7852ebc088
Before llvm20, (void)__sync_fetch_and_add(...) always generates locked xadd insns. In linux kernel upstream discussion [1], it is found that for arm64 architecture, the original semantics of (void)__sync_fetch_and_add(...), i.e., __atomic_fetch_add(...), is preferred in order for jit to emit proper native barrier insns. In llvm commits [2] and [3], (void)__sync_fetch_and_add(...) will generate the following insns: - for cpu v1/v2: locked xadd insns to keep backward compatibility - for cpu v3/v4: __atomic_fetch_add() insns To ensure proper barrier semantics for (void)__sync_fetch_and_add(...), cpu v3/v4 is recommended. This patch enables cpu=v3 as the default cpu version. For users wanting to use cpu v1, -mcpu=v1 needs to be explicitly added to clang/llc command line. [1] https://lore.kernel.org/bpf/ZqqiQQWRnz7H93Hc@google.com/T/#mb68d67bc8f39e35a0c3db52468b9de59b79f021f [2] https://github.com/llvm/llvm-project/pull/101428 [3] https://github.com/llvm/llvm-project/pull/106494
37 lines
1.0 KiB
LLVM
37 lines
1.0 KiB
LLVM
; RUN: llc -mtriple=bpfel -mcpu=v1 -filetype=obj -o - %s | llvm-objdump --no-print-imm-hex -d - | FileCheck %s
|
|
|
|
; Source Code:
|
|
; int test(int a, int b) {
|
|
; int s = 0;
|
|
; while (a < b) { s++; a += s; b -= s; }
|
|
; return s;
|
|
; }
|
|
|
|
define i32 @test(i32, i32) local_unnamed_addr #0 {
|
|
; CHECK-LABEL: <test>:
|
|
%3 = icmp slt i32 %0, %1
|
|
br i1 %3, label %4, label %13
|
|
|
|
; <label>:4: ; preds = %2
|
|
br label %5
|
|
; CHECK: if r4 s>= r3 goto +10 <test+0x90>
|
|
|
|
; <label>:5: ; preds = %4, %5
|
|
%6 = phi i32 [ %9, %5 ], [ 0, %4 ]
|
|
%7 = phi i32 [ %11, %5 ], [ %1, %4 ]
|
|
%8 = phi i32 [ %10, %5 ], [ %0, %4 ]
|
|
%9 = add nuw nsw i32 %6, 1
|
|
%10 = add nsw i32 %9, %8
|
|
%11 = sub nsw i32 %7, %9
|
|
%12 = icmp slt i32 %10, %11
|
|
br i1 %12, label %5, label %13
|
|
; CHECK: r1 = r3
|
|
; CHECK: if r2 s> r3 goto -10 <test+0x40>
|
|
|
|
; <label>:13: ; preds = %5, %2
|
|
%14 = phi i32 [ 0, %2 ], [ %9, %5 ]
|
|
ret i32 %14
|
|
; CHECK: exit
|
|
}
|
|
attributes #0 = { norecurse nounwind readnone }
|