257 lines · plain
1; RUN: opt < %s -passes=pgo-memop-opt -verify-dom-info -pgo-memop-count-threshold=90 -pgo-memop-percent-threshold=15 --pgo-memop-optimize-memcmp-bcmp -S | FileCheck %s --check-prefix=MEMOP_OPT2; RUN: opt < %s -passes=pgo-memop-opt -verify-dom-info -pgo-memop-count-threshold=90 -pgo-memop-percent-threshold=15 -pgo-memop-optimize-memcmp-bcmp -pass-remarks-with-hotness -pass-remarks-output=%t.opt.yaml -S | FileCheck %s --check-prefix=MEMOP_OPT3; RUN: FileCheck %s -input-file=%t.opt.yaml --check-prefix=YAML4; RUN: FileCheck %s -input-file=%t.opt.yaml --check-prefix=YAML5 6 7target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"8target triple = "x86_64-unknown-linux-gnu"9 10define void @foo(ptr %dst, ptr %src, ptr %dst2, ptr %src2, ptr %a, i32 %n) !prof !27 {11entry:12 br label %for.cond13 14for.cond:15 %i.0 = phi i32 [ 0, %entry ], [ %inc5, %for.inc4 ]16 %cmp = icmp slt i32 %i.0, %n17 br i1 %cmp, label %for.body, label %for.end6, !prof !2818 19for.body:20 br label %for.cond121 22for.cond1:23 %j.0 = phi i32 [ 0, %for.body ], [ %inc, %for.inc ]24 %idx.ext = sext i32 %i.0 to i6425 %add.ptr = getelementptr inbounds i32, ptr %a, i64 %idx.ext26 %0 = load i32, ptr %add.ptr, align 427 %cmp2 = icmp slt i32 %j.0, %028 br i1 %cmp2, label %for.body3, label %for.end, !prof !2929 30for.body3:31 %add = add nsw i32 %i.0, 132 %conv = sext i32 %add to i6433 call void @llvm.memcpy.p0.p0.i64(ptr %dst, ptr %src, i64 %conv, i1 false), !prof !3034 call void @llvm.memcpy.p0.p0.i64(ptr %dst2, ptr %src2, i64 %conv, i1 false), !prof !3135 br label %for.inc36 37; MEMOP_OPT: switch i64 %conv, label %[[DEFAULT_LABEL:.*]] [38; MEMOP_OPT: i64 0, label %[[CASE_1_LABEL:.*]]39; MEMOP_OPT: ], !prof [[SWITCH_BW:![0-9]+]] 40; MEMOP_OPT: [[CASE_1_LABEL]]:41; MEMOP_OPT: call void @llvm.memcpy.p0.p0.i64(ptr %dst, ptr %src, i64 0, i1 false)42; MEMOP_OPT: br label %[[MERGE_LABEL:.*]]43; MEMOP_OPT: [[DEFAULT_LABEL]]:44; MEMOP_OPT: call void @llvm.memcpy.p0.p0.i64(ptr %dst, ptr %src, i64 %conv, i1 false), !prof [[NEWVP:![0-9]+]]45; MEMOP_OPT: br label %[[MERGE_LABEL]]46; MEMOP_OPT: [[MERGE_LABEL]]:47; MEMOP_OPT: switch i64 %conv, label %[[DEFAULT_LABEL2:.*]] [48; MEMOP_OPT: i64 0, label %[[CASE_1_LABEL2:.*]]49; MEMOP_OPT: ], !prof [[SWITCH_BW:![0-9]+]] 50; MEMOP_OPT: [[CASE_1_LABEL2]]:51; MEMOP_OPT: call void @llvm.memcpy.p0.p0.i64(ptr %dst2, ptr %src2, i64 0, i1 false)52; MEMOP_OPT: br label %[[MERGE_LABEL2:.*]]53; MEMOP_OPT: [[DEFAULT_LABEL2]]:54; MEMOP_OPT: call void @llvm.memcpy.p0.p0.i64(ptr %dst2, ptr %src2, i64 %conv, i1 false), !prof [[NEWVP]]55; MEMOP_OPT: br label %[[MERGE_LABEL2]]56; MEMOP_OPT: [[MERGE_LABEL2]]:57; MEMOP_OPT: br label %for.inc58 59for.inc:60 %inc = add nsw i32 %j.0, 161 br label %for.cond162 63for.end:64 br label %for.inc465 66for.inc4:67 %inc5 = add nsw i32 %i.0, 168 br label %for.cond69 70for.end6:71 ret void72}73 74declare void @consume(i32 %v1, i32 %v2)75 76define void @foo_memcmp_bcmp(ptr %dst, ptr %src, ptr %dst2, ptr %src2, ptr %a, i32 %n) !prof !27 {77entry:78 br label %for.cond79 80for.cond:81 %i.0 = phi i32 [ 0, %entry ], [ %inc5, %for.inc4 ]82 %cmp = icmp slt i32 %i.0, %n83 br i1 %cmp, label %for.body, label %for.end6, !prof !2884 85for.body:86 br label %for.cond187 88for.cond1:89 %j.0 = phi i32 [ 0, %for.body ], [ %inc, %for.inc ]90 %idx.ext = sext i32 %i.0 to i6491 %add.ptr = getelementptr inbounds i32, ptr %a, i64 %idx.ext92 %0 = load i32, ptr %add.ptr, align 493 %cmp2 = icmp slt i32 %j.0, %094 br i1 %cmp2, label %for.body3, label %for.end, !prof !2995 96for.body3:97 %add = add nsw i32 %i.0, 198 %conv = sext i32 %add to i6499 %memcmp = call i32 @memcmp(ptr %dst, ptr %src, i64 %conv), !prof !30100 %bcmp = call i32 @bcmp(ptr %dst2, ptr %src2, i64 %conv), !prof !31101 call void @consume(i32 %memcmp, i32 %bcmp)102 br label %for.inc103 104; MEMOP_OPT: switch i64 %conv, label %[[DEFAULT_LABEL:.*]] [105; MEMOP_OPT: i64 0, label %[[CASE_1_LABEL:.*]]106; MEMOP_OPT: ], !prof [[SWITCH_BW:![0-9]+]]107; MEMOP_OPT: [[CASE_1_LABEL]]:108; MEMOP_OPT: %[[RV:.*]] = call i32 @memcmp(ptr %dst, ptr %src, i64 0)109; MEMOP_OPT: br label %[[MERGE_LABEL:.*]]110; MEMOP_OPT: [[DEFAULT_LABEL]]:111; MEMOP_OPT: %[[RVD:.*]] = call i32 @memcmp(ptr %dst, ptr %src, i64 %conv), !prof [[NEWVP:![0-9]+]]112; MEMOP_OPT: br label %[[MERGE_LABEL]]113; MEMOP_OPT: [[MERGE_LABEL]]:114; MEMOP_OPT: %[[PHI:.*]] = phi i32 [ %[[RVD]], %[[DEFAULT_LABEL]] ], [ %[[RV]], %[[CASE_1_LABEL]] ]115; MEMOP_OPT: switch i64 %conv, label %[[DEFAULT_LABEL2:.*]] [116; MEMOP_OPT: i64 0, label %[[CASE_1_LABEL2:.*]]117; MEMOP_OPT: ], !prof [[SWITCH_BW:![0-9]+]]118; MEMOP_OPT: [[CASE_1_LABEL2]]:119; MEMOP_OPT: %[[RV2:.*]] = call i32 @bcmp(ptr %dst2, ptr %src2, i64 0)120; MEMOP_OPT: br label %[[MERGE_LABEL2:.*]]121; MEMOP_OPT: [[DEFAULT_LABEL2]]:122; MEMOP_OPT: %[[RVD2:.*]] = call i32 @bcmp(ptr %dst2, ptr %src2, i64 %conv), !prof [[NEWVP]]123; MEMOP_OPT: br label %[[MERGE_LABEL2]]124; MEMOP_OPT: [[MERGE_LABEL2]]:125; MEMOP_OPT: %[[PHI2:.*]] = phi i32 [ %[[RVD2]], %[[DEFAULT_LABEL2]] ], [ %[[RV2]], %[[CASE_1_LABEL2]] ]126; MEMOP_OPT: call void @consume(i32 %[[PHI]], i32 %[[PHI2]])127; MEMOP_OPT: br label %for.inc128 129for.inc:130 %inc = add nsw i32 %j.0, 1131 br label %for.cond1132 133for.end:134 br label %for.inc4135 136for.inc4:137 %inc5 = add nsw i32 %i.0, 1138 br label %for.cond139 140for.end6:141 ret void142}143 144; MEMOP_OPT: [[SWITCH_BW]] = !{!"branch_weights", i32 457, i32 99}145; Should be 457 total left (original total count 556, minus 99 from specialized146; value 0, which is removed from VP array. This should preserve all unpromoted values.147; MEMOP_OPT: [[NEWVP]] = !{!"VP", i32 1, i64 457, i64 2, i64 88, i64 3, i64 77, i64 9, i64 72, i64 4, i64 66, i64 5, i64 55, i64 6, i64 44, i64 7, i64 33, i64 8, i64 22}148 149!llvm.module.flags = !{!0}150 151!0 = !{i32 1, !"ProfileSummary", !1}152!1 = !{!2, !3, !4, !5, !6, !7, !8, !9}153!2 = !{!"ProfileFormat", !"InstrProf"}154!3 = !{!"TotalCount", i64 579}155!4 = !{!"MaxCount", i64 556}156!5 = !{!"MaxInternalCount", i64 20}157!6 = !{!"MaxFunctionCount", i64 556}158!7 = !{!"NumCounts", i64 6}159!8 = !{!"NumFunctions", i64 3}160!9 = !{!"DetailedSummary", !10}161!10 = !{!11, !12, !13, !14, !15, !16, !16, !17, !17, !18, !19, !20, !21, !22, !23, !24, !25, !26}162!11 = !{i32 10000, i64 556, i32 1}163!12 = !{i32 100000, i64 556, i32 1}164!13 = !{i32 200000, i64 556, i32 1}165!14 = !{i32 300000, i64 556, i32 1}166!15 = !{i32 400000, i64 556, i32 1}167!16 = !{i32 500000, i64 556, i32 1}168!17 = !{i32 600000, i64 556, i32 1}169!18 = !{i32 700000, i64 556, i32 1}170!19 = !{i32 800000, i64 556, i32 1}171!20 = !{i32 900000, i64 556, i32 1}172!21 = !{i32 950000, i64 556, i32 1}173!22 = !{i32 990000, i64 20, i32 2}174!23 = !{i32 999000, i64 1, i32 5}175!24 = !{i32 999900, i64 1, i32 5}176!25 = !{i32 999990, i64 1, i32 5}177!26 = !{i32 999999, i64 1, i32 5}178!27 = !{!"function_entry_count", i64 1}179!28 = !{!"branch_weights", i32 20, i32 1}180!29 = !{!"branch_weights", i32 556, i32 20}181!30 = !{!"VP", i32 1, i64 556, i64 0, i64 99, i64 2, i64 88, i64 3, i64 77, i64 9, i64 72, i64 4, i64 66, i64 5, i64 55, i64 6, i64 44, i64 7, i64 33, i64 8, i64 22}182!31 = !{!"VP", i32 1, i64 556, i64 0, i64 99, i64 2, i64 88, i64 3, i64 77, i64 9, i64 72, i64 4, i64 66, i64 5, i64 55, i64 6, i64 44, i64 7, i64 33, i64 8, i64 22}183 184declare void @llvm.lifetime.start(ptr nocapture)185 186declare void @llvm.memcpy.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1)187 188declare i32 @memcmp(ptr, ptr, i64)189declare i32 @bcmp(ptr, ptr, i64)190 191declare void @llvm.lifetime.end(ptr nocapture)192 193; YAML: --- !Passed194; YAML-NEXT: Pass: pgo-memop-opt195; YAML-NEXT: Name: memopt-opt196; YAML-NEXT: Function: foo197; YAML-NEXT: Hotness: 0198; YAML-NEXT: Args:199; YAML-NEXT: - String: 'optimized '200; YAML-NEXT: - Memop: memcpy201; YAML-NEXT: - String: ' with count '202; YAML-NEXT: - Count: '99'203; YAML-NEXT: - String: ' out of '204; YAML-NEXT: - Total: '556'205; YAML-NEXT: - String: ' for '206; YAML-NEXT: - Versions: '1'207; YAML-NEXT: - String: ' versions'208; YAML-NEXT: ...209; YAML-NEXT: --- !Passed210; YAML-NEXT: Pass: pgo-memop-opt211; YAML-NEXT: Name: memopt-opt212; YAML-NEXT: Function: foo213; YAML-NEXT: Hotness: 0214; YAML-NEXT: Args:215; YAML-NEXT: - String: 'optimized '216; YAML-NEXT: - Memop: memcpy217; YAML-NEXT: - String: ' with count '218; YAML-NEXT: - Count: '99'219; YAML-NEXT: - String: ' out of '220; YAML-NEXT: - Total: '556'221; YAML-NEXT: - String: ' for '222; YAML-NEXT: - Versions: '1'223; YAML-NEXT: - String: ' versions'224; YAML-NEXT: ...225; YAML-NEXT: --- !Passed226; YAML-NEXT: Pass: pgo-memop-opt227; YAML-NEXT: Name: memopt-opt228; YAML-NEXT: Function: foo_memcmp_bcmp229; YAML-NEXT: Hotness: 0230; YAML-NEXT: Args:231; YAML-NEXT: - String: 'optimized '232; YAML-NEXT: - Memop: memcmp233; YAML-NEXT: - String: ' with count '234; YAML-NEXT: - Count: '99'235; YAML-NEXT: - String: ' out of '236; YAML-NEXT: - Total: '556'237; YAML-NEXT: - String: ' for '238; YAML-NEXT: - Versions: '1'239; YAML-NEXT: - String: ' versions'240; YAML-NEXT: ...241; YAML-NEXT: --- !Passed242; YAML-NEXT: Pass: pgo-memop-opt243; YAML-NEXT: Name: memopt-opt244; YAML-NEXT: Function: foo_memcmp_bcmp245; YAML-NEXT: Hotness: 0246; YAML-NEXT: Args:247; YAML-NEXT: - String: 'optimized '248; YAML-NEXT: - Memop: bcmp249; YAML-NEXT: - String: ' with count '250; YAML-NEXT: - Count: '99'251; YAML-NEXT: - String: ' out of '252; YAML-NEXT: - Total: '556'253; YAML-NEXT: - String: ' for '254; YAML-NEXT: - Versions: '1'255; YAML-NEXT: - String: ' versions'256; YAML-NEXT: ...257