2014-02-22 02:15:18 +08:00
|
|
|
; RUN: opt < %s -analyze -basicaa -da -da-delinearize=false | FileCheck %s
|
|
|
|
|
; RUN: opt < %s -analyze -basicaa -da -da-delinearize | FileCheck %s -check-prefix=DELIN
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
; ModuleID = 'GCD.bc'
|
|
|
|
|
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-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
|
|
|
|
|
target triple = "x86_64-apple-macosx10.6.0"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
;; for (long int i = 0; i < 100; i++)
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (long int j = 0; j < 100; j++) {
|
|
|
|
|
;; A[2*i - 4*j] = i;
|
|
|
|
|
;; *B++ = A[6*i + 8*j];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd0(i32* %A, i32* %B) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd0'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - flow [=> *|<]!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-28 03:12:26 +08:00
|
|
|
; CHECK: da analyze - none!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd0'
|
2014-05-08 02:01:20 +08:00
|
|
|
; DELIN: da analyze - output [* *]!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - flow [=> *|<]!
|
|
|
|
|
; DELIN: da analyze - confused!
|
2014-05-08 02:01:20 +08:00
|
|
|
; DELIN: da analyze - input [* *]!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader: ; preds = %entry, %for.inc8
|
|
|
|
|
%B.addr.04 = phi i32* [ %B, %entry ], [ %scevgep, %for.inc8 ]
|
|
|
|
|
%i.03 = phi i64 [ 0, %entry ], [ %inc9, %for.inc8 ]
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.cond1.preheader, %for.body3
|
|
|
|
|
%j.02 = phi i64 [ 0, %for.cond1.preheader ], [ %inc, %for.body3 ]
|
|
|
|
|
%B.addr.11 = phi i32* [ %B.addr.04, %for.cond1.preheader ], [ %incdec.ptr, %for.body3 ]
|
|
|
|
|
%conv = trunc i64 %i.03 to i32
|
|
|
|
|
%mul = shl nsw i64 %i.03, 1
|
|
|
|
|
%mul4 = shl nsw i64 %j.02, 2
|
|
|
|
|
%sub = sub nsw i64 %mul, %mul4
|
[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
|
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %sub
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %conv, i32* %arrayidx, align 4
|
|
|
|
|
%mul5 = mul nsw i64 %i.03, 6
|
|
|
|
|
%mul6 = shl nsw i64 %j.02, 3
|
|
|
|
|
%add = add nsw i64 %mul5, %mul6
|
[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
|
|
|
%arrayidx7 = getelementptr inbounds i32, i32* %A, i64 %add
|
2015-02-28 05:17:42 +08:00
|
|
|
%0 = load i32, i32* %arrayidx7, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.11, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %0, i32* %B.addr.11, align 4
|
|
|
|
|
%inc = add nsw i64 %j.02, 1
|
|
|
|
|
%exitcond = icmp ne i64 %inc, 100
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc8
|
|
|
|
|
|
|
|
|
|
for.inc8: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.04, i64 100
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
%inc9 = add nsw i64 %i.03, 1
|
|
|
|
|
%exitcond5 = icmp ne i64 %inc9, 100
|
|
|
|
|
br i1 %exitcond5, label %for.cond1.preheader, label %for.end10
|
|
|
|
|
|
|
|
|
|
for.end10: ; preds = %for.inc8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
;; for (long int i = 0; i < 100; i++)
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (long int j = 0; j < 100; j++) {
|
|
|
|
|
;; A[2*i - 4*j] = i;
|
|
|
|
|
;; *B++ = A[6*i + 8*j + 1];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd1(i32* %A, i32* %B) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd1'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - none!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-28 03:12:26 +08:00
|
|
|
; CHECK: da analyze - none!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd1'
|
2014-05-08 02:01:20 +08:00
|
|
|
; DELIN: da analyze - output [* *]!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
; DELIN: da analyze - confused!
|
2014-05-08 02:01:20 +08:00
|
|
|
; DELIN: da analyze - input [* *]!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader: ; preds = %entry, %for.inc9
|
|
|
|
|
%B.addr.04 = phi i32* [ %B, %entry ], [ %scevgep, %for.inc9 ]
|
|
|
|
|
%i.03 = phi i64 [ 0, %entry ], [ %inc10, %for.inc9 ]
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.cond1.preheader, %for.body3
|
|
|
|
|
%j.02 = phi i64 [ 0, %for.cond1.preheader ], [ %inc, %for.body3 ]
|
|
|
|
|
%B.addr.11 = phi i32* [ %B.addr.04, %for.cond1.preheader ], [ %incdec.ptr, %for.body3 ]
|
|
|
|
|
%conv = trunc i64 %i.03 to i32
|
|
|
|
|
%mul = shl nsw i64 %i.03, 1
|
|
|
|
|
%mul4 = shl nsw i64 %j.02, 2
|
|
|
|
|
%sub = sub nsw i64 %mul, %mul4
|
[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
|
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %sub
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %conv, i32* %arrayidx, align 4
|
|
|
|
|
%mul5 = mul nsw i64 %i.03, 6
|
|
|
|
|
%mul6 = shl nsw i64 %j.02, 3
|
|
|
|
|
%add = add nsw i64 %mul5, %mul6
|
|
|
|
|
%add7 = or i64 %add, 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
|
|
|
%arrayidx8 = getelementptr inbounds i32, i32* %A, i64 %add7
|
2015-02-28 05:17:42 +08:00
|
|
|
%0 = load i32, i32* %arrayidx8, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.11, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %0, i32* %B.addr.11, align 4
|
|
|
|
|
%inc = add nsw i64 %j.02, 1
|
|
|
|
|
%exitcond = icmp ne i64 %inc, 100
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc9
|
|
|
|
|
|
|
|
|
|
for.inc9: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.04, i64 100
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
%inc10 = add nsw i64 %i.03, 1
|
|
|
|
|
%exitcond5 = icmp ne i64 %inc10, 100
|
|
|
|
|
br i1 %exitcond5, label %for.cond1.preheader, label %for.end11
|
|
|
|
|
|
|
|
|
|
for.end11: ; preds = %for.inc9
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
;; for (long int i = 0; i < 100; i++)
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (long int j = 0; j < 100; j++) {
|
|
|
|
|
;; A[2*i - 4*j + 1] = i;
|
|
|
|
|
;; *B++ = A[6*i + 8*j];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd2(i32* %A, i32* %B) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd2'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - none!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-28 03:12:26 +08:00
|
|
|
; CHECK: da analyze - none!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd2'
|
2014-05-08 02:01:20 +08:00
|
|
|
; DELIN: da analyze - output [* *]!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
; DELIN: da analyze - confused!
|
2014-05-08 02:01:20 +08:00
|
|
|
; DELIN: da analyze - input [* *]!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader: ; preds = %entry, %for.inc9
|
|
|
|
|
%B.addr.04 = phi i32* [ %B, %entry ], [ %scevgep, %for.inc9 ]
|
|
|
|
|
%i.03 = phi i64 [ 0, %entry ], [ %inc10, %for.inc9 ]
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.cond1.preheader, %for.body3
|
|
|
|
|
%j.02 = phi i64 [ 0, %for.cond1.preheader ], [ %inc, %for.body3 ]
|
|
|
|
|
%B.addr.11 = phi i32* [ %B.addr.04, %for.cond1.preheader ], [ %incdec.ptr, %for.body3 ]
|
|
|
|
|
%conv = trunc i64 %i.03 to i32
|
|
|
|
|
%mul = shl nsw i64 %i.03, 1
|
|
|
|
|
%mul4 = shl nsw i64 %j.02, 2
|
|
|
|
|
%sub = sub nsw i64 %mul, %mul4
|
|
|
|
|
%add5 = or i64 %sub, 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
|
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %add5
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %conv, i32* %arrayidx, align 4
|
|
|
|
|
%mul5 = mul nsw i64 %i.03, 6
|
|
|
|
|
%mul6 = shl nsw i64 %j.02, 3
|
|
|
|
|
%add7 = add nsw i64 %mul5, %mul6
|
[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
|
|
|
%arrayidx8 = getelementptr inbounds i32, i32* %A, i64 %add7
|
2015-02-28 05:17:42 +08:00
|
|
|
%0 = load i32, i32* %arrayidx8, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.11, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %0, i32* %B.addr.11, align 4
|
|
|
|
|
%inc = add nsw i64 %j.02, 1
|
|
|
|
|
%exitcond = icmp ne i64 %inc, 100
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc9
|
|
|
|
|
|
|
|
|
|
for.inc9: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.04, i64 100
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
%inc10 = add nsw i64 %i.03, 1
|
|
|
|
|
%exitcond6 = icmp ne i64 %inc10, 100
|
|
|
|
|
br i1 %exitcond6, label %for.cond1.preheader, label %for.end11
|
|
|
|
|
|
|
|
|
|
for.end11: ; preds = %for.inc9
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
;; for (long int i = 0; i < 100; i++)
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (long int j = 0; j < 100; j++) {
|
|
|
|
|
;; A[i + 2*j] = i;
|
|
|
|
|
;; *B++ = A[i + 2*j - 1];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd3(i32* %A, i32* %B) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd3'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - flow [<> *]!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-28 03:12:26 +08:00
|
|
|
; CHECK: da analyze - none!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd3'
|
|
|
|
|
; DELIN: da analyze - output [* *]!
|
|
|
|
|
; DELIN: da analyze - flow [<> *]!
|
|
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - input [* *]!
|
|
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader: ; preds = %entry, %for.inc7
|
|
|
|
|
%B.addr.04 = phi i32* [ %B, %entry ], [ %scevgep, %for.inc7 ]
|
|
|
|
|
%i.03 = phi i64 [ 0, %entry ], [ %inc8, %for.inc7 ]
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.cond1.preheader, %for.body3
|
|
|
|
|
%j.02 = phi i64 [ 0, %for.cond1.preheader ], [ %inc, %for.body3 ]
|
|
|
|
|
%B.addr.11 = phi i32* [ %B.addr.04, %for.cond1.preheader ], [ %incdec.ptr, %for.body3 ]
|
|
|
|
|
%conv = trunc i64 %i.03 to i32
|
|
|
|
|
%mul = shl nsw i64 %j.02, 1
|
|
|
|
|
%add = add nsw i64 %i.03, %mul
|
[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
|
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %add
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %conv, i32* %arrayidx, align 4
|
|
|
|
|
%mul4 = shl nsw i64 %j.02, 1
|
|
|
|
|
%add5 = add nsw i64 %i.03, %mul4
|
|
|
|
|
%sub = add nsw i64 %add5, -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
|
|
|
%arrayidx6 = getelementptr inbounds i32, i32* %A, i64 %sub
|
2015-02-28 05:17:42 +08:00
|
|
|
%0 = load i32, i32* %arrayidx6, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.11, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %0, i32* %B.addr.11, align 4
|
|
|
|
|
%inc = add nsw i64 %j.02, 1
|
|
|
|
|
%exitcond = icmp ne i64 %inc, 100
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc7
|
|
|
|
|
|
|
|
|
|
for.inc7: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.04, i64 100
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
%inc8 = add nsw i64 %i.03, 1
|
|
|
|
|
%exitcond5 = icmp ne i64 %inc8, 100
|
|
|
|
|
br i1 %exitcond5, label %for.cond1.preheader, label %for.end9
|
|
|
|
|
|
|
|
|
|
for.end9: ; preds = %for.inc7
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (long int i = 0; i < 100; i++)
|
|
|
|
|
;; for (long int j = 0; j < 100; j++) {
|
|
|
|
|
;; A[5*i + 10*j*M + 9*M*N] = i;
|
|
|
|
|
;; *B++ = A[15*i + 20*j*M - 21*N*M + 4];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd4(i32* %A, i32* %B, i64 %M, i64 %N) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd4'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - none!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-28 03:12:26 +08:00
|
|
|
; CHECK: da analyze - none!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd4'
|
2015-08-19 10:56:36 +08:00
|
|
|
; DELIN: da analyze - none!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
; DELIN: da analyze - confused!
|
2015-08-19 10:56:36 +08:00
|
|
|
; DELIN: da analyze - none!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader: ; preds = %entry, %for.inc17
|
|
|
|
|
%B.addr.04 = phi i32* [ %B, %entry ], [ %scevgep, %for.inc17 ]
|
|
|
|
|
%i.03 = phi i64 [ 0, %entry ], [ %inc18, %for.inc17 ]
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.cond1.preheader, %for.body3
|
|
|
|
|
%j.02 = phi i64 [ 0, %for.cond1.preheader ], [ %inc, %for.body3 ]
|
|
|
|
|
%B.addr.11 = phi i32* [ %B.addr.04, %for.cond1.preheader ], [ %incdec.ptr, %for.body3 ]
|
|
|
|
|
%conv = trunc i64 %i.03 to i32
|
|
|
|
|
%mul = mul nsw i64 %i.03, 5
|
|
|
|
|
%mul4 = mul nsw i64 %j.02, 10
|
|
|
|
|
%mul5 = mul nsw i64 %mul4, %M
|
|
|
|
|
%add = add nsw i64 %mul, %mul5
|
|
|
|
|
%mul6 = mul nsw i64 %M, 9
|
|
|
|
|
%mul7 = mul nsw i64 %mul6, %N
|
|
|
|
|
%add8 = add nsw i64 %add, %mul7
|
[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
|
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %add8
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %conv, i32* %arrayidx, align 4
|
|
|
|
|
%mul9 = mul nsw i64 %i.03, 15
|
|
|
|
|
%mul10 = mul nsw i64 %j.02, 20
|
|
|
|
|
%mul11 = mul nsw i64 %mul10, %M
|
|
|
|
|
%add12 = add nsw i64 %mul9, %mul11
|
|
|
|
|
%mul13 = mul nsw i64 %N, 21
|
|
|
|
|
%mul14 = mul nsw i64 %mul13, %M
|
|
|
|
|
%sub = sub nsw i64 %add12, %mul14
|
|
|
|
|
%add15 = add nsw i64 %sub, 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
|
|
|
%arrayidx16 = getelementptr inbounds i32, i32* %A, i64 %add15
|
2015-02-28 05:17:42 +08:00
|
|
|
%0 = load i32, i32* %arrayidx16, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.11, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %0, i32* %B.addr.11, align 4
|
|
|
|
|
%inc = add nsw i64 %j.02, 1
|
|
|
|
|
%exitcond = icmp ne i64 %inc, 100
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc17
|
|
|
|
|
|
|
|
|
|
for.inc17: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.04, i64 100
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
%inc18 = add nsw i64 %i.03, 1
|
|
|
|
|
%exitcond5 = icmp ne i64 %inc18, 100
|
|
|
|
|
br i1 %exitcond5, label %for.cond1.preheader, label %for.end19
|
|
|
|
|
|
|
|
|
|
for.end19: ; preds = %for.inc17
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (long int i = 0; i < 100; i++)
|
|
|
|
|
;; for (long int j = 0; j < 100; j++) {
|
|
|
|
|
;; A[5*i + 10*j*M + 9*M*N] = i;
|
|
|
|
|
;; *B++ = A[15*i + 20*j*M - 21*N*M + 5];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd5(i32* %A, i32* %B, i64 %M, i64 %N) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd5'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - flow [<> *]!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-28 03:12:26 +08:00
|
|
|
; CHECK: da analyze - none!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd5'
|
2015-08-19 10:56:36 +08:00
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
; DELIN: da analyze - flow [> *]!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - confused!
|
2015-08-19 10:56:36 +08:00
|
|
|
; DELIN: da analyze - none!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader: ; preds = %entry, %for.inc17
|
|
|
|
|
%B.addr.04 = phi i32* [ %B, %entry ], [ %scevgep, %for.inc17 ]
|
|
|
|
|
%i.03 = phi i64 [ 0, %entry ], [ %inc18, %for.inc17 ]
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.cond1.preheader, %for.body3
|
|
|
|
|
%j.02 = phi i64 [ 0, %for.cond1.preheader ], [ %inc, %for.body3 ]
|
|
|
|
|
%B.addr.11 = phi i32* [ %B.addr.04, %for.cond1.preheader ], [ %incdec.ptr, %for.body3 ]
|
|
|
|
|
%conv = trunc i64 %i.03 to i32
|
|
|
|
|
%mul = mul nsw i64 %i.03, 5
|
|
|
|
|
%mul4 = mul nsw i64 %j.02, 10
|
|
|
|
|
%mul5 = mul nsw i64 %mul4, %M
|
|
|
|
|
%add = add nsw i64 %mul, %mul5
|
|
|
|
|
%mul6 = mul nsw i64 %M, 9
|
|
|
|
|
%mul7 = mul nsw i64 %mul6, %N
|
|
|
|
|
%add8 = add nsw i64 %add, %mul7
|
[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
|
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %add8
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %conv, i32* %arrayidx, align 4
|
|
|
|
|
%mul9 = mul nsw i64 %i.03, 15
|
|
|
|
|
%mul10 = mul nsw i64 %j.02, 20
|
|
|
|
|
%mul11 = mul nsw i64 %mul10, %M
|
|
|
|
|
%add12 = add nsw i64 %mul9, %mul11
|
|
|
|
|
%mul13 = mul nsw i64 %N, 21
|
|
|
|
|
%mul14 = mul nsw i64 %mul13, %M
|
|
|
|
|
%sub = sub nsw i64 %add12, %mul14
|
|
|
|
|
%add15 = add nsw i64 %sub, 5
|
[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
|
|
|
%arrayidx16 = getelementptr inbounds i32, i32* %A, i64 %add15
|
2015-02-28 05:17:42 +08:00
|
|
|
%0 = load i32, i32* %arrayidx16, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.11, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %0, i32* %B.addr.11, align 4
|
|
|
|
|
%inc = add nsw i64 %j.02, 1
|
|
|
|
|
%exitcond = icmp ne i64 %inc, 100
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc17
|
|
|
|
|
|
|
|
|
|
for.inc17: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.04, i64 100
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
%inc18 = add nsw i64 %i.03, 1
|
|
|
|
|
%exitcond5 = icmp ne i64 %inc18, 100
|
|
|
|
|
br i1 %exitcond5, label %for.cond1.preheader, label %for.end19
|
|
|
|
|
|
|
|
|
|
for.end19: ; preds = %for.inc17
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (long int i = 0; i < n; i++)
|
|
|
|
|
;; for (long int j = 0; j < n; j++) {
|
|
|
|
|
;; A[2*i][4*j] = i;
|
|
|
|
|
;; *B++ = A[8*i][6*j + 1];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd6(i64 %n, i32* %A, i32* %B) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
%cmp4 = icmp sgt i64 %n, 0
|
|
|
|
|
br i1 %cmp4, label %for.cond1.preheader.preheader, label %for.end12
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd6'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - none!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd6'
|
2015-08-19 10:56:36 +08:00
|
|
|
; DELIN: da analyze - none!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
; DELIN: da analyze - confused!
|
2015-08-19 10:56:36 +08:00
|
|
|
; DELIN: da analyze - none!
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - output [* *]!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader.preheader: ; preds = %entry
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
|
|
|
|
for.cond1.preheader: ; preds = %for.cond1.preheader.preheader, %for.inc10
|
|
|
|
|
%i.06 = phi i64 [ %inc11, %for.inc10 ], [ 0, %for.cond1.preheader.preheader ]
|
|
|
|
|
%B.addr.05 = phi i32* [ %B.addr.1.lcssa, %for.inc10 ], [ %B, %for.cond1.preheader.preheader ]
|
|
|
|
|
%cmp21 = icmp sgt i64 %n, 0
|
|
|
|
|
br i1 %cmp21, label %for.body3.preheader, label %for.inc10
|
|
|
|
|
|
|
|
|
|
for.body3.preheader: ; preds = %for.cond1.preheader
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.body3.preheader, %for.body3
|
|
|
|
|
%j.03 = phi i64 [ %inc, %for.body3 ], [ 0, %for.body3.preheader ]
|
|
|
|
|
%B.addr.12 = phi i32* [ %incdec.ptr, %for.body3 ], [ %B.addr.05, %for.body3.preheader ]
|
|
|
|
|
%conv = trunc i64 %i.06 to i32
|
|
|
|
|
%mul = shl nsw i64 %j.03, 2
|
|
|
|
|
%mul4 = shl nsw i64 %i.06, 1
|
|
|
|
|
%0 = mul nsw i64 %mul4, %n
|
|
|
|
|
%arrayidx.sum = add i64 %0, %mul
|
[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
|
|
|
%arrayidx5 = getelementptr inbounds i32, i32* %A, i64 %arrayidx.sum
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %conv, i32* %arrayidx5, align 4
|
|
|
|
|
%mul6 = mul nsw i64 %j.03, 6
|
|
|
|
|
%add7 = or i64 %mul6, 1
|
|
|
|
|
%mul7 = shl nsw i64 %i.06, 3
|
|
|
|
|
%1 = mul nsw i64 %mul7, %n
|
|
|
|
|
%arrayidx8.sum = add i64 %1, %add7
|
[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
|
|
|
%arrayidx9 = getelementptr inbounds i32, i32* %A, i64 %arrayidx8.sum
|
2015-02-28 05:17:42 +08:00
|
|
|
%2 = load i32, i32* %arrayidx9, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.12, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %2, i32* %B.addr.12, align 4
|
|
|
|
|
%inc = add nsw i64 %j.03, 1
|
|
|
|
|
%exitcond = icmp ne i64 %inc, %n
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc10.loopexit
|
|
|
|
|
|
|
|
|
|
for.inc10.loopexit: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.05, i64 %n
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
br label %for.inc10
|
|
|
|
|
|
|
|
|
|
for.inc10: ; preds = %for.inc10.loopexit, %for.cond1.preheader
|
|
|
|
|
%B.addr.1.lcssa = phi i32* [ %B.addr.05, %for.cond1.preheader ], [ %scevgep, %for.inc10.loopexit ]
|
|
|
|
|
%inc11 = add nsw i64 %i.06, 1
|
|
|
|
|
%exitcond8 = icmp ne i64 %inc11, %n
|
|
|
|
|
br i1 %exitcond8, label %for.cond1.preheader, label %for.end12.loopexit
|
|
|
|
|
|
|
|
|
|
for.end12.loopexit: ; preds = %for.inc10
|
|
|
|
|
br label %for.end12
|
|
|
|
|
|
|
|
|
|
for.end12: ; preds = %for.end12.loopexit, %entry
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (int i = 0; i < n; i++)
|
|
|
|
|
;; for (int j = 0; j < n; j++) {
|
|
|
|
|
;; A[2*i][4*j] = i;
|
|
|
|
|
;; *B++ = A[8*i][6*j + 1];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd7(i32 %n, i32* %A, i32* %B) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
%0 = zext i32 %n to i64
|
|
|
|
|
%cmp4 = icmp sgt i32 %n, 0
|
|
|
|
|
br i1 %cmp4, label %for.cond1.preheader.preheader, label %for.end15
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd7'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - flow [* *|<]!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd7'
|
|
|
|
|
; DELIN: da analyze - output [* *]!
|
|
|
|
|
; DELIN: da analyze - flow [* *|<]!
|
|
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - input [* *]!
|
|
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - output [* *]!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader.preheader: ; preds = %entry
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
|
|
|
|
for.cond1.preheader: ; preds = %for.cond1.preheader.preheader, %for.inc13
|
|
|
|
|
%indvars.iv8 = phi i64 [ 0, %for.cond1.preheader.preheader ], [ %indvars.iv.next9, %for.inc13 ]
|
|
|
|
|
%B.addr.05 = phi i32* [ %B.addr.1.lcssa, %for.inc13 ], [ %B, %for.cond1.preheader.preheader ]
|
|
|
|
|
%1 = add i32 %n, -1
|
|
|
|
|
%2 = zext i32 %1 to i64
|
|
|
|
|
%3 = add i64 %2, 1
|
|
|
|
|
%cmp21 = icmp sgt i32 %n, 0
|
|
|
|
|
br i1 %cmp21, label %for.body3.preheader, label %for.inc13
|
|
|
|
|
|
|
|
|
|
for.body3.preheader: ; preds = %for.cond1.preheader
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.body3.preheader, %for.body3
|
|
|
|
|
%indvars.iv = phi i64 [ 0, %for.body3.preheader ], [ %indvars.iv.next, %for.body3 ]
|
|
|
|
|
%B.addr.12 = phi i32* [ %incdec.ptr, %for.body3 ], [ %B.addr.05, %for.body3.preheader ]
|
|
|
|
|
%4 = trunc i64 %indvars.iv to i32
|
|
|
|
|
%mul = shl nsw i32 %4, 2
|
|
|
|
|
%idxprom = sext i32 %mul to i64
|
|
|
|
|
%5 = trunc i64 %indvars.iv8 to i32
|
|
|
|
|
%mul4 = shl nsw i32 %5, 1
|
|
|
|
|
%idxprom5 = sext i32 %mul4 to i64
|
|
|
|
|
%6 = mul nsw i64 %idxprom5, %0
|
|
|
|
|
%arrayidx.sum = add i64 %6, %idxprom
|
[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
|
|
|
%arrayidx6 = getelementptr inbounds i32, i32* %A, i64 %arrayidx.sum
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
%7 = trunc i64 %indvars.iv8 to i32
|
|
|
|
|
store i32 %7, i32* %arrayidx6, align 4
|
|
|
|
|
%8 = trunc i64 %indvars.iv to i32
|
|
|
|
|
%mul7 = mul nsw i32 %8, 6
|
|
|
|
|
%add7 = or i32 %mul7, 1
|
|
|
|
|
%idxprom8 = sext i32 %add7 to i64
|
|
|
|
|
%9 = trunc i64 %indvars.iv8 to i32
|
|
|
|
|
%mul9 = shl nsw i32 %9, 3
|
|
|
|
|
%idxprom10 = sext i32 %mul9 to i64
|
|
|
|
|
%10 = mul nsw i64 %idxprom10, %0
|
|
|
|
|
%arrayidx11.sum = add i64 %10, %idxprom8
|
[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
|
|
|
%arrayidx12 = getelementptr inbounds i32, i32* %A, i64 %arrayidx11.sum
|
2015-02-28 05:17:42 +08:00
|
|
|
%11 = load i32, i32* %arrayidx12, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.12, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %11, i32* %B.addr.12, align 4
|
|
|
|
|
%indvars.iv.next = add i64 %indvars.iv, 1
|
|
|
|
|
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
|
|
|
|
|
%exitcond = icmp ne i32 %lftr.wideiv, %n
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc13.loopexit
|
|
|
|
|
|
|
|
|
|
for.inc13.loopexit: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.05, i64 %3
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
br label %for.inc13
|
|
|
|
|
|
|
|
|
|
for.inc13: ; preds = %for.inc13.loopexit, %for.cond1.preheader
|
|
|
|
|
%B.addr.1.lcssa = phi i32* [ %B.addr.05, %for.cond1.preheader ], [ %scevgep, %for.inc13.loopexit ]
|
|
|
|
|
%indvars.iv.next9 = add i64 %indvars.iv8, 1
|
|
|
|
|
%lftr.wideiv10 = trunc i64 %indvars.iv.next9 to i32
|
|
|
|
|
%exitcond11 = icmp ne i32 %lftr.wideiv10, %n
|
|
|
|
|
br i1 %exitcond11, label %for.cond1.preheader, label %for.end15.loopexit
|
|
|
|
|
|
|
|
|
|
for.end15.loopexit: ; preds = %for.inc13
|
|
|
|
|
br label %for.end15
|
|
|
|
|
|
|
|
|
|
for.end15: ; preds = %for.end15.loopexit, %entry
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (int i = 0; i < n; i++)
|
|
|
|
|
;; for (int j = 0; j < n; j++) {
|
|
|
|
|
;; A[n*2*i + 4*j] = i;
|
|
|
|
|
;; *B++ = A[n*8*i + 6*j + 1];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd8(i32 %n, i32* %A, i32* %B) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
%cmp4 = icmp sgt i32 %n, 0
|
|
|
|
|
br i1 %cmp4, label %for.cond1.preheader.preheader, label %for.end15
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd8'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - none!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd8'
|
|
|
|
|
; DELIN: da analyze - output [* *]!
|
|
|
|
|
; DELIN: da analyze - none!
|
|
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - input [* *]!
|
|
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - output [* *]!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader.preheader: ; preds = %entry
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
|
|
|
|
for.cond1.preheader: ; preds = %for.cond1.preheader.preheader, %for.inc13
|
|
|
|
|
%i.06 = phi i32 [ %inc14, %for.inc13 ], [ 0, %for.cond1.preheader.preheader ]
|
|
|
|
|
%B.addr.05 = phi i32* [ %B.addr.1.lcssa, %for.inc13 ], [ %B, %for.cond1.preheader.preheader ]
|
|
|
|
|
%0 = add i32 %n, -1
|
|
|
|
|
%1 = zext i32 %0 to i64
|
|
|
|
|
%2 = add i64 %1, 1
|
|
|
|
|
%cmp21 = icmp sgt i32 %n, 0
|
|
|
|
|
br i1 %cmp21, label %for.body3.preheader, label %for.inc13
|
|
|
|
|
|
|
|
|
|
for.body3.preheader: ; preds = %for.cond1.preheader
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.body3.preheader, %for.body3
|
|
|
|
|
%indvars.iv = phi i64 [ 0, %for.body3.preheader ], [ %indvars.iv.next, %for.body3 ]
|
|
|
|
|
%B.addr.12 = phi i32* [ %incdec.ptr, %for.body3 ], [ %B.addr.05, %for.body3.preheader ]
|
|
|
|
|
%mul = shl nsw i32 %n, 1
|
|
|
|
|
%mul4 = mul nsw i32 %mul, %i.06
|
|
|
|
|
%3 = trunc i64 %indvars.iv to i32
|
|
|
|
|
%mul5 = shl nsw i32 %3, 2
|
|
|
|
|
%add = add nsw i32 %mul4, %mul5
|
|
|
|
|
%idxprom = sext i32 %add to i64
|
[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
|
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %idxprom
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %i.06, i32* %arrayidx, align 4
|
|
|
|
|
%mul6 = shl nsw i32 %n, 3
|
|
|
|
|
%mul7 = mul nsw i32 %mul6, %i.06
|
|
|
|
|
%4 = trunc i64 %indvars.iv to i32
|
|
|
|
|
%mul8 = mul nsw i32 %4, 6
|
|
|
|
|
%add9 = add nsw i32 %mul7, %mul8
|
|
|
|
|
%add10 = or i32 %add9, 1
|
|
|
|
|
%idxprom11 = sext i32 %add10 to i64
|
[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
|
|
|
%arrayidx12 = getelementptr inbounds i32, i32* %A, i64 %idxprom11
|
2015-02-28 05:17:42 +08:00
|
|
|
%5 = load i32, i32* %arrayidx12, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.12, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %5, i32* %B.addr.12, align 4
|
|
|
|
|
%indvars.iv.next = add i64 %indvars.iv, 1
|
|
|
|
|
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
|
|
|
|
|
%exitcond = icmp ne i32 %lftr.wideiv, %n
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc13.loopexit
|
|
|
|
|
|
|
|
|
|
for.inc13.loopexit: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.05, i64 %2
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
br label %for.inc13
|
|
|
|
|
|
|
|
|
|
for.inc13: ; preds = %for.inc13.loopexit, %for.cond1.preheader
|
|
|
|
|
%B.addr.1.lcssa = phi i32* [ %B.addr.05, %for.cond1.preheader ], [ %scevgep, %for.inc13.loopexit ]
|
|
|
|
|
%inc14 = add nsw i32 %i.06, 1
|
|
|
|
|
%exitcond7 = icmp ne i32 %inc14, %n
|
|
|
|
|
br i1 %exitcond7, label %for.cond1.preheader, label %for.end15.loopexit
|
|
|
|
|
|
|
|
|
|
for.end15.loopexit: ; preds = %for.inc13
|
|
|
|
|
br label %for.end15
|
|
|
|
|
|
|
|
|
|
for.end15: ; preds = %for.end15.loopexit, %entry
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2012-11-13 20:12:02 +08:00
|
|
|
;; for (unsigned i = 0; i < n; i++)
|
|
|
|
|
;; for (unsigned j = 0; j < n; j++) {
|
|
|
|
|
;; A[2*i][4*j] = i;
|
|
|
|
|
;; *B++ = A[8*i][6*j + 1];
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
|
|
|
|
|
define void @gcd9(i32 %n, i32* %A, i32* %B) nounwind uwtable ssp {
|
|
|
|
|
entry:
|
|
|
|
|
%0 = zext i32 %n to i64
|
|
|
|
|
%cmp4 = icmp eq i32 %n, 0
|
|
|
|
|
br i1 %cmp4, label %for.end15, label %for.cond1.preheader.preheader
|
|
|
|
|
|
2013-11-13 06:47:30 +08:00
|
|
|
; CHECK: 'Dependence Analysis' for function 'gcd9'
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - flow [* *|<]!
|
|
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - input [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
; CHECK: da analyze - confused!
|
2012-11-27 14:41:46 +08:00
|
|
|
; CHECK: da analyze - output [* *]!
|
2012-11-13 20:12:02 +08:00
|
|
|
|
2014-02-22 02:15:18 +08:00
|
|
|
; DELIN: 'Dependence Analysis' for function 'gcd9'
|
|
|
|
|
; DELIN: da analyze - output [* *]!
|
|
|
|
|
; DELIN: da analyze - flow [* *|<]!
|
|
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - input [* *]!
|
|
|
|
|
; DELIN: da analyze - confused!
|
|
|
|
|
; DELIN: da analyze - output [* *]!
|
|
|
|
|
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
for.cond1.preheader.preheader: ; preds = %entry
|
|
|
|
|
br label %for.cond1.preheader
|
|
|
|
|
|
|
|
|
|
for.cond1.preheader: ; preds = %for.cond1.preheader.preheader, %for.inc13
|
|
|
|
|
%indvars.iv8 = phi i64 [ 0, %for.cond1.preheader.preheader ], [ %indvars.iv.next9, %for.inc13 ]
|
|
|
|
|
%B.addr.05 = phi i32* [ %B.addr.1.lcssa, %for.inc13 ], [ %B, %for.cond1.preheader.preheader ]
|
|
|
|
|
%1 = add i32 %n, -1
|
|
|
|
|
%2 = zext i32 %1 to i64
|
|
|
|
|
%3 = add i64 %2, 1
|
|
|
|
|
%cmp21 = icmp eq i32 %n, 0
|
|
|
|
|
br i1 %cmp21, label %for.inc13, label %for.body3.preheader
|
|
|
|
|
|
|
|
|
|
for.body3.preheader: ; preds = %for.cond1.preheader
|
|
|
|
|
br label %for.body3
|
|
|
|
|
|
|
|
|
|
for.body3: ; preds = %for.body3.preheader, %for.body3
|
|
|
|
|
%indvars.iv = phi i64 [ 0, %for.body3.preheader ], [ %indvars.iv.next, %for.body3 ]
|
|
|
|
|
%B.addr.12 = phi i32* [ %incdec.ptr, %for.body3 ], [ %B.addr.05, %for.body3.preheader ]
|
|
|
|
|
%4 = trunc i64 %indvars.iv to i32
|
|
|
|
|
%mul = shl i32 %4, 2
|
|
|
|
|
%idxprom = zext i32 %mul to i64
|
|
|
|
|
%5 = trunc i64 %indvars.iv8 to i32
|
|
|
|
|
%mul4 = shl i32 %5, 1
|
|
|
|
|
%idxprom5 = zext i32 %mul4 to i64
|
|
|
|
|
%6 = mul nsw i64 %idxprom5, %0
|
|
|
|
|
%arrayidx.sum = add i64 %6, %idxprom
|
[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
|
|
|
%arrayidx6 = getelementptr inbounds i32, i32* %A, i64 %arrayidx.sum
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
%7 = trunc i64 %indvars.iv8 to i32
|
|
|
|
|
store i32 %7, i32* %arrayidx6, align 4
|
|
|
|
|
%8 = trunc i64 %indvars.iv to i32
|
|
|
|
|
%mul7 = mul i32 %8, 6
|
|
|
|
|
%add7 = or i32 %mul7, 1
|
|
|
|
|
%idxprom8 = zext i32 %add7 to i64
|
|
|
|
|
%9 = trunc i64 %indvars.iv8 to i32
|
|
|
|
|
%mul9 = shl i32 %9, 3
|
|
|
|
|
%idxprom10 = zext i32 %mul9 to i64
|
|
|
|
|
%10 = mul nsw i64 %idxprom10, %0
|
|
|
|
|
%arrayidx11.sum = add i64 %10, %idxprom8
|
[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
|
|
|
%arrayidx12 = getelementptr inbounds i32, i32* %A, i64 %arrayidx11.sum
|
2015-02-28 05:17:42 +08:00
|
|
|
%11 = load i32, i32* %arrayidx12, 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
|
|
|
%incdec.ptr = getelementptr inbounds i32, i32* %B.addr.12, i64 1
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
store i32 %11, i32* %B.addr.12, align 4
|
|
|
|
|
%indvars.iv.next = add i64 %indvars.iv, 1
|
|
|
|
|
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
|
|
|
|
|
%exitcond = icmp ne i32 %lftr.wideiv, %n
|
|
|
|
|
br i1 %exitcond, label %for.body3, label %for.inc13.loopexit
|
|
|
|
|
|
|
|
|
|
for.inc13.loopexit: ; preds = %for.body3
|
[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
|
|
|
%scevgep = getelementptr i32, i32* %B.addr.05, i64 %3
|
dependence analysis
Patch from Preston Briggs <preston.briggs@gmail.com>.
This is an updated version of the dependence-analysis patch, including an MIV
test based on Banerjee's inequalities.
It's a fairly complete implementation of the paper
Practical Dependence Testing
Gina Goff, Ken Kennedy, and Chau-Wen Tseng
PLDI 1991
It cannot yet propagate constraints between coupled RDIV subscripts (discussed
in Section 5.3.2 of the paper).
It's organized as a FunctionPass with a single entry point that supports testing
for dependence between two instructions in a function. If there's no dependence,
it returns null. If there's a dependence, it returns a pointer to a Dependence
which can be queried about details (what kind of dependence, is it loop
independent, direction and distance vector entries, etc). I haven't included
every imaginable feature, but there's a good selection that should be adequate
for supporting many loop transformations. Of course, it can be extended as
necessary.
Included in the patch file are many test cases, commented with C code showing
the loops and array references.
llvm-svn: 165708
2012-10-11 15:32:34 +08:00
|
|
|
br label %for.inc13
|
|
|
|
|
|
|
|
|
|
for.inc13: ; preds = %for.inc13.loopexit, %for.cond1.preheader
|
|
|
|
|
%B.addr.1.lcssa = phi i32* [ %B.addr.05, %for.cond1.preheader ], [ %scevgep, %for.inc13.loopexit ]
|
|
|
|
|
%indvars.iv.next9 = add i64 %indvars.iv8, 1
|
|
|
|
|
%lftr.wideiv10 = trunc i64 %indvars.iv.next9 to i32
|
|
|
|
|
%exitcond11 = icmp ne i32 %lftr.wideiv10, %n
|
|
|
|
|
br i1 %exitcond11, label %for.cond1.preheader, label %for.end15.loopexit
|
|
|
|
|
|
|
|
|
|
for.end15.loopexit: ; preds = %for.inc13
|
|
|
|
|
br label %for.end15
|
|
|
|
|
|
|
|
|
|
for.end15: ; preds = %for.end15.loopexit, %entry
|
|
|
|
|
ret void
|
|
|
|
|
}
|