maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/polly/test/CodeGen/multiple_sai_fro_same_base_...

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

45 lines
2.7 KiB
LLVM
Raw Normal View History

[BlockGenerator] Fix PHI merges for MK_Arrays. Value merging is only necessary for scalars when they are used outside of the scop. While an array's base pointer can be used after the scop, it gets an extra ScopArrayInfo of type MK_Value. We used to generate phi's for both of them, where one was assuming the reault of the other phi would be the original value, because it has already been replaced by the previous phi. This resulted in IR that the current IR verifier allows, but is probably illegal. This reduces the number of LNT test-suite fails with -polly-position=before-vectorizer -polly-process-unprofitable from 16 to 10. Also see llvm.org/PR26718. llvm-svn: 262629
2016-03-04 01:20:43 +08:00
; RUN: opt %loadPolly -polly-position=before-vectorizer -polly-scops -analyze < %s | FileCheck %s --check-prefix=SCOP
; RUN: opt %loadPolly -polly-position=before-vectorizer -polly-codegen -S < %s | FileCheck %s --check-prefix=IR
; The IR has two ScopArrayInfo for the value %next.0. This used to produce two
; phi nodes in polly.merge_new_and_old, one illegaly using the result of the
; other. There must be only one merge phi, no need to generate them for arrays
; of type MK_Array.
; Derived from test-suite/MultiSource/Applications/siod/slib.c
%struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962 = type { i16, i16, %union.anon.1.289.321.337.353.481.545.593.625.817.897.913.961 }
%union.anon.1.289.321.337.353.481.545.593.625.817.897.913.961 = type { %struct.anon.0.288.320.336.352.480.544.592.624.816.896.912.960 }
%struct.anon.0.288.320.336.352.480.544.592.624.816.896.912.960 = type { %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962*, %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962* }
define void @leval_or() {
entry:
br label %while.cond
while.cond: ; preds = %sw.bb1.i30, %cond.end.i28, %entry
%next.0 = phi %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962* [ null, %entry ], [ %1, %sw.bb1.i30 ], [ null, %cond.end.i28 ]
br i1 undef, label %cond.end.i28, label %if.then
if.then: ; preds = %while.cond
ret void
cond.end.i28: ; preds = %while.cond
%type.i24 = getelementptr inbounds %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962, %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962* %next.0, i64 0, i32 1
%0 = load i16, i16* %type.i24, align 2
br i1 false, label %sw.bb1.i30, label %while.cond
sw.bb1.i30: ; preds = %cond.end.i28
%cdr.i29 = getelementptr inbounds %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962, %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962* %next.0, i64 0, i32 2, i32 0, i32 1
%1 = load %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962*, %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962** %cdr.i29, align 8
br label %while.cond
}
; SCOP: Arrays {
; SCOP-NEXT: %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962* MemRef_next_0;
[FIX] Ensure SAI objects for exit PHIs If all exiting blocks of a SCoP are error blocks and therefor not represented we will not generate accesses and consequently no SAI objects for exit PHIs. However, they are needed in the code generation to generate the merge PHIs between the original and optimized region. With this patch we enusre that the SAI objects for exit PHIs exist even if all exiting blocks turn out to be eror blocks. This fixes the crash reported in PR27207. llvm-svn: 265393
2016-04-05 21:44:21 +08:00
; SCOP-NEXT: i16 MemRef_next_0[*];
[BlockGenerator] Fix PHI merges for MK_Arrays. Value merging is only necessary for scalars when they are used outside of the scop. While an array's base pointer can be used after the scop, it gets an extra ScopArrayInfo of type MK_Value. We used to generate phi's for both of them, where one was assuming the reault of the other phi would be the original value, because it has already been replaced by the previous phi. This resulted in IR that the current IR verifier allows, but is probably illegal. This reduces the number of LNT test-suite fails with -polly-position=before-vectorizer -polly-process-unprofitable from 16 to 10. Also see llvm.org/PR26718. llvm-svn: 262629
2016-03-04 01:20:43 +08:00
; SCOP-NEXT: }
; IR: polly.merge_new_and_old:
; IR-NEXT: %next.0.ph.merge = phi %struct.obj.2.290.322.338.354.482.546.594.626.818.898.914.962* [ %next.0.ph.final_reload, %polly.exiting ], [ %next.0.ph, %while.cond.region_exiting ]
; IR-NEXT: %indvar.next = add i64 %indvar, 1
; IR-NEXT: br label %while.cond