[llvm] 36b4b9d - [X86] Support immediate folding for CCMP/CTEST (#86616)

[via llvm-commits]

Author: Freddy Ye
Date: 2024-03-28T18:54:32+08:00
New Revision: 36b4b9d988ff1e82758fd5a462120086d21989e7

URL: https://github.com/llvm/llvm-project/commit/36b4b9d988ff1e82758fd5a462120086d21989e7
DIFF: https://github.com/llvm/llvm-project/commit/36b4b9d988ff1e82758fd5a462120086d21989e7.diff

LOG: [X86] Support immediate folding for CCMP/CTEST (#86616)

E.g.
%0:gr32 = MOV32ri 81
CTEST32rr %0, %1, 2, 10, implicit-def $eflags, implicit $eflags
=>
CTEST32ri %1, 81, 2, 10, implicit-def $eflags, implicit $eflags

Added: 
    llvm/test/CodeGen/X86/apx/foldimmediate.mir

Modified: 
    llvm/lib/Target/X86/X86InstrInfo.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index eb42a4b2119d5e..f24334312c116a 100644
--- a/llvm/lib/Target/X86/X86InstrInfo.cpp
+++ b/llvm/lib/Target/X86/X86InstrInfo.cpp
@@ -5595,9 +5595,13 @@ static unsigned convertALUrr2ALUri(unsigned Opc) {
   case X86::FROM:                                                              \
     return X86::TO;
     FROM_TO(TEST64rr, TEST64ri32)
+    FROM_TO(CTEST64rr, CTEST64ri32)
     FROM_TO(CMP64rr, CMP64ri32)
+    FROM_TO(CCMP64rr, CCMP64ri32)
     FROM_TO(TEST32rr, TEST32ri)
+    FROM_TO(CTEST32rr, CTEST32ri)
     FROM_TO(CMP32rr, CMP32ri)
+    FROM_TO(CCMP32rr, CCMP32ri)
 #undef FROM_TO
   }
 }
@@ -5697,7 +5701,8 @@ bool X86InstrInfo::foldImmediateImpl(MachineInstr &UseMI, MachineInstr *DefMI,
       UseMI.findRegisterUseOperandIdx(Reg) != 2)
     return false;
   // For CMP instructions the immediate can only be at index 1.
-  if ((NewOpc == X86::CMP64ri32 || NewOpc == X86::CMP32ri) &&
+  if (((NewOpc == X86::CMP64ri32 || NewOpc == X86::CMP32ri) ||
+       (NewOpc == X86::CCMP64ri32 || NewOpc == X86::CCMP32ri)) &&
       UseMI.findRegisterUseOperandIdx(Reg) != 1)
     return false;
 
@@ -5742,7 +5747,7 @@ bool X86InstrInfo::foldImmediateImpl(MachineInstr &UseMI, MachineInstr *DefMI,
       unsigned Op1 = 1, Op2 = CommuteAnyOperandIndex;
       unsigned ImmOpNum = 2;
       if (!UseMI.getOperand(0).isDef()) {
-        Op1 = 0; // TEST, CMP
+        Op1 = 0; // TEST, CMP, CTEST, CCMP
         ImmOpNum = 1;
       }
       if (Opc == TargetOpcode::COPY)

diff  --git a/llvm/test/CodeGen/X86/apx/foldimmediate.mir b/llvm/test/CodeGen/X86/apx/foldimmediate.mir
new file mode 100644
index 00000000000000..310fc64841f7a5
--- /dev/null
+++ b/llvm/test/CodeGen/X86/apx/foldimmediate.mir
@@ -0,0 +1,70 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 3
+# RUN: llc -mtriple=x86_64-- -run-pass=peephole-opt %s -o - | FileCheck %s
+--- |
+  define void @foldImmediate() { ret void }
+...
+---
+# Check that immediates can be folded into ALU instructions.
+name: foldImmediate
+registers:
+  - { id: 0, class: gr32 }
+  - { id: 1, class: gr32 }
+  - { id: 2, class: gr32 }
+  - { id: 3, class: gr32 }
+  - { id: 4, class: gr32 }
+  - { id: 5, class: gr32 }
+  - { id: 6, class: gr32 }
+  - { id: 7, class: gr64 }
+  - { id: 8, class: gr64 }
+  - { id: 9, class: gr64 }
+  - { id: 10, class: gr64 }
+  - { id: 11, class: gr64 }
+  - { id: 12, class: gr64 }
+  - { id: 13, class: gr64 }
+  - { id: 14, class: gr64 }
+  - { id: 15, class: gr64 }
+  - { id: 16, class: gr32 }
+  - { id: 17, class: gr64 }
+  - { id: 18, class: gr32 }
+
+body: |
+  bb.0:
+    liveins: $rdi, $rsi
+
+    ; CHECK-LABEL: name: foldImmediate
+    ; CHECK: liveins: $rdi, $rsi
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[MOV32ri:%[0-9]+]]:gr32 = MOV32ri 81
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gr32 = COPY $edi
+    ; CHECK-NEXT: CTEST32ri [[COPY]], 81, 2, 10, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit $eflags
+    ; CHECK-NEXT: CCMP32ri [[COPY]], 81, 2, 10, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit $eflags
+    ; CHECK-NEXT: [[SUBREG_TO_REG:%[0-9]+]]:gr64 = SUBREG_TO_REG 0, killed [[MOV32ri]], %subreg.sub_32bit
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gr64 = COPY $rsi
+    ; CHECK-NEXT: CTEST64ri32 [[COPY1]], 81, 2, 10, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit $eflags
+    ; CHECK-NEXT: CCMP64ri32 [[COPY1]], 81, 2, 10, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit $eflags
+    ; CHECK-NEXT: CCMP64rr [[SUBREG_TO_REG]], [[COPY1]], 2, 10, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit $eflags
+    %0 = MOV32ri 81
+    %1 = COPY $edi
+
+    CTEST32rr %0, %1, 2, 10, implicit-def $eflags, implicit $eflags
+    NOOP implicit $eflags
+
+    CCMP32rr %1, %0, 2, 10, implicit-def $eflags, implicit $eflags
+    NOOP implicit $eflags
+
+    %7 = SUBREG_TO_REG 0, killed %0:gr32, %subreg.sub_32bit
+    %8 = COPY $rsi
+
+    CTEST64rr %8, %7, 2, 10, implicit-def $eflags, implicit $eflags
+    NOOP implicit $eflags
+
+    CCMP64rr %8, %7, 2, 10, implicit-def $eflags, implicit $eflags
+    NOOP implicit $eflags
+    CCMP64rr %7, %8, 2, 10, implicit-def $eflags, implicit $eflags
+    NOOP implicit $eflags
+...