Simplify operand folding when rematerializing a load.

We already know how to fold a reload from a frameindex without
analyzing the load instruction. Generalize this to handle any
frameindex load. This streamlines the logic for rematerializing loads
from stack arguments. As a side effect, it allows stackmaps to record
a stack argument location without spilling it.

Verified no effect on codegen for llvm test-suite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194497 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 12 insertions, 12 deletions

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index eda285b861..75c6c2a755 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -4292,6 +4292,12 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
                                                   MachineInstr *MI,
                                            const SmallVectorImpl<unsigned> &Ops,
                                                   MachineInstr *LoadMI) const {
+  // If loading from a FrameIndex, fold directly from the FrameIndex.
+  unsigned NumOps = LoadMI->getDesc().getNumOperands();
+  int FrameIndex;
+  if (isLoadFromStackSlot(LoadMI, FrameIndex))
+    return foldMemoryOperandImpl(MF, MI, Ops, FrameIndex);
+
   // Check switch flag
   if (NoFusing) return NULL;
 
@@ -4417,7 +4423,6 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
       return NULL;
 
     // Folding a normal load. Just copy the load's address operands.
-    unsigned NumOps = LoadMI->getDesc().getNumOperands();
     for (unsigned i = NumOps - X86::AddrNumOperands; i != NumOps; ++i)
       MOs.push_back(LoadMI->getOperand(i));
     break;
diff --git a/test/CodeGen/X86/stackmap.ll b/test/CodeGen/X86/stackmap.ll
index 9274ddf606..335aaf0db4 100644
--- a/test/CodeGen/X86/stackmap.ll
+++ b/test/CodeGen/X86/stackmap.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=x86_64-apple-darwin | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -disable-fp-elim | FileCheck %s
 ;
 ; Note: Print verbose stackmaps using -debug-only=stackmaps.
 
@@ -182,21 +182,16 @@ entry:
 ; CHECK:      .long 11
 ; CHECK-NEXT: .long L{{.*}}-_spilledValue
 ; CHECK-NEXT: .short 0
-; CHECK-NEXT: .short 10
+; CHECK-NEXT: .short 17
 ;
-; Check that at least one is a spilled entry (Indirect).
+; Check that at least one is a spilled entry from RBP.
+; Location: Indirect RBP + ...
 ; CHECK: .byte 3
 ; CHECK: .byte 0
+; CHECK: .short 6
 define void @spilledValue(i64 %arg0, i64 %arg1, i64 %arg2, i64 %arg3, i64 %arg4, i64 %l0, i64 %l1, i64 %l2, i64 %l3, i64 %l4, i64 %l5, i64 %l6, i64 %l7, i64 %l8, i64 %l9, i64 %l10, i64 %l11, i64 %l12, i64 %l13, i64 %l14, i64 %l15, i64 %l16) {
 entry:
-  %resolveCall = inttoptr i64 -559038737 to i8*
-  call void (i32, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i32 11, i32 15, i8* %resolveCall, i32 5, i64 %arg0, i64 %arg1, i64 %arg2, i64 %arg3, i64 %arg4, i64 %l0, i64 %l1, i64 %l2, i64 %l3, i64 %l4, i64 %l5, i64 %l6, i64 %l7, i64 %l8, i64 %l9)
-
-; FIXME: The Spiller needs to be able to fold all rematted loads! This
-; can be seen by adding %l15 to the stackmap.
-; <rdar:/15202984> [JS] Ran out of registers during register allocation
-;  %resolveCall = inttoptr i64 -559038737 to i8*
-;  call void (i32, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i32 12, i32 15, i8* %resolveCall, i32 5, i64 %arg0, i64 %arg1, i64 %arg2, i64 %arg3, i64 %arg4, i64 %l0, i64 %l1, i64 %l2, i64 %l3, i64 %l4, i64 %l5, i64 %l6, i64 %l7, i64 %l8, i64 %l9, i64 %l10, i64 %l11, i64 %l12, i64 %l13, i64 %l14, i64 %l15, i64 %l16)
+  call void (i32, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i32 11, i32 15, i8* null, i32 5, i64 %arg0, i64 %arg1, i64 %arg2, i64 %arg3, i64 %arg4, i64 %l0, i64 %l1, i64 %l2, i64 %l3, i64 %l4, i64 %l5, i64 %l6, i64 %l7, i64 %l8, i64 %l9, i64 %l10, i64 %l11, i64 %l12, i64 %l13, i64 %l14, i64 %l15, i64 %l16)
   ret void
 }