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
44 lines
1.3 KiB
LLVM
44 lines
1.3 KiB
LLVM
; RUN: llc < %s -march=bpfel -mcpu=v1 -verify-machineinstrs | FileCheck %s
|
|
; RUN: llc < %s -march=bpfeb -mcpu=v1 -verify-machineinstrs | FileCheck %s
|
|
|
|
; Source code:
|
|
; struct test_t1
|
|
; {
|
|
; short a;
|
|
; short b;
|
|
; char c;
|
|
; };
|
|
;
|
|
; struct test_t1 g;
|
|
; int test()
|
|
; {
|
|
; struct test_t1 t1[] = {{50, 500, 5}, {60, 600, 6}, {70, 700, 7}, {80, 800, 8} };
|
|
;
|
|
; g = t1[1];
|
|
; return 0;
|
|
; }
|
|
|
|
%struct.test_t1 = type { i16, i16, i8 }
|
|
|
|
@test.t1 = private unnamed_addr constant [4 x %struct.test_t1] [%struct.test_t1 { i16 50, i16 500, i8 5 }, %struct.test_t1 { i16 60, i16 600, i8 6 }, %struct.test_t1 { i16 70, i16 700, i8 7 }, %struct.test_t1 { i16 80, i16 800, i8 8 }], align 2
|
|
@g = common local_unnamed_addr global %struct.test_t1 zeroinitializer, align 2
|
|
|
|
; Function Attrs: nounwind
|
|
define i32 @test() local_unnamed_addr #0 {
|
|
; CHECK-LABEL: test:
|
|
entry:
|
|
tail call void @llvm.memcpy.p0.p0.i64(ptr align 2 @g, ptr align 2 getelementptr inbounds ([4 x %struct.test_t1], ptr @test.t1, i64 0, i64 1), i64 6, i1 false)
|
|
; CHECK: r2 = 600
|
|
; CHECK: *(u16 *)(r1 + 2) = r2
|
|
; CHECK: r2 = 60
|
|
; CHECK: *(u16 *)(r1 + 0) = r2
|
|
ret i32 0
|
|
}
|
|
; CHECK: .section .rodata,"a",@progbits
|
|
|
|
; Function Attrs: argmemonly nounwind
|
|
declare void @llvm.memcpy.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1) #1
|
|
|
|
attributes #0 = { nounwind }
|
|
attributes #1 = { argmemonly nounwind }
|