This is a follow-up on r174446, now taking Atom processors into

account. Atoms use LEA for updating SP in prologs/epilogs, and the
exact LEA opcode depends on the data model.

Also reapplying the test case which was added and then reverted
(because of Atom failures), this time specifying explicitly the CPU in
addition to the triple. The test case now checks all variations (data
mode, cpu Atom vs. Core).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@174542 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86FrameLowering.cpp b/lib/Target/X86/X86FrameLowering.cpp
index 09019617f585baf36b3f4743d6f399bffab47962..84b1c106a0afe783bd7b6f0d42daf9e5eb141cca 100644 (file)
--- a/lib/Target/X86/X86FrameLowering.cpp
+++ b/lib/Target/X86/X86FrameLowering.cpp
@@ -67,8 +67,8 @@ static unsigned getSUBriOpcode(unsigned isLP64, int64_t Imm) {
   }
 }
 
-static unsigned getADDriOpcode(unsigned isLP64, int64_t Imm) {
-  if (isLP64) {
+static unsigned getADDriOpcode(unsigned IsLP64, int64_t Imm) {
+  if (IsLP64) {
     if (isInt<8>(Imm))
       return X86::ADD64ri8;
     return X86::ADD64ri32;
@@ -79,8 +79,8 @@ static unsigned getADDriOpcode(unsigned isLP64, int64_t Imm) {
   }
 }
 
-static unsigned getLEArOpcode(unsigned is64Bit) {
-  return is64Bit ? X86::LEA64r : X86::LEA32r;
+static unsigned getLEArOpcode(unsigned IsLP64) {
+  return IsLP64 ? X86::LEA64r : X86::LEA32r;
 }
 
 /// findDeadCallerSavedReg - Return a caller-saved register that isn't live
@@ -151,7 +151,7 @@ void emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
   uint64_t Offset = isSub ? -NumBytes : NumBytes;
   unsigned Opc;
   if (UseLEA)
-    Opc = getLEArOpcode(Is64Bit);
+    Opc = getLEArOpcode(IsLP64);
   else
     Opc = isSub
       ? getSUBriOpcode(IsLP64, Offset)
@@ -1083,7 +1083,7 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
     if (RegInfo->needsStackRealignment(MF))
       MBBI = FirstCSPop;
     if (CSSize != 0) {
-      unsigned Opc = getLEArOpcode(Is64Bit);
+      unsigned Opc = getLEArOpcode(IsLP64);
       addRegOffset(BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr),
                    FramePtr, false, -CSSize);
     } else {

diff --git a/test/CodeGen/X86/stack-update-frame-opcode.ll b/test/CodeGen/X86/stack-update-frame-opcode.ll
new file mode 100644 (file)
index 0000000..9533ec8
--- /dev/null
+++ b/test/CodeGen/X86/stack-update-frame-opcode.ll
@@ -0,0 +1,31 @@
+; RUN: llc -mtriple=x86_64-pc-linux -mcpu=corei7 < %s | FileCheck -check-prefix=CORE_LP64 %s
+; RUN: llc -mtriple=x86_64-pc-linux -mcpu=atom < %s | FileCheck -check-prefix=ATOM_LP64 %s
+; RUN: llc -mtriple=x86_64-pc-linux-gnux32 -mcpu=corei7 < %s | FileCheck -check-prefix=CORE_ILP32 %s
+; RUN: llc -mtriple=x86_64-pc-linux-gnux32 -mcpu=atom < %s | FileCheck -check-prefix=ATOM_ILP32 %s
+
+define i32 @bar(i32 %a) nounwind {
+entry:
+  %arr = alloca [400 x i32], align 16
+
+; There is a 2x2 variation matrix here:
+; Atoms use LEA to update the SP. Opcode bitness depends on data model.
+; Cores use sub/add to update the SP. Opcode titness depends on data model.
+
+; CORE_LP64: subq $1608
+; CORE_ILP32: subl $1608
+; ATOM_LP64: leaq -1608
+; ATOM_ILP32: leal -1608
+
+  %arraydecay = getelementptr inbounds [400 x i32]* %arr, i64 0, i64 0
+  %call = call i32 @foo(i32 %a, i32* %arraydecay) nounwind
+  ret i32 %call
+
+; CORE_LP64: addq $1608
+; CORE_ILP32: addl $1608
+; ATOM_LP64: leaq 1608
+; ATOM_ILP32: leal 1608
+
+}
+
+declare i32 @foo(i32, i32*)
+