2013-05-06 16:17:29 +00:00
|
|
|
; Test the handling of base + 12-bit displacement addresses for large frames,
|
2013-10-31 12:14:17 +00:00
|
|
|
; in cases where no 20-bit form exists. The tests here assume z10 register
|
|
|
|
|
; pressure, without the high words being available.
|
2013-05-06 16:17:29 +00:00
|
|
|
;
|
2013-10-31 12:14:17 +00:00
|
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | \
|
|
|
|
|
; RUN: FileCheck -check-prefix=CHECK-NOFP %s
|
|
|
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 -disable-fp-elim | \
|
|
|
|
|
; RUN: FileCheck -check-prefix=CHECK-FP %s
|
2013-05-06 16:17:29 +00:00
|
|
|
|
|
|
|
|
; This file tests what happens when a displacement is converted from
|
|
|
|
|
; being relative to the start of a frame object to being relative to
|
|
|
|
|
; the frame itself. In some cases the test is only possible if two
|
|
|
|
|
; objects are allocated.
|
|
|
|
|
;
|
|
|
|
|
; Rather than rely on a particular order for those objects, the tests
|
|
|
|
|
; instead allocate two objects of the same size and apply the test to
|
|
|
|
|
; both of them. For consistency, all tests follow this model, even if
|
|
|
|
|
; one object would actually be enough.
|
|
|
|
|
|
|
|
|
|
; First check the highest in-range offset after conversion, which is 4092
|
|
|
|
|
; for word-addressing instructions like MVHI.
|
|
|
|
|
;
|
2013-07-05 13:11:52 +00:00
|
|
|
; The last in-range doubleword offset is 4088. Since the frame has two
|
|
|
|
|
; emergency spill slots at 160(%r15), the amount that we need to allocate
|
|
|
|
|
; in order to put another object at offset 4088 is (4088 - 176) / 4 = 978
|
2013-05-06 16:17:29 +00:00
|
|
|
; words.
|
|
|
|
|
define void @f1() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f1:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: mvhi 4092(%r15), 42
|
|
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f1:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: mvhi 4092(%r11), 42
|
|
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [978 x i32], align 8
|
|
|
|
|
%region2 = alloca [978 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [978 x i32], [978 x i32]* %region1, i64 0, i64 1
|
|
|
|
|
%ptr2 = getelementptr inbounds [978 x i32], [978 x i32]* %region2, i64 0, i64 1
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; Test the first out-of-range offset. We cannot use an index register here.
|
|
|
|
|
define void @f2() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f2:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay %r1, 4096(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 0(%r1), 42
|
|
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f2:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: lay %r1, 4096(%r11)
|
|
|
|
|
; CHECK-FP: mvhi 0(%r1), 42
|
|
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [978 x i32], align 8
|
|
|
|
|
%region2 = alloca [978 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [978 x i32], [978 x i32]* %region1, i64 0, i64 2
|
|
|
|
|
%ptr2 = getelementptr inbounds [978 x i32], [978 x i32]* %region2, i64 0, i64 2
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; Test the next offset after that.
|
|
|
|
|
define void @f3() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f3:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay %r1, 4096(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 4(%r1), 42
|
|
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f3:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: lay %r1, 4096(%r11)
|
|
|
|
|
; CHECK-FP: mvhi 4(%r1), 42
|
|
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [978 x i32], align 8
|
|
|
|
|
%region2 = alloca [978 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [978 x i32], [978 x i32]* %region1, i64 0, i64 3
|
|
|
|
|
%ptr2 = getelementptr inbounds [978 x i32], [978 x i32]* %region2, i64 0, i64 3
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; Add 4096 bytes (1024 words) to the size of each object and repeat.
|
|
|
|
|
define void @f4() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f4:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay %r1, 4096(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 4092(%r1), 42
|
|
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f4:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: lay %r1, 4096(%r11)
|
|
|
|
|
; CHECK-FP: mvhi 4092(%r1), 42
|
|
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [2002 x i32], align 8
|
|
|
|
|
%region2 = alloca [2002 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [2002 x i32], [2002 x i32]* %region1, i64 0, i64 1
|
|
|
|
|
%ptr2 = getelementptr inbounds [2002 x i32], [2002 x i32]* %region2, i64 0, i64 1
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; ...as above.
|
|
|
|
|
define void @f5() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f5:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay %r1, 8192(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 0(%r1), 42
|
|
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f5:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: lay %r1, 8192(%r11)
|
|
|
|
|
; CHECK-FP: mvhi 0(%r1), 42
|
|
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [2002 x i32], align 8
|
|
|
|
|
%region2 = alloca [2002 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [2002 x i32], [2002 x i32]* %region1, i64 0, i64 2
|
|
|
|
|
%ptr2 = getelementptr inbounds [2002 x i32], [2002 x i32]* %region2, i64 0, i64 2
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; ...as above.
|
|
|
|
|
define void @f6() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f6:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay %r1, 8192(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 4(%r1), 42
|
|
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f6:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: lay %r1, 8192(%r11)
|
|
|
|
|
; CHECK-FP: mvhi 4(%r1), 42
|
|
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [2002 x i32], align 8
|
|
|
|
|
%region2 = alloca [2002 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [2002 x i32], [2002 x i32]* %region1, i64 0, i64 3
|
|
|
|
|
%ptr2 = getelementptr inbounds [2002 x i32], [2002 x i32]* %region2, i64 0, i64 3
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; Now try an offset of 4092 from the start of the object, with the object
|
2013-07-05 13:11:52 +00:00
|
|
|
; being at offset 8192. This time we need objects of (8192 - 176) / 4 = 2004
|
2013-05-06 16:17:29 +00:00
|
|
|
; words.
|
|
|
|
|
define void @f7() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f7:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay %r1, 8192(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 4092(%r1), 42
|
|
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f7:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: lay %r1, 8192(%r11)
|
|
|
|
|
; CHECK-FP: mvhi 4092(%r1), 42
|
|
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [2004 x i32], align 8
|
|
|
|
|
%region2 = alloca [2004 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [2004 x i32], [2004 x i32]* %region1, i64 0, i64 1023
|
|
|
|
|
%ptr2 = getelementptr inbounds [2004 x i32], [2004 x i32]* %region2, i64 0, i64 1023
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; Keep the object-relative offset the same but bump the size of the
|
|
|
|
|
; objects by one doubleword.
|
|
|
|
|
define void @f8() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f8:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay %r1, 12288(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 4(%r1), 42
|
|
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f8:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: lay %r1, 12288(%r11)
|
|
|
|
|
; CHECK-FP: mvhi 4(%r1), 42
|
|
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [2006 x i32], align 8
|
|
|
|
|
%region2 = alloca [2006 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [2006 x i32], [2006 x i32]* %region1, i64 0, i64 1023
|
|
|
|
|
%ptr2 = getelementptr inbounds [2006 x i32], [2006 x i32]* %region2, i64 0, i64 1023
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; Check a case where the original displacement is out of range. The backend
|
2013-08-23 11:18:53 +00:00
|
|
|
; should force STY to be used instead.
|
2013-05-06 16:17:29 +00:00
|
|
|
define void @f9() {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f9:
|
2013-08-23 11:18:53 +00:00
|
|
|
; CHECK-NOFP: lhi [[TMP:%r[0-5]]], 42
|
|
|
|
|
; CHECK-NOFP: sty [[TMP]], 12296(%r15)
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f9:
|
2013-08-23 11:18:53 +00:00
|
|
|
; CHECK-FP: lhi [[TMP:%r[0-5]]], 42
|
|
|
|
|
; CHECK-FP: sty [[TMP]], 12296(%r11)
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: br %r14
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [2006 x i32], align 8
|
|
|
|
|
%region2 = alloca [2006 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [2006 x i32], [2006 x i32]* %region1, i64 0, i64 1024
|
|
|
|
|
%ptr2 = getelementptr inbounds [2006 x i32], [2006 x i32]* %region2, i64 0, i64 1024
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2013-07-05 13:11:52 +00:00
|
|
|
; Repeat f2 in a case that needs the emergency spill slots (because all
|
2013-05-06 16:17:29 +00:00
|
|
|
; call-clobbered registers are live and no call-saved ones have been
|
|
|
|
|
; allocated).
|
|
|
|
|
define void @f10(i32 *%vptr) {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f10:
|
2013-07-05 13:11:52 +00:00
|
|
|
; CHECK-NOFP: stg [[REGISTER:%r[1-9][0-4]?]], [[OFFSET:160|168]](%r15)
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay [[REGISTER]], 4096(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 0([[REGISTER]]), 42
|
2013-07-05 13:11:52 +00:00
|
|
|
; CHECK-NOFP: lg [[REGISTER]], [[OFFSET]](%r15)
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: br %r14
|
|
|
|
|
;
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-FP-LABEL: f10:
|
2013-07-05 13:11:52 +00:00
|
|
|
; CHECK-FP: stg [[REGISTER:%r[1-9][0-4]?]], [[OFFSET:160|168]](%r11)
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: lay [[REGISTER]], 4096(%r11)
|
|
|
|
|
; CHECK-FP: mvhi 0([[REGISTER]]), 42
|
2013-07-05 13:11:52 +00:00
|
|
|
; CHECK-FP: lg [[REGISTER]], [[OFFSET]](%r11)
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-FP: br %r14
|
2015-02-27 21:17:42 +00:00
|
|
|
%i0 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i1 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i3 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i4 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i5 = load volatile i32 , i32 *%vptr
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [978 x i32], align 8
|
|
|
|
|
%region2 = alloca [978 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [978 x i32], [978 x i32]* %region1, i64 0, i64 2
|
|
|
|
|
%ptr2 = getelementptr inbounds [978 x i32], [978 x i32]* %region2, i64 0, i64 2
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
store volatile i32 %i0, i32 *%vptr
|
|
|
|
|
store volatile i32 %i1, i32 *%vptr
|
|
|
|
|
store volatile i32 %i3, i32 *%vptr
|
|
|
|
|
store volatile i32 %i4, i32 *%vptr
|
|
|
|
|
store volatile i32 %i5, i32 *%vptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2013-07-05 13:11:52 +00:00
|
|
|
; And again with maximum register pressure. The only spill slots that the
|
|
|
|
|
; NOFP case needs are the emergency ones, so the offsets are the same as for f2.
|
2013-12-10 10:36:34 +00:00
|
|
|
; The FP case needs to spill an extra register and is too dependent on
|
|
|
|
|
; register allocation heuristics for a stable test.
|
2013-05-06 16:17:29 +00:00
|
|
|
define void @f11(i32 *%vptr) {
|
2013-07-14 06:24:09 +00:00
|
|
|
; CHECK-NOFP-LABEL: f11:
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: stmg %r6, %r15,
|
2013-07-05 13:11:52 +00:00
|
|
|
; CHECK-NOFP: stg [[REGISTER:%r[1-9][0-4]?]], [[OFFSET:160|168]](%r15)
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lay [[REGISTER]], 4096(%r15)
|
|
|
|
|
; CHECK-NOFP: mvhi 0([[REGISTER]]), 42
|
2013-07-05 13:11:52 +00:00
|
|
|
; CHECK-NOFP: lg [[REGISTER]], [[OFFSET]](%r15)
|
2013-05-06 16:17:29 +00:00
|
|
|
; CHECK-NOFP: lmg %r6, %r15,
|
|
|
|
|
; CHECK-NOFP: br %r14
|
2015-02-27 21:17:42 +00:00
|
|
|
%i0 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i1 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i3 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i4 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i5 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i6 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i7 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i8 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i9 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i10 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i11 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i12 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i13 = load volatile i32 , i32 *%vptr
|
|
|
|
|
%i14 = load volatile i32 , i32 *%vptr
|
2013-07-05 13:11:52 +00:00
|
|
|
%region1 = alloca [978 x i32], align 8
|
|
|
|
|
%region2 = alloca [978 x i32], align 8
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
|
|
|
%ptr1 = getelementptr inbounds [978 x i32], [978 x i32]* %region1, i64 0, i64 2
|
|
|
|
|
%ptr2 = getelementptr inbounds [978 x i32], [978 x i32]* %region2, i64 0, i64 2
|
2013-05-06 16:17:29 +00:00
|
|
|
store volatile i32 42, i32 *%ptr1
|
|
|
|
|
store volatile i32 42, i32 *%ptr2
|
|
|
|
|
store volatile i32 %i0, i32 *%vptr
|
|
|
|
|
store volatile i32 %i1, i32 *%vptr
|
|
|
|
|
store volatile i32 %i3, i32 *%vptr
|
|
|
|
|
store volatile i32 %i4, i32 *%vptr
|
|
|
|
|
store volatile i32 %i5, i32 *%vptr
|
|
|
|
|
store volatile i32 %i6, i32 *%vptr
|
|
|
|
|
store volatile i32 %i7, i32 *%vptr
|
|
|
|
|
store volatile i32 %i8, i32 *%vptr
|
|
|
|
|
store volatile i32 %i9, i32 *%vptr
|
|
|
|
|
store volatile i32 %i10, i32 *%vptr
|
|
|
|
|
store volatile i32 %i11, i32 *%vptr
|
|
|
|
|
store volatile i32 %i12, i32 *%vptr
|
|
|
|
|
store volatile i32 %i13, i32 *%vptr
|
|
|
|
|
store volatile i32 %i14, i32 *%vptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
