Files
clang-p2996/llvm/test/Transforms/PhaseOrdering/X86/speculation-vs-tbaa.ll
Nikita Popov a105877646 [InstCombine] Remove some of the complexity-based canonicalization (#91185)
The idea behind this canonicalization is that it allows us to handle less
patterns, because we know that some will be canonicalized away. This is
indeed very useful to e.g. know that constants are always on the right.

However, this is only useful if the canonicalization is actually
reliable. This is the case for constants, but not for arguments: Moving
these to the right makes it look like the "more complex" expression is
guaranteed to be on the left, but this is not actually the case in
practice. It fails as soon as you replace the argument with another
instruction.

The end result is that it looks like things correctly work in tests,
while they actually don't. We use the "thwart complexity-based
canonicalization" trick to handle this in tests, but it's often a
challenge for new contributors to get this right, and based on the
regressions this PR originally exposed, we clearly don't get this right
in many cases.

For this reason, I think that it's better to remove this complexity
canonicalization. It will make it much easier to write tests for
commuted cases and make sure that they are handled.
2024-08-21 12:02:54 +02:00

97 lines
4.9 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes="default<O1>" -S < %s | FileCheck --check-prefixes=O1 %s
; RUN: opt -passes="default<O2>" -S < %s | FileCheck --check-prefixes=O23 %s
; RUN: opt -passes="default<O3>" -S < %s | FileCheck --check-prefixes=O23 %s
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
; We should retain the TBAA on the load here, not lose it.
define void @licm(ptr align 8 dereferenceable(8) %_M_start.i, i64 %numElem) {
; O1-LABEL: @licm(
; O1-NEXT: entry:
; O1-NEXT: [[CMP1_NOT:%.*]] = icmp eq i64 [[NUMELEM:%.*]], 0
; O1-NEXT: br i1 [[CMP1_NOT]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY_LR_PH:%.*]]
; O1: for.body.lr.ph:
; O1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[_M_START_I:%.*]], align 8, !tbaa [[TBAA3:![0-9]+]]
; O1-NEXT: br label [[FOR_BODY:%.*]]
; O1: for.body:
; O1-NEXT: [[K_02:%.*]] = phi i64 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
; O1-NEXT: [[ADD_PTR_I:%.*]] = getelementptr inbounds double, ptr [[TMP0]], i64 [[K_02]]
; O1-NEXT: store double 2.000000e+00, ptr [[ADD_PTR_I]], align 8, !tbaa [[TBAA8:![0-9]+]]
; O1-NEXT: [[INC]] = add nuw i64 [[K_02]], 1
; O1-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[NUMELEM]]
; O1-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]]
; O1: for.cond.cleanup:
; O1-NEXT: ret void
;
; O23-LABEL: @licm(
; O23-NEXT: entry:
; O23-NEXT: [[CMP1_NOT:%.*]] = icmp eq i64 [[NUMELEM:%.*]], 0
; O23-NEXT: br i1 [[CMP1_NOT]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY_LR_PH:%.*]]
; O23: for.body.lr.ph:
; O23-NEXT: [[TMP0:%.*]] = load ptr, ptr [[_M_START_I:%.*]], align 8, !tbaa [[TBAA3:![0-9]+]]
; O23-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[NUMELEM]], 4
; O23-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY_PREHEADER:%.*]], label [[VECTOR_PH:%.*]]
; O23: vector.ph:
; O23-NEXT: [[N_VEC:%.*]] = and i64 [[NUMELEM]], -4
; O23-NEXT: br label [[VECTOR_BODY:%.*]]
; O23: vector.body:
; O23-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; O23-NEXT: [[TMP1:%.*]] = getelementptr inbounds double, ptr [[TMP0]], i64 [[INDEX]]
; O23-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 16
; O23-NEXT: store <2 x double> <double 2.000000e+00, double 2.000000e+00>, ptr [[TMP1]], align 8, !tbaa [[TBAA8:![0-9]+]]
; O23-NEXT: store <2 x double> <double 2.000000e+00, double 2.000000e+00>, ptr [[TMP2]], align 8, !tbaa [[TBAA8]]
; O23-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; O23-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; O23-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; O23: middle.block:
; O23-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[NUMELEM]], [[N_VEC]]
; O23-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY_PREHEADER]]
; O23: for.body.preheader:
; O23-NEXT: [[K_02_PH:%.*]] = phi i64 [ 0, [[FOR_BODY_LR_PH]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
; O23-NEXT: br label [[FOR_BODY:%.*]]
; O23: for.body:
; O23-NEXT: [[K_02:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[K_02_PH]], [[FOR_BODY_PREHEADER]] ]
; O23-NEXT: [[ADD_PTR_I:%.*]] = getelementptr inbounds double, ptr [[TMP0]], i64 [[K_02]]
; O23-NEXT: store double 2.000000e+00, ptr [[ADD_PTR_I]], align 8, !tbaa [[TBAA8]]
; O23-NEXT: [[INC]] = add nuw i64 [[K_02]], 1
; O23-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[NUMELEM]]
; O23-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
; O23: for.cond.cleanup:
; O23-NEXT: ret void
;
entry:
br label %for.cond
for.cond: ; preds = %for.body, %entry
%k.0 = phi i64 [ 0, %entry ], [ %inc, %for.body ]
%cmp = icmp ult i64 %k.0, %numElem
br i1 %cmp, label %for.body, label %for.cond.cleanup
for.body: ; preds = %for.cond
%0 = load ptr, ptr %_M_start.i, align 8, !tbaa !3
%add.ptr.i = getelementptr inbounds double, ptr %0, i64 %k.0
store double 2.000000e+00, ptr %add.ptr.i, align 8, !tbaa !8
%inc = add nuw i64 %k.0, 1
br label %for.cond
for.cond.cleanup: ; preds = %for.cond
ret void
}
!llvm.module.flags = !{!0, !1}
!llvm.ident = !{!2}
!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{i32 7, !"uwtable", i32 1}
!2 = !{!"clang version 15.0.0 (https://github.com/llvm/llvm-project.git fc510998f7c287df2bc1304673e0cd8452d50b31)"}
!3 = !{!4, !5, i64 0}
!4 = !{!"_ZTSNSt12_Vector_baseIdSaIdEE17_Vector_impl_dataE", !5, i64 0, !5, i64 8, !5, i64 16}
!5 = !{!"any pointer", !6, i64 0}
!6 = !{!"omnipotent char", !7, i64 0}
!7 = !{!"Simple C++ TBAA"}
!8 = !{!9, !9, i64 0}
!9 = !{!"double", !6, i64 0}