[clang] [llvm] [X86] Support branch hint (PR #97721)

[Shengchen Kan via cfe-commits]

================
@@ -0,0 +1,95 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-linux -branch-hint -branch-hint-probability-threshold=60 -show-mc-encoding | FileCheck %s
+
+; Design: Insert "ds # encoding: [0x3e]" for condition branches who has high probability to take.
+; Source code:
+; int XX = 0;
+; int bar(int x);
+; void foo(int a, int b, char* c) {
+;   if (a == 0) {
+;     XX += 1;
+;   } else {
+;     if (c == 0)
+;       XX += 2;
+;   }
+;   for(int i=0; i<b; i++)
+;     bar(XX);
+; }
+
+ at XX = dso_local local_unnamed_addr global i32 0, align 4
+
+; Function Attrs: nounwind
+define dso_local void @foo(i32 noundef %a, i32 noundef %b, i8* noundef readnone %c) local_unnamed_addr #0 {
+; CHECK-LABEL: foo:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    pushq %rbx # encoding: [0x53]
+; CHECK-NEXT:    movl %esi, %ebx # encoding: [0x89,0xf3]
+; CHECK-NEXT:    testl %edi, %edi # encoding: [0x85,0xff]
+; CHECK-NEXT:    je .LBB0_1 # encoding: [0x74,A]
+; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB0_1-1, kind: FK_PCRel_1
+; CHECK-NEXT:  # %bb.2: # %if.else
+; CHECK-NEXT:    testq %rdx, %rdx # encoding: [0x48,0x85,0xd2]
+; CHECK-NEXT:    ds # encoding: [0x3e]
+; CHECK-NEXT:    jne .LBB0_5 # encoding: [0x75,A]
+; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB0_5-1, kind: FK_PCRel_1
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    movl $2, %eax # encoding: [0xb8,0x02,0x00,0x00,0x00]
+; CHECK-NEXT:    jmp .LBB0_4 # encoding: [0xeb,A]
+; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB0_4-1, kind: FK_PCRel_1
+; CHECK-NEXT:  .LBB0_1:
+; CHECK-NEXT:    movl $1, %eax # encoding: [0xb8,0x01,0x00,0x00,0x00]
+; CHECK-NEXT:  .LBB0_4: # %if.end4.sink.split
+; CHECK-NEXT:    addl %eax, XX(%rip) # encoding: [0x01,0x05,A,A,A,A]
+; CHECK-NEXT:    # fixup A - offset: 2, value: XX-4, kind: reloc_riprel_4byte
+; CHECK-NEXT:  .LBB0_5: # %if.end4
+; CHECK-NEXT:    testl %ebx, %ebx # encoding: [0x85,0xdb]
+; CHECK-NEXT:    jle .LBB0_7 # encoding: [0x7e,A]
+; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB0_7-1, kind: FK_PCRel_1
+; CHECK-NEXT:    .p2align 4, 0x90
+; CHECK-NEXT:  .LBB0_6: # %for.body
+; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:    movl XX(%rip), %edi # encoding: [0x8b,0x3d,A,A,A,A]
+; CHECK-NEXT:    # fixup A - offset: 2, value: XX-4, kind: reloc_riprel_4byte_relax
+; CHECK-NEXT:    callq bar # encoding: [0xe8,A,A,A,A]
+; CHECK-NEXT:    # fixup A - offset: 1, value: bar-4, kind: reloc_branch_4byte_pcrel
+; CHECK-NEXT:    decl %ebx # encoding: [0xff,0xcb]
+; CHECK-NEXT:    ds # encoding: [0x3e]
+; CHECK-NEXT:    jne .LBB0_6 # encoding: [0x75,A]
+; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB0_6-1, kind: FK_PCRel_1
+; CHECK-NEXT:  .LBB0_7: # %for.cond.cleanup
+; CHECK-NEXT:    popq %rbx # encoding: [0x5b]
+; CHECK-NEXT:    retq # encoding: [0xc3]
+entry:
+  %cmp = icmp eq i32 %a, 0
+  br i1 %cmp, label %if.end4.sink.split, label %if.else
+
+if.else:                                          ; preds = %entry
+  %cmp1 = icmp eq i8* %c, null
+  br i1 %cmp1, label %if.end4.sink.split, label %if.end4
+
+if.end4.sink.split:                               ; preds = %if.else, %entry
+  %.sink10 = phi i32 [ 1, %entry ], [ 2, %if.else ]
+  %0 = load i32, i32* @XX, align 4
+  %add3 = add nsw i32 %0, %.sink10
+  store i32 %add3, i32* @XX, align 4
+  br label %if.end4
+
+if.end4:                                          ; preds = %if.end4.sink.split, %if.else
+  %cmp58 = icmp sgt i32 %b, 0
+  br i1 %cmp58, label %for.body, label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %for.body, %if.end4
+  ret void
+
+for.body:                                         ; preds = %if.end4, %for.body
+  %i.09 = phi i32 [ %inc, %for.body ], [ 0, %if.end4 ]
+  %1 = load i32, i32* @XX, align 4
+  %call = tail call i32 @bar(i32 noundef %1) #0
+  %inc = add nuw nsw i32 %i.09, 1
+  %exitcond.not = icmp eq i32 %inc, %b
+  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+declare dso_local i32 @bar(i32 noundef) local_unnamed_addr #0
----------------
KanRobert wrote:

drop `dso_local` `local_unnamed_addr #0`

https://github.com/llvm/llvm-project/pull/97721