[AArch64][GlobalISel] Optimize away a Not feeding a brcond by using tbz instead of...

Usually brconds are fed by compares, but not always, in which case we would
miss this fold.

Differential Revision: https://reviews.llvm.org/D86413

diff --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
index 5e5f902..a8d6818 100644 (file)
--- a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -41,6 +41,7 @@
 #define DEBUG_TYPE "aarch64-isel"
 
 using namespace llvm;
+using namespace MIPatternMatch;
 
 namespace {
 
@@ -1883,7 +1884,7 @@ bool AArch64InstructionSelector::select(MachineInstr &I) {
       return false;
     }
 
-    const Register CondReg = I.getOperand(0).getReg();
+    Register CondReg = I.getOperand(0).getReg();
     MachineBasicBlock *DestMBB = I.getOperand(1).getMBB();
 
     // Speculation tracking/SLH assumes that optimized TB(N)Z/CB(N)Z
@@ -1893,7 +1894,19 @@ bool AArch64InstructionSelector::select(MachineInstr &I) {
       return true;
 
     if (ProduceNonFlagSettingCondBr) {
-      auto MIB = BuildMI(MBB, I, I.getDebugLoc(), TII.get(AArch64::TBNZW))
+      unsigned BOpc = AArch64::TBNZW;
+      // Try to fold a not, i.e. a xor, cond, 1.
+      Register XorSrc;
+      int64_t Cst;
+      if (mi_match(CondReg, MRI,
+                   m_GTrunc(m_GXor(m_Reg(XorSrc), m_ICst(Cst)))) &&
+          Cst == 1) {
+        CondReg = XorSrc;
+        BOpc = AArch64::TBZW;
+        if (MRI.getType(XorSrc).getSizeInBits() > 32)
+          BOpc = AArch64::TBZX;
+      }
+      auto MIB = BuildMI(MBB, I, I.getDebugLoc(), TII.get(BOpc))
                      .addUse(CondReg)
                      .addImm(/*bit offset=*/0)
                      .addMBB(DestMBB);

diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/select-brcond-of-not.mir b/llvm/test/CodeGen/AArch64/GlobalISel/select-brcond-of-not.mir
new file mode 100644 (file)
index 0000000..41fe50d
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/select-brcond-of-not.mir
@@ -0,0 +1,76 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=aarch64-- -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s
+---
+name:            condbr_of_not
+legalized:       true
+regBankSelected: true
+liveins:
+  - { reg: '$x0' }
+body:             |
+  ; CHECK-LABEL: name: condbr_of_not
+  ; CHECK: bb.0:
+  ; CHECK:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
+  ; CHECK:   [[LDRBBui:%[0-9]+]]:gpr32 = LDRBBui [[COPY]], 0 :: (load 1)
+  ; CHECK:   TBZW [[LDRBBui]], 0, %bb.2
+  ; CHECK: bb.1:
+  ; CHECK:   RET_ReallyLR
+  ; CHECK: bb.2:
+  ; CHECK:   RET_ReallyLR
+  bb.1:
+    successors: %bb.2, %bb.3
+    liveins: $x0
+
+    %0:gpr(p0) = COPY $x0
+    %8:gpr(s8) = G_LOAD %0(p0) :: (load 1)
+    %4:gpr(s32) = G_ANYEXT %8(s8)
+    %5:gpr(s32) = G_CONSTANT i32 1
+    %6:gpr(s32) = G_XOR %4, %5
+    %3:gpr(s1) = G_TRUNC %6(s32)
+    G_BRCOND %3(s1), %bb.3
+
+  bb.2:
+    RET_ReallyLR
+
+  bb.3:
+    RET_ReallyLR
+
+...
+---
+name:            condbr_of_not_64
+legalized:       true
+regBankSelected: true
+liveins:
+  - { reg: '$x0' }
+body:             |
+  ; CHECK-LABEL: name: condbr_of_not_64
+  ; CHECK: bb.0:
+  ; CHECK:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
+  ; CHECK:   [[LDRBBui:%[0-9]+]]:gpr32 = LDRBBui [[COPY]], 0 :: (load 1)
+  ; CHECK:   [[SUBREG_TO_REG:%[0-9]+]]:gpr64all = SUBREG_TO_REG 0, [[LDRBBui]], %subreg.sub_32
+  ; CHECK:   [[COPY1:%[0-9]+]]:gpr64 = COPY [[SUBREG_TO_REG]]
+  ; CHECK:   TBZX [[COPY1]], 0, %bb.2
+  ; CHECK: bb.1:
+  ; CHECK:   RET_ReallyLR
+  ; CHECK: bb.2:
+  ; CHECK:   RET_ReallyLR
+  bb.1:
+    successors: %bb.2, %bb.3
+    liveins: $x0
+
+    %0:gpr(p0) = COPY $x0
+    %8:gpr(s8) = G_LOAD %0(p0) :: (load 1)
+    %4:gpr(s64) = G_ANYEXT %8(s8)
+    %5:gpr(s64) = G_CONSTANT i64 1
+    %6:gpr(s64) = G_XOR %4, %5
+    %3:gpr(s1) = G_TRUNC %6(s64)
+    G_BRCOND %3(s1), %bb.3
+
+  bb.2:
+    RET_ReallyLR
+
+  bb.3:
+    RET_ReallyLR
+
+...