[AArch64][GlobalISel] Fix atomic truncating stores from generating in…

…valid copies. If the source reg is a 64b vreg, then we need to emit a subreg copy to a 32b gpr before we select sub-64b variants like STLRW.
llvm · Nov 10, 2021 · af4dc63 · af4dc63
1 parent d71bb6a
commit af4dc63
Show file tree

Hide file tree

Showing 2 changed files with 158 additions and 0 deletions.
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -2770,6 +2770,14 @@ bool AArch64InstructionSelector::select(MachineInstr &I) {
       } else {
         static unsigned Opcodes[] = {AArch64::STLRB, AArch64::STLRH,
                                      AArch64::STLRW, AArch64::STLRX};
+        Register ValReg = LdSt.getReg(0);
+        if (MRI.getType(ValReg).getSizeInBits() == 64 && MemSizeInBits != 64) {
+          // Emit a subreg copy of 32 bits.
+          Register NewVal = MRI.createVirtualRegister(&AArch64::GPR32RegClass);
+          MIB.buildInstr(TargetOpcode::COPY, {NewVal}, {})
+              .addReg(I.getOperand(0).getReg(), 0, AArch64::sub_32);
+          I.getOperand(0).setReg(NewVal);
+        }
         I.setDesc(TII.get(Opcodes[Log2_32(MemSizeInBytes)]));
       }
       constrainSelectedInstRegOperands(I, TII, TRI, RBI);

diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/select-truncstore-atomic.mir b/llvm/test/CodeGen/AArch64/GlobalISel/select-truncstore-atomic.mir
@@ -0,0 +1,150 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=aarch64-- -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=1 %s -o - | FileCheck %s
+
+---
+name:            truncstore_atomic_32
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+liveins:
+  - { reg: '$x0' }
+  - { reg: '$w1' }
+body:             |
+  ; CHECK-LABEL: name: truncstore_atomic_32
+  ; CHECK: bb.0:
+  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; CHECK-NEXT:   liveins: $w1, $x0
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:gpr32 = COPY $w1
+  ; CHECK-NEXT:   TBNZW [[COPY1]], 0, %bb.2
+  ; CHECK-NEXT:   B %bb.1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[MOVi32imm:%[0-9]+]]:gpr32 = MOVi32imm 4
+  ; CHECK-NEXT:   [[SUBREG_TO_REG:%[0-9]+]]:gpr64all = SUBREG_TO_REG 0, [[MOVi32imm]], %subreg.sub_32
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:gpr32 = COPY [[SUBREG_TO_REG]].sub_32
+  ; CHECK-NEXT:   STLRW [[COPY2]], [[COPY]] :: (store release (s32))
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   RET_ReallyLR
+  bb.1:
+    liveins: $w1, $x0
+
+    %0:gpr(p0) = COPY $x0
+    %3:gpr(s32) = COPY $w1
+    %2:gpr(s8) = G_TRUNC %3(s32)
+    %4:gpr(s8) = G_ASSERT_ZEXT %2, 1
+    %1:gpr(s1) = G_TRUNC %4(s8)
+    G_BRCOND %1(s1), %bb.3
+    G_BR %bb.2
+
+  bb.2:
+    %8:gpr(s64) = G_CONSTANT i64 4
+    G_STORE %8(s64), %0(p0) :: (store release (s32))
+
+  bb.3:
+    RET_ReallyLR
+
+...
+---
+name:            truncstore_atomic_16
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+liveins:
+  - { reg: '$x0' }
+  - { reg: '$w1' }
+body:             |
+  ; CHECK-LABEL: name: truncstore_atomic_16
+  ; CHECK: bb.0:
+  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; CHECK-NEXT:   liveins: $w1, $x0
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:gpr32 = COPY $w1
+  ; CHECK-NEXT:   TBNZW [[COPY1]], 0, %bb.2
+  ; CHECK-NEXT:   B %bb.1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[MOVi32imm:%[0-9]+]]:gpr32 = MOVi32imm 4
+  ; CHECK-NEXT:   [[SUBREG_TO_REG:%[0-9]+]]:gpr64all = SUBREG_TO_REG 0, [[MOVi32imm]], %subreg.sub_32
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:gpr32 = COPY [[SUBREG_TO_REG]].sub_32
+  ; CHECK-NEXT:   STLRH [[COPY2]], [[COPY]] :: (store release (s16))
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   RET_ReallyLR
+  bb.1:
+    liveins: $w1, $x0
+
+    %0:gpr(p0) = COPY $x0
+    %3:gpr(s32) = COPY $w1
+    %2:gpr(s8) = G_TRUNC %3(s32)
+    %4:gpr(s8) = G_ASSERT_ZEXT %2, 1
+    %1:gpr(s1) = G_TRUNC %4(s8)
+    G_BRCOND %1(s1), %bb.3
+    G_BR %bb.2
+
+  bb.2:
+    %8:gpr(s64) = G_CONSTANT i64 4
+    G_STORE %8(s64), %0(p0) :: (store release (s16))
+
+  bb.3:
+    RET_ReallyLR
+
+...
+---
+name:            truncstore_atomic_8
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+liveins:
+  - { reg: '$x0' }
+  - { reg: '$w1' }
+body:             |
+  ; CHECK-LABEL: name: truncstore_atomic_8
+  ; CHECK: bb.0:
+  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; CHECK-NEXT:   liveins: $w1, $x0
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:gpr32 = COPY $w1
+  ; CHECK-NEXT:   TBNZW [[COPY1]], 0, %bb.2
+  ; CHECK-NEXT:   B %bb.1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[MOVi32imm:%[0-9]+]]:gpr32 = MOVi32imm 4
+  ; CHECK-NEXT:   [[SUBREG_TO_REG:%[0-9]+]]:gpr64all = SUBREG_TO_REG 0, [[MOVi32imm]], %subreg.sub_32
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:gpr32 = COPY [[SUBREG_TO_REG]].sub_32
+  ; CHECK-NEXT:   STLRB [[COPY2]], [[COPY]] :: (store release (s8))
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   RET_ReallyLR
+  bb.1:
+    liveins: $w1, $x0
+
+    %0:gpr(p0) = COPY $x0
+    %3:gpr(s32) = COPY $w1
+    %2:gpr(s8) = G_TRUNC %3(s32)
+    %4:gpr(s8) = G_ASSERT_ZEXT %2, 1
+    %1:gpr(s1) = G_TRUNC %4(s8)
+    G_BRCOND %1(s1), %bb.3
+    G_BR %bb.2
+
+  bb.2:
+    %8:gpr(s64) = G_CONSTANT i64 4
+    G_STORE %8(s64), %0(p0) :: (store release (s8))
+
+  bb.3:
+    RET_ReallyLR
+
+...