2016-06-24 20:23:17 +08:00
|
|
|
; RUN: llc -mtriple=mips-linux -relocation-model=static < %s | FileCheck --check-prefixes=ALL,O32,O32-BE %s
|
|
|
|
|
; RUN: llc -mtriple=mipsel-linux -relocation-model=static < %s | FileCheck --check-prefixes=ALL,O32,O32-LE %s
|
2014-08-01 17:17:39 +08:00
|
|
|
|
2016-06-24 20:23:17 +08:00
|
|
|
; RUN-TODO: llc -march=mips64 -relocation-model=static -target-abi o32 < %s | FileCheck --check-prefixes=ALL,O32 %s
|
|
|
|
|
; RUN-TODO: llc -march=mips64el -relocation-model=static -target-abi o32 < %s | FileCheck --check-prefixes=ALL,O32 %s
|
2014-08-01 17:17:39 +08:00
|
|
|
|
2016-06-24 20:23:17 +08:00
|
|
|
; RUN: llc -mtriple=mips64-linux -relocation-model=static -target-abi n32 < %s | FileCheck --check-prefixes=ALL,NEW,N32,NEW-BE %s
|
|
|
|
|
; RUN: llc -mtriple=mips64el-linux -relocation-model=static -target-abi n32 < %s | FileCheck --check-prefixes=ALL,NEW,N32,NEW-LE %s
|
2014-08-01 17:17:39 +08:00
|
|
|
|
2016-06-24 20:23:17 +08:00
|
|
|
; RUN: llc -march=mips64 -relocation-model=static -target-abi n64 < %s | FileCheck --check-prefixes=ALL,NEW,N64,NEW-BE %s
|
|
|
|
|
; RUN: llc -march=mips64el -relocation-model=static -target-abi n64 < %s | FileCheck --check-prefixes=ALL,NEW,N64,NEW-LE %s
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
@hwords = global [3 x i16] zeroinitializer, align 1
|
|
|
|
|
@words = global [3 x i32] zeroinitializer, align 1
|
|
|
|
|
@dwords = global [3 x i64] zeroinitializer, align 1
|
|
|
|
|
|
|
|
|
|
define void @fn_i16_dotdotdot_i16(i16 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i16_dotdotdot_i16:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
; O32-DAG: sw $5, 12([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 12 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the 4 byte slot for the first
|
|
|
|
|
; fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 12
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]]
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(hwords)
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(hwords)
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(hwords)(
|
|
|
|
|
|
|
|
|
|
; ALL-DAG: sh [[ARG1]], 2([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA2]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; ALL-DAG: sh [[ARG2]], 4([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i16
|
[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
|
|
|
%e1 = getelementptr [3 x i16], [3 x i16]* @hwords, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i16 %arg1, i16* %e1, align 2
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i16
|
[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
|
|
|
%e2 = getelementptr [3 x i16], [3 x i16]* @hwords, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i16 %arg2, i16* %e2, align 2
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @fn_i16_dotdotdot_i32(i16 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i16_dotdotdot_i32:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
; O32-DAG: sw $5, 12([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 12 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the 4 byte slot for the first
|
|
|
|
|
; fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 12
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]]
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(words)
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(words)
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(words)(
|
|
|
|
|
|
|
|
|
|
; ALL-DAG: sw [[ARG1]], 4([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA2]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; ALL-DAG: sw [[ARG2]], 8([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i32
|
[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
|
|
|
%e1 = getelementptr [3 x i32], [3 x i32]* @words, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i32 %arg1, i32* %e1, align 4
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i32
|
[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
|
|
|
%e2 = getelementptr [3 x i32], [3 x i32]* @words, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i32 %arg2, i32* %e2, align 4
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @fn_i16_dotdotdot_i64(i16 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i16_dotdotdot_i64:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
; O32-DAG: sw $5, 12([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 12 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the 4 byte slot for the first
|
|
|
|
|
; fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 12
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] (and realign pointer for O32)
|
|
|
|
|
; O32: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP0:\$[0-9]+]], [[VA]], 7
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP1:\$[0-9]+]], $zero, -8
|
|
|
|
|
; O32-DAG: and [[VA_TMP2:\$[0-9]+]], [[VA_TMP0]], [[VA_TMP1]]
|
|
|
|
|
; O32-DAG: ori [[VA2:\$[0-9]+]], [[VA_TMP2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion and copy it to the global.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(dwords)
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA_TMP2]])
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[ARG1]], 8([[GV]])
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA_TMP2]])
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[ARG1]], 12([[GV]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(dwords)
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(dwords)(
|
|
|
|
|
; NEW-DAG: ld [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-DAG: sd [[ARG1]], 8([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; FIXME: We're still aligned from the last one but CodeGen doesn't spot that.
|
|
|
|
|
; O32: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP0:\$[0-9]+]], [[VA]], 7
|
|
|
|
|
; O32-DAG: and [[VA_TMP2:\$[0-9]+]], [[VA_TMP0]], [[VA_TMP1]]
|
|
|
|
|
; O32-DAG: ori [[VA2:\$[0-9]+]], [[VA_TMP2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion and copy it to the global.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; O32-DAG: sw [[ARG2]], 16([[GV]])
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA_TMP2]])
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[ARG2]], 20([[GV]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: ld [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-DAG: sd [[ARG2]], 16([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i64
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%e1 = getelementptr [3 x i64], [3 x i64]* @dwords, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i64 %arg1, i64* %e1, align 8
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i64
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%e2 = getelementptr [3 x i64], [3 x i64]* @dwords, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i64 %arg2, i64* %e2, align 8
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @fn_i32_dotdotdot_i16(i32 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i32_dotdotdot_i16:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
; O32-DAG: sw $5, 12([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 12 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the 4 byte slot for the first
|
|
|
|
|
; fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 12
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]]
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(hwords)
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(hwords)
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(hwords)(
|
|
|
|
|
|
|
|
|
|
; ALL-DAG: sh [[ARG1]], 2([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA2]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; ALL-DAG: sh [[ARG2]], 4([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i16
|
[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
|
|
|
%e1 = getelementptr [3 x i16], [3 x i16]* @hwords, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i16 %arg1, i16* %e1, align 2
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i16
|
[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
|
|
|
%e2 = getelementptr [3 x i16], [3 x i16]* @hwords, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i16 %arg2, i16* %e2, align 2
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @fn_i32_dotdotdot_i32(i32 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i32_dotdotdot_i32:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
; O32-DAG: sw $5, 12([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 12 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the 4 byte slot for the first
|
|
|
|
|
; fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 12
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]]
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(words)
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(words)
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(words)(
|
|
|
|
|
|
|
|
|
|
; ALL-DAG: sw [[ARG1]], 4([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA2]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; ALL-DAG: sw [[ARG2]], 8([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i32
|
[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
|
|
|
%e1 = getelementptr [3 x i32], [3 x i32]* @words, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i32 %arg1, i32* %e1, align 4
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i32
|
[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
|
|
|
%e2 = getelementptr [3 x i32], [3 x i32]* @words, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i32 %arg2, i32* %e2, align 4
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @fn_i32_dotdotdot_i64(i32 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i32_dotdotdot_i64:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
; O32-DAG: sw $5, 12([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 12 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the 4 byte slot for the first
|
|
|
|
|
; fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 12
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] (and realign pointer for O32)
|
|
|
|
|
; O32: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP0:\$[0-9]+]], [[VA]], 7
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP1:\$[0-9]+]], $zero, -8
|
|
|
|
|
; O32-DAG: and [[VA_TMP2:\$[0-9]+]], [[VA_TMP0]], [[VA_TMP1]]
|
|
|
|
|
; O32-DAG: ori [[VA2:\$[0-9]+]], [[VA_TMP2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion and copy it to the global.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(dwords)
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA_TMP2]])
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[ARG1]], 8([[GV]])
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA_TMP2]])
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[ARG1]], 12([[GV]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(dwords)
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(dwords)(
|
|
|
|
|
; NEW-DAG: ld [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-DAG: sd [[ARG1]], 8([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; FIXME: We're still aligned from the last one but CodeGen doesn't spot that.
|
|
|
|
|
; O32: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP0:\$[0-9]+]], [[VA]], 7
|
|
|
|
|
; O32-DAG: and [[VA_TMP2:\$[0-9]+]], [[VA_TMP0]], [[VA_TMP1]]
|
|
|
|
|
; O32-DAG: ori [[VA2:\$[0-9]+]], [[VA_TMP2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion and copy it to the global.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; O32-DAG: sw [[ARG2]], 16([[GV]])
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 4
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA_TMP2]])
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[ARG2]], 20([[GV]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: ld [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-DAG: sd [[ARG2]], 16([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i64
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%e1 = getelementptr [3 x i64], [3 x i64]* @dwords, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i64 %arg1, i64* %e1, align 8
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i64
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%e2 = getelementptr [3 x i64], [3 x i64]* @dwords, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i64 %arg2, i64* %e2, align 8
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @fn_i64_dotdotdot_i16(i64 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i64_dotdotdot_i16:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 16 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the two 4 byte slots for the
|
|
|
|
|
; first fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 16
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]]
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(hwords)
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(hwords)
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(hwords)(
|
|
|
|
|
|
|
|
|
|
; ALL-DAG: sh [[ARG1]], 2([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA2]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; ALL-DAG: sh [[ARG2]], 4([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i16
|
[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
|
|
|
%e1 = getelementptr [3 x i16], [3 x i16]* @hwords, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i16 %arg1, i16* %e1, align 2
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i16
|
[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
|
|
|
%e2 = getelementptr [3 x i16], [3 x i16]* @hwords, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i16 %arg2, i16* %e2, align 2
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @fn_i64_dotdotdot_i32(i64 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i64_dotdotdot_i32:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 16 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the two 4 byte slots for the
|
|
|
|
|
; first fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 16
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]]
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(words)
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(words)
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(words)(
|
|
|
|
|
|
|
|
|
|
; ALL-DAG: sw [[ARG1]], 4([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; O32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
|
|
|
|
|
; NEW-LE-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-BE-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA2]])
|
|
|
|
|
|
|
|
|
|
; Copy the arg to the global
|
|
|
|
|
; ALL-DAG: sw [[ARG2]], 8([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i32
|
[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
|
|
|
%e1 = getelementptr [3 x i32], [3 x i32]* @words, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i32 %arg1, i32* %e1, align 4
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i32
|
[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
|
|
|
%e2 = getelementptr [3 x i32], [3 x i32]* @words, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i32 %arg2, i32* %e2, align 4
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define void @fn_i64_dotdotdot_i64(i64 %a, ...) {
|
|
|
|
|
entry:
|
|
|
|
|
; ALL-LABEL: fn_i64_dotdotdot_i64:
|
|
|
|
|
|
|
|
|
|
; Set up the stack with an 8-byte local area. N32/N64 must also make room for
|
|
|
|
|
; the argument save area (56 bytes).
|
|
|
|
|
; O32: addiu [[SP:\$sp]], $sp, -8
|
|
|
|
|
; N32: addiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
; N64: daddiu [[SP:\$sp]], $sp, -64
|
|
|
|
|
|
|
|
|
|
; Save variable argument portion on the stack
|
|
|
|
|
; O32-DAG: sw $7, 20([[SP]])
|
|
|
|
|
; O32-DAG: sw $6, 16([[SP]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: sd $11, 56([[SP]])
|
|
|
|
|
; NEW-DAG: sd $10, 48([[SP]])
|
|
|
|
|
; NEW-DAG: sd $9, 40([[SP]])
|
|
|
|
|
; NEW-DAG: sd $8, 32([[SP]])
|
|
|
|
|
; NEW-DAG: sd $7, 24([[SP]])
|
|
|
|
|
; NEW-DAG: sd $6, 16([[SP]])
|
|
|
|
|
; NEW-DAG: sd $5, 8([[SP]])
|
|
|
|
|
|
|
|
|
|
; Initialize variable argument pointer.
|
|
|
|
|
; For O32, the offset is 16 due to the 4 bytes used to store local variables,
|
|
|
|
|
; 4 bytes padding to maintain stack alignment, and the two 4 byte slots for the
|
|
|
|
|
; first fixed argument.
|
|
|
|
|
; For N32/N64, it is only 8 since the fixed arguments do not reserve stack
|
|
|
|
|
; space.
|
|
|
|
|
; O32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 16
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: daddiu [[VA:\$[0-9]+]], [[SP]], 8
|
|
|
|
|
; N64-DAG: sd [[VA]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Store [[VA]]
|
|
|
|
|
; O32-DAG: sw [[VA]], 0([[SP]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 1
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] (and realign pointer for O32)
|
|
|
|
|
; O32: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP0:\$[0-9]+]], [[VA]], 7
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP1:\$[0-9]+]], $zero, -8
|
|
|
|
|
; O32-DAG: and [[VA_TMP2:\$[0-9]+]], [[VA_TMP0]], [[VA_TMP1]]
|
|
|
|
|
; O32-DAG: ori [[VA2:\$[0-9]+]], [[VA_TMP2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: addiu [[VA2:\$[0-9]+]], [[VA]], 8
|
|
|
|
|
; N64-DAG: sd [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the first argument from the variable portion and copy it to the global.
|
|
|
|
|
; This has used the stack pointer directly rather than the [[VA]] we just set
|
|
|
|
|
; up.
|
|
|
|
|
; Big-endian mode for N32/N64 must add an additional 4 to the offset due to byte
|
|
|
|
|
; order.
|
|
|
|
|
; O32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(dwords)
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; O32-DAG: sw [[ARG1]], 8([[GV]])
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
; O32-DAG: lw [[ARG1:\$[0-9]+]], 4([[VA_TMP2]])
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[ARG1]], 12([[GV]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: addiu [[GV:\$[0-9]+]], ${{[0-9]+}}, %lo(dwords)
|
|
|
|
|
; N64-DAG: ld [[GV:\$[0-9]+]], %got_disp(dwords)(
|
|
|
|
|
; NEW-DAG: ld [[ARG1:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; NEW-DAG: sd [[ARG1]], 8([[GV]])
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
; ALL: teqi $zero, 2
|
2014-08-01 17:17:39 +08:00
|
|
|
|
|
|
|
|
; Increment [[VA]] again.
|
|
|
|
|
; FIXME: We're still aligned from the last one but CodeGen doesn't spot that.
|
|
|
|
|
; O32: lw [[VA:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; O32-DAG: addiu [[VA_TMP0:\$[0-9]+]], [[VA]], 7
|
|
|
|
|
; O32-DAG: and [[VA_TMP2:\$[0-9]+]], [[VA_TMP0]], [[VA_TMP1]]
|
|
|
|
|
; O32-DAG: ori [[VA2:\$[0-9]+]], [[VA_TMP2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA2]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N32-DAG: lw [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; N64-DAG: ld [[VA2:\$[0-9]+]], 0([[SP]])
|
|
|
|
|
; N64-DAG: daddiu [[VA3:\$[0-9]+]], [[VA2]], 8
|
|
|
|
|
; N64-DAG: sd [[VA3]], 0([[SP]])
|
|
|
|
|
|
|
|
|
|
; Load the second argument from the variable portion and copy it to the global.
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 0([[VA]])
|
|
|
|
|
; O32-DAG: sw [[ARG2]], 16([[GV]])
|
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled.
Retrying after fixing after removing load-store factoring through
token factors in favor of improved token factor operand pruning
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and
merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, dsanders, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 289659
2016-12-14 23:44:26 +08:00
|
|
|
; O32-DAG: addiu [[VA3:\$[0-9]+]], [[VA2]], 4
|
|
|
|
|
; O32-DAG: sw [[VA3]], 0([[SP]])
|
|
|
|
|
; O32-DAG: lw [[ARG2:\$[0-9]+]], 4([[VA_TMP2]])
|
2014-08-01 17:17:39 +08:00
|
|
|
; O32-DAG: sw [[ARG2]], 20([[GV]])
|
|
|
|
|
|
|
|
|
|
; NEW-DAG: ld [[ARG2:\$[0-9]+]], 0([[VA2]])
|
|
|
|
|
; NEW-DAG: sd [[ARG2]], 16([[GV]])
|
|
|
|
|
|
|
|
|
|
%ap = alloca i8*, align 8
|
|
|
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
|
|
|
call void @llvm.va_start(i8* %ap2)
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 1", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg1 = va_arg i8** %ap, i64
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%e1 = getelementptr [3 x i64], [3 x i64]* @dwords, i32 0, i32 1
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i64 %arg1, i64* %e1, align 8
|
|
|
|
|
|
2015-11-26 18:26:18 +08:00
|
|
|
call void asm sideeffect "teqi $$zero, 2", ""()
|
2014-08-01 17:17:39 +08:00
|
|
|
%arg2 = va_arg i8** %ap, i64
|
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%e2 = getelementptr [3 x i64], [3 x i64]* @dwords, i32 0, i32 2
|
2014-08-01 17:17:39 +08:00
|
|
|
store volatile i64 %arg2, i64* %e2, align 8
|
|
|
|
|
|
|
|
|
|
call void @llvm.va_end(i8* %ap2)
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
declare void @llvm.va_start(i8*)
|
|
|
|
|
declare void @llvm.va_end(i8*)
|