[ARM] Fix "Constant pool entry out of range!" in Thumb1 mode

[apazos]

Summary

This patch fixes PR36658, "Constant pool entry out of range!" in Thumb1 mode.

In ARMConstantIslands::optimizeThumb2JumpTables() in Thumb1 mode, adjustBBOffsetsAfter() is not calculating postOffset correctly by properly accounting for the padding that is required for the constant pool that immediately follows the jump table branch instruction.

[tstellar]

Is this a candidate for 6.0.1?

[apazos]

This bug was uncovered when running SPEC workload in Thumb1 mode.
Thumb1 mode does not seem to get widely tested.
There are other issues we want to fix in Thumb1 mode, for example, modelling TBB/TBH as pseudoinstructions of size 2 does not seem correct (should be 8 and 10).
But that will be done in a separate future patch.
We need to fix post offset computation first.
Tim, do you agree with the fix and do you want to make it a candidate for 6.0.1?

[t.p.northover]

Looks like it's clearly an improvement, and very low risk so it'd just as well go into 6.0 if there's time.

llvm/trunk/lib/Target/ARM/ARMComputeBlockSize.cpp

switch(MI->getOpcode()) {
     case ARM::t2LEApcrel:
     case ARM::t2LDRpci:
     // optimizeThumb2Branches.
     case ARM::t2B:
     case ARM::t2Bcc:
     case ARM::tBcc:
     // optimizeThumb2JumpTables.
     case ARM::t2BR_JT:
+    case ARM::tBR_JTr:
       return true;
   }
   return false;
 }
 
 void computeBlockSize(MachineFunction *MF, MachineBasicBlock *MBB,
                       BasicBlockInfo &BBI) {
   const ARMBaseInstrInfo *TII =

llvm/trunk/test/CodeGen/Thumb/PR36658.mir

+# REQUIRES: asserts
+# RUN: llc -run-pass arm-cp-islands %s -o - | FileCheck %s
+#
+# This is a reduced test made to expose a bug in
+# ARMConstantIslandPass in Thumb1 mode, see PR36658.
+
+# Verify optimized JT code uses TBB instructions.
+# CHECK-LABEL: bb.7.entry:
+# CHECK: tTBB_JT $pc, killed $r2, %jump-table.1, 0
+# CHECK-LABEL: bb.8:
+# CHECK: JUMPTABLE_TBB 0, %jump-table.1, 44
+
+# CHECK-LABEL: bb.11.entry:
+# CHECK: $r1 = tMOVSr $r0, implicit-def dead $cpsr
+# CHECK: tTBB_JT $pc, killed $r2, %jump-table.0, 1
+# CHECK-LABEL: bb.12:
+# CHECK: JUMPTABLE_TBB 1, %jump-table.0, 44
+
+--- |
+  ; ModuleID = 'PR36658.ll'
+  source_filename = "PR36658.ll"
+  target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
+  target triple = "thumbv5e-none-linux-gnueabi"
+
+  declare i32 @foo1(...)
+
+  declare i32 @foo2(i32)
+
+  declare i32 @foo3(i32*)
+
+  ; Function Attrs: nounwind optsize
+  define internal fastcc i32 @foo4(i32* nocapture %ignore_ptr) #0 {
+  entry:
+    %call = tail call i32 @foo3(i32* undef)
+    switch i32 %call, label %sw.epilog [
+      i32 120, label %sw.bb
+      i32 48, label %sw.bb73
+      i32 49, label %sw.bb73
+      i32 50, label %sw.bb73
+      i32 51, label %sw.bb73
+      i32 52, label %sw.bb73
+      i32 53, label %sw.bb73
+      i32 54, label %sw.bb73
+      i32 55, label %sw.bb73
+      i32 92, label %cleanup
+      i32 39, label %cleanup
+      i32 34, label %cleanup
+      i32 10, label %sw.bb91
+      i32 110, label %sw.bb93
+      i32 116, label %sw.bb94
+      i32 114, label %sw.bb95
+      i32 102, label %sw.bb96
+      i32 98, label %sw.bb97
+      i32 97, label %sw.bb98
+      i32 118, label %sw.bb106
+      i32 101, label %sw.bb107
+      i32 69, label %sw.bb107
+      i32 63, label %cleanup
+    ]
+
+  sw.bb:                                            ; preds = %entry
+    br label %while.cond
+
+  while.cond:                                       ; preds = %while.cond, %sw.bb
+    %call5 = tail call i32 @foo3(i32* null)
+    br label %while.cond
+
+  sw.bb73:                                          ; preds = %entry, %entry, %entry, %entry, %entry, %entry, %entry, %entry
+    %0 = and i32 %call, -8
+    %1 = icmp eq i32 %0, 48
+    br i1 %1, label %while.body83.preheader, label %while.end88
+
+  while.body83.preheader:                           ; preds = %sw.bb73
+    br label %while.body83
+
+  while.body83:                                     ; preds = %while.body83.preheader, %while.body83
+    %call87 = tail call i32 @foo3(i32* null)
+    br label %while.body83
+
+  while.end88:                                      ; preds = %sw.bb73
+    %call89 = tail call i32 @foo2(i32 %call)
+    unreachable
+
+  sw.bb91:                                          ; preds = %entry
+    store i32 1, i32* %ignore_ptr, align 4
+    br label %cleanup
+
+  sw.bb93:                                          ; preds = %entry
+    br label %cleanup
+
+  sw.bb94:                                          ; preds = %entry
+    br label %cleanup
+
+  sw.bb95:                                          ; preds = %entry
+    br label %cleanup
+
+  sw.bb96:                                          ; preds = %entry
+    br label %cleanup
+
+  sw.bb97:                                          ; preds = %entry
+    br label %cleanup
+
+  sw.bb98:                                          ; preds = %entry
+    br label %cleanup
+
+  sw.bb106:                                         ; preds = %entry
+    br label %cleanup
+
+  sw.bb107:                                         ; preds = %entry, %entry
+    br i1 undef, label %cleanup, label %if.then109
+
+  if.then109:                                       ; preds = %sw.bb107
+    %call110 = tail call i32 bitcast (i32 (...)* @foo1 to i32 (i8*, i32)*)(i8* undef, i32 %call)
+    unreachable
+
+  sw.epilog:                                        ; preds = %entry
+    %call.off = add i32 %call, -32
+    unreachable
+
+  cleanup:                                          ; preds = %sw.bb107, %sw.bb106, %sw.bb98, %sw.bb97, %sw.bb96, %sw.bb95, %sw.bb94, %sw.bb93, %sw.bb91, %entry, %entry, %entry, %entry
+    %retval.0 = phi i32 [ 11, %sw.bb106 ], [ 7, %sw.bb98 ], [ 8, %sw.bb97 ], [ 12, %sw.bb96 ], [ 13, %sw.bb95 ], [ 9, %sw.bb94 ], [ 10, %sw.bb93 ], [ 0, %sw.bb91 ], [ %call, %entry ], [ %call, %entry ], [ %call, %entry ], [ 27, %sw.bb107 ], [ %call, %entry ]
+    ret i32 %retval.0
+  }
+
+  ; Function Attrs: nounwind
+  declare void @llvm.stackprotector(i8*, i8**) #1
+
+  attributes #0 = { nounwind optsize }
+  attributes #1 = { nounwind }
+
+...
+---
+name:            foo4
+alignment:       1
+tracksRegLiveness: true
+liveins:
+  - { reg: '$r0' }
+frameInfo:
+  stackSize:       8
+  maxAlignment:    4
+  adjustsStack:    true
+  hasCalls:        true
+  maxCallFrameSize: 0
+stack:
+  - { id: 0, type: spill-slot, offset: -4, size: 4, alignment: 4, stack-id: 0,
+      callee-saved-register: '$lr', callee-saved-restored: false }
+  - { id: 1, type: spill-slot, offset: -8, size: 4, alignment: 4, stack-id: 0,
+      callee-saved-register: '$r4' }
+jumpTable:
+  kind:            inline
+  entries:
+    - id:              0
+      blocks:          [ '%bb.28', '%bb.26', '%bb.26', '%bb.26', '%bb.26',
+                         '%bb.24', '%bb.23', '%bb.26', '%bb.26', '%bb.12',
+                         '%bb.22' ]
+    - id:              1
+      blocks:          [ '%bb.19', '%bb.26', '%bb.26', '%bb.26', '%bb.21',
+                         '%bb.26', '%bb.20', '%bb.26', '%bb.25', '%bb.26',
+                         '%bb.15' ]
+body:             |
+  bb.0.entry:
+    successors: %bb.1(0x42c8590b), %bb.9(0x3d37a6f5)
+    liveins: $r0, $r4, $lr
+
+    frame-setup tPUSH 14, $noreg, killed $r4, killed $lr, implicit-def $sp, implicit $sp
+    frame-setup CFI_INSTRUCTION def_cfa_offset 8
+    frame-setup CFI_INSTRUCTION offset $lr, -4
+    frame-setup CFI_INSTRUCTION offset $r4, -8
+    $r4 = tMOVSr $r0, implicit-def dead $cpsr
+    tBL 14, $noreg, @foo3, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit undef $r0, implicit-def $sp, implicit-def $r0
+    $r1 = tMOVSr $r0, implicit-def dead $cpsr
+    tCMPi8 $r0, 68, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.9, 12, killed $cpsr
+
+  bb.1.entry:
+    successors: %bb.2(0x20000000), %bb.7(0x60000000)
+    liveins: $r0, $r1, $r4
+
+    tCMPi8 renamable $r1, 47, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.2, 13, killed $cpsr
+
+  bb.7.entry:
+    successors: %bb.16(0x71c71c72), %bb.8(0x0e38e38e)
+    liveins: $r0, $r1
+
+    $r2 = tMOVSr $r1, implicit-def dead $cpsr
+    renamable $r2, dead $cpsr = tSUBi8 killed renamable $r2, 48, 14, $noreg
+    tCMPi8 killed renamable $r2, 8, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.8, 2, killed $cpsr
+
+  bb.16.sw.bb73:
+    successors: %bb.17(0x7fffffff), %bb.18(0x00000001)
+    liveins: $r0, $r1
+
+    renamable $r2, dead $cpsr = tMOVi8 7, 14, $noreg
+    renamable $r1, dead $cpsr = tBIC killed renamable $r1, killed renamable $r2, 14, $noreg
+    tCMPi8 killed renamable $r1, 48, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.18, 1, killed $cpsr
+
+  bb.17.while.body83:
+    renamable $r0, dead $cpsr = tMOVi8 0, 14, $noreg
+    tBL 14, $noreg, @foo3, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def dead $r0
+    tB %bb.17, 14, $noreg
+
+  bb.9.entry:
+    successors: %bb.10(0x45d1745d), %bb.29(0x3a2e8ba3)
+    liveins: $r0, $r1
+
+    $r2 = tMOVSr $r1, implicit-def dead $cpsr
+    renamable $r2, dead $cpsr = tSUBi8 killed renamable $r2, 92, 14, $noreg
+    tCMPi8 renamable $r2, 10, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.29, 9, killed $cpsr
+
+  bb.10.entry:
+    successors: %bb.11(0x15555555), %bb.14(0x6aaaaaab)
+    liveins: $r0, $r1
+
+    $r2 = tMOVSr $r1, implicit-def dead $cpsr
+    renamable $r2, dead $cpsr = tSUBi8 killed renamable $r2, 110, 14, $noreg
+    tCMPi8 renamable $r2, 10, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.11, 8, killed $cpsr
+
+  bb.14.entry:
+    successors: %bb.19(0x1999999a), %bb.26(0x00000000), %bb.21(0x1999999a), %bb.20(0x1999999a), %bb.25(0x1999999a), %bb.15(0x1999999a)
+    liveins: $r2
+
+    renamable $r0, dead $cpsr = tLSLri killed renamable $r2, 2, 14, $noreg
+    renamable $r1 = tLEApcrelJT %jump-table.1, 14, $noreg
+    renamable $r0 = tLDRr killed renamable $r1, killed renamable $r0, 14, $noreg :: (load 4 from jump-table)
+    tBR_JTr killed renamable $r0, %jump-table.1
+
+  bb.19.sw.bb93:
+    renamable $r1, dead $cpsr = tMOVi8 10, 14, $noreg
+    tB %bb.28, 14, $noreg
+
+  bb.15.while.cond:
+    renamable $r0, dead $cpsr = tMOVi8 0, 14, $noreg
+    tBL 14, $noreg, @foo3, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def dead $r0
+    tB %bb.15, 14, $noreg
+
+  bb.29.entry:
+    successors: %bb.28(0x1999999a), %bb.26(0x00000000), %bb.24(0x1999999a), %bb.23(0x1999999a), %bb.12(0x1999999a), %bb.22(0x1999999a)
+    liveins: $r0, $r2
+
+    renamable $r1, dead $cpsr = tLSLri killed renamable $r2, 2, 14, $noreg
+    renamable $r2 = tLEApcrelJT %jump-table.0, 14, $noreg
+    renamable $r2 = tLDRr killed renamable $r2, killed renamable $r1, 14, $noreg :: (load 4 from jump-table)
+    $r1 = tMOVSr $r0, implicit-def dead $cpsr
+    tBR_JTr killed renamable $r2, %jump-table.0
+
+  bb.24.sw.bb98:
+    renamable $r1, dead $cpsr = tMOVi8 7, 14, $noreg
+    tB %bb.28, 14, $noreg
+
+  bb.2.entry:
+    successors: %bb.27(0x2aaaaaab), %bb.3(0x55555555)
+    liveins: $r0, $r1, $r4
+
+    tCMPi8 renamable $r1, 10, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.27, 0, killed $cpsr
+
+  bb.3.entry:
+    successors: %bb.4, %bb.5
+    liveins: $r0, $r1
+
+    tCMPi8 renamable $r1, 34, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.5, 1, killed $cpsr
+
+  bb.4:
+    liveins: $r0
+
+    $r1 = tMOVSr killed $r0, implicit-def dead $cpsr
+    tB %bb.28, 14, $noreg
+
+  bb.25.sw.bb106:
+    renamable $r1, dead $cpsr = tMOVi8 11, 14, $noreg
+    tB %bb.28, 14, $noreg
+
+  bb.23.sw.bb97:
+    renamable $r1, dead $cpsr = tMOVi8 8, 14, $noreg
+    tB %bb.28, 14, $noreg
+
+  bb.27.sw.bb91:
+    liveins: $r4
+
+    renamable $r0, dead $cpsr = tMOVi8 1, 14, $noreg
+    tSTRi killed renamable $r0, killed renamable $r4, 0, 14, $noreg :: (store 4 into %ir.ignore_ptr)
+    renamable $r1, dead $cpsr = tMOVi8 0, 14, $noreg
+    tB %bb.28, 14, $noreg
+
+  bb.21.sw.bb95:
+    renamable $r1, dead $cpsr = tMOVi8 13, 14, $noreg
+    tB %bb.28, 14, $noreg
+
+  bb.20.sw.bb94:
+    renamable $r1, dead $cpsr = tMOVi8 9, 14, $noreg
+    tB %bb.28, 14, $noreg
+
+  bb.5.entry:
+    liveins: $r0, $r1
+
+    tCMPi8 killed renamable $r1, 39, 14, $noreg, implicit-def $cpsr
+    tB %bb.6, 14, $noreg
+
+  bb.11.entry:
+    successors: %bb.12(0x80000000), %bb.26(0x00000000)
+    liveins: $r0, $r1
+
+    tCMPi8 killed renamable $r1, 69, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.26, 1, killed $cpsr
+
+  bb.12.sw.bb107:
+    successors: %bb.28(0x7fffffff), %bb.13(0x00000001)
+    liveins: $r0
+
+    renamable $r1, dead $cpsr = tMOVi8 27, 14, $noreg
+    renamable $r2, dead $cpsr = tMOVi8 0, 14, $noreg
+    tCMPi8 killed renamable $r2, 0, 14, $noreg, implicit-def $cpsr
+    tBcc %bb.28, 1, killed $cpsr
+
+  bb.13.if.then109:
+    successors:
+    liveins: $r0
+
+    $r1 = tMOVSr killed $r0, implicit-def dead $cpsr
+    tBL 14, $noreg, @foo1, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit undef $r0, implicit $r1, implicit-def $sp, implicit-def dead $r0
+
+  bb.8.entry:
+    liveins: $r0, $r1
+
+    tCMPi8 killed renamable $r1, 63, 14, $noreg, implicit-def $cpsr
+
+  bb.6.entry:
+    successors: %bb.28(0x80000000), %bb.26(0x00000000)
+    liveins: $cpsr, $r0
+
+    tPUSH 14, $noreg, killed $r0, implicit-def $sp, implicit $sp
+    tPOP 14, $noreg, def $r1, implicit-def $sp, implicit $sp
+    tBcc %bb.28, 0, killed $cpsr
+
+  bb.26.sw.epilog:
+    successors:
+
+
+  bb.22.sw.bb96:
+    renamable $r1, dead $cpsr = tMOVi8 12, 14, $noreg
+
+  bb.28.cleanup:
+    liveins: $r1
+
+    $r0 = tMOVSr killed $r1, implicit-def dead $cpsr
+    tPOP_RET 14, $noreg, def $r4, def $pc, implicit-def $sp, implicit $sp, implicit $r0
+
+  bb.18.while.end88:
+    liveins: $r0
+
+    tBL 14, $noreg, @foo2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def dead $r0
+
+...