad(LD->getExtensionType(), N->getDebugLoc(),
+ LD->getValueType(0),
+ Chain, Ptr, LD->getSrcValue(),
+ LD->getSrcValueOffset(), LD->getMemoryVT(),
+ LD->isVolatile(), LD->isNonTemporal(), Align);
+ }
+ }
+
if (CombinerAA) {
// Walk up chain skipping non-aliasing memory nodes.
SDValue BetterChain = FindBetterChain(N, Chain);
@@ -5250,17 +5250,6 @@
SDValue Value = ST->getValue();
SDValue Ptr = ST->getBasePtr();
- // Try to infer better alignment information than the store already has.
- if (OptLevel != CodeGenOpt::None && ST->isUnindexed()) {
- if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
- if (Align > ST->getAlignment())
- return DAG.getTruncStore(Chain, N->getDebugLoc(), Value,
- Ptr, ST->getSrcValue(),
- ST->getSrcValueOffset(), ST->getMemoryVT(),
- ST->isVolatile(), ST->isNonTemporal(), Align);
- }
- }
-
// If this is a store of a bit convert, store the input value if the
// resultant store does not need a higher alignment than the original.
if (Value.getOpcode() == ISD::BIT_CONVERT && !ST->isTruncatingStore() &&
@@ -5351,6 +5340,17 @@
}
}
+ // Try to infer better alignment information than the store already has.
+ if (OptLevel != CodeGenOpt::None && ST->isUnindexed()) {
+ if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
+ if (Align > ST->getAlignment())
+ return DAG.getTruncStore(Chain, N->getDebugLoc(), Value,
+ Ptr, ST->getSrcValue(),
+ ST->getSrcValueOffset(), ST->getMemoryVT(),
+ ST->isVolatile(), ST->isNonTemporal(), Align);
+ }
+ }
+
if (CombinerAA) {
// Walk up chain skipping non-aliasing memory nodes.
SDValue BetterChain = FindBetterChain(N, Chain);