[ARM] Optimise Thumb1 frame record creation to not clobber LR

When setting up frame records using r11 as the frame pointer for Thumb1,
we currently use LR as a temporary. This requires us to re-load it in
cases where LR is live into the function, so it is better to use an
argument register if one is free.

Author: ostannard

llvm/lib/Target/ARM/Thumb1FrameLowering.cpp

@@ -1125,7 +1125,22 @@ bool Thumb1FrameLowering::spillCalleeSavedRegisters(
       SpilledGPRs.insert(Reg);
   }
 
-  pushRegsToStack(MBB, MI, TII, FrameRecord, {ARM::LR}, UsedLRAsTemp);
+  // Determine intermediate registers which can be used for pushing the frame
+  // record:
+  // - Unused argument registers
+  // - LR: This is possible because the first PUSH will save it on the stack,
+  //       so it is free to be used as a temporary for the second. However, it
+  //       is possible for LR to be live-in to the function, in which case we
+  //       will need to restore it later in the prologue, so we only use this
+  //       if there are no free argument registers.
+  std::set<Register> FrameRecordCopyRegs;
+  for (unsigned ArgReg : {ARM::R0, ARM::R1, ARM::R2, ARM::R3})
+    if (!MF.getRegInfo().isLiveIn(ArgReg))
+      FrameRecordCopyRegs.insert(ArgReg);
+  if (FrameRecordCopyRegs.empty())
+    FrameRecordCopyRegs.insert(ARM::LR);
+
+  pushRegsToStack(MBB, MI, TII, FrameRecord, FrameRecordCopyRegs, UsedLRAsTemp);
 
   // Determine intermediate registers which can be used for pushing high regs:
   // - Spilled low regs

llvm/test/CodeGen/Thumb/frame-chain.ll

@@ -23,9 +23,9 @@ define dso_local noundef i32 @leaf(i32 noundef %0) {
 ; LEAF-FP-AAPCS:       @ %bb.0:
 ; LEAF-FP-AAPCS-NEXT:    .save {lr}
 ; LEAF-FP-AAPCS-NEXT:    push {lr}
-; LEAF-FP-AAPCS-NEXT:    mov lr, r11
+; LEAF-FP-AAPCS-NEXT:    mov r3, r11
 ; LEAF-FP-AAPCS-NEXT:    .save {r11}
-; LEAF-FP-AAPCS-NEXT:    push {lr}
+; LEAF-FP-AAPCS-NEXT:    push {r3}
 ; LEAF-FP-AAPCS-NEXT:    .setfp r11, sp
 ; LEAF-FP-AAPCS-NEXT:    mov r11, sp
 ; LEAF-FP-AAPCS-NEXT:    .pad #4
@@ -80,9 +80,9 @@ define dso_local noundef i32 @non_leaf(i32 noundef %0) {
 ; FP-AAPCS:       @ %bb.0:
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
 ; FP-AAPCS-NEXT:    .pad #8
@@ -161,9 +161,9 @@ define dso_local void @required_fp(i32 %0, i32 %1) {
 ; FP-AAPCS:       @ %bb.0:
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
 ; FP-AAPCS-NEXT:    .save {r4, r6}
@@ -227,9 +227,9 @@ define dso_local void @required_fp(i32 %0, i32 %1) {
 ; NOFP-AAPCS:       @ %bb.0:
 ; NOFP-AAPCS-NEXT:    .save {lr}
 ; NOFP-AAPCS-NEXT:    push {lr}
-; NOFP-AAPCS-NEXT:    mov lr, r11
+; NOFP-AAPCS-NEXT:    mov r3, r11
 ; NOFP-AAPCS-NEXT:    .save {r11}
-; NOFP-AAPCS-NEXT:    push {lr}
+; NOFP-AAPCS-NEXT:    push {r3}
 ; NOFP-AAPCS-NEXT:    .setfp r11, sp
 ; NOFP-AAPCS-NEXT:    mov r11, sp
 ; NOFP-AAPCS-NEXT:    .save {r4, r6}

llvm/test/CodeGen/Thumb/returnaddress.ll

@@ -26,13 +26,11 @@ define void @ra_call() {
 ; FP-AAPCS:       @ %bb.0: @ %entry
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
-; FP-AAPCS-NEXT:    ldr r3, [sp, #4]
-; FP-AAPCS-NEXT:    mov lr, r3
 ; FP-AAPCS-NEXT:    mov r0, lr
 ; FP-AAPCS-NEXT:    bl sink_ptr
 ; FP-AAPCS-NEXT:    pop {r0}
@@ -63,13 +61,11 @@ define ptr @ra_return() {
 ; FP-AAPCS:       @ %bb.0: @ %entry
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
-; FP-AAPCS-NEXT:    ldr r3, [sp, #4]
-; FP-AAPCS-NEXT:    mov lr, r3
 ; FP-AAPCS-NEXT:    mov r0, lr
 ; FP-AAPCS-NEXT:    pop {r1}
 ; FP-AAPCS-NEXT:    mov r11, r1
@@ -104,15 +100,13 @@ define ptr @callee_saved_low() {
 ; FP-AAPCS:       @ %bb.0: @ %entry
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
 ; FP-AAPCS-NEXT:    .save {r4, r5}
 ; FP-AAPCS-NEXT:    push {r4, r5}
-; FP-AAPCS-NEXT:    ldr r5, [sp, #12]
-; FP-AAPCS-NEXT:    mov lr, r5
 ; FP-AAPCS-NEXT:    mov r0, lr
 ; FP-AAPCS-NEXT:    @APP
 ; FP-AAPCS-NEXT:    @NO_APP
@@ -163,17 +157,15 @@ define ptr @callee_saved_high() {
 ; FP-AAPCS:       @ %bb.0: @ %entry
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
 ; FP-AAPCS-NEXT:    mov r3, r9
 ; FP-AAPCS-NEXT:    mov r2, r8
 ; FP-AAPCS-NEXT:    .save {r8, r9}
 ; FP-AAPCS-NEXT:    push {r2, r3}
-; FP-AAPCS-NEXT:    ldr r3, [sp, #12]
-; FP-AAPCS-NEXT:    mov lr, r3
 ; FP-AAPCS-NEXT:    mov r0, lr
 ; FP-AAPCS-NEXT:    @APP
 ; FP-AAPCS-NEXT:    @NO_APP
@@ -237,15 +229,13 @@ define ptr @large_alloca() {
 ; FP-AAPCS:       @ %bb.0: @ %entry
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
 ; FP-AAPCS-NEXT:    .save {r4, r7}
 ; FP-AAPCS-NEXT:    push {r4, r7}
-; FP-AAPCS-NEXT:    ldr r7, [sp, #12]
-; FP-AAPCS-NEXT:    mov lr, r7
 ; FP-AAPCS-NEXT:    ldr r7, .LCPI4_0
 ; FP-AAPCS-NEXT:    .pad #2000
 ; FP-AAPCS-NEXT:    add sp, r7
@@ -318,15 +308,13 @@ define ptr @var_alloca(i32 %size) {
 ; FP-AAPCS:       @ %bb.0: @ %entry
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
 ; FP-AAPCS-NEXT:    .save {r4, r6}
 ; FP-AAPCS-NEXT:    push {r4, r6}
-; FP-AAPCS-NEXT:    ldr r6, [sp, #12]
-; FP-AAPCS-NEXT:    mov lr, r6
 ; FP-AAPCS-NEXT:    mov r6, sp
 ; FP-AAPCS-NEXT:    mov r4, lr
 ; FP-AAPCS-NEXT:    adds r0, r0, #7
@@ -433,9 +421,9 @@ define ptr @ra_depth_1() {
 ; FP-AAPCS:       @ %bb.0: @ %entry
 ; FP-AAPCS-NEXT:    .save {lr}
 ; FP-AAPCS-NEXT:    push {lr}
-; FP-AAPCS-NEXT:    mov lr, r11
+; FP-AAPCS-NEXT:    mov r3, r11
 ; FP-AAPCS-NEXT:    .save {r11}
-; FP-AAPCS-NEXT:    push {lr}
+; FP-AAPCS-NEXT:    push {r3}
 ; FP-AAPCS-NEXT:    .setfp r11, sp
 ; FP-AAPCS-NEXT:    mov r11, sp
 ; FP-AAPCS-NEXT:    mov r0, r11