logical AND and OR in if-conditionals can turn to multiple branch instructions #162041

manik-muk · 2025-10-06T04:47:01Z

github-actions · 2025-10-06T04:47:22Z

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

llvmbot · 2025-10-06T04:47:51Z

@llvm/pr-subscribers-backend-x86

Author: None (manik-muk)

Changes

Addresses #160612

Full diff: https://github.com/llvm/llvm-project/pull/162041.diff

5 Files Affected:

(modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+18)
(modified) llvm/test/CodeGen/X86/2007-08-09-IllegalX86-64Asm.ll (+53-56)
(added) llvm/test/CodeGen/X86/issue-160612.c (+19)
(added) llvm/test/CodeGen/X86/issue-160612.ll (+74)
(modified) llvm/test/CodeGen/X86/x86-shrink-wrap-unwind.ll (+36-40)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index cdc97faf394ca..997d04b3ccbe6 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -3625,6 +3625,24 @@ X86TargetLowering::getJumpConditionMergingParams(Instruction::BinaryOps Opc,
       match(Lhs, m_SpecificICmp(ICmpInst::ICMP_EQ, m_Value(), m_Value())) &&
       match(Rhs, m_SpecificICmp(ICmpInst::ICMP_EQ, m_Value(), m_Value())))
     BaseCost += 1;
+  
+  // For OR conditions with EQ comparisons, prefer splitting into branches
+  // (unless CCMP is available). OR+EQ cannot be optimized via bitwise ops,
+  // unlike OR+NE which becomes (P|Q)!=0. Similarly, don't split signed
+  // comparisons (SLT, SGT) that can be optimized.
+  if (BaseCost >= 0 && !Subtarget.hasCCMP() && Opc == Instruction::Or) {
+    auto *LCmp = dyn_cast<ICmpInst>(Lhs);
+    auto *RCmp = dyn_cast<ICmpInst>(Rhs);
+    if (LCmp && RCmp) {
+      ICmpInst::Predicate LPred = LCmp->getPredicate();
+      ICmpInst::Predicate RPred = RCmp->getPredicate();
+      // Split OR+EQ patterns as they don't have clever optimizations
+      if (LPred == ICmpInst::ICMP_EQ && RPred == ICmpInst::ICMP_EQ) {
+        return {-1, -1, -1};
+      }
+    }
+  }
+  
   return {BaseCost, BrMergingLikelyBias.getValue(),
           BrMergingUnlikelyBias.getValue()};
 }
diff --git a/llvm/test/CodeGen/X86/2007-08-09-IllegalX86-64Asm.ll b/llvm/test/CodeGen/X86/2007-08-09-IllegalX86-64Asm.ll
index 28b4541c1bfc7..7bdc4e19a1cf6 100644
--- a/llvm/test/CodeGen/X86/2007-08-09-IllegalX86-64Asm.ll
+++ b/llvm/test/CodeGen/X86/2007-08-09-IllegalX86-64Asm.ll
@@ -44,7 +44,7 @@ define ptr @ubyte_divmod(ptr %a, ptr %b) {
 ; CHECK-NEXT:    leaq {{[0-9]+}}(%rsp), %rsi
 ; CHECK-NEXT:    callq __ubyte_convert_to_ctype
 ; CHECK-NEXT:    testl %eax, %eax
-; CHECK-NEXT:    js LBB0_6
+; CHECK-NEXT:    js LBB0_4
 ; CHECK-NEXT:  ## %bb.1: ## %cond_next.i
 ; CHECK-NEXT:    leaq {{[0-9]+}}(%rsp), %rsi
 ; CHECK-NEXT:    movq %rbx, %rdi
@@ -53,84 +53,81 @@ define ptr @ubyte_divmod(ptr %a, ptr %b) {
 ; CHECK-NEXT:    sarl $31, %ecx
 ; CHECK-NEXT:    andl %eax, %ecx
 ; CHECK-NEXT:    cmpl $-2, %ecx
-; CHECK-NEXT:    je LBB0_10
+; CHECK-NEXT:    je LBB0_8
 ; CHECK-NEXT:  ## %bb.2: ## %cond_next.i
 ; CHECK-NEXT:    cmpl $-1, %ecx
-; CHECK-NEXT:    jne LBB0_3
-; CHECK-NEXT:  LBB0_8: ## %bb4
+; CHECK-NEXT:    jne LBB0_6
+; CHECK-NEXT:  LBB0_3: ## %bb4
 ; CHECK-NEXT:    movq _PyArray_API@GOTPCREL(%rip), %rax
 ; CHECK-NEXT:    movq (%rax), %rax
 ; CHECK-NEXT:    movq 16(%rax), %rax
-; CHECK-NEXT:    jmp LBB0_9
-; CHECK-NEXT:  LBB0_6: ## %_ubyte_convert2_to_ctypes.exit
+; CHECK-NEXT:    jmp LBB0_10
+; CHECK-NEXT:  LBB0_4: ## %_ubyte_convert2_to_ctypes.exit
 ; CHECK-NEXT:    cmpl $-2, %eax
-; CHECK-NEXT:    je LBB0_10
-; CHECK-NEXT:  ## %bb.7: ## %_ubyte_convert2_to_ctypes.exit
-; CHECK-NEXT:    cmpl $-1, %eax
 ; CHECK-NEXT:    je LBB0_8
-; CHECK-NEXT:  LBB0_3: ## %bb35
+; CHECK-NEXT:  ## %bb.5: ## %_ubyte_convert2_to_ctypes.exit
+; CHECK-NEXT:    cmpl $-1, %eax
+; CHECK-NEXT:    je LBB0_3
+; CHECK-NEXT:  LBB0_6: ## %bb35
 ; CHECK-NEXT:    movq _PyUFunc_API@GOTPCREL(%rip), %r14
 ; CHECK-NEXT:    movq (%r14), %rax
 ; CHECK-NEXT:    callq *216(%rax)
 ; CHECK-NEXT:    movzbl {{[0-9]+}}(%rsp), %edx
 ; CHECK-NEXT:    testb %dl, %dl
-; CHECK-NEXT:    je LBB0_4
-; CHECK-NEXT:  ## %bb.12: ## %cond_false.i
-; CHECK-NEXT:    setne %dil
+; CHECK-NEXT:    je LBB0_11
+; CHECK-NEXT:  ## %bb.7: ## %cond_false.i
 ; CHECK-NEXT:    movzbl {{[0-9]+}}(%rsp), %esi
 ; CHECK-NEXT:    movzbl %sil, %ecx
 ; CHECK-NEXT:    movl %ecx, %eax
 ; CHECK-NEXT:    divb %dl
 ; CHECK-NEXT:    movl %eax, %r15d
 ; CHECK-NEXT:    testb %cl, %cl
-; CHECK-NEXT:    setne %al
-; CHECK-NEXT:    testb %dil, %al
-; CHECK-NEXT:    jne LBB0_5
-; CHECK-NEXT:  LBB0_13: ## %cond_true.i200
-; CHECK-NEXT:    testb %dl, %dl
-; CHECK-NEXT:    jne LBB0_15
-; CHECK-NEXT:  ## %bb.14: ## %cond_true14.i
-; CHECK-NEXT:    movl $4, %edi
-; CHECK-NEXT:    callq _feraiseexcept
-; CHECK-NEXT:  LBB0_15: ## %ubyte_ctype_remainder.exit
-; CHECK-NEXT:    xorl %ebx, %ebx
-; CHECK-NEXT:    jmp LBB0_16
-; CHECK-NEXT:  LBB0_10: ## %bb17
+; CHECK-NEXT:    jne LBB0_12
+; CHECK-NEXT:    jmp LBB0_14
+; CHECK-NEXT:  LBB0_8: ## %bb17
 ; CHECK-NEXT:    callq _PyErr_Occurred
 ; CHECK-NEXT:    testq %rax, %rax
-; CHECK-NEXT:    jne LBB0_23
-; CHECK-NEXT:  ## %bb.11: ## %cond_next
+; CHECK-NEXT:    jne LBB0_27
+; CHECK-NEXT:  ## %bb.9: ## %cond_next
 ; CHECK-NEXT:    movq _PyArray_API@GOTPCREL(%rip), %rax
 ; CHECK-NEXT:    movq (%rax), %rax
 ; CHECK-NEXT:    movq 80(%rax), %rax
-; CHECK-NEXT:  LBB0_9: ## %bb4
+; CHECK-NEXT:  LBB0_10: ## %bb4
 ; CHECK-NEXT:    movq 96(%rax), %rax
 ; CHECK-NEXT:    movq %r14, %rdi
 ; CHECK-NEXT:    movq %rbx, %rsi
 ; CHECK-NEXT:    callq *40(%rax)
-; CHECK-NEXT:    jmp LBB0_24
-; CHECK-NEXT:  LBB0_4: ## %cond_true.i
+; CHECK-NEXT:    jmp LBB0_28
+; CHECK-NEXT:  LBB0_11: ## %cond_true.i
 ; CHECK-NEXT:    movl $4, %edi
 ; CHECK-NEXT:    callq _feraiseexcept
 ; CHECK-NEXT:    movzbl {{[0-9]+}}(%rsp), %edx
 ; CHECK-NEXT:    movzbl {{[0-9]+}}(%rsp), %esi
+; CHECK-NEXT:    xorl %r15d, %r15d
 ; CHECK-NEXT:    testb %sil, %sil
-; CHECK-NEXT:    sete %al
+; CHECK-NEXT:    je LBB0_14
+; CHECK-NEXT:  LBB0_12: ## %cond_false.i
 ; CHECK-NEXT:    testb %dl, %dl
-; CHECK-NEXT:    sete %cl
-; CHECK-NEXT:    xorl %r15d, %r15d
-; CHECK-NEXT:    orb %al, %cl
-; CHECK-NEXT:    jne LBB0_13
-; CHECK-NEXT:  LBB0_5: ## %cond_next17.i
+; CHECK-NEXT:    je LBB0_14
+; CHECK-NEXT:  ## %bb.13: ## %cond_next17.i
 ; CHECK-NEXT:    movzbl %sil, %eax
 ; CHECK-NEXT:    divb %dl
 ; CHECK-NEXT:    movzbl %ah, %ebx
-; CHECK-NEXT:  LBB0_16: ## %ubyte_ctype_remainder.exit
+; CHECK-NEXT:    jmp LBB0_18
+; CHECK-NEXT:  LBB0_14: ## %cond_true.i200
+; CHECK-NEXT:    testb %dl, %dl
+; CHECK-NEXT:    jne LBB0_17
+; CHECK-NEXT:  ## %bb.16: ## %cond_true14.i
+; CHECK-NEXT:    movl $4, %edi
+; CHECK-NEXT:    callq _feraiseexcept
+; CHECK-NEXT:  LBB0_17: ## %ubyte_ctype_remainder.exit
+; CHECK-NEXT:    xorl %ebx, %ebx
+; CHECK-NEXT:  LBB0_18: ## %ubyte_ctype_remainder.exit
 ; CHECK-NEXT:    movq (%r14), %rax
 ; CHECK-NEXT:    callq *224(%rax)
 ; CHECK-NEXT:    testl %eax, %eax
-; CHECK-NEXT:    je LBB0_19
-; CHECK-NEXT:  ## %bb.17: ## %cond_true61
+; CHECK-NEXT:    je LBB0_21
+; CHECK-NEXT:  ## %bb.19: ## %cond_true61
 ; CHECK-NEXT:    movl %eax, %ebp
 ; CHECK-NEXT:    movq (%r14), %rax
 ; CHECK-NEXT:    movq _.str5@GOTPCREL(%rip), %rdi
@@ -139,8 +136,8 @@ define ptr @ubyte_divmod(ptr %a, ptr %b) {
 ; CHECK-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
 ; CHECK-NEXT:    callq *200(%rax)
 ; CHECK-NEXT:    testl %eax, %eax
-; CHECK-NEXT:    js LBB0_23
-; CHECK-NEXT:  ## %bb.18: ## %cond_next73
+; CHECK-NEXT:    js LBB0_27
+; CHECK-NEXT:  ## %bb.20: ## %cond_next73
 ; CHECK-NEXT:    movl $1, {{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    movq (%r14), %rax
 ; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %rsi
@@ -149,13 +146,13 @@ define ptr @ubyte_divmod(ptr %a, ptr %b) {
 ; CHECK-NEXT:    movl %ebp, %edx
 ; CHECK-NEXT:    callq *232(%rax)
 ; CHECK-NEXT:    testl %eax, %eax
-; CHECK-NEXT:    jne LBB0_23
-; CHECK-NEXT:  LBB0_19: ## %cond_next89
+; CHECK-NEXT:    jne LBB0_27
+; CHECK-NEXT:  LBB0_21: ## %cond_next89
 ; CHECK-NEXT:    movl $2, %edi
 ; CHECK-NEXT:    callq _PyTuple_New
 ; CHECK-NEXT:    testq %rax, %rax
-; CHECK-NEXT:    je LBB0_23
-; CHECK-NEXT:  ## %bb.20: ## %cond_next97
+; CHECK-NEXT:    je LBB0_27
+; CHECK-NEXT:  ## %bb.22: ## %cond_next97
 ; CHECK-NEXT:    movq %rax, %r14
 ; CHECK-NEXT:    movq _PyArray_API@GOTPCREL(%rip), %r12
 ; CHECK-NEXT:    movq (%r12), %rax
@@ -163,8 +160,8 @@ define ptr @ubyte_divmod(ptr %a, ptr %b) {
 ; CHECK-NEXT:    xorl %esi, %esi
 ; CHECK-NEXT:    callq *304(%rdi)
 ; CHECK-NEXT:    testq %rax, %rax
-; CHECK-NEXT:    je LBB0_21
-; CHECK-NEXT:  ## %bb.25: ## %cond_next135
+; CHECK-NEXT:    je LBB0_25
+; CHECK-NEXT:  ## %bb.23: ## %cond_next135
 ; CHECK-NEXT:    movb %r15b, 16(%rax)
 ; CHECK-NEXT:    movq %rax, 24(%r14)
 ; CHECK-NEXT:    movq (%r12), %rax
@@ -172,22 +169,22 @@ define ptr @ubyte_divmod(ptr %a, ptr %b) {
 ; CHECK-NEXT:    xorl %esi, %esi
 ; CHECK-NEXT:    callq *304(%rdi)
 ; CHECK-NEXT:    testq %rax, %rax
-; CHECK-NEXT:    je LBB0_21
-; CHECK-NEXT:  ## %bb.26: ## %cond_next182
+; CHECK-NEXT:    je LBB0_25
+; CHECK-NEXT:  ## %bb.24: ## %cond_next182
 ; CHECK-NEXT:    movb %bl, 16(%rax)
 ; CHECK-NEXT:    movq %rax, 32(%r14)
 ; CHECK-NEXT:    movq %r14, %rax
-; CHECK-NEXT:    jmp LBB0_24
-; CHECK-NEXT:  LBB0_21: ## %cond_true113
+; CHECK-NEXT:    jmp LBB0_28
+; CHECK-NEXT:  LBB0_25: ## %cond_true113
 ; CHECK-NEXT:    decq (%r14)
-; CHECK-NEXT:    jne LBB0_23
-; CHECK-NEXT:  ## %bb.22: ## %cond_true126
+; CHECK-NEXT:    jne LBB0_27
+; CHECK-NEXT:  ## %bb.26: ## %cond_true126
 ; CHECK-NEXT:    movq 8(%r14), %rax
 ; CHECK-NEXT:    movq %r14, %rdi
 ; CHECK-NEXT:    callq *48(%rax)
-; CHECK-NEXT:  LBB0_23: ## %UnifiedReturnBlock
+; CHECK-NEXT:  LBB0_27: ## %UnifiedReturnBlock
 ; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:  LBB0_24: ## %UnifiedReturnBlock
+; CHECK-NEXT:  LBB0_28: ## %UnifiedReturnBlock
 ; CHECK-NEXT:    addq $32, %rsp
 ; CHECK-NEXT:    popq %rbx
 ; CHECK-NEXT:    popq %r12
diff --git a/llvm/test/CodeGen/X86/issue-160612.c b/llvm/test/CodeGen/X86/issue-160612.c
new file mode 100644
index 0000000000000..c7b231cc81b6c
--- /dev/null
+++ b/llvm/test/CodeGen/X86/issue-160612.c
@@ -0,0 +1,19 @@
+extern void subroutine_foo(void);
+extern void subroutine_bar(void);
+
+void func_a(int x, int y) {
+    if (x == 0 || y == 0)
+        subroutine_foo();
+    else
+        subroutine_bar();
+}
+
+void func_b(int x, int y) {
+    if (x == 0)
+        subroutine_foo();
+    else if (y == 0)
+        subroutine_foo();
+    else
+        subroutine_bar();
+}
+
diff --git a/llvm/test/CodeGen/X86/issue-160612.ll b/llvm/test/CodeGen/X86/issue-160612.ll
new file mode 100644
index 0000000000000..6572c421b7fe6
--- /dev/null
+++ b/llvm/test/CodeGen/X86/issue-160612.ll
@@ -0,0 +1,74 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -O2 | FileCheck %s
+
+; Test for issue #160612: OR conditions in branches should use multiple branches
+; instead of materializing booleans with SETCC when no special optimizations apply.
+
+declare void @subroutine_foo()
+declare void @subroutine_bar()
+
+; Original issue: (x == 0 || y == 0) was generating SETCC + TEST + BRANCH
+; instead of using two conditional branches directly.
+define void @func_a(i32 noundef %x, i32 noundef %y) {
+; CHECK-LABEL: func_a:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    testl %edi, %edi
+; CHECK-NEXT:    je subroutine_foo@PLT # TAILCALL
+; CHECK-NEXT:  # %bb.1: # %entry
+; CHECK-NEXT:    testl %esi, %esi
+; CHECK-NEXT:    jne subroutine_bar@PLT # TAILCALL
+; CHECK-NEXT:  # %bb.2: # %if.then
+; CHECK-NEXT:    jmp subroutine_foo@PLT # TAILCALL
+entry:
+  %cmp = icmp eq i32 %x, 0
+  %cmp1 = icmp eq i32 %y, 0
+  %or.cond = or i1 %cmp, %cmp1
+  br i1 %or.cond, label %if.then, label %if.else
+
+if.then:
+  tail call void @subroutine_foo()
+  br label %if.end
+
+if.else:
+  tail call void @subroutine_bar()
+  br label %if.end
+
+if.end:
+  ret void
+}
+
+; Reference implementation that already generated optimal code.
+; This should continue to generate the same optimal code.
+define void @func_b(i32 noundef %x, i32 noundef %y) {
+; CHECK-LABEL: func_b:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    testl %edi, %edi
+; CHECK-NEXT:    je subroutine_foo@PLT # TAILCALL
+; CHECK-NEXT:  # %bb.1: # %if.else
+; CHECK-NEXT:    testl %esi, %esi
+; CHECK-NEXT:    je subroutine_foo@PLT # TAILCALL
+; CHECK-NEXT:  # %bb.2: # %if.else3
+; CHECK-NEXT:    jmp subroutine_bar@PLT # TAILCALL
+entry:
+  %cmp = icmp eq i32 %x, 0
+  br i1 %cmp, label %if.then, label %if.else
+
+if.then:
+  tail call void @subroutine_foo()
+  br label %if.end4
+
+if.else:
+  %cmp1 = icmp eq i32 %y, 0
+  br i1 %cmp1, label %if.then2, label %if.else3
+
+if.then2:
+  tail call void @subroutine_foo()
+  br label %if.end4
+
+if.else3:
+  tail call void @subroutine_bar()
+  br label %if.end4
+
+if.end4:
+  ret void
+}
diff --git a/llvm/test/CodeGen/X86/x86-shrink-wrap-unwind.ll b/llvm/test/CodeGen/X86/x86-shrink-wrap-unwind.ll
index 3349d31cad4b9..02d4d88a21682 100644
--- a/llvm/test/CodeGen/X86/x86-shrink-wrap-unwind.ll
+++ b/llvm/test/CodeGen/X86/x86-shrink-wrap-unwind.ll
@@ -181,40 +181,38 @@ define zeroext i1 @segmentedStack(ptr readonly %vk1, ptr readonly %vk2, i64 %key
 ; CHECK-LABEL: segmentedStack:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    cmpq %gs:816, %rsp
-; CHECK-NEXT:    jbe LBB3_6
+; CHECK-NEXT:    jbe LBB3_7
 ; CHECK-NEXT:  LBB3_1: ## %entry
 ; CHECK-NEXT:    pushq %rax
 ; CHECK-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-NEXT:    testq %rdi, %rdi
-; CHECK-NEXT:    sete %al
-; CHECK-NEXT:    testq %rsi, %rsi
-; CHECK-NEXT:    sete %cl
-; CHECK-NEXT:    orb %al, %cl
 ; CHECK-NEXT:    movq %rdi, %rax
 ; CHECK-NEXT:    orq %rsi, %rax
 ; CHECK-NEXT:    sete %al
-; CHECK-NEXT:    testb %cl, %cl
-; CHECK-NEXT:    jne LBB3_4
-; CHECK-NEXT:  ## %bb.2: ## %if.end4.i
+; CHECK-NEXT:    testq %rdi, %rdi
+; CHECK-NEXT:    je LBB3_5
+; CHECK-NEXT:  ## %bb.2: ## %entry
+; CHECK-NEXT:    testq %rsi, %rsi
+; CHECK-NEXT:    je LBB3_5
+; CHECK-NEXT:  ## %bb.3: ## %if.end4.i
 ; CHECK-NEXT:    movq 8(%rdi), %rdx
 ; CHECK-NEXT:    cmpq 8(%rsi), %rdx
-; CHECK-NEXT:    jne LBB3_5
-; CHECK-NEXT:  ## %bb.3: ## %land.rhs.i.i
+; CHECK-NEXT:    jne LBB3_6
+; CHECK-NEXT:  ## %bb.4: ## %land.rhs.i.i
 ; CHECK-NEXT:    movq (%rsi), %rsi
 ; CHECK-NEXT:    movq (%rdi), %rdi
 ; CHECK-NEXT:    callq _memcmp
 ; CHECK-NEXT:    testl %eax, %eax
 ; CHECK-NEXT:    sete %al
-; CHECK-NEXT:  LBB3_4: ## %__go_ptr_strings_equal.exit
+; CHECK-NEXT:  LBB3_5: ## %__go_ptr_strings_equal.exit
 ; CHECK-NEXT:    ## kill: def $al killed $al killed $eax
 ; CHECK-NEXT:    popq %rcx
 ; CHECK-NEXT:    retq
-; CHECK-NEXT:  LBB3_5:
+; CHECK-NEXT:  LBB3_6:
 ; CHECK-NEXT:    xorl %eax, %eax
 ; CHECK-NEXT:    ## kill: def $al killed $al killed $eax
 ; CHECK-NEXT:    popq %rcx
 ; CHECK-NEXT:    retq
-; CHECK-NEXT:  LBB3_6:
+; CHECK-NEXT:  LBB3_7:
 ; CHECK-NEXT:    movl $8, %r10d
 ; CHECK-NEXT:    movl $0, %r11d
 ; CHECK-NEXT:    callq ___morestack
@@ -224,43 +222,41 @@ define zeroext i1 @segmentedStack(ptr readonly %vk1, ptr readonly %vk2, i64 %key
 ; NOCOMPACTUNWIND-LABEL: segmentedStack:
 ; NOCOMPACTUNWIND:       # %bb.0:
 ; NOCOMPACTUNWIND-NEXT:    cmpq %fs:112, %rsp
-; NOCOMPACTUNWIND-NEXT:    jbe .LBB3_6
+; NOCOMPACTUNWIND-NEXT:    jbe .LBB3_7
 ; NOCOMPACTUNWIND-NEXT:  .LBB3_1: # %entry
 ; NOCOMPACTUNWIND-NEXT:    pushq %rax
 ; NOCOMPACTUNWIND-NEXT:    .cfi_def_cfa_offset 16
-; NOCOMPACTUNWIND-NEXT:    testq %rdi, %rdi
-; NOCOMPACTUNWIND-NEXT:    sete %al
-; NOCOMPACTUNWIND-NEXT:    testq %rsi, %rsi
-; NOCOMPACTUNWIND-NEXT:    sete %cl
-; NOCOMPACTUNWIND-NEXT:    orb %al, %cl
 ; NOCOMPACTUNWIND-NEXT:    movq %rdi, %rax
 ; NOCOMPACTUNWIND-NEXT:    orq %rsi, %rax
 ; NOCOMPACTUNWIND-NEXT:    sete %al
-; NOCOMPACTUNWIND-NEXT:    testb %cl, %cl
-; NOCOMPACTUNWIND-NEXT:    jne .LBB3_4
-; NOCOMPACTUNWIND-NEXT:  # %bb.2: # %if.end4.i
+; NOCOMPACTUNWIND-NEXT:    testq %rdi, %rdi
+; NOCOMPACTUNWIND-NEXT:    je .LBB3_5
+; NOCOMPACTUNWIND-NEXT:  # %bb.2: # %entry
+; NOCOMPACTUNWIND-NEXT:    testq %rsi, %rsi
+; NOCOMPACTUNWIND-NEXT:    je .LBB3_5
+; NOCOMPACTUNWIND-NEXT:  # %bb.3: # %if.end4.i
 ; NOCOMPACTUNWIND-NEXT:    movq 8(%rdi), %rdx
 ; NOCOMPACTUNWIND-NEXT:    cmpq 8(%rsi), %rdx
-; NOCOMPACTUNWIND-NEXT:    jne .LBB3_5
-; NOCOMPACTUNWIND-NEXT:  # %bb.3: # %land.rhs.i.i
+; NOCOMPACTUNWIND-NEXT:    jne .LBB3_6
+; NOCOMPACTUNWIND-NEXT:  # %bb.4: # %land.rhs.i.i
 ; NOCOMPACTUNWIND-NEXT:    movq (%rsi), %rsi
 ; NOCOMPACTUNWIND-NEXT:    movq (%rdi), %rdi
 ; NOCOMPACTUNWIND-NEXT:    callq memcmp@PLT
 ; NOCOMPACTUNWIND-NEXT:    testl %eax, %eax
 ; NOCOMPACTUNWIND-NEXT:    sete %al
-; NOCOMPACTUNWIND-NEXT:  .LBB3_4: # %__go_ptr_strings_equal.exit
+; NOCOMPACTUNWIND-NEXT:  .LBB3_5: # %__go_ptr_strings_equal.exit
 ; NOCOMPACTUNWIND-NEXT:    # kill: def $al killed $al killed $eax
 ; NOCOMPACTUNWIND-NEXT:    popq %rcx
 ; NOCOMPACTUNWIND-NEXT:    .cfi_def_cfa_offset 8
 ; NOCOMPACTUNWIND-NEXT:    retq
-; NOCOMPACTUNWIND-NEXT:  .LBB3_5:
+; NOCOMPACTUNWIND-NEXT:  .LBB3_6:
 ; NOCOMPACTUNWIND-NEXT:    .cfi_def_cfa_offset 16
 ; NOCOMPACTUNWIND-NEXT:    xorl %eax, %eax
 ; NOCOMPACTUNWIND-NEXT:    # kill: def $al killed $al killed $eax
 ; NOCOMPACTUNWIND-NEXT:    popq %rcx
 ; NOCOMPACTUNWIND-NEXT:    .cfi_def_cfa_offset 8
 ; NOCOMPACTUNWIND-NEXT:    retq
-; NOCOMPACTUNWIND-NEXT:  .LBB3_6:
+; NOCOMPACTUNWIND-NEXT:  .LBB3_7:
 ; NOCOMPACTUNWIND-NEXT:    movl $8, %r10d
 ; NOCOMPACTUNWIND-NEXT:    movl $0, %r11d
 ; NOCOMPACTUNWIND-NEXT:    callq __morestack
@@ -317,13 +313,13 @@ define void @with_nounwind(i1 %cond) nounwind personality ptr @my_personality {
 ; CHECK-NEXT:    popq %rax
 ; CHECK-NEXT:    retq
 ; CHECK-NEXT:  LBB4_1: ## %throw
-; CHECK-NEXT:  Ltmp0:
+; CHECK-NEXT:  Ltmp0: ## EH_LABEL
 ; CHECK-NEXT:    callq _throw_exception
-; CHECK-NEXT:  Ltmp1:
+; CHECK-NEXT:  Ltmp1: ## EH_LABEL
 ; CHECK-NEXT:  ## %bb.2: ## %unreachable
 ; CHECK-NEXT:    ud2
 ; CHECK-NEXT:  LBB4_3: ## %landing
-; CHECK-NEXT:  Ltmp2:
+; CHECK-NEXT:  Ltmp2: ## EH_LABEL
 ; CHECK-NEXT:    popq %rax
 ; CHECK-NEXT:    retq
 ; CHECK-NEXT:  Lfunc_end0:
@@ -340,12 +336,12 @@ define void @with_nounwind(i1 %cond) nounwind personality ptr @my_personality {
 ; NOCOMPACTUNWIND-NEXT:    retq
 ; NOCOMPACTUNWIND-NEXT:  .LBB4_1: # %throw
 ; NOCOMPACTUNWIND-NEXT:    .cfi_def_cfa_offset 16
-; NOCOMPACTUNWIND-NEXT:  .Ltmp0:
+; NOCOMPACTUNWIND-NEXT:  .Ltmp0: # EH_LABEL
 ; NOCOMPACTUNWIND-NEXT:    callq throw_exception@PLT
-; NOCOMPACTUNWIND-NEXT:  .Ltmp1:
+; NOCOMPACTUNWIND-NEXT:  .Ltmp1: # EH_LABEL
 ; NOCOMPACTUNWIND-NEXT:  # %bb.2: # %unreachable
 ; NOCOMPACTUNWIND-NEXT:  .LBB4_3: # %landing
-; NOCOMPACTUNWIND-NEXT:  .Ltmp2:
+; NOCOMPACTUNWIND-NEXT:  .Ltmp2: # EH_LABEL
 ; NOCOMPACTUNWIND-NEXT:    popq %rax
 ; NOCOMPACTUNWIND-NEXT:    .cfi_def_cfa_offset 8
 ; NOCOMPACTUNWIND-NEXT:    retq
@@ -379,9 +375,9 @@ define void @with_nounwind_same_succ(i1 %cond) nounwind personality ptr @my_pers
 ; CHECK-NEXT:  ## %bb.1: ## %throw
 ; CHECK-NEXT:    pushq %rax
 ; CHECK-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-NEXT:  Ltmp3:
+; CHECK-NEXT:  Ltmp3: ## EH_LABEL
 ; CHECK-NEXT:    callq _throw_exception
-; CHECK-NEXT:  Ltmp4:
+; CHECK-NEXT:  Ltmp4: ## EH_LABEL
 ; CHECK-NEXT:  LBB5_3: ## %fallthrough
 ; CHECK-NEXT:    ## InlineAsm Start
 ; CHECK-NEXT:    nop
@@ -390,7 +386,7 @@ define void @with_nounwind_same_succ(i1 %cond) nounwind personality ptr @my_pers
 ; CHECK-NEXT:  LBB5_4: ## %return
 ; CHECK-NEXT:    retq
 ; CHECK-NEXT:  LBB5_2: ## %landing
-; CHECK-NEXT:  Ltmp5:
+; CHECK-NEXT:  Ltmp5: ## EH_LABEL
 ; CHECK-NEXT:    jmp LBB5_3
 ; CHECK-NEXT:  Lfunc_end1:
 ;
@@ -401,9 +397,9 @@ define void @with_nounwind_same_succ(i1 %cond) nounwind personality ptr @my_pers
 ; NOCOMPACTUNWIND-NEXT:  # %bb.1: # %throw
 ; NOCOMPACTUNWIND-NEXT:    pushq %rax
 ; NOCOMPACTUNWIND-NEXT:    .cfi_def_cfa_offset 16
-; NOCOMPACTUNWIND-NEXT:  .Ltmp3:
+; NOCOMPACTUNWIND-NEXT:  .Ltmp3: # EH_LABEL
 ; NOCOMPACTUNWIND-NEXT:    callq throw_exception@PLT
-; NOCOMPACTUNWIND-NEXT:  .Ltmp4:
+; NOCOMPACTUNWIND-NEXT:  .Ltmp4: # EH_LABEL
 ; NOCOMPACTUNWIND-NEXT:  .LBB5_3: # %fallthrough
 ; NOCOMPACTUNWIND-NEXT:    #APP
 ; NOCOMPACTUNWIND-NEXT:    nop
@@ -414,7 +410,7 @@ define void @with_nounwind_same_succ(i1 %cond) nounwind personality ptr @my_pers
 ; NOCOMPACTUNWIND-NEXT:    retq
 ; NOCOMPACTUNWIND-NEXT:  .LBB5_2: # %landing
 ; NOCOMPACTUNWIND-NEXT:    .cfi_def_cfa_offset 16
-; NOCOMPACTUNWIND-NEXT:  .Ltmp5:
+; NOCOMPACTUNWIND-NEXT:  .Ltmp5: # EH_LABEL
 ; NOCOMPACTUNWIND-NEXT:    jmp .LBB5_3
 entry:
   br i1 %cond, label %throw, label %return

github-actions · 2025-10-06T06:34:38Z

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff origin/main HEAD --extensions cpp,c -- llvm/test/CodeGen/X86/issue-160612.c llvm/lib/Target/X86/X86ISelLowering.cpp

⚠️
The reproduction instructions above might return results for more than one PR
in a stack if you are using a stacked PR workflow. You can limit the results by
changing origin/main to the base branch/commit you want to compare against.
⚠️

View the diff from clang-format here.

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index c68d7ce95..b23579c98 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -3625,7 +3625,7 @@ X86TargetLowering::getJumpConditionMergingParams(Instruction::BinaryOps Opc,
       match(Lhs, m_SpecificICmp(ICmpInst::ICMP_EQ, m_Value(), m_Value())) &&
       match(Rhs, m_SpecificICmp(ICmpInst::ICMP_EQ, m_Value(), m_Value())))
     BaseCost += 1;
-  
+
   // For OR conditions with EQ comparisons, prefer splitting into branches
   // (unless CCMP is available). OR+EQ cannot be optimized via bitwise ops,
   // unlike OR+NE which becomes (P|Q)!=0. Similarly, don't split signed
@@ -3642,7 +3642,7 @@ X86TargetLowering::getJumpConditionMergingParams(Instruction::BinaryOps Opc,
       }
     }
   }
-  
+
   return {BaseCost, BrMergingLikelyBias.getValue(),
           BrMergingUnlikelyBias.getValue()};
 }

abhishek-kaushik22 · 2025-10-06T06:37:28Z

llvm/test/CodeGen/X86/issue-160612.c

@@ -0,0 +1,19 @@
+extern void subroutine_foo(void);


This should be deleted?

abhishek-kaushik22 · 2025-10-06T07:07:59Z

llvm/lib/Target/X86/X86ISelLowering.cpp

+  // (unless CCMP is available). OR+EQ cannot be optimized via bitwise ops,
+  // unlike OR+NE which becomes (P|Q)!=0. Similarly, don't split signed
+  // comparisons (SLT, SGT) that can be optimized.
+  if (BaseCost >= 0 && !Subtarget.hasCCMP() && Opc == Instruction::Or) {


Can we use a match here like the previous check?

if (BaseCost >= 0 && !Subtarget.hasCCMP() && Opc == Instruction::Or && match(Lhs, m_SpecificICmp(ICmpInst::ICMP_EQ, m_Value(), m_Value())) && match(Rhs, m_SpecificICmp(ICmpInst::ICMP_EQ, m_Value(), m_Value()))) return {-1,-1,-1};

added optimization

1f963ab

llvmbot added the backend:X86 label Oct 6, 2025

abhishek-kaushik22 linked an issue Oct 6, 2025 that may be closed by this pull request

x86 optimization: logical AND and OR in if-conditionals can turn to multiple branch instructions #160612

Open

abhishek-kaushik22 requested changes Oct 6, 2025

View reviewed changes

llvm/test/CodeGen/X86/issue-160612.c

@@ -0,0 +1,19 @@

extern void subroutine_foo(void);

Copy link

Contributor

abhishek-kaushik22 Oct 6, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This should be deleted?

abhishek-kaushik22 requested changes Oct 6, 2025

View reviewed changes

RKSimon self-requested a review October 6, 2025 07:41

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

logical AND and OR in if-conditionals can turn to multiple branch instructions #162041

logical AND and OR in if-conditionals can turn to multiple branch instructions #162041

manik-muk commented Oct 6, 2025

Uh oh!

github-actions bot commented Oct 6, 2025

Uh oh!

llvmbot commented Oct 6, 2025

Uh oh!

github-actions bot commented Oct 6, 2025

Uh oh!

abhishek-kaushik22 Oct 6, 2025

Uh oh!

abhishek-kaushik22 Oct 6, 2025

Uh oh!

Uh oh!

logical AND and OR in if-conditionals can turn to multiple branch instructions #162041

Are you sure you want to change the base?

logical AND and OR in if-conditionals can turn to multiple branch instructions #162041

Conversation

manik-muk commented Oct 6, 2025

Uh oh!

github-actions bot commented Oct 6, 2025

Uh oh!

llvmbot commented Oct 6, 2025

Uh oh!

github-actions bot commented Oct 6, 2025

Uh oh!

abhishek-kaushik22 Oct 6, 2025

Choose a reason for hiding this comment

Uh oh!

abhishek-kaushik22 Oct 6, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!