|
| 1 | +; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme,+sve -verify-machineinstrs --pass-remarks-analysis=sme -o /dev/null < %s 2>&1 | FileCheck %s |
| 2 | + |
| 3 | +declare void @normal_callee() |
| 4 | +declare void @streaming_callee() "aarch64_pstate_sm_enabled" |
| 5 | +declare void @streaming_compatible_callee() "aarch64_pstate_sm_compatible" |
| 6 | + |
| 7 | +; CHECK: remark: <unknown>:0:0: call from 'normal_caller_streaming_callee' to 'streaming_callee' requires a streaming mode transition |
| 8 | +define void @normal_caller_streaming_callee() nounwind { |
| 9 | + call void @streaming_callee() |
| 10 | + ret void; |
| 11 | +} |
| 12 | + |
| 13 | +; CHECK: remark: <unknown>:0:0: call from 'streaming_caller_normal_callee' to 'normal_callee' requires a streaming mode transition |
| 14 | +define void @streaming_caller_normal_callee() nounwind "aarch64_pstate_sm_enabled" { |
| 15 | + call void @normal_callee() |
| 16 | + ret void; |
| 17 | +} |
| 18 | + |
| 19 | +; CHECK-NOT: streaming_caller_streaming_callee |
| 20 | +define void @streaming_caller_streaming_callee() nounwind "aarch64_pstate_sm_enabled" { |
| 21 | + call void @streaming_callee() |
| 22 | + ret void; |
| 23 | +} |
| 24 | + |
| 25 | +; CHECK-NOT: streaming_caller_streaming_compatible_callee |
| 26 | +define void @streaming_caller_streaming_compatible_callee() nounwind "aarch64_pstate_sm_enabled" { |
| 27 | + call void @streaming_compatible_callee() |
| 28 | + ret void; |
| 29 | +} |
| 30 | + |
| 31 | +; CHECK: remark: <unknown>:0:0: call from 'call_to_function_pointer_streaming_enabled' to 'unknown callee' requires a streaming mode transition |
| 32 | +define void @call_to_function_pointer_streaming_enabled(ptr %p) nounwind { |
| 33 | + call void %p() "aarch64_pstate_sm_enabled" |
| 34 | + ret void |
| 35 | +} |
| 36 | + |
| 37 | +; CHECK: remark: <unknown>:0:0: call from 'smstart_clobber_simdfp' to 'streaming_callee' requires a streaming mode transition |
| 38 | +define <4 x i32> @smstart_clobber_simdfp(<4 x i32> %x) nounwind { |
| 39 | + call void @streaming_callee() |
| 40 | + ret <4 x i32> %x; |
| 41 | +} |
| 42 | + |
| 43 | +; CHECK: remark: <unknown>:0:0: call from 'smstart_clobber_sve' to 'streaming_callee' requires a streaming mode transition |
| 44 | +define <vscale x 4 x i32> @smstart_clobber_sve(<vscale x 4 x i32> %x) nounwind { |
| 45 | + call void @streaming_callee() |
| 46 | + ret <vscale x 4 x i32> %x; |
| 47 | +} |
| 48 | + |
| 49 | +; CHECK: remark: <unknown>:0:0: call from 'smstart_clobber_sve_duplicate' to 'streaming_callee' requires a streaming mode transition |
| 50 | +; CHECK: remark: <unknown>:0:0: call from 'smstart_clobber_sve_duplicate' to 'streaming_callee' requires a streaming mode transition |
| 51 | +define <vscale x 4 x i32> @smstart_clobber_sve_duplicate(<vscale x 4 x i32> %x) nounwind { |
| 52 | + call void @streaming_callee() |
| 53 | + call void @streaming_callee() |
| 54 | + ret <vscale x 4 x i32> %x; |
| 55 | +} |
| 56 | + |
| 57 | +; CHECK: remark: <unknown>:0:0: call from 'call_to_intrinsic_without_chain' to 'cos' requires a streaming mode transition |
| 58 | +define double @call_to_intrinsic_without_chain(double %x) nounwind "aarch64_pstate_sm_enabled" { |
| 59 | +entry: |
| 60 | + %res = call fast double @llvm.cos.f64(double %x) |
| 61 | + %res.fadd = fadd fast double %res, %x |
| 62 | + ret double %res.fadd |
| 63 | +} |
| 64 | + |
| 65 | +declare double @llvm.cos.f64(double) |
| 66 | + |
| 67 | +; CHECK: remark: <unknown>:0:0: call from 'disable_tailcallopt' to 'streaming_callee' requires a streaming mode transition |
| 68 | +define void @disable_tailcallopt() nounwind { |
| 69 | + tail call void @streaming_callee() |
| 70 | + ret void; |
| 71 | +} |
| 72 | + |
| 73 | +; CHECK: remark: <unknown>:0:0: call from 'call_to_non_streaming_pass_sve_objects' to 'foo' requires a streaming mode transition |
| 74 | +define i8 @call_to_non_streaming_pass_sve_objects(ptr nocapture noundef readnone %ptr) #0 { |
| 75 | +entry: |
| 76 | + %Data1 = alloca <vscale x 16 x i8>, align 16 |
| 77 | + %Data2 = alloca <vscale x 16 x i8>, align 16 |
| 78 | + %Data3 = alloca <vscale x 16 x i8>, align 16 |
| 79 | + %0 = tail call i64 @llvm.aarch64.sme.cntsb() |
| 80 | + call void @foo(ptr noundef nonnull %Data1, ptr noundef nonnull %Data2, ptr noundef nonnull %Data3, i64 noundef %0) |
| 81 | + %1 = load <vscale x 16 x i8>, ptr %Data1, align 16 |
| 82 | + %vecext = extractelement <vscale x 16 x i8> %1, i64 0 |
| 83 | + ret i8 %vecext |
| 84 | +} |
| 85 | + |
| 86 | +declare i64 @llvm.aarch64.sme.cntsb() |
| 87 | + |
| 88 | +declare void @foo(ptr noundef, ptr noundef, ptr noundef, i64 noundef) |
| 89 | + |
| 90 | +attributes #0 = { nounwind vscale_range(1,16) "aarch64_pstate_sm_enabled" } |
0 commit comments