forked from OSchip/llvm-project
[LoopInterchange] Add remark for calls preventing interchanging.
It also updates test/Transforms/LoopInterchange/call-instructions.ll to use accesses where we can prove dependence after D35430. Reviewers: sebpop, karthikthecool, blitz.opensource Reviewed By: sebpop Differential Revision: https://reviews.llvm.org/D45206 llvm-svn: 329111
This commit is contained in:
parent
9114eb40b5
commit
9467ccf447
|
@ -985,6 +985,13 @@ bool LoopInterchangeLegality::canInterchangeLoops(unsigned InnerLoopId,
|
|||
continue;
|
||||
DEBUG(dbgs() << "Loops with call instructions cannot be interchanged "
|
||||
<< "safely.");
|
||||
ORE->emit([&]() {
|
||||
return OptimizationRemarkMissed(DEBUG_TYPE, "CallInst",
|
||||
CI->getDebugLoc(),
|
||||
CI->getParent())
|
||||
<< "Cannot interchange loops due to call instruction.";
|
||||
});
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
|
|
|
@ -1,5 +1,5 @@
|
|||
; REQUIRES: asserts
|
||||
; RUN: opt < %s -basicaa -loop-interchange -verify-dom-info -S -debug 2>&1 | FileCheck %s
|
||||
; RUN: opt < %s -basicaa -loop-interchange -pass-remarks-missed='loop-interchange' -pass-remarks-output=%t
|
||||
; RUN: FileCheck --input-file=%t %s
|
||||
|
||||
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
|
||||
target triple = "x86_64-unknown-linux-gnu"
|
||||
|
@ -13,42 +13,37 @@ declare void @bar(i64 %a) readnone
|
|||
;; Not safe to interchange, because the called function `foo` is not marked as
|
||||
;; readnone, so it could introduce dependences.
|
||||
;;
|
||||
;; for(int i=0;i<N;i++) {
|
||||
;; for(int j=1;j<N;j++) {
|
||||
;; for(int i=0;i<100;i++) {
|
||||
;; for(int j=1;j<100;j++) {
|
||||
;; foo(i);
|
||||
;; A[j][i] = A[j][i]+k;
|
||||
;; }
|
||||
;; }
|
||||
|
||||
; CHECK: Not interchanging loops. Cannot prove legality.
|
||||
; CHECK: --- !Missed
|
||||
; CHECK-NEXT: Pass: loop-interchange
|
||||
; CHECK-NEXT: Name: CallInst
|
||||
; CHECK-NEXT: Function: interchange_01
|
||||
; CHECK-NEXT: Args:
|
||||
; CHECK-NEXT - String: Cannot interchange loops due to call instruction.
|
||||
|
||||
define void @interchange_01(i32 %k, i32 %N) {
|
||||
define void @interchange_01(i32 %k) {
|
||||
entry:
|
||||
%cmp21 = icmp sgt i32 %N, 0
|
||||
br i1 %cmp21, label %for1.ph, label %exit
|
||||
|
||||
for1.ph:
|
||||
%cmp219 = icmp sgt i32 %N, 1
|
||||
%0 = add i32 %N, -1
|
||||
br label %for1.header
|
||||
|
||||
for1.header:
|
||||
%indvars.iv23 = phi i64 [ 0, %for1.ph ], [ %indvars.iv.next24, %for1.inc10 ]
|
||||
br i1 %cmp219, label %for2.ph, label %for1.inc10
|
||||
|
||||
for2.ph:
|
||||
%indvars.iv23 = phi i64 [ 0, %entry ], [ %indvars.iv.next24, %for1.inc10 ]
|
||||
br label %for2
|
||||
|
||||
for2:
|
||||
%indvars.iv = phi i64 [ %indvars.iv.next, %for2 ], [ 1, %for2.ph ]
|
||||
%indvars.iv = phi i64 [ %indvars.iv.next, %for2 ], [ 1, %for1.header ]
|
||||
call void @foo(i64 %indvars.iv23)
|
||||
%arrayidx5 = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* @A, i64 0, i64 %indvars.iv, i64 %indvars.iv23
|
||||
%1 = load i32, i32* %arrayidx5
|
||||
%add = add nsw i32 %1, %k
|
||||
%lv = load i32, i32* %arrayidx5
|
||||
%add = add nsw i32 %lv, %k
|
||||
store i32 %add, i32* %arrayidx5
|
||||
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
|
||||
%lftr.wideiv = trunc i64 %indvars.iv to i32
|
||||
%exitcond = icmp eq i32 %lftr.wideiv, %0
|
||||
%exitcond = icmp eq i64 %indvars.iv, 99
|
||||
br i1 %exitcond, label %for2.loopexit , label %for2
|
||||
|
||||
for2.loopexit:
|
||||
|
@ -56,8 +51,7 @@ for2.loopexit:
|
|||
|
||||
for1.inc10:
|
||||
%indvars.iv.next24 = add nuw nsw i64 %indvars.iv23, 1
|
||||
%lftr.wideiv25 = trunc i64 %indvars.iv23 to i32
|
||||
%exitcond26 = icmp eq i32 %lftr.wideiv25, %0
|
||||
%exitcond26 = icmp eq i64 %indvars.iv23, 99
|
||||
br i1 %exitcond26, label %for1.loopexit, label %for1.header
|
||||
|
||||
for1.loopexit:
|
||||
|
@ -71,42 +65,38 @@ exit:
|
|||
;; Safe to interchange, because the called function `bar` is marked as readnone,
|
||||
;; so it cannot introduce dependences.
|
||||
;;
|
||||
;; for(int i=0;i<N;i++) {
|
||||
;; for(int j=1;j<N;j++) {
|
||||
;; for(int i=0;i<100;i++) {
|
||||
;; for(int j=1;j<100;j++) {
|
||||
;; bar(i);
|
||||
;; A[j][i] = A[j][i]+k;
|
||||
;; }
|
||||
;; }
|
||||
|
||||
; CHECK: Not interchanging loops. Cannot prove legality.
|
||||
; CHECK: --- !Passed
|
||||
; CHECK-NEXT: Pass: loop-interchange
|
||||
; CHECK-NEXT: Name: Interchanged
|
||||
; CHECK-NEXT: Function: interchange_02
|
||||
; CHECK-NEXT: Args:
|
||||
; CHECK-NEXT: - String: Loop interchanged with enclosing loop.
|
||||
; CHECK-NEXT: ...
|
||||
|
||||
define void @interchange_02(i32 %k, i32 %N) {
|
||||
define void @interchange_02(i32 %k) {
|
||||
entry:
|
||||
%cmp21 = icmp sgt i32 %N, 0
|
||||
br i1 %cmp21, label %for1.ph, label %exit
|
||||
|
||||
for1.ph:
|
||||
%cmp219 = icmp sgt i32 %N, 1
|
||||
%0 = add i32 %N, -1
|
||||
br label %for1.header
|
||||
|
||||
for1.header:
|
||||
%indvars.iv23 = phi i64 [ 0, %for1.ph ], [ %indvars.iv.next24, %for1.inc10 ]
|
||||
br i1 %cmp219, label %for2.ph, label %for1.inc10
|
||||
|
||||
for2.ph:
|
||||
%indvars.iv23 = phi i64 [ 0, %entry ], [ %indvars.iv.next24, %for1.inc10 ]
|
||||
br label %for2
|
||||
|
||||
for2:
|
||||
%indvars.iv = phi i64 [ %indvars.iv.next, %for2 ], [ 1, %for2.ph ]
|
||||
%indvars.iv = phi i64 [ %indvars.iv.next, %for2 ], [ 1, %for1.header ]
|
||||
call void @bar(i64 %indvars.iv23)
|
||||
%arrayidx5 = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* @A, i64 0, i64 %indvars.iv, i64 %indvars.iv23
|
||||
%1 = load i32, i32* %arrayidx5
|
||||
%add = add nsw i32 %1, %k
|
||||
%lv = load i32, i32* %arrayidx5
|
||||
%add = add nsw i32 %lv, %k
|
||||
store i32 %add, i32* %arrayidx5
|
||||
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
|
||||
%lftr.wideiv = trunc i64 %indvars.iv to i32
|
||||
%exitcond = icmp eq i32 %lftr.wideiv, %0
|
||||
%exitcond = icmp eq i64 %indvars.iv, 99
|
||||
br i1 %exitcond, label %for2.loopexit , label %for2
|
||||
|
||||
for2.loopexit:
|
||||
|
@ -114,8 +104,7 @@ for2.loopexit:
|
|||
|
||||
for1.inc10:
|
||||
%indvars.iv.next24 = add nuw nsw i64 %indvars.iv23, 1
|
||||
%lftr.wideiv25 = trunc i64 %indvars.iv23 to i32
|
||||
%exitcond26 = icmp eq i32 %lftr.wideiv25, %0
|
||||
%exitcond26 = icmp eq i64 %indvars.iv23, 99
|
||||
br i1 %exitcond26, label %for1.loopexit, label %for1.header
|
||||
|
||||
for1.loopexit:
|
||||
|
|
Loading…
Reference in New Issue