maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll

50 lines
1.8 KiB
LLVM
Raw Normal View History

[LoopUnroll] Switch the default value of -unroll-runtime-epilog back to its original value. As agreed in post-commit review of r265388, I'm switching the flag to its original value until the 90% runtime performance regression on SingleSource/Benchmarks/Stanford/Bubblesort is addressed. llvm-svn: 277524
2016-08-03 05:24:14 +08:00
; RUN: opt < %s -S -unroll-runtime -unroll-count=2 -loop-unroll -unroll-runtime-epilog=true | FileCheck %s -check-prefix=EPILOG
The patch fixes PR27392. Summary: It is incorrect to compare TripCount (which is BECount + 1) with extraiters (or Count) to check if we should enter unrolled loop or not, because TripCount can potentially overflow (when BECount is max unsigned integer). While comparing BECount with (Count - 1) is overflow safe and therefore correct. Reviewer: hfinkel Differential Revision: http://reviews.llvm.org/D19256 From: Evgeny Stupachenko <evstupac@gmail.com> llvm-svn: 267662
2016-04-27 11:04:54 +08:00
; RUN: opt < %s -S -unroll-runtime -unroll-count=2 -loop-unroll -unroll-runtime-epilog=false | FileCheck %s -check-prefix=PROLOG
Fix a trip-count overflow issue in LoopUnroll. Currently LoopUnroll generates a prologue loop before the main loop body to execute first N%UnrollFactor iterations. Also, this loop is used if trip-count can overflow - it's determined by a runtime check. However, we've been mistakenly optimizing this loop to a linear code for UnrollFactor = 2, not taking into account that it also serves as a safe version of the loop if its trip-count overflows. llvm-svn: 222451
2014-11-21 04:19:55 +08:00
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
Partial fix for bug 22589 Don't spend the entire iteration space in the scalar loop prologue if computing the trip count overflows. This change also gets rid of the backedge check in the prologue loop and the extra check for overflowing trip-count. Differential Revision: http://reviews.llvm.org/D7715 llvm-svn: 229731
2015-02-19 03:32:25 +08:00
; This test case documents how runtime loop unrolling handles the case
; when the backedge-count is -1.
; If %N, the backedge-taken count, is -1 then %0 unsigned-overflows
; and is 0. %xtraiter too is 0, signifying that the total trip-count
; is divisible by 2. The prologue then branches to the unrolled loop
; and executes the 2^32 iterations there, in groups of 2.
The patch fixes PR27392. Summary: It is incorrect to compare TripCount (which is BECount + 1) with extraiters (or Count) to check if we should enter unrolled loop or not, because TripCount can potentially overflow (when BECount is max unsigned integer). While comparing BECount with (Count - 1) is overflow safe and therefore correct. Reviewer: hfinkel Differential Revision: http://reviews.llvm.org/D19256 From: Evgeny Stupachenko <evstupac@gmail.com> llvm-svn: 267662
2016-04-27 11:04:54 +08:00
; EPILOG: entry:
Partial fix for bug 22589 Don't spend the entire iteration space in the scalar loop prologue if computing the trip count overflows. This change also gets rid of the backedge check in the prologue loop and the extra check for overflowing trip-count. Differential Revision: http://reviews.llvm.org/D7715 llvm-svn: 229731
2015-02-19 03:32:25 +08:00
The patch fixes PR27392. Summary: It is incorrect to compare TripCount (which is BECount + 1) with extraiters (or Count) to check if we should enter unrolled loop or not, because TripCount can potentially overflow (when BECount is max unsigned integer). While comparing BECount with (Count - 1) is overflow safe and therefore correct. Reviewer: hfinkel Differential Revision: http://reviews.llvm.org/D19256 From: Evgeny Stupachenko <evstupac@gmail.com> llvm-svn: 267662
2016-04-27 11:04:54 +08:00
; EPILOG-NEXT: %0 = add i32 %N, 1
; EPILOG-NEXT: %xtraiter = and i32 %0, 1
; EPILOG-NEXT: %1 = icmp ult i32 %N, 1
; EPILOG-NEXT: br i1 %1, label %while.end.unr-lcssa, label %entry.new
; EPILOG: while.body:
Partial fix for bug 22589 Don't spend the entire iteration space in the scalar loop prologue if computing the trip count overflows. This change also gets rid of the backedge check in the prologue loop and the extra check for overflowing trip-count. Differential Revision: http://reviews.llvm.org/D7715 llvm-svn: 229731
2015-02-19 03:32:25 +08:00
The patch fixes PR27392. Summary: It is incorrect to compare TripCount (which is BECount + 1) with extraiters (or Count) to check if we should enter unrolled loop or not, because TripCount can potentially overflow (when BECount is max unsigned integer). While comparing BECount with (Count - 1) is overflow safe and therefore correct. Reviewer: hfinkel Differential Revision: http://reviews.llvm.org/D19256 From: Evgeny Stupachenko <evstupac@gmail.com> llvm-svn: 267662
2016-04-27 11:04:54 +08:00
; EPILOG: %lcmp.mod = icmp ne i32 %xtraiter, 0
; EPILOG-NEXT: br i1 %lcmp.mod, label %while.body.epil.preheader, label %while.end
; EPILOG: while.body.epil:
Partial fix for bug 22589 Don't spend the entire iteration space in the scalar loop prologue if computing the trip count overflows. This change also gets rid of the backedge check in the prologue loop and the extra check for overflowing trip-count. Differential Revision: http://reviews.llvm.org/D7715 llvm-svn: 229731
2015-02-19 03:32:25 +08:00
The patch fixes PR27392. Summary: It is incorrect to compare TripCount (which is BECount + 1) with extraiters (or Count) to check if we should enter unrolled loop or not, because TripCount can potentially overflow (when BECount is max unsigned integer). While comparing BECount with (Count - 1) is overflow safe and therefore correct. Reviewer: hfinkel Differential Revision: http://reviews.llvm.org/D19256 From: Evgeny Stupachenko <evstupac@gmail.com> llvm-svn: 267662
2016-04-27 11:04:54 +08:00
; PROLOG: entry:
; PROLOG-NEXT: %0 = add i32 %N, 1
; PROLOG-NEXT: %xtraiter = and i32 %0, 1
; PROLOG-NEXT: %lcmp.mod = icmp ne i32 %xtraiter, 0
; PROLOG-NEXT: br i1 %lcmp.mod, label %while.body.prol.preheader, label %while.body.prol.loopexit
; PROLOG: while.body.prol:
; PROLOG: %1 = icmp ult i32 %N, 1
; PROLOG-NEXT: br i1 %1, label %while.end, label %entry.new
; PROLOG: while.body:
Fix a trip-count overflow issue in LoopUnroll. Currently LoopUnroll generates a prologue loop before the main loop body to execute first N%UnrollFactor iterations. Also, this loop is used if trip-count can overflow - it's determined by a runtime check. However, we've been mistakenly optimizing this loop to a linear code for UnrollFactor = 2, not taking into account that it also serves as a safe version of the loop if its trip-count overflows. llvm-svn: 222451
2014-11-21 04:19:55 +08:00
; Function Attrs: nounwind readnone ssp uwtable
Partial fix for bug 22589 Don't spend the entire iteration space in the scalar loop prologue if computing the trip count overflows. This change also gets rid of the backedge check in the prologue loop and the extra check for overflowing trip-count. Differential Revision: http://reviews.llvm.org/D7715 llvm-svn: 229731
2015-02-19 03:32:25 +08:00
define i32 @foo(i32 %N) {
Fix a trip-count overflow issue in LoopUnroll. Currently LoopUnroll generates a prologue loop before the main loop body to execute first N%UnrollFactor iterations. Also, this loop is used if trip-count can overflow - it's determined by a runtime check. However, we've been mistakenly optimizing this loop to a linear code for UnrollFactor = 2, not taking into account that it also serves as a safe version of the loop if its trip-count overflows. llvm-svn: 222451
2014-11-21 04:19:55 +08:00
entry:
br label %while.body
while.body: ; preds = %while.body, %entry
%i = phi i32 [ 0, %entry ], [ %inc, %while.body ]
%cmp = icmp eq i32 %i, %N
%inc = add i32 %i, 1
br i1 %cmp, label %while.end, label %while.body
while.end: ; preds = %while.body
ret i32 %i
}