Revert r185257 (InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms)
I'm reverting this commit because:
1. As discussed during review, it needs to be rewritten (to avoid creating and
then deleting instructions).
2. This is causing optimizer crashes. Specifically, I'm seeing things like
this:
While deleting: i1 %
Use still stuck around after Def is destroyed: <badref> = select i1 <badref>, i32 0, i32 1
opt: /src/llvm-trunk/lib/IR/Value.cpp:79: virtual llvm::Value::~Value(): Assertion `use_empty() && "Uses remain when a value is destroyed!"' failed.
I'd guess that these will go away once we're no longer creating/deleting
instructions here, but just in case, I'm adding a regression test.
Because the code is bring rewritten, I've just XFAIL'd the original regression test. Original commit message:
InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms
Real world code sometimes has the denominator of a 'udiv' be a
'select'. LLVM can handle such cases but only when the 'select'
operands are symmetric in structure (both select operands are a constant
power of two or a left shift, etc.). This falls apart if we are dealt a
'udiv' where the code is not symetric or if the select operands lead us
to more select instructions.
Instead, we should treat the LHS and each select operand as a distinct
divide operation and try to optimize them independently. If we can
to simplify each operation, then we can replace the 'udiv' with, say, a
'lshr' that has a new select with a bunch of new operands for the
select.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185415 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-02 05:21:11 +00:00
|
|
|
; RUN: opt -instcombine < %s
|
|
|
|
target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
|
|
|
|
target triple = "powerpc64-unknown-linux-gnu"
|
|
|
|
|
|
|
|
%struct.S0.0.1.2.3.4.13.22.31.44.48.53.54.55.56.58.59.60.66.68.70.74.77.106.107.108.109.110.113.117.118.128.129 = type <{ i64 }>
|
|
|
|
|
|
|
|
; Function Attrs: nounwind
|
|
|
|
define void @main() #0 {
|
|
|
|
entry:
|
|
|
|
%l_819.i.i = alloca %struct.S0.0.1.2.3.4.13.22.31.44.48.53.54.55.56.58.59.60.66.68.70.74.77.106.107.108.109.110.113.117.118.128.129, align 8
|
|
|
|
br i1 undef, label %land.lhs.true, label %for.cond.i
|
|
|
|
|
|
|
|
land.lhs.true: ; preds = %entry
|
|
|
|
br label %for.cond.i
|
|
|
|
|
|
|
|
for.cond.i: ; preds = %land.lhs.true, %entry
|
[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
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%0 = getelementptr inbounds %struct.S0.0.1.2.3.4.13.22.31.44.48.53.54.55.56.58.59.60.66.68.70.74.77.106.107.108.109.110.113.117.118.128.129, %struct.S0.0.1.2.3.4.13.22.31.44.48.53.54.55.56.58.59.60.66.68.70.74.77.106.107.108.109.110.113.117.118.128.129* %l_819.i.i, i64 0, i32 0
|
Revert r185257 (InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms)
I'm reverting this commit because:
1. As discussed during review, it needs to be rewritten (to avoid creating and
then deleting instructions).
2. This is causing optimizer crashes. Specifically, I'm seeing things like
this:
While deleting: i1 %
Use still stuck around after Def is destroyed: <badref> = select i1 <badref>, i32 0, i32 1
opt: /src/llvm-trunk/lib/IR/Value.cpp:79: virtual llvm::Value::~Value(): Assertion `use_empty() && "Uses remain when a value is destroyed!"' failed.
I'd guess that these will go away once we're no longer creating/deleting
instructions here, but just in case, I'm adding a regression test.
Because the code is bring rewritten, I've just XFAIL'd the original regression test. Original commit message:
InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms
Real world code sometimes has the denominator of a 'udiv' be a
'select'. LLVM can handle such cases but only when the 'select'
operands are symmetric in structure (both select operands are a constant
power of two or a left shift, etc.). This falls apart if we are dealt a
'udiv' where the code is not symetric or if the select operands lead us
to more select instructions.
Instead, we should treat the LHS and each select operand as a distinct
divide operation and try to optimize them independently. If we can
to simplify each operation, then we can replace the 'udiv' with, say, a
'lshr' that has a new select with a bunch of new operands for the
select.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185415 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-02 05:21:11 +00:00
|
|
|
br label %for.cond.i6.i.i
|
|
|
|
|
|
|
|
for.cond.i6.i.i: ; preds = %for.body.i8.i.i, %for.cond.i
|
|
|
|
br i1 undef, label %for.body.i8.i.i, label %lbl_707.i.i.i
|
|
|
|
|
|
|
|
for.body.i8.i.i: ; preds = %for.cond.i6.i.i
|
|
|
|
br label %for.cond.i6.i.i
|
|
|
|
|
|
|
|
lbl_707.i.i.i: ; preds = %for.cond.i6.i.i
|
|
|
|
br i1 undef, label %lor.rhs.i.i.i, label %lor.end.i.i.i
|
|
|
|
|
|
|
|
lor.rhs.i.i.i: ; preds = %lbl_707.i.i.i
|
|
|
|
br label %lor.end.i.i.i
|
|
|
|
|
|
|
|
lor.end.i.i.i: ; preds = %lor.rhs.i.i.i, %lbl_707.i.i.i
|
|
|
|
br label %for.cond1.i.i.i.i
|
|
|
|
|
|
|
|
for.cond1.i.i.i.i: ; preds = %for.body4.i.i.i.i, %lor.end.i.i.i
|
|
|
|
br i1 undef, label %for.body4.i.i.i.i, label %func_39.exit.i.i
|
|
|
|
|
|
|
|
for.body4.i.i.i.i: ; preds = %for.cond1.i.i.i.i
|
|
|
|
br label %for.cond1.i.i.i.i
|
|
|
|
|
|
|
|
func_39.exit.i.i: ; preds = %for.cond1.i.i.i.i
|
2015-02-27 21:17:42 +00:00
|
|
|
%l_8191.sroa.0.0.copyload.i.i = load i64, i64* %0, align 1
|
Revert r185257 (InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms)
I'm reverting this commit because:
1. As discussed during review, it needs to be rewritten (to avoid creating and
then deleting instructions).
2. This is causing optimizer crashes. Specifically, I'm seeing things like
this:
While deleting: i1 %
Use still stuck around after Def is destroyed: <badref> = select i1 <badref>, i32 0, i32 1
opt: /src/llvm-trunk/lib/IR/Value.cpp:79: virtual llvm::Value::~Value(): Assertion `use_empty() && "Uses remain when a value is destroyed!"' failed.
I'd guess that these will go away once we're no longer creating/deleting
instructions here, but just in case, I'm adding a regression test.
Because the code is bring rewritten, I've just XFAIL'd the original regression test. Original commit message:
InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms
Real world code sometimes has the denominator of a 'udiv' be a
'select'. LLVM can handle such cases but only when the 'select'
operands are symmetric in structure (both select operands are a constant
power of two or a left shift, etc.). This falls apart if we are dealt a
'udiv' where the code is not symetric or if the select operands lead us
to more select instructions.
Instead, we should treat the LHS and each select operand as a distinct
divide operation and try to optimize them independently. If we can
to simplify each operation, then we can replace the 'udiv' with, say, a
'lshr' that has a new select with a bunch of new operands for the
select.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185415 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-02 05:21:11 +00:00
|
|
|
br label %for.cond1.i.i.i
|
|
|
|
|
|
|
|
for.cond1.i.i.i: ; preds = %safe_div_func_uint32_t_u_u.exit.i.i.i, %func_39.exit.i.i
|
|
|
|
br i1 undef, label %for.cond7.i.i.i, label %func_11.exit.i
|
|
|
|
|
|
|
|
for.cond7.i.i.i: ; preds = %for.end30.i.i.i, %for.cond1.i.i.i
|
|
|
|
%storemerge.i.i.i = phi i32 [ %sub.i.i.i, %for.end30.i.i.i ], [ 4, %for.cond1.i.i.i ]
|
|
|
|
br i1 undef, label %for.cond22.i.i.i, label %for.end32.i.i.i
|
|
|
|
|
|
|
|
for.cond22.i.i.i: ; preds = %for.body25.i.i.i, %for.cond7.i.i.i
|
|
|
|
br i1 undef, label %for.body25.i.i.i, label %for.end30.i.i.i
|
|
|
|
|
|
|
|
for.body25.i.i.i: ; preds = %for.cond22.i.i.i
|
|
|
|
br label %for.cond22.i.i.i
|
|
|
|
|
|
|
|
for.end30.i.i.i: ; preds = %for.cond22.i.i.i
|
|
|
|
%sub.i.i.i = add nsw i32 0, -1
|
|
|
|
br label %for.cond7.i.i.i
|
|
|
|
|
|
|
|
for.end32.i.i.i: ; preds = %for.cond7.i.i.i
|
|
|
|
%conv33.i.i.i = trunc i64 %l_8191.sroa.0.0.copyload.i.i to i32
|
|
|
|
%xor.i.i.i.i = xor i32 %storemerge.i.i.i, -701565022
|
|
|
|
%sub.i.i.i.i = sub nsw i32 0, %storemerge.i.i.i
|
|
|
|
%xor3.i.i.i.i = xor i32 %sub.i.i.i.i, %storemerge.i.i.i
|
|
|
|
%and4.i.i.i.i = and i32 %xor.i.i.i.i, %xor3.i.i.i.i
|
|
|
|
%cmp.i.i.i.i = icmp slt i32 %and4.i.i.i.i, 0
|
|
|
|
%sub5.i.i.i.i = sub nsw i32 -701565022, %storemerge.i.i.i
|
|
|
|
%.sub5.i.i.i.i = select i1 %cmp.i.i.i.i, i32 -701565022, i32 %sub5.i.i.i.i
|
|
|
|
br i1 undef, label %safe_div_func_uint32_t_u_u.exit.i.i.i, label %cond.false.i.i.i.i
|
|
|
|
|
|
|
|
cond.false.i.i.i.i: ; preds = %for.end32.i.i.i
|
|
|
|
%div.i.i.i.i = udiv i32 %conv33.i.i.i, %.sub5.i.i.i.i
|
|
|
|
br label %safe_div_func_uint32_t_u_u.exit.i.i.i
|
|
|
|
|
|
|
|
safe_div_func_uint32_t_u_u.exit.i.i.i: ; preds = %cond.false.i.i.i.i, %for.end32.i.i.i
|
|
|
|
%cond.i.i.i.i = phi i32 [ %div.i.i.i.i, %cond.false.i.i.i.i ], [ %conv33.i.i.i, %for.end32.i.i.i ]
|
|
|
|
%cmp35.i.i.i = icmp ne i32 %cond.i.i.i.i, -7
|
|
|
|
br label %for.cond1.i.i.i
|
|
|
|
|
|
|
|
func_11.exit.i: ; preds = %for.cond1.i.i.i
|
|
|
|
br i1 undef, label %for.body, label %for.end
|
|
|
|
|
|
|
|
for.body: ; preds = %func_11.exit.i
|
|
|
|
unreachable
|
|
|
|
|
|
|
|
for.end: ; preds = %func_11.exit.i
|
|
|
|
br label %for.cond15
|
|
|
|
|
|
|
|
for.cond15: ; preds = %for.cond19, %for.end
|
|
|
|
br i1 undef, label %for.cond19, label %for.end45
|
|
|
|
|
|
|
|
for.cond19: ; preds = %for.cond15
|
|
|
|
br label %for.cond15
|
|
|
|
|
|
|
|
for.end45: ; preds = %for.cond15
|
|
|
|
unreachable
|
|
|
|
}
|
|
|
|
|
|
|
|
attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
|