[llvm] a7dafea - [SDAG] Allow folding stack slots into sincos/frexp in more cases (#118117)

via llvm-commits llvm-commits at lists.llvm.org
Tue Dec 17 02:54:21 PST 2024


Author: Benjamin Maxwell
Date: 2024-12-17T10:54:17Z
New Revision: a7dafea384a519342b2fbe210ed101c1e67f3be7

URL: https://github.com/llvm/llvm-project/commit/a7dafea384a519342b2fbe210ed101c1e67f3be7
DIFF: https://github.com/llvm/llvm-project/commit/a7dafea384a519342b2fbe210ed101c1e67f3be7.diff

LOG: [SDAG] Allow folding stack slots into sincos/frexp in more cases (#118117)

This adds a new helper `canFoldStoreIntoLibCallOutputPointers()` to
check that it is safe to fold a store into a node that will expand to a
library call that takes output pointers. This requires checking for two
(independent) properties:

1. The store is not within a CALLSEQ_START..CALLSEQ_END pair
* If it is, the expansion would lead to nested call sequences (which is
invalid)
2. The node does not appear as a predecessor to the store
* If it does, attempting to merge the store into the call would result
in a cycle in the DAG

These two properties are checked as part of the same traversal in
`canFoldStoreIntoLibCallOutputPointers()`

Added: 
    llvm/test/CodeGen/X86/llvm.sincos.ll

Modified: 
    llvm/include/llvm/CodeGen/SelectionDAG.h
    llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
    llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
    llvm/test/CodeGen/AArch64/sincos-stack-slots.ll
    llvm/test/CodeGen/PowerPC/f128-arith.ll
    llvm/test/CodeGen/RISCV/llvm.frexp.ll
    llvm/test/CodeGen/X86/llvm.frexp.ll
    llvm/test/CodeGen/X86/sincos-stack-args.ll

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/CodeGen/SelectionDAG.h b/llvm/include/llvm/CodeGen/SelectionDAG.h
index e97e01839f73b4..d63e1c559122b6 100644
--- a/llvm/include/llvm/CodeGen/SelectionDAG.h
+++ b/llvm/include/llvm/CodeGen/SelectionDAG.h
@@ -455,6 +455,9 @@ class SelectionDAG {
   // Maximum depth for recursive analysis such as computeKnownBits, etc.
   static constexpr unsigned MaxRecursionDepth = 6;
 
+  // Returns the maximum steps for SDNode->hasPredecessor() like searches.
+  static unsigned getHasPredecessorMaxSteps();
+
   explicit SelectionDAG(const TargetMachine &TM, CodeGenOptLevel);
   SelectionDAG(const SelectionDAG &) = delete;
   SelectionDAG &operator=(const SelectionDAG &) = delete;

diff  --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 4fe5e7416c7871..10fc8eecaff907 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -153,12 +153,6 @@ static cl::opt<bool> EnableVectorFCopySignExtendRound(
     "combiner-vector-fcopysign-extend-round", cl::Hidden, cl::init(false),
     cl::desc(
         "Enable merging extends and rounds into FCOPYSIGN on vector types"));
-
-static cl::opt<unsigned int>
-    MaxSteps("has-predecessor-max-steps", cl::Hidden, cl::init(8192),
-             cl::desc("DAG combiner limit number of steps when searching DAG "
-                      "for predecessor nodes"));
-
 namespace {
 
   class DAGCombiner {
@@ -18929,6 +18923,7 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) {
   // can be folded with this one. We should do this to avoid having to keep
   // a copy of the original base pointer.
   SmallVector<SDNode *, 16> OtherUses;
+  unsigned MaxSteps = SelectionDAG::getHasPredecessorMaxSteps();
   if (isa<ConstantSDNode>(Offset))
     for (SDNode::use_iterator UI = BasePtr->use_begin(),
                               UE = BasePtr->use_end();
@@ -19093,6 +19088,7 @@ static bool shouldCombineToPostInc(SDNode *N, SDValue Ptr, SDNode *PtrUse,
     return false;
 
   SmallPtrSet<const SDNode *, 32> Visited;
+  unsigned MaxSteps = SelectionDAG::getHasPredecessorMaxSteps();
   for (SDNode *Use : BasePtr->uses()) {
     if (Use == Ptr.getNode())
       continue;
@@ -19139,6 +19135,7 @@ static SDNode *getPostIndexedLoadStoreOp(SDNode *N, bool &IsLoad,
   // 2) Op must be independent of N, i.e. Op is neither a predecessor
   //    nor a successor of N. Otherwise, if Op is folded that would
   //    create a cycle.
+  unsigned MaxSteps = SelectionDAG::getHasPredecessorMaxSteps();
   for (SDNode *Op : Ptr->uses()) {
     // Check for #1.
     if (!shouldCombineToPostInc(N, Ptr, Op, BasePtr, Offset, AM, DAG, TLI))

diff  --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 34214550f3a12b..0fb5c4d5c4cb9b 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -111,10 +111,17 @@ static cl::opt<int> MaxLdStGlue("ldstmemcpy-glue-max",
        cl::desc("Number limit for gluing ld/st of memcpy."),
        cl::Hidden, cl::init(0));
 
+static cl::opt<unsigned>
+    MaxSteps("has-predecessor-max-steps", cl::Hidden, cl::init(8192),
+             cl::desc("DAG combiner limit number of steps when searching DAG "
+                      "for predecessor nodes"));
+
 static void NewSDValueDbgMsg(SDValue V, StringRef Msg, SelectionDAG *G) {
   LLVM_DEBUG(dbgs() << Msg; V.getNode()->dump(G););
 }
 
+unsigned SelectionDAG::getHasPredecessorMaxSteps() { return MaxSteps; }
+
 //===----------------------------------------------------------------------===//
 //                              ConstantFPSDNode Class
 //===----------------------------------------------------------------------===//
@@ -2474,6 +2481,51 @@ SDValue SelectionDAG::getPartialReduceAdd(SDLoc DL, EVT ReducedTy, SDValue Op1,
   return Subvectors[0];
 }
 
+/// Given a store node \p StoreNode, return true if it is safe to fold that node
+/// into \p FPNode, which expands to a library call with output pointers.
+static bool canFoldStoreIntoLibCallOutputPointers(StoreSDNode *StoreNode,
+                                                  SDNode *FPNode) {
+  SmallVector<const SDNode *, 8> Worklist;
+  SmallVector<const SDNode *, 8> DeferredNodes;
+  SmallPtrSet<const SDNode *, 16> Visited;
+
+  // Skip FPNode use by StoreNode (that's the use we want to fold into FPNode).
+  for (SDValue Op : StoreNode->ops())
+    if (Op.getNode() != FPNode)
+      Worklist.push_back(Op.getNode());
+
+  unsigned MaxSteps = SelectionDAG::getHasPredecessorMaxSteps();
+  while (!Worklist.empty()) {
+    const SDNode *Node = Worklist.pop_back_val();
+    auto [_, Inserted] = Visited.insert(Node);
+    if (!Inserted)
+      continue;
+
+    if (MaxSteps > 0 && Visited.size() >= MaxSteps)
+      return false;
+
+    // Reached the FPNode (would result in a cycle).
+    // OR Reached CALLSEQ_START (would result in nested call sequences).
+    if (Node == FPNode || Node->getOpcode() == ISD::CALLSEQ_START)
+      return false;
+
+    if (Node->getOpcode() == ISD::CALLSEQ_END) {
+      // Defer looking into call sequences (so we can check we're outside one).
+      // We still need to look through these for the predecessor check.
+      DeferredNodes.push_back(Node);
+      continue;
+    }
+
+    for (SDValue Op : Node->ops())
+      Worklist.push_back(Op.getNode());
+  }
+
+  // True if we're outside a call sequence and don't have the FPNode as a
+  // predecessor. No cycles or nested call sequences possible.
+  return !SDNode::hasPredecessorHelper(FPNode, Visited, DeferredNodes,
+                                       MaxSteps);
+}
+
 bool SelectionDAG::expandMultipleResultFPLibCall(
     RTLIB::Libcall LC, SDNode *Node, SmallVectorImpl<SDValue> &Results,
     std::optional<unsigned> CallRetResNo) {
@@ -2502,11 +2554,7 @@ bool SelectionDAG::expandMultipleResultFPLibCall(
 
   // Find users of the node that store the results (and share input chains). The
   // destination pointers can be used instead of creating stack allocations.
-  // FIXME: This should allow stores with the same chains (not just the entry
-  // chain), but there's a risk the store is within a (CALLSEQ_START,
-  // CALLSEQ_END) pair, which after this expansion will lead to nested call
-  // sequences.
-  SDValue InChain = getEntryNode();
+  SDValue StoresInChain;
   SmallVector<StoreSDNode *, 2> ResultStores(NumResults);
   for (SDNode *User : Node->uses()) {
     if (!ISD::isNormalStore(User))
@@ -2514,14 +2562,27 @@ bool SelectionDAG::expandMultipleResultFPLibCall(
     auto *ST = cast<StoreSDNode>(User);
     SDValue StoreValue = ST->getValue();
     unsigned ResNo = StoreValue.getResNo();
+    // Ensure the store corresponds to an output pointer.
+    if (CallRetResNo == ResNo)
+      continue;
+    // Ensure the store to the default address space and not atomic or volatile.
+    if (!ST->isSimple() || ST->getAddressSpace() != 0)
+      continue;
+    // Ensure all store chains are the same (so they don't alias).
+    if (StoresInChain && ST->getChain() != StoresInChain)
+      continue;
+    // Ensure the store is properly aligned.
     Type *StoreType = StoreValue.getValueType().getTypeForEVT(Ctx);
-    if (CallRetResNo == ResNo || !ST->isSimple() ||
-        ST->getAddressSpace() != 0 ||
-        ST->getAlign() <
-            getDataLayout().getABITypeAlign(StoreType->getScalarType()) ||
-        ST->getChain() != InChain)
+    if (ST->getAlign() <
+        getDataLayout().getABITypeAlign(StoreType->getScalarType()))
+      continue;
+    // Avoid:
+    //  1. Creating cyclic dependencies.
+    //  2. Expanding the node to a call within a call sequence.
+    if (!canFoldStoreIntoLibCallOutputPointers(ST, Node))
       continue;
     ResultStores[ResNo] = ST;
+    StoresInChain = ST->getChain();
   }
 
   TargetLowering::ArgListTy Args;
@@ -2563,6 +2624,7 @@ bool SelectionDAG::expandMultipleResultFPLibCall(
   Type *RetType = CallRetResNo.has_value()
                       ? Node->getValueType(*CallRetResNo).getTypeForEVT(Ctx)
                       : Type::getVoidTy(Ctx);
+  SDValue InChain = StoresInChain ? StoresInChain : getEntryNode();
   SDValue Callee = getExternalSymbol(VD ? VD->getVectorFnName().data() : LCName,
                                      TLI->getPointerTy(getDataLayout()));
   TargetLowering::CallLoweringInfo CLI(*this);

diff  --git a/llvm/test/CodeGen/AArch64/sincos-stack-slots.ll b/llvm/test/CodeGen/AArch64/sincos-stack-slots.ll
index 8ef8b5d13b62d4..c5fef61c96af3a 100644
--- a/llvm/test/CodeGen/AArch64/sincos-stack-slots.ll
+++ b/llvm/test/CodeGen/AArch64/sincos-stack-slots.ll
@@ -253,3 +253,37 @@ entry:
   store double %cos, ptr %out_cos, align 4
   ret void
 }
+
+declare void @foo(ptr, ptr)
+
+define void @can_fold_with_call_in_chain(float %x, ptr noalias %a, ptr noalias %b) {
+; CHECK-LABEL: can_fold_with_call_in_chain:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    str d8, [sp, #-32]! // 8-byte Folded Spill
+; CHECK-NEXT:    str x30, [sp, #8] // 8-byte Folded Spill
+; CHECK-NEXT:    stp x20, x19, [sp, #16] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    .cfi_offset w19, -8
+; CHECK-NEXT:    .cfi_offset w20, -16
+; CHECK-NEXT:    .cfi_offset w30, -24
+; CHECK-NEXT:    .cfi_offset b8, -32
+; CHECK-NEXT:    mov x19, x1
+; CHECK-NEXT:    mov x20, x0
+; CHECK-NEXT:    fmov s8, s0
+; CHECK-NEXT:    bl foo
+; CHECK-NEXT:    fmov s0, s8
+; CHECK-NEXT:    mov x0, x20
+; CHECK-NEXT:    mov x1, x19
+; CHECK-NEXT:    bl sincosf
+; CHECK-NEXT:    ldp x20, x19, [sp, #16] // 16-byte Folded Reload
+; CHECK-NEXT:    ldr x30, [sp, #8] // 8-byte Folded Reload
+; CHECK-NEXT:    ldr d8, [sp], #32 // 8-byte Folded Reload
+; CHECK-NEXT:    ret
+entry:
+  %sin = tail call float @llvm.sin.f32(float %x)
+  %cos = tail call float @llvm.cos.f32(float %x)
+  call void @foo(ptr %a, ptr %b)
+  store float %sin, ptr %a, align 4
+  store float %cos, ptr %b, align 4
+  ret void
+}

diff  --git a/llvm/test/CodeGen/PowerPC/f128-arith.ll b/llvm/test/CodeGen/PowerPC/f128-arith.ll
index 35e5d61947ead7..decc4a38f7ccd4 100644
--- a/llvm/test/CodeGen/PowerPC/f128-arith.ll
+++ b/llvm/test/CodeGen/PowerPC/f128-arith.ll
@@ -1365,45 +1365,33 @@ define dso_local fp128 @qpFREXP(ptr %a, ptr %b) {
 ; CHECK-LABEL: qpFREXP:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    mflr r0
-; CHECK-NEXT:    .cfi_def_cfa_offset 64
+; CHECK-NEXT:    stdu r1, -32(r1)
+; CHECK-NEXT:    std r0, 48(r1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    .cfi_offset r30, -16
-; CHECK-NEXT:    std r30, -16(r1) # 8-byte Folded Spill
-; CHECK-NEXT:    stdu r1, -64(r1)
-; CHECK-NEXT:    std r0, 80(r1)
-; CHECK-NEXT:    addi r5, r1, 44
-; CHECK-NEXT:    mr r30, r4
 ; CHECK-NEXT:    lxv v2, 0(r3)
+; CHECK-NEXT:    mr r5, r4
 ; CHECK-NEXT:    bl frexpf128
 ; CHECK-NEXT:    nop
-; CHECK-NEXT:    lwz r3, 44(r1)
-; CHECK-NEXT:    stw r3, 0(r30)
-; CHECK-NEXT:    addi r1, r1, 64
+; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
-; CHECK-NEXT:    ld r30, -16(r1) # 8-byte Folded Reload
 ; CHECK-NEXT:    mtlr r0
 ; CHECK-NEXT:    blr
 ;
 ; CHECK-P8-LABEL: qpFREXP:
 ; CHECK-P8:       # %bb.0: # %entry
 ; CHECK-P8-NEXT:    mflr r0
-; CHECK-P8-NEXT:    .cfi_def_cfa_offset 64
+; CHECK-P8-NEXT:    stdu r1, -32(r1)
+; CHECK-P8-NEXT:    std r0, 48(r1)
+; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    .cfi_offset r30, -16
-; CHECK-P8-NEXT:    std r30, -16(r1) # 8-byte Folded Spill
-; CHECK-P8-NEXT:    stdu r1, -64(r1)
-; CHECK-P8-NEXT:    std r0, 80(r1)
-; CHECK-P8-NEXT:    addi r5, r1, 44
-; CHECK-P8-NEXT:    mr r30, r4
 ; CHECK-P8-NEXT:    lxvd2x vs0, 0, r3
+; CHECK-P8-NEXT:    mr r5, r4
 ; CHECK-P8-NEXT:    xxswapd v2, vs0
 ; CHECK-P8-NEXT:    bl frexpf128
 ; CHECK-P8-NEXT:    nop
-; CHECK-P8-NEXT:    lwz r3, 44(r1)
-; CHECK-P8-NEXT:    stw r3, 0(r30)
-; CHECK-P8-NEXT:    addi r1, r1, 64
+; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
-; CHECK-P8-NEXT:    ld r30, -16(r1) # 8-byte Folded Reload
 ; CHECK-P8-NEXT:    mtlr r0
 ; CHECK-P8-NEXT:    blr
 entry:

diff  --git a/llvm/test/CodeGen/RISCV/llvm.frexp.ll b/llvm/test/CodeGen/RISCV/llvm.frexp.ll
index 74dec76a02e892..4a77b4d32cdda6 100644
--- a/llvm/test/CodeGen/RISCV/llvm.frexp.ll
+++ b/llvm/test/CodeGen/RISCV/llvm.frexp.ll
@@ -543,50 +543,42 @@ define i32 @test_frexp_f32_i32_only_use_exp(float %a) nounwind {
 define { <4 x float>, <4 x i32> } @test_frexp_v4f32_v4i32(<4 x float> %a) nounwind {
 ; RV32IFD-LABEL: test_frexp_v4f32_v4i32:
 ; RV32IFD:       # %bb.0:
-; RV32IFD-NEXT:    addi sp, sp, -64
-; RV32IFD-NEXT:    sw ra, 60(sp) # 4-byte Folded Spill
-; RV32IFD-NEXT:    sw s0, 56(sp) # 4-byte Folded Spill
-; RV32IFD-NEXT:    fsd fs0, 48(sp) # 8-byte Folded Spill
-; RV32IFD-NEXT:    fsd fs1, 40(sp) # 8-byte Folded Spill
-; RV32IFD-NEXT:    fsd fs2, 32(sp) # 8-byte Folded Spill
-; RV32IFD-NEXT:    fsd fs3, 24(sp) # 8-byte Folded Spill
+; RV32IFD-NEXT:    addi sp, sp, -48
+; RV32IFD-NEXT:    sw ra, 44(sp) # 4-byte Folded Spill
+; RV32IFD-NEXT:    sw s0, 40(sp) # 4-byte Folded Spill
+; RV32IFD-NEXT:    fsd fs0, 32(sp) # 8-byte Folded Spill
+; RV32IFD-NEXT:    fsd fs1, 24(sp) # 8-byte Folded Spill
+; RV32IFD-NEXT:    fsd fs2, 16(sp) # 8-byte Folded Spill
+; RV32IFD-NEXT:    fsd fs3, 8(sp) # 8-byte Folded Spill
 ; RV32IFD-NEXT:    fmv.s fs0, fa3
 ; RV32IFD-NEXT:    fmv.s fs1, fa2
 ; RV32IFD-NEXT:    fmv.s fs2, fa1
 ; RV32IFD-NEXT:    mv s0, a0
-; RV32IFD-NEXT:    addi a0, sp, 8
+; RV32IFD-NEXT:    addi a0, a0, 16
 ; RV32IFD-NEXT:    call frexpf
 ; RV32IFD-NEXT:    fmv.s fs3, fa0
-; RV32IFD-NEXT:    addi a0, sp, 12
+; RV32IFD-NEXT:    addi a0, s0, 20
 ; RV32IFD-NEXT:    fmv.s fa0, fs2
 ; RV32IFD-NEXT:    call frexpf
 ; RV32IFD-NEXT:    fmv.s fs2, fa0
-; RV32IFD-NEXT:    addi a0, sp, 16
+; RV32IFD-NEXT:    addi a0, s0, 24
 ; RV32IFD-NEXT:    fmv.s fa0, fs1
 ; RV32IFD-NEXT:    call frexpf
 ; RV32IFD-NEXT:    fmv.s fs1, fa0
-; RV32IFD-NEXT:    addi a0, sp, 20
+; RV32IFD-NEXT:    addi a0, s0, 28
 ; RV32IFD-NEXT:    fmv.s fa0, fs0
 ; RV32IFD-NEXT:    call frexpf
-; RV32IFD-NEXT:    lw a0, 8(sp)
-; RV32IFD-NEXT:    lw a1, 12(sp)
-; RV32IFD-NEXT:    lw a2, 16(sp)
-; RV32IFD-NEXT:    lw a3, 20(sp)
-; RV32IFD-NEXT:    sw a0, 16(s0)
-; RV32IFD-NEXT:    sw a1, 20(s0)
-; RV32IFD-NEXT:    sw a2, 24(s0)
-; RV32IFD-NEXT:    sw a3, 28(s0)
 ; RV32IFD-NEXT:    fsw fs3, 0(s0)
 ; RV32IFD-NEXT:    fsw fs2, 4(s0)
 ; RV32IFD-NEXT:    fsw fs1, 8(s0)
 ; RV32IFD-NEXT:    fsw fa0, 12(s0)
-; RV32IFD-NEXT:    lw ra, 60(sp) # 4-byte Folded Reload
-; RV32IFD-NEXT:    lw s0, 56(sp) # 4-byte Folded Reload
-; RV32IFD-NEXT:    fld fs0, 48(sp) # 8-byte Folded Reload
-; RV32IFD-NEXT:    fld fs1, 40(sp) # 8-byte Folded Reload
-; RV32IFD-NEXT:    fld fs2, 32(sp) # 8-byte Folded Reload
-; RV32IFD-NEXT:    fld fs3, 24(sp) # 8-byte Folded Reload
-; RV32IFD-NEXT:    addi sp, sp, 64
+; RV32IFD-NEXT:    lw ra, 44(sp) # 4-byte Folded Reload
+; RV32IFD-NEXT:    lw s0, 40(sp) # 4-byte Folded Reload
+; RV32IFD-NEXT:    fld fs0, 32(sp) # 8-byte Folded Reload
+; RV32IFD-NEXT:    fld fs1, 24(sp) # 8-byte Folded Reload
+; RV32IFD-NEXT:    fld fs2, 16(sp) # 8-byte Folded Reload
+; RV32IFD-NEXT:    fld fs3, 8(sp) # 8-byte Folded Reload
+; RV32IFD-NEXT:    addi sp, sp, 48
 ; RV32IFD-NEXT:    ret
 ;
 ; RV64IFD-LABEL: test_frexp_v4f32_v4i32:
@@ -639,52 +631,44 @@ define { <4 x float>, <4 x i32> } @test_frexp_v4f32_v4i32(<4 x float> %a) nounwi
 ;
 ; RV32IZFINXZDINX-LABEL: test_frexp_v4f32_v4i32:
 ; RV32IZFINXZDINX:       # %bb.0:
-; RV32IZFINXZDINX-NEXT:    addi sp, sp, -48
-; RV32IZFINXZDINX-NEXT:    sw ra, 44(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s0, 40(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s1, 36(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s2, 32(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s3, 28(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s4, 24(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    addi sp, sp, -32
+; RV32IZFINXZDINX-NEXT:    sw ra, 28(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s0, 24(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s1, 20(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s2, 16(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s3, 12(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s4, 8(sp) # 4-byte Folded Spill
 ; RV32IZFINXZDINX-NEXT:    mv s0, a4
 ; RV32IZFINXZDINX-NEXT:    mv s1, a3
 ; RV32IZFINXZDINX-NEXT:    mv s2, a2
 ; RV32IZFINXZDINX-NEXT:    mv a2, a1
 ; RV32IZFINXZDINX-NEXT:    mv s3, a0
-; RV32IZFINXZDINX-NEXT:    addi a1, sp, 8
+; RV32IZFINXZDINX-NEXT:    addi a1, a0, 16
 ; RV32IZFINXZDINX-NEXT:    mv a0, a2
 ; RV32IZFINXZDINX-NEXT:    call frexpf
 ; RV32IZFINXZDINX-NEXT:    mv s4, a0
-; RV32IZFINXZDINX-NEXT:    addi a1, sp, 12
+; RV32IZFINXZDINX-NEXT:    addi a1, s3, 20
 ; RV32IZFINXZDINX-NEXT:    mv a0, s2
 ; RV32IZFINXZDINX-NEXT:    call frexpf
 ; RV32IZFINXZDINX-NEXT:    mv s2, a0
-; RV32IZFINXZDINX-NEXT:    addi a1, sp, 16
+; RV32IZFINXZDINX-NEXT:    addi a1, s3, 24
 ; RV32IZFINXZDINX-NEXT:    mv a0, s1
 ; RV32IZFINXZDINX-NEXT:    call frexpf
 ; RV32IZFINXZDINX-NEXT:    mv s1, a0
-; RV32IZFINXZDINX-NEXT:    addi a1, sp, 20
+; RV32IZFINXZDINX-NEXT:    addi a1, s3, 28
 ; RV32IZFINXZDINX-NEXT:    mv a0, s0
 ; RV32IZFINXZDINX-NEXT:    call frexpf
-; RV32IZFINXZDINX-NEXT:    lw a1, 8(sp)
-; RV32IZFINXZDINX-NEXT:    lw a2, 12(sp)
-; RV32IZFINXZDINX-NEXT:    lw a3, 16(sp)
-; RV32IZFINXZDINX-NEXT:    lw a4, 20(sp)
-; RV32IZFINXZDINX-NEXT:    sw a1, 16(s3)
-; RV32IZFINXZDINX-NEXT:    sw a2, 20(s3)
-; RV32IZFINXZDINX-NEXT:    sw a3, 24(s3)
-; RV32IZFINXZDINX-NEXT:    sw a4, 28(s3)
 ; RV32IZFINXZDINX-NEXT:    sw s4, 0(s3)
 ; RV32IZFINXZDINX-NEXT:    sw s2, 4(s3)
 ; RV32IZFINXZDINX-NEXT:    sw s1, 8(s3)
 ; RV32IZFINXZDINX-NEXT:    sw a0, 12(s3)
-; RV32IZFINXZDINX-NEXT:    lw ra, 44(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s0, 40(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s1, 36(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s2, 32(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s3, 28(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s4, 24(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    addi sp, sp, 48
+; RV32IZFINXZDINX-NEXT:    lw ra, 28(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s4, 8(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    addi sp, sp, 32
 ; RV32IZFINXZDINX-NEXT:    ret
 ;
 ; RV64IZFINXZDINX-LABEL: test_frexp_v4f32_v4i32:
@@ -1096,41 +1080,34 @@ define <4 x float> @test_frexp_v4f32_v4i32_only_use_fract(<4 x float> %a) nounwi
 define <4 x i32> @test_frexp_v4f32_v4i32_only_use_exp(<4 x float> %a) nounwind {
 ; RV32IFD-LABEL: test_frexp_v4f32_v4i32_only_use_exp:
 ; RV32IFD:       # %bb.0:
-; RV32IFD-NEXT:    addi sp, sp, -48
-; RV32IFD-NEXT:    sw ra, 44(sp) # 4-byte Folded Spill
-; RV32IFD-NEXT:    sw s0, 40(sp) # 4-byte Folded Spill
-; RV32IFD-NEXT:    fsd fs0, 32(sp) # 8-byte Folded Spill
-; RV32IFD-NEXT:    fsd fs1, 24(sp) # 8-byte Folded Spill
-; RV32IFD-NEXT:    fsd fs2, 16(sp) # 8-byte Folded Spill
-; RV32IFD-NEXT:    fmv.s fs0, fa3
-; RV32IFD-NEXT:    fmv.s fs1, fa2
-; RV32IFD-NEXT:    fmv.s fs2, fa1
+; RV32IFD-NEXT:    addi sp, sp, -32
+; RV32IFD-NEXT:    sw ra, 28(sp) # 4-byte Folded Spill
+; RV32IFD-NEXT:    sw s0, 24(sp) # 4-byte Folded Spill
+; RV32IFD-NEXT:    fsd fs0, 16(sp) # 8-byte Folded Spill
+; RV32IFD-NEXT:    fsd fs1, 8(sp) # 8-byte Folded Spill
+; RV32IFD-NEXT:    fsd fs2, 0(sp) # 8-byte Folded Spill
+; RV32IFD-NEXT:    fmv.s fs0, fa2
+; RV32IFD-NEXT:    fmv.s fs1, fa1
+; RV32IFD-NEXT:    fmv.s fs2, fa0
 ; RV32IFD-NEXT:    mv s0, a0
-; RV32IFD-NEXT:    mv a0, sp
+; RV32IFD-NEXT:    addi a0, a0, 12
+; RV32IFD-NEXT:    fmv.s fa0, fa3
 ; RV32IFD-NEXT:    call frexpf
-; RV32IFD-NEXT:    addi a0, sp, 4
-; RV32IFD-NEXT:    fmv.s fa0, fs2
+; RV32IFD-NEXT:    addi a0, s0, 8
+; RV32IFD-NEXT:    fmv.s fa0, fs0
 ; RV32IFD-NEXT:    call frexpf
-; RV32IFD-NEXT:    addi a0, sp, 8
+; RV32IFD-NEXT:    addi a0, s0, 4
 ; RV32IFD-NEXT:    fmv.s fa0, fs1
 ; RV32IFD-NEXT:    call frexpf
-; RV32IFD-NEXT:    addi a0, sp, 12
-; RV32IFD-NEXT:    fmv.s fa0, fs0
+; RV32IFD-NEXT:    fmv.s fa0, fs2
+; RV32IFD-NEXT:    mv a0, s0
 ; RV32IFD-NEXT:    call frexpf
-; RV32IFD-NEXT:    lw a0, 0(sp)
-; RV32IFD-NEXT:    lw a1, 4(sp)
-; RV32IFD-NEXT:    lw a2, 8(sp)
-; RV32IFD-NEXT:    lw a3, 12(sp)
-; RV32IFD-NEXT:    sw a0, 0(s0)
-; RV32IFD-NEXT:    sw a1, 4(s0)
-; RV32IFD-NEXT:    sw a2, 8(s0)
-; RV32IFD-NEXT:    sw a3, 12(s0)
-; RV32IFD-NEXT:    lw ra, 44(sp) # 4-byte Folded Reload
-; RV32IFD-NEXT:    lw s0, 40(sp) # 4-byte Folded Reload
-; RV32IFD-NEXT:    fld fs0, 32(sp) # 8-byte Folded Reload
-; RV32IFD-NEXT:    fld fs1, 24(sp) # 8-byte Folded Reload
-; RV32IFD-NEXT:    fld fs2, 16(sp) # 8-byte Folded Reload
-; RV32IFD-NEXT:    addi sp, sp, 48
+; RV32IFD-NEXT:    lw ra, 28(sp) # 4-byte Folded Reload
+; RV32IFD-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload
+; RV32IFD-NEXT:    fld fs0, 16(sp) # 8-byte Folded Reload
+; RV32IFD-NEXT:    fld fs1, 8(sp) # 8-byte Folded Reload
+; RV32IFD-NEXT:    fld fs2, 0(sp) # 8-byte Folded Reload
+; RV32IFD-NEXT:    addi sp, sp, 32
 ; RV32IFD-NEXT:    ret
 ;
 ; RV64IFD-LABEL: test_frexp_v4f32_v4i32_only_use_exp:
@@ -1174,43 +1151,34 @@ define <4 x i32> @test_frexp_v4f32_v4i32_only_use_exp(<4 x float> %a) nounwind {
 ;
 ; RV32IZFINXZDINX-LABEL: test_frexp_v4f32_v4i32_only_use_exp:
 ; RV32IZFINXZDINX:       # %bb.0:
-; RV32IZFINXZDINX-NEXT:    addi sp, sp, -48
-; RV32IZFINXZDINX-NEXT:    sw ra, 44(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s0, 40(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s1, 36(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s2, 32(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    sw s3, 28(sp) # 4-byte Folded Spill
-; RV32IZFINXZDINX-NEXT:    mv s0, a4
-; RV32IZFINXZDINX-NEXT:    mv s1, a3
-; RV32IZFINXZDINX-NEXT:    mv s2, a2
-; RV32IZFINXZDINX-NEXT:    mv a2, a1
+; RV32IZFINXZDINX-NEXT:    addi sp, sp, -32
+; RV32IZFINXZDINX-NEXT:    sw ra, 28(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s0, 24(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s1, 20(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s2, 16(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    sw s3, 12(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT:    mv s0, a3
+; RV32IZFINXZDINX-NEXT:    mv s1, a2
+; RV32IZFINXZDINX-NEXT:    mv s2, a1
 ; RV32IZFINXZDINX-NEXT:    mv s3, a0
-; RV32IZFINXZDINX-NEXT:    addi a1, sp, 12
-; RV32IZFINXZDINX-NEXT:    mv a0, a2
+; RV32IZFINXZDINX-NEXT:    addi a1, a0, 12
+; RV32IZFINXZDINX-NEXT:    mv a0, a4
 ; RV32IZFINXZDINX-NEXT:    call frexpf
-; RV32IZFINXZDINX-NEXT:    addi a1, sp, 16
-; RV32IZFINXZDINX-NEXT:    mv a0, s2
+; RV32IZFINXZDINX-NEXT:    addi a1, s3, 8
+; RV32IZFINXZDINX-NEXT:    mv a0, s0
 ; RV32IZFINXZDINX-NEXT:    call frexpf
-; RV32IZFINXZDINX-NEXT:    addi a1, sp, 20
+; RV32IZFINXZDINX-NEXT:    addi a1, s3, 4
 ; RV32IZFINXZDINX-NEXT:    mv a0, s1
 ; RV32IZFINXZDINX-NEXT:    call frexpf
-; RV32IZFINXZDINX-NEXT:    addi a1, sp, 24
-; RV32IZFINXZDINX-NEXT:    mv a0, s0
+; RV32IZFINXZDINX-NEXT:    mv a0, s2
+; RV32IZFINXZDINX-NEXT:    mv a1, s3
 ; RV32IZFINXZDINX-NEXT:    call frexpf
-; RV32IZFINXZDINX-NEXT:    lw a0, 12(sp)
-; RV32IZFINXZDINX-NEXT:    lw a1, 16(sp)
-; RV32IZFINXZDINX-NEXT:    lw a2, 20(sp)
-; RV32IZFINXZDINX-NEXT:    lw a3, 24(sp)
-; RV32IZFINXZDINX-NEXT:    sw a0, 0(s3)
-; RV32IZFINXZDINX-NEXT:    sw a1, 4(s3)
-; RV32IZFINXZDINX-NEXT:    sw a2, 8(s3)
-; RV32IZFINXZDINX-NEXT:    sw a3, 12(s3)
-; RV32IZFINXZDINX-NEXT:    lw ra, 44(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s0, 40(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s1, 36(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s2, 32(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    lw s3, 28(sp) # 4-byte Folded Reload
-; RV32IZFINXZDINX-NEXT:    addi sp, sp, 48
+; RV32IZFINXZDINX-NEXT:    lw ra, 28(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT:    addi sp, sp, 32
 ; RV32IZFINXZDINX-NEXT:    ret
 ;
 ; RV64IZFINXZDINX-LABEL: test_frexp_v4f32_v4i32_only_use_exp:

diff  --git a/llvm/test/CodeGen/X86/llvm.frexp.ll b/llvm/test/CodeGen/X86/llvm.frexp.ll
index cd560ad627de4c..96de34519556d0 100644
--- a/llvm/test/CodeGen/X86/llvm.frexp.ll
+++ b/llvm/test/CodeGen/X86/llvm.frexp.ll
@@ -325,28 +325,27 @@ define { <4 x float>, <4 x i32> } @test_frexp_v4f32_v4i32(<4 x float> %a) {
 ;
 ; WIN32-LABEL: test_frexp_v4f32_v4i32:
 ; WIN32:       # %bb.0:
-; WIN32-NEXT:    pushl %edi
 ; WIN32-NEXT:    pushl %esi
-; WIN32-NEXT:    subl $60, %esp
+; WIN32-NEXT:    subl $44, %esp
 ; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; WIN32-NEXT:    leal 24(%esi), %eax
 ; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    fstpl (%esp)
 ; WIN32-NEXT:    calll _frexp
 ; WIN32-NEXT:    fstpl {{[-0-9]+}}(%e{{[sb]}}p) # 8-byte Folded Spill
-; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; WIN32-NEXT:    leal 20(%esi), %eax
 ; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    fstpl (%esp)
 ; WIN32-NEXT:    calll _frexp
 ; WIN32-NEXT:    fstpl {{[-0-9]+}}(%e{{[sb]}}p) # 8-byte Folded Spill
-; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; WIN32-NEXT:    leal 16(%esi), %eax
 ; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    fstpl (%esp)
 ; WIN32-NEXT:    calll _frexp
-; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; WIN32-NEXT:    leal 28(%esi), %eax
 ; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    fstpl (%esp)
@@ -361,22 +360,13 @@ define { <4 x float>, <4 x i32> } @test_frexp_v4f32_v4i32(<4 x float> %a) {
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
-; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; WIN32-NEXT:    movl %edi, 28(%esi)
-; WIN32-NEXT:    movl %edx, 24(%esi)
-; WIN32-NEXT:    movl %ecx, 20(%esi)
-; WIN32-NEXT:    movl %eax, 16(%esi)
 ; WIN32-NEXT:    fstps 12(%esi)
 ; WIN32-NEXT:    fstps 8(%esi)
 ; WIN32-NEXT:    fstps 4(%esi)
 ; WIN32-NEXT:    fstps (%esi)
 ; WIN32-NEXT:    movl %esi, %eax
-; WIN32-NEXT:    addl $60, %esp
+; WIN32-NEXT:    addl $44, %esp
 ; WIN32-NEXT:    popl %esi
-; WIN32-NEXT:    popl %edi
 ; WIN32-NEXT:    retl
   %result = call { <4 x float>, <4 x i32> } @llvm.frexp.v4f32.v4i32(<4 x float> %a)
   ret { <4 x float>, <4 x i32> } %result
@@ -499,46 +489,35 @@ define <4 x i32> @test_frexp_v4f32_v4i32_only_use_exp(<4 x float> %a) {
 ;
 ; WIN32-LABEL: test_frexp_v4f32_v4i32_only_use_exp:
 ; WIN32:       # %bb.0:
-; WIN32-NEXT:    pushl %edi
 ; WIN32-NEXT:    pushl %esi
-; WIN32-NEXT:    subl $28, %esp
+; WIN32-NEXT:    subl $12, %esp
 ; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; WIN32-NEXT:    leal 8(%esi), %eax
 ; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    fstpl (%esp)
 ; WIN32-NEXT:    calll _frexp
 ; WIN32-NEXT:    fstp %st(0)
-; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; WIN32-NEXT:    leal 4(%esi), %eax
 ; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    fstpl (%esp)
 ; WIN32-NEXT:    calll _frexp
 ; WIN32-NEXT:    fstp %st(0)
-; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; WIN32-NEXT:    leal 12(%esi), %eax
 ; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    fstpl (%esp)
 ; WIN32-NEXT:    calll _frexp
 ; WIN32-NEXT:    fstp %st(0)
-; WIN32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; WIN32-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; WIN32-NEXT:    movl %esi, {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    flds {{[0-9]+}}(%esp)
 ; WIN32-NEXT:    fstpl (%esp)
 ; WIN32-NEXT:    calll _frexp
 ; WIN32-NEXT:    fstp %st(0)
-; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; WIN32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; WIN32-NEXT:    movl %edi, 12(%esi)
-; WIN32-NEXT:    movl %edx, 8(%esi)
-; WIN32-NEXT:    movl %ecx, 4(%esi)
-; WIN32-NEXT:    movl %eax, (%esi)
 ; WIN32-NEXT:    movl %esi, %eax
-; WIN32-NEXT:    addl $28, %esp
+; WIN32-NEXT:    addl $12, %esp
 ; WIN32-NEXT:    popl %esi
-; WIN32-NEXT:    popl %edi
 ; WIN32-NEXT:    retl
   %result = call { <4 x float>, <4 x i32> } @llvm.frexp.v4f32.v4i32(<4 x float> %a)
   %result.1 = extractvalue { <4 x float>, <4 x i32> } %result, 1

diff  --git a/llvm/test/CodeGen/X86/llvm.sincos.ll b/llvm/test/CodeGen/X86/llvm.sincos.ll
new file mode 100644
index 00000000000000..a429314630e56b
--- /dev/null
+++ b/llvm/test/CodeGen/X86/llvm.sincos.ll
@@ -0,0 +1,162 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --no_x86_scrub_sp --version 5
+; RUN: llc < %s -mtriple=i386-unknown-linux-gnu  | FileCheck %s
+
+define void @test_sincos_v4f32(<4 x float> %x, ptr noalias %out_sin, ptr noalias %out_cos) {
+; CHECK-LABEL: test_sincos_v4f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pushl %edi
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    pushl %esi
+; CHECK-NEXT:    .cfi_def_cfa_offset 12
+; CHECK-NEXT:    subl $52, %esp
+; CHECK-NEXT:    .cfi_def_cfa_offset 64
+; CHECK-NEXT:    .cfi_offset %esi, -12
+; CHECK-NEXT:    .cfi_offset %edi, -8
+; CHECK-NEXT:    movl 84(%esp), %esi
+; CHECK-NEXT:    flds 76(%esp)
+; CHECK-NEXT:    fstps {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; CHECK-NEXT:    flds 64(%esp)
+; CHECK-NEXT:    fstps {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; CHECK-NEXT:    flds 72(%esp)
+; CHECK-NEXT:    fstps {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; CHECK-NEXT:    flds 68(%esp)
+; CHECK-NEXT:    movl 80(%esp), %edi
+; CHECK-NEXT:    leal 40(%esp), %eax
+; CHECK-NEXT:    movl %eax, 8(%esp)
+; CHECK-NEXT:    leal 4(%edi), %eax
+; CHECK-NEXT:    movl %eax, 4(%esp)
+; CHECK-NEXT:    fstps (%esp)
+; CHECK-NEXT:    calll sincosf
+; CHECK-NEXT:    leal 44(%esp), %eax
+; CHECK-NEXT:    movl %eax, 8(%esp)
+; CHECK-NEXT:    leal 8(%edi), %eax
+; CHECK-NEXT:    movl %eax, 4(%esp)
+; CHECK-NEXT:    flds {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
+; CHECK-NEXT:    fstps (%esp)
+; CHECK-NEXT:    calll sincosf
+; CHECK-NEXT:    leal 36(%esp), %eax
+; CHECK-NEXT:    movl %eax, 8(%esp)
+; CHECK-NEXT:    movl %edi, 4(%esp)
+; CHECK-NEXT:    flds {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
+; CHECK-NEXT:    fstps (%esp)
+; CHECK-NEXT:    calll sincosf
+; CHECK-NEXT:    leal 48(%esp), %eax
+; CHECK-NEXT:    movl %eax, 8(%esp)
+; CHECK-NEXT:    addl $12, %edi
+; CHECK-NEXT:    movl %edi, 4(%esp)
+; CHECK-NEXT:    flds {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
+; CHECK-NEXT:    fstps (%esp)
+; CHECK-NEXT:    calll sincosf
+; CHECK-NEXT:    flds 36(%esp)
+; CHECK-NEXT:    flds 40(%esp)
+; CHECK-NEXT:    flds 44(%esp)
+; CHECK-NEXT:    flds 48(%esp)
+; CHECK-NEXT:    fstps 12(%esi)
+; CHECK-NEXT:    fstps 8(%esi)
+; CHECK-NEXT:    fstps 4(%esi)
+; CHECK-NEXT:    fstps (%esi)
+; CHECK-NEXT:    addl $52, %esp
+; CHECK-NEXT:    .cfi_def_cfa_offset 12
+; CHECK-NEXT:    popl %esi
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    popl %edi
+; CHECK-NEXT:    .cfi_def_cfa_offset 4
+; CHECK-NEXT:    retl
+  %result = call { <4 x float>, <4 x float> } @llvm.sincos.v4f32(<4 x float> %x)
+  %result.0 = extractvalue { <4 x float>, <4 x float> } %result, 0
+  %result.1 = extractvalue { <4 x float>, <4 x float> } %result, 1
+  store <4 x float> %result.0, ptr %out_sin, align 4
+  store <4 x float> %result.1, ptr %out_cos, align 4
+  ret void
+}
+
+define void @test_sincos_v2f64(<2 x double> %x, ptr noalias %out_sin, ptr noalias %out_cos) {
+; CHECK-LABEL: test_sincos_v2f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pushl %edi
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    pushl %esi
+; CHECK-NEXT:    .cfi_def_cfa_offset 12
+; CHECK-NEXT:    subl $52, %esp
+; CHECK-NEXT:    .cfi_def_cfa_offset 64
+; CHECK-NEXT:    .cfi_offset %esi, -12
+; CHECK-NEXT:    .cfi_offset %edi, -8
+; CHECK-NEXT:    movl 84(%esp), %esi
+; CHECK-NEXT:    fldl 72(%esp)
+; CHECK-NEXT:    fstpl {{[-0-9]+}}(%e{{[sb]}}p) # 8-byte Folded Spill
+; CHECK-NEXT:    fldl 64(%esp)
+; CHECK-NEXT:    movl 80(%esp), %edi
+; CHECK-NEXT:    leal 24(%esp), %eax
+; CHECK-NEXT:    movl %eax, 12(%esp)
+; CHECK-NEXT:    movl %edi, 8(%esp)
+; CHECK-NEXT:    fstpl (%esp)
+; CHECK-NEXT:    calll sincos
+; CHECK-NEXT:    leal 32(%esp), %eax
+; CHECK-NEXT:    movl %eax, 12(%esp)
+; CHECK-NEXT:    addl $8, %edi
+; CHECK-NEXT:    movl %edi, 8(%esp)
+; CHECK-NEXT:    fldl {{[-0-9]+}}(%e{{[sb]}}p) # 8-byte Folded Reload
+; CHECK-NEXT:    fstpl (%esp)
+; CHECK-NEXT:    calll sincos
+; CHECK-NEXT:    fldl 24(%esp)
+; CHECK-NEXT:    fldl 32(%esp)
+; CHECK-NEXT:    fstpl 8(%esi)
+; CHECK-NEXT:    fstpl (%esi)
+; CHECK-NEXT:    addl $52, %esp
+; CHECK-NEXT:    .cfi_def_cfa_offset 12
+; CHECK-NEXT:    popl %esi
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    popl %edi
+; CHECK-NEXT:    .cfi_def_cfa_offset 4
+; CHECK-NEXT:    retl
+  %result = call { <2 x double>, <2 x double> } @llvm.sincos.v2f64(<2 x double> %x)
+  %result.0 = extractvalue { <2 x double>, <2 x double> } %result, 0
+  %result.1 = extractvalue { <2 x double>, <2 x double> } %result, 1
+  store <2 x double> %result.0, ptr %out_sin, align 8
+  store <2 x double> %result.1, ptr %out_cos, align 8
+  ret void
+}
+
+declare void @foo(ptr, ptr)
+
+define void @can_fold_with_call_in_chain(float %x, ptr noalias %a, ptr noalias %b) {
+; CHECK-LABEL: can_fold_with_call_in_chain:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    pushl %edi
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    pushl %esi
+; CHECK-NEXT:    .cfi_def_cfa_offset 12
+; CHECK-NEXT:    subl $20, %esp
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    .cfi_offset %esi, -12
+; CHECK-NEXT:    .cfi_offset %edi, -8
+; CHECK-NEXT:    flds 32(%esp)
+; CHECK-NEXT:    fstps {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; CHECK-NEXT:    movl 36(%esp), %edi
+; CHECK-NEXT:    movl 40(%esp), %esi
+; CHECK-NEXT:    movl %esi, 4(%esp)
+; CHECK-NEXT:    movl %edi, (%esp)
+; CHECK-NEXT:    calll foo at PLT
+; CHECK-NEXT:    leal 16(%esp), %eax
+; CHECK-NEXT:    movl %eax, 8(%esp)
+; CHECK-NEXT:    movl %edi, 4(%esp)
+; CHECK-NEXT:    flds {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
+; CHECK-NEXT:    fstps (%esp)
+; CHECK-NEXT:    calll sincosf
+; CHECK-NEXT:    flds 16(%esp)
+; CHECK-NEXT:    fstps (%esi)
+; CHECK-NEXT:    addl $20, %esp
+; CHECK-NEXT:    .cfi_def_cfa_offset 12
+; CHECK-NEXT:    popl %esi
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    popl %edi
+; CHECK-NEXT:    .cfi_def_cfa_offset 4
+; CHECK-NEXT:    retl
+entry:
+  %sin = tail call float @llvm.sin.f32(float %x)
+  %cos = tail call float @llvm.cos.f32(float %x)
+  call void @foo(ptr %a, ptr %b)
+  store float %sin, ptr %a, align 4
+  store float %cos, ptr %b, align 4
+  ret void
+}

diff  --git a/llvm/test/CodeGen/X86/sincos-stack-args.ll b/llvm/test/CodeGen/X86/sincos-stack-args.ll
index 9fb3a6769fda11..fe15330833334d 100644
--- a/llvm/test/CodeGen/X86/sincos-stack-args.ll
+++ b/llvm/test/CodeGen/X86/sincos-stack-args.ll
@@ -4,8 +4,11 @@
 
 declare double @g(double, double)
 
-define double @f(double %a) {
-; CHECK-LABEL: f:
+; Though not visible within the IR, this will lower to an FSINCOS node, with
+; store users, that are within a (callseq_start, callseq_end) pair. In this
+; case, the stores cannot be folded into the sincos call.
+define double @negative_sincos_with_stores_within_call_sequence(double %a) {
+; CHECK-LABEL: negative_sincos_with_stores_within_call_sequence:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    subl $44, %esp
 ; CHECK-NEXT:    .cfi_def_cfa_offset 48


        


More information about the llvm-commits mailing list