87 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly<no-default-opts>' -polly-codegen-perf-monitoring -S < %s | FileCheck %s2 3; void f(long A[], long N) {4; long i;5; if (true)6; for (i = 0; i < N; ++i)7; A[i] = i;8; }9 10target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"11target triple = "x86_64-unknown-linux-gnu"12 13define void @f(ptr %A, i64 %N) nounwind {14entry:15 fence seq_cst16 br label %next17 18next:19 br i1 true, label %for.i, label %return20 21for.i:22 %indvar = phi i64 [ 0, %next], [ %indvar.next, %for.i ]23 %scevgep = getelementptr i64, ptr %A, i64 %indvar24 store i64 %indvar, ptr %scevgep25 %indvar.next = add nsw i64 %indvar, 126 %exitcond = icmp eq i64 %indvar.next, %N27 br i1 %exitcond, label %return, label %for.i28 29return:30 fence seq_cst31 ret void32}33 34; CHECK: @__polly_perf_cycles_total_start = weak thread_local(initialexec) constant i64 035; CHECK-NEXT: @__polly_perf_initialized = weak thread_local(initialexec) constant i1 false36; CHECK-NEXT: @__polly_perf_cycles_in_scops = weak thread_local(initialexec) constant i64 037; CHECK-NEXT: @__polly_perf_cycles_in_scop_start = weak thread_local(initialexec) constant i64 038 39; CHECK: polly.split_new_and_old: ; preds = %entry40; CHECK-NEXT: %0 = call { i64, i32 } @llvm.x86.rdtscp()41; CHECK-NEXT: %1 = extractvalue { i64, i32 } %0, 042; CHECK-NEXT: store volatile i64 %1, ptr @__polly_perf_cycles_in_scop_start43 44; CHECK: polly.merge_new_and_old: ; preds = %polly.exiting, %return.region_exiting45; CHECK-NEXT: %6 = load volatile i64, ptr @__polly_perf_cycles_in_scop_start46; CHECK-NEXT: %7 = call { i64, i32 } @llvm.x86.rdtscp()47; CHECK-NEXT: %8 = extractvalue { i64, i32 } %7, 048; CHECK-NEXT: %9 = sub i64 %8, %649; CHECK-NEXT: %10 = load volatile i64, ptr @__polly_perf_cycles_in_scops50; CHECK-NEXT: %11 = add i64 %10, %951; CHECK-NEXT: store volatile i64 %11, ptr @__polly_perf_cycles_in_scops52 53 54; CHECK: define weak_odr void @__polly_perf_final() {55; CHECK-NEXT: start:56; CHECK-NEXT: %0 = call { i64, i32 } @llvm.x86.rdtscp()57; CHECK-NEXT: %1 = extractvalue { i64, i32 } %0, 058; CHECK-NEXT: %2 = load volatile i64, ptr @__polly_perf_cycles_total_start59; CHECK-NEXT: %3 = sub i64 %1, %260; CHECK-NEXT: %4 = load volatile i64, ptr @__polly_perf_cycles_in_scops61; CHECK-NEXT: %5 = call i32 (...) @printf(ptr @1, ptr addrspace(4) @0)62; CHECK-NEXT: %6 = call i32 @fflush(ptr null)63; CHECK-NEXT: %7 = call i32 (...) @printf(ptr @3, ptr addrspace(4) @2)64; CHECK-NEXT: %8 = call i32 @fflush(ptr null)65; CHECK-NEXT: %9 = call i32 (...) @printf(ptr @6, ptr addrspace(4) @4, i64 %3, ptr addrspace(4) @5)66; CHECK-NEXT: %10 = call i32 @fflush(ptr null)67; CHECK-NEXT: %11 = call i32 (...) @printf(ptr @9, ptr addrspace(4) @7, i64 %4, ptr addrspace(4) @8)68; CHECK-NEXT: %12 = call i32 @fflush(ptr null)69 70 71; CHECK: define weak_odr void @__polly_perf_init() {72; CHECK-NEXT: start:73; CHECK-NEXT: %0 = load i1, ptr @__polly_perf_initialized74; CHECK-NEXT: br i1 %0, label %earlyreturn, label %initbb75 76; CHECK: earlyreturn: ; preds = %start77; CHECK-NEXT: ret void78 79; CHECK: initbb: ; preds = %start80; CHECK-NEXT: store i1 true, ptr @__polly_perf_initialized81; CHECK-NEXT: %1 = call i32 @atexit(ptr @__polly_perf_final)82; CHECK-NEXT: %2 = call { i64, i32 } @llvm.x86.rdtscp()83; CHECK-NEXT: %3 = extractvalue { i64, i32 } %2, 084; CHECK-NEXT: store volatile i64 %3, ptr @__polly_perf_cycles_total_start85; CHECK-NEXT: ret void86; CHECK-NEXT: }87