2016-07-19 21:35:11 +08:00
|
|
|
; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple -verify-machineinstrs | FileCheck %s
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st1lane_16b(<16 x i8> %A, i8* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st1lane_16b
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st1.b
|
|
|
|
|
%tmp = extractelement <16 x i8> %A, i32 1
|
|
|
|
|
store i8 %tmp, i8* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_16b(<16 x i8> %A, i8* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_16b
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.b { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i8, i8* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <16 x i8> %A, i32 1
|
|
|
|
|
store i8 %tmp, i8* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_16b(<16 x i8> %A, i8* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_16b
|
2015-01-06 01:10:26 +08:00
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.b { v0 }[0], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i8, i8* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <16 x i8> %A, i32 0
|
|
|
|
|
store i8 %tmp, i8* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-03-29 18:18:08 +08:00
|
|
|
define void @st1lane_8h(<8 x i16> %A, i16* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st1lane_8h
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st1.h
|
|
|
|
|
%tmp = extractelement <8 x i16> %A, i32 1
|
|
|
|
|
store i16 %tmp, i16* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_8h(<8 x i16> %A, i16* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_8h
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.h { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i16, i16* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <8 x i16> %A, i32 1
|
|
|
|
|
store i16 %tmp, i16* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_8h(<8 x i16> %A, i16* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_8h
|
|
|
|
|
; CHECK: str h0, [x0, x1, lsl #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
|
|
|
%ptr = getelementptr i16, i16* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <8 x i16> %A, i32 0
|
|
|
|
|
store i16 %tmp, i16* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-03-29 18:18:08 +08:00
|
|
|
define void @st1lane_4s(<4 x i32> %A, i32* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st1lane_4s
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st1.s
|
|
|
|
|
%tmp = extractelement <4 x i32> %A, i32 1
|
|
|
|
|
store i32 %tmp, i32* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_4s(<4 x i32> %A, i32* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_4s
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.s { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i32, i32* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <4 x i32> %A, i32 1
|
|
|
|
|
store i32 %tmp, i32* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_4s(<4 x i32> %A, i32* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_4s
|
|
|
|
|
; CHECK: str s0, [x0, x1, lsl #2]
|
[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
|
|
|
%ptr = getelementptr i32, i32* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <4 x i32> %A, i32 0
|
|
|
|
|
store i32 %tmp, i32* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-04-01 18:37:09 +08:00
|
|
|
define void @st1lane_4s_float(<4 x float> %A, float* %D) {
|
|
|
|
|
; CHECK-LABEL: st1lane_4s_float
|
|
|
|
|
; CHECK: st1.s
|
|
|
|
|
%tmp = extractelement <4 x float> %A, i32 1
|
|
|
|
|
store float %tmp, float* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_4s_float(<4 x float> %A, float* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_4s_float
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.s { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr float, float* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <4 x float> %A, i32 1
|
|
|
|
|
store float %tmp, float* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_4s_float(<4 x float> %A, float* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_4s_float
|
|
|
|
|
; CHECK: str s0, [x0, x1, lsl #2]
|
[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
|
|
|
%ptr = getelementptr float, float* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <4 x float> %A, i32 0
|
|
|
|
|
store float %tmp, float* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-03-29 18:18:08 +08:00
|
|
|
define void @st1lane_2d(<2 x i64> %A, i64* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st1lane_2d
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st1.d
|
|
|
|
|
%tmp = extractelement <2 x i64> %A, i32 1
|
|
|
|
|
store i64 %tmp, i64* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_2d(<2 x i64> %A, i64* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_2d
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.d { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i64, i64* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <2 x i64> %A, i32 1
|
|
|
|
|
store i64 %tmp, i64* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_2d(<2 x i64> %A, i64* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_2d
|
|
|
|
|
; CHECK: str d0, [x0, x1, lsl #3]
|
[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
|
|
|
%ptr = getelementptr i64, i64* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <2 x i64> %A, i32 0
|
|
|
|
|
store i64 %tmp, i64* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-04-01 18:37:09 +08:00
|
|
|
define void @st1lane_2d_double(<2 x double> %A, double* %D) {
|
|
|
|
|
; CHECK-LABEL: st1lane_2d_double
|
|
|
|
|
; CHECK: st1.d
|
|
|
|
|
%tmp = extractelement <2 x double> %A, i32 1
|
|
|
|
|
store double %tmp, double* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_2d_double(<2 x double> %A, double* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_2d_double
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.d { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr double, double* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <2 x double> %A, i32 1
|
|
|
|
|
store double %tmp, double* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_2d_double(<2 x double> %A, double* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_2d_double
|
|
|
|
|
; CHECK: str d0, [x0, x1, lsl #3]
|
[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
|
|
|
%ptr = getelementptr double, double* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <2 x double> %A, i32 0
|
|
|
|
|
store double %tmp, double* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-04-01 18:37:09 +08:00
|
|
|
define void @st1lane_8b(<8 x i8> %A, i8* %D) {
|
|
|
|
|
; CHECK-LABEL: st1lane_8b
|
|
|
|
|
; CHECK: st1.b
|
|
|
|
|
%tmp = extractelement <8 x i8> %A, i32 1
|
|
|
|
|
store i8 %tmp, i8* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_8b(<8 x i8> %A, i8* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_8b
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.b { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i8, i8* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <8 x i8> %A, i32 1
|
|
|
|
|
store i8 %tmp, i8* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_8b(<8 x i8> %A, i8* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_8b
|
2015-01-06 01:10:26 +08:00
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.b { v0 }[0], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i8, i8* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <8 x i8> %A, i32 0
|
|
|
|
|
store i8 %tmp, i8* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-04-01 18:37:09 +08:00
|
|
|
define void @st1lane_4h(<4 x i16> %A, i16* %D) {
|
|
|
|
|
; CHECK-LABEL: st1lane_4h
|
|
|
|
|
; CHECK: st1.h
|
|
|
|
|
%tmp = extractelement <4 x i16> %A, i32 1
|
|
|
|
|
store i16 %tmp, i16* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_4h(<4 x i16> %A, i16* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_4h
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.h { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i16, i16* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <4 x i16> %A, i32 1
|
|
|
|
|
store i16 %tmp, i16* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_4h(<4 x i16> %A, i16* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_4h
|
|
|
|
|
; CHECK: str h0, [x0, x1, lsl #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
|
|
|
%ptr = getelementptr i16, i16* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <4 x i16> %A, i32 0
|
|
|
|
|
store i16 %tmp, i16* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-04-01 18:37:09 +08:00
|
|
|
define void @st1lane_2s(<2 x i32> %A, i32* %D) {
|
|
|
|
|
; CHECK-LABEL: st1lane_2s
|
|
|
|
|
; CHECK: st1.s
|
|
|
|
|
%tmp = extractelement <2 x i32> %A, i32 1
|
|
|
|
|
store i32 %tmp, i32* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_2s(<2 x i32> %A, i32* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_2s
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.s { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr i32, i32* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <2 x i32> %A, i32 1
|
|
|
|
|
store i32 %tmp, i32* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_2s(<2 x i32> %A, i32* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_2s
|
|
|
|
|
; CHECK: str s0, [x0, x1, lsl #2]
|
[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
|
|
|
%ptr = getelementptr i32, i32* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <2 x i32> %A, i32 0
|
|
|
|
|
store i32 %tmp, i32* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-04-01 18:37:09 +08:00
|
|
|
define void @st1lane_2s_float(<2 x float> %A, float* %D) {
|
|
|
|
|
; CHECK-LABEL: st1lane_2s_float
|
|
|
|
|
; CHECK: st1.s
|
|
|
|
|
%tmp = extractelement <2 x float> %A, i32 1
|
|
|
|
|
store float %tmp, float* %D
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2015-01-06 01:10:26 +08:00
|
|
|
define void @st1lane_ro_2s_float(<2 x float> %A, float* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane_ro_2s_float
|
|
|
|
|
; CHECK: add x[[XREG:[0-9]+]], x0, x1
|
|
|
|
|
; CHECK: st1.s { v0 }[1], [x[[XREG]]]
|
[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
|
|
|
%ptr = getelementptr float, float* %D, i64 %offset
|
2015-01-06 01:10:26 +08:00
|
|
|
%tmp = extractelement <2 x float> %A, i32 1
|
|
|
|
|
store float %tmp, float* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
define void @st1lane0_ro_2s_float(<2 x float> %A, float* %D, i64 %offset) {
|
|
|
|
|
; CHECK-LABEL: st1lane0_ro_2s_float
|
|
|
|
|
; CHECK: str s0, [x0, x1, lsl #2]
|
[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
|
|
|
%ptr = getelementptr float, float* %D, i64 %offset
|
[AArch64] Improve codegen of store lane 0 instructions by directly storing the subregister.
For 0-lane stores, we used to generate code similar to:
fmov w8, s0
str w8, [x0, x1, lsl #2]
instead of:
str s0, [x0, x1, lsl #2]
To correct that: for store lane 0 patterns, directly match to STR <subreg>0.
Byte-sized instructions don't have the special case for a 0 index,
because FPR8s are defined to have untyped content.
rdar://16372710
Differential Revision: http://reviews.llvm.org/D6772
llvm-svn: 225181
2015-01-06 01:02:28 +08:00
|
|
|
%tmp = extractelement <2 x float> %A, i32 0
|
|
|
|
|
store float %tmp, float* %ptr
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-03-29 18:18:08 +08:00
|
|
|
define void @st2lane_16b(<16 x i8> %A, <16 x i8> %B, i8* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2lane_16b
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st2.b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2lane.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, i64 1, i8* %D)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st2lane_8h(<8 x i16> %A, <8 x i16> %B, i16* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2lane_8h
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st2.h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2lane.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, i64 1, i16* %D)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st2lane_4s(<4 x i32> %A, <4 x i32> %B, i32* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2lane_4s
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st2.s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2lane.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, i64 1, i32* %D)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st2lane_2d(<2 x i64> %A, <2 x i64> %B, i64* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2lane_2d
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st2.d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2lane.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, i64 1, i64* %D)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2lane.v16i8.p0i8(<16 x i8>, <16 x i8>, i64, i8*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st2lane.v8i16.p0i16(<8 x i16>, <8 x i16>, i64, i16*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st2lane.v4i32.p0i32(<4 x i32>, <4 x i32>, i64, i32*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st2lane.v2i64.p0i64(<2 x i64>, <2 x i64>, i64, i64*) nounwind readnone
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st3lane_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, i8* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3lane_16b
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st3.b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3lane.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, i64 1, i8* %D)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3lane_8h(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, i16* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3lane_8h
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st3.h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3lane.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, i64 1, i16* %D)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3lane_4s(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, i32* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3lane_4s
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st3.s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3lane.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, i64 1, i32* %D)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3lane_2d(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, i64* %D) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3lane_2d
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st3.d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3lane.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, i64 1, i64* %D)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st3lane.v16i8.p0i8(<16 x i8>, <16 x i8>, <16 x i8>, i64, i8*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st3lane.v8i16.p0i16(<8 x i16>, <8 x i16>, <8 x i16>, i64, i16*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st3lane.v4i32.p0i32(<4 x i32>, <4 x i32>, <4 x i32>, i64, i32*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st3lane.v2i64.p0i64(<2 x i64>, <2 x i64>, <2 x i64>, i64, i64*) nounwind readnone
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st4lane_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, i8* %E) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4lane_16b
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st4.b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4lane.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, i64 1, i8* %E)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4lane_8h(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, <8 x i16> %D, i16* %E) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4lane_8h
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st4.h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4lane.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, <8 x i16> %D, i64 1, i16* %E)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4lane_4s(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> %D, i32* %E) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4lane_4s
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st4.s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4lane.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> %D, i64 1, i32* %E)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4lane_2d(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, <2 x i64> %D, i64* %E) {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4lane_2d
|
2014-03-29 18:18:08 +08:00
|
|
|
; CHECK: st4.d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4lane.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, <2 x i64> %D, i64 1, i64* %E)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st4lane.v16i8.p0i8(<16 x i8>, <16 x i8>, <16 x i8>, <16 x i8>, i64, i8*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st4lane.v8i16.p0i16(<8 x i16>, <8 x i16>, <8 x i16>, <8 x i16>, i64, i16*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st4lane.v4i32.p0i32(<4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, i64, i32*) nounwind readnone
|
|
|
|
|
declare void @llvm.aarch64.neon.st4lane.v2i64.p0i64(<2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, i64, i64*) nounwind readnone
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
define void @st2_8b(<8 x i8> %A, <8 x i8> %B, i8* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2_8b
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st2.8b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2.v8i8.p0i8(<8 x i8> %A, <8 x i8> %B, i8* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3_8b(<8 x i8> %A, <8 x i8> %B, <8 x i8> %C, i8* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3_8b
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st3.8b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3.v8i8.p0i8(<8 x i8> %A, <8 x i8> %B, <8 x i8> %C, i8* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4_8b(<8 x i8> %A, <8 x i8> %B, <8 x i8> %C, <8 x i8> %D, i8* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4_8b
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st4.8b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4.v8i8.p0i8(<8 x i8> %A, <8 x i8> %B, <8 x i8> %C, <8 x i8> %D, i8* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2.v8i8.p0i8(<8 x i8>, <8 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st3.v8i8.p0i8(<8 x i8>, <8 x i8>, <8 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st4.v8i8.p0i8(<8 x i8>, <8 x i8>, <8 x i8>, <8 x i8>, i8*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st2_16b(<16 x i8> %A, <16 x i8> %B, i8* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2_16b
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st2.16b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, i8* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, i8* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3_16b
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st3.16b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, i8* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, i8* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4_16b
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st4.16b
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, i8* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2.v16i8.p0i8(<16 x i8>, <16 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st3.v16i8.p0i8(<16 x i8>, <16 x i8>, <16 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st4.v16i8.p0i8(<16 x i8>, <16 x i8>, <16 x i8>, <16 x i8>, i8*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st2_4h(<4 x i16> %A, <4 x i16> %B, i16* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2_4h
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st2.4h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2.v4i16.p0i16(<4 x i16> %A, <4 x i16> %B, i16* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3_4h(<4 x i16> %A, <4 x i16> %B, <4 x i16> %C, i16* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3_4h
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st3.4h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3.v4i16.p0i16(<4 x i16> %A, <4 x i16> %B, <4 x i16> %C, i16* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4_4h(<4 x i16> %A, <4 x i16> %B, <4 x i16> %C, <4 x i16> %D, i16* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4_4h
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st4.4h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4.v4i16.p0i16(<4 x i16> %A, <4 x i16> %B, <4 x i16> %C, <4 x i16> %D, i16* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2.v4i16.p0i16(<4 x i16>, <4 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st3.v4i16.p0i16(<4 x i16>, <4 x i16>, <4 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st4.v4i16.p0i16(<4 x i16>, <4 x i16>, <4 x i16>, <4 x i16>, i16*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st2_8h(<8 x i16> %A, <8 x i16> %B, i16* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2_8h
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st2.8h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, i16* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3_8h(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, i16* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3_8h
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st3.8h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, i16* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4_8h(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, <8 x i16> %D, i16* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4_8h
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st4.8h
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, <8 x i16> %D, i16* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2.v8i16.p0i16(<8 x i16>, <8 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st3.v8i16.p0i16(<8 x i16>, <8 x i16>, <8 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st4.v8i16.p0i16(<8 x i16>, <8 x i16>, <8 x i16>, <8 x i16>, i16*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st2_2s(<2 x i32> %A, <2 x i32> %B, i32* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2_2s
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st2.2s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2.v2i32.p0i32(<2 x i32> %A, <2 x i32> %B, i32* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3_2s(<2 x i32> %A, <2 x i32> %B, <2 x i32> %C, i32* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3_2s
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st3.2s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3.v2i32.p0i32(<2 x i32> %A, <2 x i32> %B, <2 x i32> %C, i32* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4_2s(<2 x i32> %A, <2 x i32> %B, <2 x i32> %C, <2 x i32> %D, i32* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4_2s
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st4.2s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4.v2i32.p0i32(<2 x i32> %A, <2 x i32> %B, <2 x i32> %C, <2 x i32> %D, i32* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2.v2i32.p0i32(<2 x i32>, <2 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st3.v2i32.p0i32(<2 x i32>, <2 x i32>, <2 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st4.v2i32.p0i32(<2 x i32>, <2 x i32>, <2 x i32>, <2 x i32>, i32*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st2_4s(<4 x i32> %A, <4 x i32> %B, i32* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2_4s
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st2.4s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, i32* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3_4s(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, i32* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3_4s
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st3.4s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, i32* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4_4s(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> %D, i32* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4_4s
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st4.4s
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> %D, i32* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2.v4i32.p0i32(<4 x i32>, <4 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st3.v4i32.p0i32(<4 x i32>, <4 x i32>, <4 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st4.v4i32.p0i32(<4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, i32*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
2015-03-14 09:43:57 +08:00
|
|
|
; If there's only one element, st2/3/4 don't make much sense, stick to st1.
|
2014-03-29 18:18:08 +08:00
|
|
|
define void @st2_1d(<1 x i64> %A, <1 x i64> %B, i64* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2_1d
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st1.1d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2.v1i64.p0i64(<1 x i64> %A, <1 x i64> %B, i64* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3_1d(<1 x i64> %A, <1 x i64> %B, <1 x i64> %C, i64* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3_1d
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st1.1d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3.v1i64.p0i64(<1 x i64> %A, <1 x i64> %B, <1 x i64> %C, i64* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4_1d(<1 x i64> %A, <1 x i64> %B, <1 x i64> %C, <1 x i64> %D, i64* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4_1d
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st1.1d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4.v1i64.p0i64(<1 x i64> %A, <1 x i64> %B, <1 x i64> %C, <1 x i64> %D, i64* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2.v1i64.p0i64(<1 x i64>, <1 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st3.v1i64.p0i64(<1 x i64>, <1 x i64>, <1 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st4.v1i64.p0i64(<1 x i64>, <1 x i64>, <1 x i64>, <1 x i64>, i64*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st2_2d(<2 x i64> %A, <2 x i64> %B, i64* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st2_2d
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st2.2d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st2.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, i64* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st3_2d(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, i64* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st3_2d
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st3.2d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st3.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, i64* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st4_2d(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, <2 x i64> %D, i64* %P) nounwind {
|
2014-04-01 18:37:09 +08:00
|
|
|
; CHECK-LABEL: st4_2d
|
2015-03-14 09:43:57 +08:00
|
|
|
; CHECK: st4.2d
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st4.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, <2 x i64> %D, i64* %P)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st2.v2i64.p0i64(<2 x i64>, <2 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st3.v2i64.p0i64(<2 x i64>, <2 x i64>, <2 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st4.v2i64.p0i64(<2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, i64*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st1x2.v8i8.p0i8(<8 x i8>, <8 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v4i16.p0i16(<4 x i16>, <4 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v2i32.p0i32(<2 x i32>, <2 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v2f32.p0f32(<2 x float>, <2 x float>, float*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v1i64.p0i64(<1 x i64>, <1 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v1f64.p0f64(<1 x double>, <1 x double>, double*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st1_x2_v8i8(<8 x i8> %A, <8 x i8> %B, i8* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v8i8:
|
|
|
|
|
; CHECK: st1.8b { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v8i8.p0i8(<8 x i8> %A, <8 x i8> %B, i8* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v4i16(<4 x i16> %A, <4 x i16> %B, i16* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v4i16:
|
|
|
|
|
; CHECK: st1.4h { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v4i16.p0i16(<4 x i16> %A, <4 x i16> %B, i16* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v2i32(<2 x i32> %A, <2 x i32> %B, i32* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v2i32:
|
|
|
|
|
; CHECK: st1.2s { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v2i32.p0i32(<2 x i32> %A, <2 x i32> %B, i32* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v2f32(<2 x float> %A, <2 x float> %B, float* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v2f32:
|
|
|
|
|
; CHECK: st1.2s { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v2f32.p0f32(<2 x float> %A, <2 x float> %B, float* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v1i64(<1 x i64> %A, <1 x i64> %B, i64* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v1i64:
|
|
|
|
|
; CHECK: st1.1d { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v1i64.p0i64(<1 x i64> %A, <1 x i64> %B, i64* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v1f64(<1 x double> %A, <1 x double> %B, double* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v1f64:
|
|
|
|
|
; CHECK: st1.1d { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v1f64.p0f64(<1 x double> %A, <1 x double> %B, double* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st1x2.v16i8.p0i8(<16 x i8>, <16 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v8i16.p0i16(<8 x i16>, <8 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v4i32.p0i32(<4 x i32>, <4 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v4f32.p0f32(<4 x float>, <4 x float>, float*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v2i64.p0i64(<2 x i64>, <2 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x2.v2f64.p0f64(<2 x double>, <2 x double>, double*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st1_x2_v16i8(<16 x i8> %A, <16 x i8> %B, i8* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v16i8:
|
|
|
|
|
; CHECK: st1.16b { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, i8* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v8i16(<8 x i16> %A, <8 x i16> %B, i16* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v8i16:
|
|
|
|
|
; CHECK: st1.8h { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, i16* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v4i32(<4 x i32> %A, <4 x i32> %B, i32* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v4i32:
|
|
|
|
|
; CHECK: st1.4s { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, i32* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v4f32(<4 x float> %A, <4 x float> %B, float* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v4f32:
|
|
|
|
|
; CHECK: st1.4s { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v4f32.p0f32(<4 x float> %A, <4 x float> %B, float* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v2i64(<2 x i64> %A, <2 x i64> %B, i64* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v2i64:
|
|
|
|
|
; CHECK: st1.2d { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, i64* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x2_v2f64(<2 x double> %A, <2 x double> %B, double* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x2_v2f64:
|
|
|
|
|
; CHECK: st1.2d { {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x2.v2f64.p0f64(<2 x double> %A, <2 x double> %B, double* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st1x3.v8i8.p0i8(<8 x i8>, <8 x i8>, <8 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v4i16.p0i16(<4 x i16>, <4 x i16>, <4 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v2i32.p0i32(<2 x i32>, <2 x i32>, <2 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v2f32.p0f32(<2 x float>, <2 x float>, <2 x float>, float*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v1i64.p0i64(<1 x i64>, <1 x i64>, <1 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v1f64.p0f64(<1 x double>, <1 x double>, <1 x double>, double*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st1_x3_v8i8(<8 x i8> %A, <8 x i8> %B, <8 x i8> %C, i8* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v8i8:
|
|
|
|
|
; CHECK: st1.8b { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v8i8.p0i8(<8 x i8> %A, <8 x i8> %B, <8 x i8> %C, i8* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v4i16(<4 x i16> %A, <4 x i16> %B, <4 x i16> %C, i16* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v4i16:
|
|
|
|
|
; CHECK: st1.4h { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v4i16.p0i16(<4 x i16> %A, <4 x i16> %B, <4 x i16> %C, i16* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v2i32(<2 x i32> %A, <2 x i32> %B, <2 x i32> %C, i32* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v2i32:
|
|
|
|
|
; CHECK: st1.2s { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v2i32.p0i32(<2 x i32> %A, <2 x i32> %B, <2 x i32> %C, i32* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v2f32(<2 x float> %A, <2 x float> %B, <2 x float> %C, float* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v2f32:
|
|
|
|
|
; CHECK: st1.2s { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v2f32.p0f32(<2 x float> %A, <2 x float> %B, <2 x float> %C, float* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v1i64(<1 x i64> %A, <1 x i64> %B, <1 x i64> %C, i64* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v1i64:
|
|
|
|
|
; CHECK: st1.1d { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v1i64.p0i64(<1 x i64> %A, <1 x i64> %B, <1 x i64> %C, i64* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v1f64(<1 x double> %A, <1 x double> %B, <1 x double> %C, double* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v1f64:
|
|
|
|
|
; CHECK: st1.1d { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v1f64.p0f64(<1 x double> %A, <1 x double> %B, <1 x double> %C, double* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st1x3.v16i8.p0i8(<16 x i8>, <16 x i8>, <16 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v8i16.p0i16(<8 x i16>, <8 x i16>, <8 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v4i32.p0i32(<4 x i32>, <4 x i32>, <4 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v4f32.p0f32(<4 x float>, <4 x float>, <4 x float>, float*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v2i64.p0i64(<2 x i64>, <2 x i64>, <2 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x3.v2f64.p0f64(<2 x double>, <2 x double>, <2 x double>, double*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st1_x3_v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, i8* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v16i8:
|
|
|
|
|
; CHECK: st1.16b { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, i8* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v8i16(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, i16* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v8i16:
|
|
|
|
|
; CHECK: st1.8h { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, i16* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v4i32(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, i32* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v4i32:
|
|
|
|
|
; CHECK: st1.4s { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, i32* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v4f32(<4 x float> %A, <4 x float> %B, <4 x float> %C, float* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v4f32:
|
|
|
|
|
; CHECK: st1.4s { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v4f32.p0f32(<4 x float> %A, <4 x float> %B, <4 x float> %C, float* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v2i64(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, i64* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v2i64:
|
|
|
|
|
; CHECK: st1.2d { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, i64* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x3_v2f64(<2 x double> %A, <2 x double> %B, <2 x double> %C, double* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x3_v2f64:
|
|
|
|
|
; CHECK: st1.2d { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x3.v2f64.p0f64(<2 x double> %A, <2 x double> %B, <2 x double> %C, double* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st1x4.v8i8.p0i8(<8 x i8>, <8 x i8>, <8 x i8>, <8 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v4i16.p0i16(<4 x i16>, <4 x i16>, <4 x i16>, <4 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v2i32.p0i32(<2 x i32>, <2 x i32>, <2 x i32>, <2 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v2f32.p0f32(<2 x float>, <2 x float>, <2 x float>, <2 x float>, float*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v1i64.p0i64(<1 x i64>, <1 x i64>, <1 x i64>, <1 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v1f64.p0f64(<1 x double>, <1 x double>, <1 x double>, <1 x double>, double*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st1_x4_v8i8(<8 x i8> %A, <8 x i8> %B, <8 x i8> %C, <8 x i8> %D, i8* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v8i8:
|
|
|
|
|
; CHECK: st1.8b { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v8i8.p0i8(<8 x i8> %A, <8 x i8> %B, <8 x i8> %C, <8 x i8> %D, i8* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v4i16(<4 x i16> %A, <4 x i16> %B, <4 x i16> %C, <4 x i16> %D, i16* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v4i16:
|
|
|
|
|
; CHECK: st1.4h { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v4i16.p0i16(<4 x i16> %A, <4 x i16> %B, <4 x i16> %C, <4 x i16> %D, i16* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v2i32(<2 x i32> %A, <2 x i32> %B, <2 x i32> %C, <2 x i32> %D, i32* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v2i32:
|
|
|
|
|
; CHECK: st1.2s { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v2i32.p0i32(<2 x i32> %A, <2 x i32> %B, <2 x i32> %C, <2 x i32> %D, i32* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v2f32(<2 x float> %A, <2 x float> %B, <2 x float> %C, <2 x float> %D, float* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v2f32:
|
|
|
|
|
; CHECK: st1.2s { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v2f32.p0f32(<2 x float> %A, <2 x float> %B, <2 x float> %C, <2 x float> %D, float* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v1i64(<1 x i64> %A, <1 x i64> %B, <1 x i64> %C, <1 x i64> %D, i64* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v1i64:
|
|
|
|
|
; CHECK: st1.1d { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v1i64.p0i64(<1 x i64> %A, <1 x i64> %B, <1 x i64> %C, <1 x i64> %D, i64* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v1f64(<1 x double> %A, <1 x double> %B, <1 x double> %C, <1 x double> %D, double* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v1f64:
|
|
|
|
|
; CHECK: st1.1d { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v1f64.p0f64(<1 x double> %A, <1 x double> %B, <1 x double> %C, <1 x double> %D, double* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2014-05-24 20:50:23 +08:00
|
|
|
declare void @llvm.aarch64.neon.st1x4.v16i8.p0i8(<16 x i8>, <16 x i8>, <16 x i8>, <16 x i8>, i8*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v8i16.p0i16(<8 x i16>, <8 x i16>, <8 x i16>, <8 x i16>, i16*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v4i32.p0i32(<4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, i32*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v4f32.p0f32(<4 x float>, <4 x float>, <4 x float>, <4 x float>, float*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v2i64.p0i64(<2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, i64*) nounwind readonly
|
|
|
|
|
declare void @llvm.aarch64.neon.st1x4.v2f64.p0f64(<2 x double>, <2 x double>, <2 x double>, <2 x double>, double*) nounwind readonly
|
2014-03-29 18:18:08 +08:00
|
|
|
|
|
|
|
|
define void @st1_x4_v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, i8* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v16i8:
|
|
|
|
|
; CHECK: st1.16b { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v16i8.p0i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, i8* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v8i16(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, <8 x i16> %D, i16* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v8i16:
|
|
|
|
|
; CHECK: st1.8h { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v8i16.p0i16(<8 x i16> %A, <8 x i16> %B, <8 x i16> %C, <8 x i16> %D, i16* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v4i32(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> %D, i32* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v4i32:
|
|
|
|
|
; CHECK: st1.4s { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v4i32.p0i32(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> %D, i32* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v4f32(<4 x float> %A, <4 x float> %B, <4 x float> %C, <4 x float> %D, float* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v4f32:
|
|
|
|
|
; CHECK: st1.4s { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v4f32.p0f32(<4 x float> %A, <4 x float> %B, <4 x float> %C, <4 x float> %D, float* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v2i64(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, <2 x i64> %D, i64* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v2i64:
|
|
|
|
|
; CHECK: st1.2d { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v2i64.p0i64(<2 x i64> %A, <2 x i64> %B, <2 x i64> %C, <2 x i64> %D, i64* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @st1_x4_v2f64(<2 x double> %A, <2 x double> %B, <2 x double> %C, <2 x double> %D, double* %addr) {
|
|
|
|
|
; CHECK-LABEL: st1_x4_v2f64:
|
|
|
|
|
; CHECK: st1.2d { {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} }, [x0]
|
2014-05-24 20:50:23 +08:00
|
|
|
call void @llvm.aarch64.neon.st1x4.v2f64.p0f64(<2 x double> %A, <2 x double> %B, <2 x double> %C, <2 x double> %D, double* %addr)
|
2014-03-29 18:18:08 +08:00
|
|
|
ret void
|
|
|
|
|
}
|