Fix for PR7193 was overly conservative. The only case where sibcall callee

address cannot be allocated a register is in 32-bit mode where the first
three arguments are marked inreg. In that case EAX, EDX, and ECX will be
used for argument passing.

This fixes PR7610.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108327 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 25 insertions, 6 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 1a63474480..937d51023b 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -2458,17 +2458,23 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
     // If the tailcall address may be in a register, then make sure it's
     // possible to register allocate for it. In 32-bit, the call address can
     // only target EAX, EDX, or ECX since the tail call must be scheduled after
-    // callee-saved registers are restored. In 64-bit, it's RAX, RCX, RDX, RSI,
-    // RDI, R8, R9, R11.
-    if (!isa<GlobalAddressSDNode>(Callee) &&
+    // callee-saved registers are restored. These happen to be the same
+    // registers used to pass 'inreg' arguments so watch out for those.
+    if (!Subtarget->is64Bit() &&
+        !isa<GlobalAddressSDNode>(Callee) &&
         !isa<ExternalSymbolSDNode>(Callee)) {
-      unsigned Limit = Subtarget->is64Bit() ? 8 : 3;
       unsigned NumInRegs = 0;
       for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
         CCValAssign &VA = ArgLocs[i];
-        if (VA.isRegLoc()) {
-          if (++NumInRegs == Limit)
+        if (!VA.isRegLoc())
+          continue;
+        unsigned Reg = VA.getLocReg();
+        switch (Reg) {
+        default: break;
+        case X86::EAX: case X86::EDX: case X86::ECX:
+          if (++NumInRegs == 3)
             return false;
+          break;
         }
       }
     }
diff --git a/test/CodeGen/X86/sibcall-4.ll b/test/CodeGen/X86/sibcall-4.ll
new file mode 100644
index 0000000000..1499e66880
--- /dev/null
+++ b/test/CodeGen/X86/sibcall-4.ll
@@ -0,0 +1,13 @@
+; RUN: llc < %s -mtriple=i386-pc-linux-gnu | FileCheck %s
+; pr7610
+
+define cc10 void @t(i32* %Base_Arg, i32* %Sp_Arg, i32* %Hp_Arg, i32 %R1_Arg) nounwind {
+cm1:
+; CHECK: t:
+; CHECK: jmpl *%eax
+  %nm3 = getelementptr i32* %Sp_Arg, i32 1
+  %nm9 = load i32* %Sp_Arg
+  %nma = inttoptr i32 %nm9 to void (i32*, i32*, i32*, i32)*
+  tail call cc10 void %nma(i32* %Base_Arg, i32* %nm3, i32* %Hp_Arg, i32 %R1_Arg) nounwind
+  ret void
+}