ARM: Fix TPsoft for Thumb mode

Reviewed at http://reviews.llvm.org/D4230

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211601 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMExpandPseudoInsts.cpp b/lib/Target/ARM/ARMExpandPseudoInsts.cpp
index 6045738e2e34f2d2371b170723d49648d0a6516c..51d3dbb5bd8eeff9a108f9f45092326e25450b71 100644 (file)
--- a/lib/Target/ARM/ARMExpandPseudoInsts.cpp
+++ b/lib/Target/ARM/ARMExpandPseudoInsts.cpp
@@ -927,10 +927,16 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB,
     }
     case ARM::tTPsoft:
     case ARM::TPsoft: {
-      MachineInstrBuilder MIB =
-        BuildMI(MBB, MBBI, MI.getDebugLoc(),
-                TII->get(Opcode == ARM::tTPsoft ? ARM::tBL : ARM::BL))
-        .addExternalSymbol("__aeabi_read_tp", 0);
+      MachineInstrBuilder MIB;
+      if (Opcode == ARM::tTPsoft)
+        MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
+                      TII->get( ARM::tBL))
+              .addImm((unsigned)ARMCC::AL).addReg(0)
+              .addExternalSymbol("__aeabi_read_tp", 0);
+      else
+        MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
+                      TII->get( ARM::BL))
+              .addExternalSymbol("__aeabi_read_tp", 0);
 
       MIB->setMemRefs(MI.memoperands_begin(), MI.memoperands_end());
       TransferImpOps(MI, MIB, MIB);

diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td
index 4788bac8f351ffd87ee4623a3a8a17c57f36dfb9..af946a90b195caf453992e92d45164d824fd9aab 100644 (file)
--- a/lib/Target/ARM/ARMInstrInfo.td
+++ b/lib/Target/ARM/ARMInstrInfo.td
@@ -5113,9 +5113,11 @@ let usesCustomInserter = 1, Uses = [R4], Defs = [R4, SP] in
 // __aeabi_read_tp preserves the registers r1-r3.
 // This is a pseudo inst so that we can get the encoding right,
 // complete with fixup for the aeabi_read_tp function.
+// TPsoft is valid for ARM mode only, in case of Thumb mode a tTPsoft pattern
+// is defined in "ARMInstrThumb.td".
 let isCall = 1,
   Defs = [R0, R12, LR, CPSR], Uses = [SP] in {
-  def TPsoft : PseudoInst<(outs), (ins), IIC_Br,
+  def TPsoft : ARMPseudoInst<(outs), (ins), 4, IIC_Br,
                [(set R0, ARMthread_pointer)]>, Sched<[WriteBr]>;
 }
 

diff --git a/test/CodeGen/Thumb2/tpsoft.ll b/test/CodeGen/Thumb2/tpsoft.ll
new file mode 100644 (file)
index 0000000..6ab8bf0
--- /dev/null
+++ b/test/CodeGen/Thumb2/tpsoft.ll
@@ -0,0 +1,54 @@
+; RUN: llc  %s -mtriple=thumbv7-linux-gnueabi -o - | \
+; RUN:    FileCheck  -check-prefix=ELFASM %s
+; RUN: llc  %s -mtriple=thumbebv7-linux-gnueabi -o - | \
+; RUN:    FileCheck  -check-prefix=ELFASM %s
+; RUN: llc  %s -mtriple=thumbv7-linux-gnueabi -filetype=obj -o - | \
+; RUN:    llvm-readobj -s -sd | FileCheck  -check-prefix=ELFOBJ -check-prefix=ELFOBJ-LE %s
+; RUN: llc  %s -mtriple=thumbebv7-linux-gnueabi -filetype=obj -o - | \
+; RUN:    llvm-readobj -s -sd | FileCheck  -check-prefix=ELFOBJ -check-prefix=ELFOBJ-BE %s
+
+;; Make sure that bl __aeabi_read_tp is materialized and fixed up correctly
+;; in the obj case.
+
+@i = external thread_local global i32
+@a = external global i8
+@b = external global [10 x i8]
+
+define arm_aapcs_vfpcc i32 @main() nounwind {
+entry:
+  %0 = load i32* @i, align 4
+  switch i32 %0, label %bb2 [
+    i32 12, label %bb
+    i32 13, label %bb1
+  ]
+
+bb:                                               ; preds = %entry
+  %1 = tail call arm_aapcs_vfpcc  i32 @foo(i8* @a) nounwind
+  ret i32 %1
+; ELFASM:              bl      __aeabi_read_tp
+
+
+; ELFOBJ:      Sections [
+; ELFOBJ:        Section {
+; ELFOBJ:          Name: .text
+; ELFOBJ-LE:          SectionData (
+;;;                  BL __aeabi_read_tp is ---------+
+;;;                                                 V
+; ELFOBJ-LE-NEXT:     0000: 2DE90048 0E487844 0168FFF7 FEFF4058
+; ELFOBJ-BE:          SectionData (
+;;;                  BL __aeabi_read_tp is ---------+
+;;;                                                 V
+; ELFOBJ-BE-NEXT:     0000: E92D4800 480E4478 6801F7FF FFFE5840
+
+
+bb1:                                              ; preds = %entry
+  %2 = tail call arm_aapcs_vfpcc  i32 @bar(i32* bitcast ([10 x i8]* @b to i32*)) nounwind
+  ret i32 %2
+
+bb2:                                              ; preds = %entry
+  ret i32 -1
+}
+
+declare arm_aapcs_vfpcc i32 @foo(i8*)
+
+declare arm_aapcs_vfpcc i32 @bar(i32*)