[X86] Support immediate folding for CCMP/CTEST

[FreddyLeaf]

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

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Freddy Ye (FreddyLeaf)

Changes

• pre-commit test
• [X86] Support Immediate Folding for CCMP/CTEST.

---

Full diff: https://github.com/llvm/llvm-project/pull/86616.diff

2 Files Affected:

• (modified) llvm/lib/Target/X86/X86InstrInfo.cpp (+4)
• (added) llvm/test/CodeGen/X86/apx/foldimmediate.mir (+143)

diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index eb42a4b2119d5e..65255c78225a7b 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
   }
 }
diff --git a/llvm/test/CodeGen/X86/apx/foldimmediate.mir b/llvm/test/CodeGen/X86/apx/foldimmediate.mir
new file mode 100644
index 00000000000000..d17dcee7cb7a27
--- /dev/null
+++ b/llvm/test/CodeGen/X86/apx/foldimmediate.mir
@@ -0,0 +1,143 @@
+# 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: [[ADD32ri:%[0-9]+]]:gr32 = ADD32ri [[COPY]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[ADD32ri]]
+    ; CHECK-NEXT: [[SUB32ri:%[0-9]+]]:gr32 = SUB32ri [[COPY]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[SUB32ri]]
+    ; CHECK-NEXT: [[AND32ri:%[0-9]+]]:gr32 = AND32ri [[COPY]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[AND32ri]]
+    ; CHECK-NEXT: [[OR32ri:%[0-9]+]]:gr32 = OR32ri [[COPY]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[OR32ri]]
+    ; CHECK-NEXT: [[XOR32ri:%[0-9]+]]:gr32 = XOR32ri [[COPY]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[XOR32ri]]
+    ; 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: [[ADC32ri:%[0-9]+]]:gr32 = ADC32ri [[COPY]], 81, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit [[ADC32ri]]
+    ; CHECK-NEXT: [[SBB32ri:%[0-9]+]]:gr32 = SBB32ri [[COPY]], 81, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit [[SBB32ri]]
+    ; 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: [[ADD64ri32_:%[0-9]+]]:gr64 = ADD64ri32 [[COPY1]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[ADD64ri32_]]
+    ; CHECK-NEXT: [[SUB64ri32_:%[0-9]+]]:gr64 = SUB64ri32 [[COPY1]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[SUB64ri32_]]
+    ; CHECK-NEXT: [[AND64ri32_:%[0-9]+]]:gr64 = AND64ri32 [[COPY1]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[AND64ri32_]]
+    ; CHECK-NEXT: [[OR64ri32_:%[0-9]+]]:gr64 = OR64ri32 [[COPY1]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[OR64ri32_]]
+    ; CHECK-NEXT: [[XOR64ri32_:%[0-9]+]]:gr64 = XOR64ri32 [[COPY1]], 81, implicit-def $eflags
+    ; CHECK-NEXT: NOOP implicit [[XOR64ri32_]]
+    ; CHECK-NEXT: [[MOV32ri64_:%[0-9]+]]:gr64 = MOV32ri64 81
+    ; CHECK-NEXT: NOOP implicit [[MOV32ri64_]]
+    ; CHECK-NEXT: CTEST64ri32 [[COPY1]], 81, 2, 10, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit $eflags
+    ; CHECK-NEXT: [[ADC64ri32_:%[0-9]+]]:gr64 = ADC64ri32 [[COPY1]], 81, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit [[ADC64ri32_]]
+    ; CHECK-NEXT: [[SBB64ri32_:%[0-9]+]]:gr64 = SBB64ri32 [[COPY1]], 81, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit [[SBB64ri32_]]
+    ; CHECK-NEXT: CCMP64ri32 [[COPY1]], 81, 2, 10, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit $eflags
+    ; CHECK-NEXT: CCMP64ri32 [[SUBREG_TO_REG]], 81, 2, 10, implicit-def $eflags, implicit $eflags
+    ; CHECK-NEXT: NOOP implicit $eflags
+    %0 = MOV32ri 81
+    %1 = COPY $edi
+    %2 = ADD32rr %0, %1, implicit-def $eflags
+    NOOP implicit %2
+
+    %3 = SUB32rr %1, %0, implicit-def $eflags
+    NOOP implicit %3
+
+    %4 = AND32rr %0, %1, implicit-def $eflags
+    NOOP implicit %4
+
+    %5 = OR32rr %0, %1, implicit-def $eflags
+    NOOP implicit %5
+
+    %6 = XOR32rr %0, %1, implicit-def $eflags
+    NOOP implicit %6
+
+    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
+
+    %16 = ADC32rr %0, %1, implicit-def $eflags, implicit $eflags
+    NOOP implicit %16
+
+    %18 = SBB32rr %1, %0, implicit-def $eflags, implicit $eflags
+    NOOP implicit %18
+
+    %7 = SUBREG_TO_REG 0, killed %0:gr32, %subreg.sub_32bit
+    %8 = COPY $rsi
+    %9 = ADD64rr %7, %8, implicit-def $eflags
+    NOOP implicit %9
+
+    %10 = SUB64rr %8, %7, implicit-def $eflags
+    NOOP implicit %10
+
+    %11 = AND64rr %8, %7, implicit-def $eflags
+    NOOP implicit %11
+
+    %12 = OR64rr %8, %7, implicit-def $eflags
+    NOOP implicit %12
+
+    %13 = XOR64rr %8, %7, implicit-def $eflags
+    NOOP implicit %13
+
+    %14 = COPY %7
+    NOOP implicit %14
+
+    CTEST64rr %8, %7, 2, 10, implicit-def $eflags, implicit $eflags
+    NOOP implicit $eflags
+
+    %15 = ADC64rr %8, %7, implicit-def $eflags, implicit $eflags
+    NOOP implicit %15
+
+    %17 = SBB64rr %8, %7, implicit-def $eflags, implicit $eflags
+    NOOP implicit %17
+
+    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
+...

[KanRobert]

Could you elaborate what immediate folding does for CCMP/CTEST?

[FreddyLeaf]

Could you elaborate what immediate folding does for CCMP/CTEST?

Done in commit message.

[KanRobert]

LGTM