[ARM][THUMB2] Allow emitting T3 types of add and sub

[dnsampaio]

Summary

This patch allows to emit thumb2 add and sub
instructions with 12 bit immediates in the
emitT2RegPlusImmediate function.

• Splitting parts of the D70680

[efriedma]

llvm/lib/Target/ARM/Thumb2InstrInfo.cpp

323

Move this assertion before the T1 check?

Maybe also easier to read as "DestReg != ARM::SP || BaseReg == ARM::SP"

llvm/test/CodeGen/Thumb2/mve-stacksplot.mir

121

I'd prefer a dedicated test; this only exercises the codepath by accident.

[dnsampaio]

Added mir test.
Moved assert before T1 test.

[dnsampaio]

• Fix existing test

[efriedma]

LGTM

llvm/lib/Target/ARM/Thumb2InstrInfo.cpp

306

Whitespace.

llvm/lib/Target/ARM/Thumb2InstrInfo.cpp

if (DestReg == ARM::SP && BaseReg != ARM::SP) {
       BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), DestReg)
           .addReg(BaseReg)
           .setMIFlags(MIFlags)
           .add(predOps(ARMCC::AL));
       BaseReg = ARM::SP;
       continue;
     }
 
-    bool HasCCOut = true;
-    if (BaseReg == ARM::SP) {
-      // sub sp, sp, #imm7
-      if (DestReg == ARM::SP && (ThisVal < ((1 << 7)-1) * 4)) {
-        assert((ThisVal & 3) == 0 && "Stack update is not multiple of 4?");
-        Opc = isSub ? ARM::tSUBspi : ARM::tADDspi;
-        BuildMI(MBB, MBBI, dl, TII.get(Opc), DestReg)
-            .addReg(BaseReg)
-            .addImm(ThisVal / 4)
-            .setMIFlags(MIFlags)
-            .add(predOps(ARMCC::AL));
-        NumBytes = 0;
-        continue;
-      }
+    assert((DestReg != ARM::SP ||BaseReg == ARM::SP) &&
+           "Writing to SP, from other register.");
+
+    // Try to use T1, as it smaller
+    if ((DestReg == ARM::SP) && (ThisVal < ((1 << 7) - 1) * 4)) {
+      break;
+    bool HasCCOut = true;
+    int ImmIsT2SO = ARM_AM::getT2SOImmVal(ThisVal);
 
-      // sub rd, sp, so_imm
-      Opc = isSub ? ARM::t2SUBri : ARM::t2ADDri;
-      if (ARM_AM::getT2SOImmVal(NumBytes) != -1) {
-        NumBytes = 0;
-      } else {
-        // FIXME: Move this to ARMAddressingModes.h?
-        unsigned RotAmt = countLeadingZeros(ThisVal);
-        ThisVal = ThisVal & ARM_AM::rotr32(0xff000000U, RotAmt);
-        NumBytes &= ~ThisVal;
-        assert(ARM_AM::getT2SOImmVal(ThisVal) != -1 &&
-               "Bit extraction didn't work?");
-      }
-    } else {
-      assert(DestReg != ARM::SP && BaseReg != ARM::SP);
-      Opc = isSub ? ARM::t2SUBri : ARM::t2ADDri;
-      if (ARM_AM::getT2SOImmVal(NumBytes) != -1) {
-        NumBytes = 0;
-      } else if (ThisVal < 4096) {
-        Opc = isSub ? ARM::t2SUBri12 : ARM::t2ADDri12;
-        HasCCOut = false;
-        NumBytes = 0;
-      } else {
-        // FIXME: Move this to ARMAddressingModes.h?
-        unsigned RotAmt = countLeadingZeros(ThisVal);
-        ThisVal = ThisVal & ARM_AM::rotr32(0xff000000U, RotAmt);
-        NumBytes &= ~ThisVal;
-        assert(ARM_AM::getT2SOImmVal(ThisVal) != -1 &&
-               "Bit extraction didn't work?");
-      }
-    }
+    // Prefer T2: sub rd, rn, so_imm | sub sp, sp, so_imm
+    if (ImmIsT2SO != -1) {
+      // Prefer T3 if can make it in a single go: subw rd, rn, imm12 | subw sp,
+      // sp, imm12
+      // Use one T2 instruction to reduce NumBytes
 
     // Build the new ADD / SUB.
     MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(Opc), DestReg)
                                   .addReg(BaseReg, RegState::Kill)
                                   .addImm(ThisVal)
                                   .add(predOps(ARMCC::AL))
                                   .setMIFlags(MIFlags);
     if (HasCCOut)

llvm/test/CodeGen/MIR/ARM/thumb2-sub-sp-t3.mir

+--- |
+  ; RUN: llc --run-pass=prologepilog -o - %s | FileCheck %s
+  ; CHECK: frame-setup CFI_INSTRUCTION def_cfa_register $r7
+  ; CHECK-NEXT: $sp = frame-setup t2SUBri12 killed $sp, 4008, 14, $noreg
+
+  target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+  target triple = "thumbv7-none-none-eabi"
+  define void @foo() #0 {
+  entry:
+    %v = alloca [4000 x i8], align 1
+    %s = alloca i8*, align 4
+    %0 = bitcast [4000 x i8]* %v to i8*
+    store i8* %0, i8** %s, align 4
+    %1 = load i8*, i8** %s, align 4
+    call void @bar(i8* %1)
+    ret void
+  }
+  declare void @bar(i8*) #1
+  ; Function Attrs: nounwind
+  declare void @llvm.stackprotector(i8*, i8**) #2
+
+  attributes #0 = { noinline nounwind optnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+armv7-a,+d32,+dsp,+fp64,+fpregs,+neon,+strict-align,+thumb-mode,+vfp2,+vfp2sp,+vfp3,+vfp3d16,+vfp3d16sp,+vfp3sp" "unsafe-fp-math"="false" "use-soft-float"="false" }
+  attributes #1 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+armv7-a,+d32,+dsp,+fp64,+fpregs,+neon,+strict-align,+thumb-mode,+vfp2,+vfp2sp,+vfp3,+vfp3d16,+vfp3d16sp,+vfp3sp" "unsafe-fp-math"="false" "use-soft-float"="false" }
+  attributes #2 = { nounwind }
+
+  !llvm.module.flags = !{!0, !1}
+  !llvm.ident = !{!2}
+
+  !0 = !{i32 1, !"wchar_size", i32 4}
+  !1 = !{i32 1, !"min_enum_size", i32 4}
+  !2 = !{!"clang version 10.0.0 ([e-mail]:llvm/llvm-project.git ee219345881bdf2c144d40731f055e7b36bc8bce)"}
+
+...
+---
+name:            foo
+alignment:       2
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       false
+registers:       []
+liveins:         []
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    4
+  adjustsStack:    true
+  hasCalls:        true
+  stackProtector:  ''
+  maxCallFrameSize: 0
+  cvBytesOfCalleeSavedRegisters: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  localFrameSize:  4004
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:      []
+stack:
+  - { id: 0, name: v, type: default, offset: 0, size: 4000, alignment: 1,
+      stack-id: default, callee-saved-register: '', callee-saved-restored: true,
+      local-offset: -4000, debug-info-variable: '', debug-info-expression: '',
+      debug-info-location: '' }
+  - { id: 1, name: s, type: default, offset: 0, size: 4, alignment: 4,
+      stack-id: default, callee-saved-register: '', callee-saved-restored: true,
+      local-offset: -4004, debug-info-variable: '', debug-info-expression: '',
+      debug-info-location: '' }
+callSites:       []
+constants:       []
+machineFunctionInfo: {}
+body:             |
+  bb.0.entry:
+    renamable $r0 = t2ADDri %stack.0.v, 0, 14, $noreg, $noreg
+    t2STRi12 killed renamable $r0, %stack.1.s, 0, 14, $noreg :: (store 4 into %ir.s)
+    renamable $r0 = t2LDRi12 %stack.1.s, 0, 14, $noreg :: (dereferenceable load 4 from %ir.s)
+    ADJCALLSTACKDOWN 0, 0, 14, $noreg, implicit-def dead $sp, implicit $sp
+    tBL 14, $noreg, @bar, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp
+    ADJCALLSTACKUP 0, 0, 14, $noreg, implicit-def dead $sp, implicit $sp
+    tBX_RET 14, $noreg
+
+...

llvm/test/CodeGen/Thumb2/large-call.ll

 ; RUN: llc < %s -O0 -verify-machineinstrs -fast-isel-abort=1 -mcpu=cortex-a8 | FileCheck %s
 target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
 target triple = "thumbv7-apple-ios0.0.0"
 
 ; This test case would clobber the outgoing call arguments by writing to the
 ; emergency spill slots at [sp, #4] or [sp, #8] without adjusting the stack
 ; pointer first.
 
 ; CHECK: main
 ; CHECK: vmov.f64
 ; Adjust SP for the large call
-; CHECK: sub sp,
+; CHECK: subw sp, sp, #3720
 ; Store to call frame + #8
 ; CHECK: vstr{{.*\[}}sp, #8]
 ; Don't clobber that store until the call.
 ; CHECK-NOT: [sp, #4]
 ; CHECK-NOT: [sp, #8]
 ; CHECK: variadic
 
 define i32 @main() ssp {

llvm/test/CodeGen/Thumb2/mve-stacksplot.mir

bb.0:
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $r11, -8
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $r10, -12
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $r9, -16
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $r8, -20
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $r7, -24
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $r6, -28
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $r5, -32
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $r4, -36
-    ; CHECK-NEXT: $sp = frame-setup t2SUBri killed $sp, 1216, 14, $noreg, $noreg
-    ; CHECK-NEXT: $sp = frame-setup tSUBspi $sp, 1, 14, $noreg
+    ; CHECK-NEXT: $sp = frame-setup t2SUBri12 killed $sp, 1220, 14, $noreg
     ; CHECK-NEXT: frame-setup CFI_INSTRUCTION def_cfa_offset 1256
     ; CHECK-NEXT: $r0 = IMPLICIT_DEF
     ; CHECK-NEXT: $r1 = IMPLICIT_DEF
     ; CHECK-NEXT: $r2 = IMPLICIT_DEF
     ; CHECK-NEXT: $r3 = IMPLICIT_DEF
     ; CHECK-NEXT: $r4 = IMPLICIT_DEF
     ; CHECK-NEXT: $r5 = IMPLICIT_DEF
     ; CHECK-NEXT: $r6 = IMPLICIT_DEF