2011-01-09 10:28:48 +08:00
|
|
|
; RUN: opt < %s -analyze -scalar-evolution | FileCheck %s
|
2009-07-25 09:22:26 +08:00
|
|
|
|
|
|
|
|
; The addrecs in this loop are analyzable only by using nsw information.
|
|
|
|
|
|
|
|
|
|
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64"
|
|
|
|
|
|
2011-01-09 10:28:48 +08:00
|
|
|
; CHECK: Classifying expressions for: @test1
|
|
|
|
|
define void @test1(double* %p) nounwind {
|
2009-07-25 09:22:26 +08:00
|
|
|
entry:
|
2015-02-28 05:17:42 +08:00
|
|
|
%tmp = load double, double* %p, align 8 ; <double> [#uses=1]
|
2009-07-25 09:22:26 +08:00
|
|
|
%tmp1 = fcmp ogt double %tmp, 2.000000e+00 ; <i1> [#uses=1]
|
|
|
|
|
br i1 %tmp1, label %bb.nph, label %return
|
|
|
|
|
|
|
|
|
|
bb.nph: ; preds = %entry
|
|
|
|
|
br label %bb
|
|
|
|
|
|
|
|
|
|
bb: ; preds = %bb1, %bb.nph
|
|
|
|
|
%i.01 = phi i32 [ %tmp8, %bb1 ], [ 0, %bb.nph ] ; <i32> [#uses=3]
|
2011-01-09 10:28:48 +08:00
|
|
|
; CHECK: %i.01
|
|
|
|
|
; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%bb>
|
2009-07-25 09:22:26 +08:00
|
|
|
%tmp2 = sext i32 %i.01 to i64 ; <i64> [#uses=1]
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%tmp3 = getelementptr double, double* %p, i64 %tmp2 ; <double*> [#uses=1]
|
2015-02-28 05:17:42 +08:00
|
|
|
%tmp4 = load double, double* %tmp3, align 8 ; <double> [#uses=1]
|
2009-07-25 09:22:26 +08:00
|
|
|
%tmp5 = fmul double %tmp4, 9.200000e+00 ; <double> [#uses=1]
|
|
|
|
|
%tmp6 = sext i32 %i.01 to i64 ; <i64> [#uses=1]
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%tmp7 = getelementptr double, double* %p, i64 %tmp6 ; <double*> [#uses=1]
|
2011-01-09 10:28:48 +08:00
|
|
|
; CHECK: %tmp7
|
|
|
|
|
; CHECK-NEXT: --> {%p,+,8}<%bb>
|
2009-07-25 09:22:26 +08:00
|
|
|
store double %tmp5, double* %tmp7, align 8
|
2009-07-28 00:11:46 +08:00
|
|
|
%tmp8 = add nsw i32 %i.01, 1 ; <i32> [#uses=2]
|
2011-01-09 10:28:48 +08:00
|
|
|
; CHECK: %tmp8
|
|
|
|
|
; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%bb>
|
2016-05-29 08:32:17 +08:00
|
|
|
%p.gep = getelementptr double, double* %p, i32 %tmp8
|
|
|
|
|
%p.val = load double, double* %p.gep
|
2009-07-25 09:22:26 +08:00
|
|
|
br label %bb1
|
|
|
|
|
|
|
|
|
|
bb1: ; preds = %bb
|
|
|
|
|
%phitmp = sext i32 %tmp8 to i64 ; <i64> [#uses=1]
|
2011-01-09 10:28:48 +08:00
|
|
|
; CHECK: %phitmp
|
2011-03-15 08:37:00 +08:00
|
|
|
; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%bb>
|
2016-05-29 08:32:17 +08:00
|
|
|
%tmp9 = getelementptr inbounds double, double* %p, i64 %phitmp ; <double*> [#uses=1]
|
2011-01-09 10:28:48 +08:00
|
|
|
; CHECK: %tmp9
|
2016-05-29 08:32:17 +08:00
|
|
|
; CHECK-NEXT: --> {(8 + %p)<nsw>,+,8}<nsw><%bb>
|
2015-02-28 05:17:42 +08:00
|
|
|
%tmp10 = load double, double* %tmp9, align 8 ; <double> [#uses=1]
|
2009-07-25 09:22:26 +08:00
|
|
|
%tmp11 = fcmp ogt double %tmp10, 2.000000e+00 ; <i1> [#uses=1]
|
|
|
|
|
br i1 %tmp11, label %bb, label %bb1.return_crit_edge
|
|
|
|
|
|
|
|
|
|
bb1.return_crit_edge: ; preds = %bb1
|
|
|
|
|
br label %return
|
|
|
|
|
|
|
|
|
|
return: ; preds = %bb1.return_crit_edge, %entry
|
|
|
|
|
ret void
|
|
|
|
|
}
|
2011-01-09 10:28:48 +08:00
|
|
|
|
|
|
|
|
; CHECK: Classifying expressions for: @test2
|
|
|
|
|
define void @test2(i32* %begin, i32* %end) ssp {
|
|
|
|
|
entry:
|
|
|
|
|
%cmp1.i.i = icmp eq i32* %begin, %end
|
|
|
|
|
br i1 %cmp1.i.i, label %_ZSt4fillIPiiEvT_S1_RKT0_.exit, label %for.body.lr.ph.i.i
|
|
|
|
|
|
|
|
|
|
for.body.lr.ph.i.i: ; preds = %entry
|
|
|
|
|
br label %for.body.i.i
|
|
|
|
|
|
|
|
|
|
for.body.i.i: ; preds = %for.body.i.i, %for.body.lr.ph.i.i
|
|
|
|
|
%__first.addr.02.i.i = phi i32* [ %begin, %for.body.lr.ph.i.i ], [ %ptrincdec.i.i, %for.body.i.i ]
|
|
|
|
|
; CHECK: %__first.addr.02.i.i
|
2013-10-28 15:30:06 +08:00
|
|
|
; CHECK-NEXT: --> {%begin,+,4}<nuw><%for.body.i.i>
|
2011-01-09 10:28:48 +08:00
|
|
|
store i32 0, i32* %__first.addr.02.i.i, align 4
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ptrincdec.i.i = getelementptr inbounds i32, i32* %__first.addr.02.i.i, i64 1
|
2011-01-09 10:28:48 +08:00
|
|
|
; CHECK: %ptrincdec.i.i
|
[SCEV] No-wrap flags are not propagated when folding "{S,+,X}+T ==> {S+T,+,X}"
Summary:
**Description**
This makes `WidenIV::widenIVUse` (IndVarSimplify.cpp) fail to widen narrow IV uses in some cases. The latter affects IndVarSimplify which may not eliminate narrow IV's when there actually exists such a possibility, thereby producing ineffective code.
When `WidenIV::widenIVUse` gets a NarrowUse such as `{(-2 + %inc.lcssa),+,1}<nsw><%for.body3>`, it first tries to get a wide recurrence for it via the `getWideRecurrence` call.
`getWideRecurrence` returns recurrence like this: `{(sext i32 (-2 + %inc.lcssa) to i64),+,1}<nsw><%for.body3>`.
Then a wide use operation is generated by `cloneIVUser`. The generated wide use is evaluated to `{(-2 + (sext i32 %inc.lcssa to i64))<nsw>,+,1}<nsw><%for.body3>`, which is different from the `getWideRecurrence` result. `cloneIVUser` sees the difference and returns nullptr.
This patch also fixes the broken LLVM tests by adding missing <nsw> entries introduced by the correction.
**Minimal reproducer:**
```
int foo(int a, int b, int c);
int baz();
void bar()
{
int arr[20];
int i = 0;
for (i = 0; i < 4; ++i)
arr[i] = baz();
for (; i < 20; ++i)
arr[i] = foo(arr[i - 4], arr[i - 3], arr[i - 2]);
}
```
**Clang command line:**
```
clang++ -mllvm -debug -S -emit-llvm -O3 --target=aarch64-linux-elf test.cpp -o test.ir
```
**Expected result:**
The ` -mllvm -debug` log shows that all the IV's for the second `for` loop have been eliminated.
Reviewers: sanjoy
Subscribers: atrick, asl, aemerson, mzolotukhin, llvm-commits
Differential Revision: http://reviews.llvm.org/D20058
llvm-svn: 270695
2016-05-25 21:01:33 +08:00
|
|
|
; CHECK-NEXT: --> {(4 + %begin)<nsw>,+,4}<nuw><%for.body.i.i>
|
2011-01-09 10:28:48 +08:00
|
|
|
%cmp.i.i = icmp eq i32* %ptrincdec.i.i, %end
|
|
|
|
|
br i1 %cmp.i.i, label %for.cond.for.end_crit_edge.i.i, label %for.body.i.i
|
|
|
|
|
|
|
|
|
|
for.cond.for.end_crit_edge.i.i: ; preds = %for.body.i.i
|
|
|
|
|
br label %_ZSt4fillIPiiEvT_S1_RKT0_.exit
|
|
|
|
|
|
|
|
|
|
_ZSt4fillIPiiEvT_S1_RKT0_.exit: ; preds = %entry, %for.cond.for.end_crit_edge.i.i
|
|
|
|
|
ret void
|
|
|
|
|
}
|
2011-02-13 11:14:49 +08:00
|
|
|
|
|
|
|
|
; Various checks for inbounds geps.
|
|
|
|
|
define void @test3(i32* %begin, i32* %end) nounwind ssp {
|
|
|
|
|
entry:
|
|
|
|
|
%cmp7.i.i = icmp eq i32* %begin, %end
|
|
|
|
|
br i1 %cmp7.i.i, label %_ZSt4fillIPiiEvT_S1_RKT0_.exit, label %for.body.i.i
|
|
|
|
|
|
|
|
|
|
for.body.i.i: ; preds = %entry, %for.body.i.i
|
|
|
|
|
%indvar.i.i = phi i64 [ %tmp, %for.body.i.i ], [ 0, %entry ]
|
|
|
|
|
; CHECK: %indvar.i.i
|
|
|
|
|
; CHECK: {0,+,1}<nuw><nsw><%for.body.i.i>
|
|
|
|
|
%tmp = add nsw i64 %indvar.i.i, 1
|
2011-03-15 08:37:00 +08:00
|
|
|
; CHECK: %tmp =
|
2011-02-13 11:14:49 +08:00
|
|
|
; CHECK: {1,+,1}<nuw><nsw><%for.body.i.i>
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ptrincdec.i.i = getelementptr inbounds i32, i32* %begin, i64 %tmp
|
2011-02-13 11:14:49 +08:00
|
|
|
; CHECK: %ptrincdec.i.i =
|
[SCEV] No-wrap flags are not propagated when folding "{S,+,X}+T ==> {S+T,+,X}"
Summary:
**Description**
This makes `WidenIV::widenIVUse` (IndVarSimplify.cpp) fail to widen narrow IV uses in some cases. The latter affects IndVarSimplify which may not eliminate narrow IV's when there actually exists such a possibility, thereby producing ineffective code.
When `WidenIV::widenIVUse` gets a NarrowUse such as `{(-2 + %inc.lcssa),+,1}<nsw><%for.body3>`, it first tries to get a wide recurrence for it via the `getWideRecurrence` call.
`getWideRecurrence` returns recurrence like this: `{(sext i32 (-2 + %inc.lcssa) to i64),+,1}<nsw><%for.body3>`.
Then a wide use operation is generated by `cloneIVUser`. The generated wide use is evaluated to `{(-2 + (sext i32 %inc.lcssa to i64))<nsw>,+,1}<nsw><%for.body3>`, which is different from the `getWideRecurrence` result. `cloneIVUser` sees the difference and returns nullptr.
This patch also fixes the broken LLVM tests by adding missing <nsw> entries introduced by the correction.
**Minimal reproducer:**
```
int foo(int a, int b, int c);
int baz();
void bar()
{
int arr[20];
int i = 0;
for (i = 0; i < 4; ++i)
arr[i] = baz();
for (; i < 20; ++i)
arr[i] = foo(arr[i - 4], arr[i - 3], arr[i - 2]);
}
```
**Clang command line:**
```
clang++ -mllvm -debug -S -emit-llvm -O3 --target=aarch64-linux-elf test.cpp -o test.ir
```
**Expected result:**
The ` -mllvm -debug` log shows that all the IV's for the second `for` loop have been eliminated.
Reviewers: sanjoy
Subscribers: atrick, asl, aemerson, mzolotukhin, llvm-commits
Differential Revision: http://reviews.llvm.org/D20058
llvm-svn: 270695
2016-05-25 21:01:33 +08:00
|
|
|
; CHECK: {(4 + %begin)<nsw>,+,4}<nsw><%for.body.i.i>
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%__first.addr.08.i.i = getelementptr inbounds i32, i32* %begin, i64 %indvar.i.i
|
2011-02-13 11:14:49 +08:00
|
|
|
; CHECK: %__first.addr.08.i.i
|
2012-04-17 14:33:57 +08:00
|
|
|
; CHECK: {%begin,+,4}<nsw><%for.body.i.i>
|
2011-02-13 11:14:49 +08:00
|
|
|
store i32 0, i32* %__first.addr.08.i.i, align 4
|
|
|
|
|
%cmp.i.i = icmp eq i32* %ptrincdec.i.i, %end
|
|
|
|
|
br i1 %cmp.i.i, label %_ZSt4fillIPiiEvT_S1_RKT0_.exit, label %for.body.i.i
|
2011-03-15 08:37:00 +08:00
|
|
|
; CHECK: Loop %for.body.i.i: backedge-taken count is ((-4 + (-1 * %begin) + %end) /u 4)
|
|
|
|
|
; CHECK: Loop %for.body.i.i: max backedge-taken count is ((-4 + (-1 * %begin) + %end) /u 4)
|
2011-02-13 11:14:49 +08:00
|
|
|
_ZSt4fillIPiiEvT_S1_RKT0_.exit: ; preds = %for.body.i.i, %entry
|
|
|
|
|
ret void
|
2011-11-29 10:16:38 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; A single AddExpr exists for (%a + %b), which is not always <nsw>.
|
|
|
|
|
; CHECK: @addnsw
|
|
|
|
|
; CHECK-NOT: --> (%a + %b)<nsw>
|
|
|
|
|
define i32 @addnsw(i32 %a, i32 %b) nounwind ssp {
|
|
|
|
|
entry:
|
|
|
|
|
%tmp = add i32 %a, %b
|
|
|
|
|
%cmp = icmp sgt i32 %tmp, 0
|
|
|
|
|
br i1 %cmp, label %greater, label %exit
|
|
|
|
|
|
|
|
|
|
greater:
|
|
|
|
|
%tmp2 = add nsw i32 %a, %b
|
|
|
|
|
br label %exit
|
|
|
|
|
|
|
|
|
|
exit:
|
|
|
|
|
%result = phi i32 [ %a, %entry ], [ %tmp2, %greater ]
|
|
|
|
|
ret i32 %result
|
|
|
|
|
}
|
2013-10-28 15:30:06 +08:00
|
|
|
|
|
|
|
|
; CHECK-LABEL: PR12375
|
[SCEV] No-wrap flags are not propagated when folding "{S,+,X}+T ==> {S+T,+,X}"
Summary:
**Description**
This makes `WidenIV::widenIVUse` (IndVarSimplify.cpp) fail to widen narrow IV uses in some cases. The latter affects IndVarSimplify which may not eliminate narrow IV's when there actually exists such a possibility, thereby producing ineffective code.
When `WidenIV::widenIVUse` gets a NarrowUse such as `{(-2 + %inc.lcssa),+,1}<nsw><%for.body3>`, it first tries to get a wide recurrence for it via the `getWideRecurrence` call.
`getWideRecurrence` returns recurrence like this: `{(sext i32 (-2 + %inc.lcssa) to i64),+,1}<nsw><%for.body3>`.
Then a wide use operation is generated by `cloneIVUser`. The generated wide use is evaluated to `{(-2 + (sext i32 %inc.lcssa to i64))<nsw>,+,1}<nsw><%for.body3>`, which is different from the `getWideRecurrence` result. `cloneIVUser` sees the difference and returns nullptr.
This patch also fixes the broken LLVM tests by adding missing <nsw> entries introduced by the correction.
**Minimal reproducer:**
```
int foo(int a, int b, int c);
int baz();
void bar()
{
int arr[20];
int i = 0;
for (i = 0; i < 4; ++i)
arr[i] = baz();
for (; i < 20; ++i)
arr[i] = foo(arr[i - 4], arr[i - 3], arr[i - 2]);
}
```
**Clang command line:**
```
clang++ -mllvm -debug -S -emit-llvm -O3 --target=aarch64-linux-elf test.cpp -o test.ir
```
**Expected result:**
The ` -mllvm -debug` log shows that all the IV's for the second `for` loop have been eliminated.
Reviewers: sanjoy
Subscribers: atrick, asl, aemerson, mzolotukhin, llvm-commits
Differential Revision: http://reviews.llvm.org/D20058
llvm-svn: 270695
2016-05-25 21:01:33 +08:00
|
|
|
; CHECK: --> {(4 + %arg)<nsw>,+,4}<nuw><%bb1>{{ U: [^ ]+ S: [^ ]+}}{{ *}}Exits: (8 + %arg)<nsw>
|
2013-10-28 15:30:06 +08:00
|
|
|
define i32 @PR12375(i32* readnone %arg) {
|
|
|
|
|
bb:
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%tmp = getelementptr inbounds i32, i32* %arg, i64 2
|
2013-10-28 15:30:06 +08:00
|
|
|
br label %bb1
|
|
|
|
|
|
|
|
|
|
bb1: ; preds = %bb1, %bb
|
|
|
|
|
%tmp2 = phi i32* [ %arg, %bb ], [ %tmp5, %bb1 ]
|
|
|
|
|
%tmp3 = phi i32 [ 0, %bb ], [ %tmp4, %bb1 ]
|
|
|
|
|
%tmp4 = add nsw i32 %tmp3, 1
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%tmp5 = getelementptr inbounds i32, i32* %tmp2, i64 1
|
2013-10-28 15:30:06 +08:00
|
|
|
%tmp6 = icmp ult i32* %tmp5, %tmp
|
|
|
|
|
br i1 %tmp6, label %bb1, label %bb7
|
|
|
|
|
|
|
|
|
|
bb7: ; preds = %bb1
|
|
|
|
|
ret i32 %tmp4
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK-LABEL: PR12376
|
2016-05-29 08:32:17 +08:00
|
|
|
; CHECK: --> {(4 + %arg)<nsw>,+,4}<nuw><%bb2>{{ U: [^ ]+ S: [^ ]+}}{{ *}}Exits: (4 + (4 * ((-1 + (-1 * %arg) + ((4 + %arg)<nsw> umax %arg1)) /u 4)) + %arg)
|
2013-10-28 15:30:06 +08:00
|
|
|
define void @PR12376(i32* nocapture %arg, i32* nocapture %arg1) {
|
|
|
|
|
bb:
|
|
|
|
|
br label %bb2
|
|
|
|
|
|
|
|
|
|
bb2: ; preds = %bb2, %bb
|
|
|
|
|
%tmp = phi i32* [ %arg, %bb ], [ %tmp4, %bb2 ]
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%tmp4 = getelementptr inbounds i32, i32* %tmp, i64 1
|
2016-05-29 08:32:17 +08:00
|
|
|
%tmp3 = icmp ult i32* %tmp4, %arg1
|
2013-10-28 15:30:06 +08:00
|
|
|
br i1 %tmp3, label %bb2, label %bb5
|
|
|
|
|
|
|
|
|
|
bb5: ; preds = %bb2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
2014-10-31 19:40:32 +08:00
|
|
|
|
|
|
|
|
declare void @f(i32)
|
|
|
|
|
|
|
|
|
|
; CHECK-LABEL: nswnowrap
|
2016-05-29 08:32:17 +08:00
|
|
|
; CHECK: --> {(1 + %v)<nsw>,+,1}<nsw><%for.body>{{ U: [^ ]+ S: [^ ]+}}{{ *}}Exits: (2 + %v)
|
|
|
|
|
define void @nswnowrap(i32 %v, i32* %buf) {
|
2014-10-31 19:40:32 +08:00
|
|
|
entry:
|
|
|
|
|
%add = add nsw i32 %v, 1
|
|
|
|
|
br label %for.body
|
|
|
|
|
|
|
|
|
|
for.body:
|
|
|
|
|
%i.04 = phi i32 [ %v, %entry ], [ %inc, %for.body ]
|
|
|
|
|
%inc = add nsw i32 %i.04, 1
|
2016-05-29 08:32:17 +08:00
|
|
|
%buf.gep = getelementptr inbounds i32, i32* %buf, i32 %inc
|
|
|
|
|
%buf.val = load i32, i32* %buf.gep
|
2014-10-31 19:40:32 +08:00
|
|
|
%cmp = icmp slt i32 %i.04, %add
|
2016-05-29 08:32:17 +08:00
|
|
|
tail call void @f(i32 %i.04)
|
2014-10-31 19:40:32 +08:00
|
|
|
br i1 %cmp, label %for.body, label %for.end
|
|
|
|
|
|
|
|
|
|
for.end:
|
|
|
|
|
ret void
|
|
|
|
|
}
|
[SCEV] No-wrap flags are not propagated when folding "{S,+,X}+T ==> {S+T,+,X}"
Summary:
**Description**
This makes `WidenIV::widenIVUse` (IndVarSimplify.cpp) fail to widen narrow IV uses in some cases. The latter affects IndVarSimplify which may not eliminate narrow IV's when there actually exists such a possibility, thereby producing ineffective code.
When `WidenIV::widenIVUse` gets a NarrowUse such as `{(-2 + %inc.lcssa),+,1}<nsw><%for.body3>`, it first tries to get a wide recurrence for it via the `getWideRecurrence` call.
`getWideRecurrence` returns recurrence like this: `{(sext i32 (-2 + %inc.lcssa) to i64),+,1}<nsw><%for.body3>`.
Then a wide use operation is generated by `cloneIVUser`. The generated wide use is evaluated to `{(-2 + (sext i32 %inc.lcssa to i64))<nsw>,+,1}<nsw><%for.body3>`, which is different from the `getWideRecurrence` result. `cloneIVUser` sees the difference and returns nullptr.
This patch also fixes the broken LLVM tests by adding missing <nsw> entries introduced by the correction.
**Minimal reproducer:**
```
int foo(int a, int b, int c);
int baz();
void bar()
{
int arr[20];
int i = 0;
for (i = 0; i < 4; ++i)
arr[i] = baz();
for (; i < 20; ++i)
arr[i] = foo(arr[i - 4], arr[i - 3], arr[i - 2]);
}
```
**Clang command line:**
```
clang++ -mllvm -debug -S -emit-llvm -O3 --target=aarch64-linux-elf test.cpp -o test.ir
```
**Expected result:**
The ` -mllvm -debug` log shows that all the IV's for the second `for` loop have been eliminated.
Reviewers: sanjoy
Subscribers: atrick, asl, aemerson, mzolotukhin, llvm-commits
Differential Revision: http://reviews.llvm.org/D20058
llvm-svn: 270695
2016-05-25 21:01:33 +08:00
|
|
|
|
|
|
|
|
; This test checks if no-wrap flags are propagated when folding {S,+,X}+T ==> {S+T,+,X}
|
|
|
|
|
; CHECK-LABEL: test4
|
|
|
|
|
; CHECK: %idxprom
|
|
|
|
|
; CHECK-NEXT: --> {(-2 + (sext i32 %arg to i64))<nsw>,+,1}<nsw><%for.body>
|
|
|
|
|
define void @test4(i32 %arg) {
|
|
|
|
|
entry:
|
|
|
|
|
%array = alloca [10 x i32], align 4
|
|
|
|
|
br label %for.body
|
|
|
|
|
|
|
|
|
|
for.body:
|
|
|
|
|
%index = phi i32 [ %inc5, %for.body ], [ %arg, %entry ]
|
|
|
|
|
%sub = add nsw i32 %index, -2
|
|
|
|
|
%idxprom = sext i32 %sub to i64
|
|
|
|
|
%arrayidx = getelementptr inbounds [10 x i32], [10 x i32]* %array, i64 0, i64 %idxprom
|
|
|
|
|
%data = load i32, i32* %arrayidx, align 4
|
|
|
|
|
%inc5 = add nsw i32 %index, 1
|
|
|
|
|
%cmp2 = icmp slt i32 %inc5, 10
|
|
|
|
|
br i1 %cmp2, label %for.body, label %for.end
|
|
|
|
|
|
|
|
|
|
for.end:
|
|
|
|
|
ret void
|
|
|
|
|
}
|