2018-10-08 23:49:19 +08:00
|
|
|
; RUN: llc -march=amdgcn -amdgpu-atomic-optimizations=false -verify-machineinstrs < %s | FileCheck -enable-var-scope -strict-whitespace -check-prefixes=GCN,SI,SICIVI %s
|
2021-01-06 12:23:39 +08:00
|
|
|
; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -amdgpu-atomic-optimizations=false -verify-machineinstrs < %s | FileCheck -enable-var-scope -strict-whitespace -check-prefixes=GCN,SICIVI,GFX89 %s
|
2018-10-08 23:49:19 +08:00
|
|
|
; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -amdgpu-atomic-optimizations=false -verify-machineinstrs < %s | FileCheck -enable-var-scope -strict-whitespace -check-prefixes=GCN,GFX9,GFX89 %s
|
2014-06-12 02:08:54 +08:00
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xchg_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_wrxchg_rtn_b64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_xchg_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw xchg i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xchg_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_wrxchg_rtn_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_xchg_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw xchg i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2019-01-17 18:49:01 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xchg_ret_f64_offset:
|
|
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
|
|
|
|
; GCN: ds_wrxchg_rtn_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
|
|
|
|
define amdgpu_kernel void @lds_atomic_xchg_ret_f64_offset(double addrspace(1)* %out, double addrspace(3)* %ptr) nounwind {
|
|
|
|
|
%gep = getelementptr double, double addrspace(3)* %ptr, i32 4
|
|
|
|
|
%result = atomicrmw xchg double addrspace(3)* %gep, double 4.0 seq_cst
|
|
|
|
|
store double %result, double addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_add_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_add_rtn_u64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_add_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw add i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_add_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI-DAG: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
|
|
|
|
; SI-DAG: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb
|
|
|
|
|
; GFX89-DAG: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 9
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
|
2015-02-27 01:08:43 +08:00
|
|
|
; GCN: ds_add_rtn_u64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}} offset:32
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: buffer_store_dwordx2 [[RESULT]],
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_add_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i64 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw add i64 addrspace(3)* %gep, i64 9 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_add1_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI-DAG: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 1{{$}}
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0{{$}}
|
|
|
|
|
; GCN: ds_add_rtn_u64 [[RESULT:v\[[0-9]+:[0-9]+\]]], {{v[0-9]+}}, v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: buffer_store_dwordx2 [[RESULT]],
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_add1_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw add i64 addrspace(3)* %ptr, i64 1 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_add1_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN: ds_add_rtn_u64 {{.*}} offset:32
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_add1_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw add i64 addrspace(3)* %gep, i64 1 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_sub_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_sub_rtn_u64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_sub_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw sub i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_sub_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_sub_rtn_u64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_sub_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw sub i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_sub1_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI-DAG: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 1{{$}}
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0{{$}}
|
|
|
|
|
; GCN: ds_sub_rtn_u64 [[RESULT:v\[[0-9]+:[0-9]+\]]], {{v[0-9]+}}, v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: buffer_store_dwordx2 [[RESULT]],
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_sub1_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw sub i64 addrspace(3)* %ptr, i64 1 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_sub1_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN: ds_sub_rtn_u64 {{.*}} offset:32
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_sub1_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw sub i64 addrspace(3)* %gep, i64 1 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_and_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_and_rtn_b64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_and_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw and i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_and_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_and_rtn_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_and_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw and i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_or_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_or_rtn_b64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_or_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw or i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_or_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_or_rtn_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_or_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw or i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xor_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_xor_rtn_b64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_xor_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw xor i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xor_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_xor_rtn_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_xor_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw xor i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; FIXME: There is no atomic nand instr
|
2016-03-24 07:23:38 +08:00
|
|
|
; XGCN-LABEL: {{^}}lds_atomic_nand_ret_i64:uction, so we somehow need to expand this.
|
2017-03-22 05:39:51 +08:00
|
|
|
; define amdgpu_kernel void @lds_atomic_nand_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
; %result = atomicrmw nand i64 addrspace(3)* %ptr, i32 4 seq_cst
|
|
|
|
|
; store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
; ret void
|
|
|
|
|
; }
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_min_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_min_rtn_i64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_min_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw min i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_min_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_min_rtn_i64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_min_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw min i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_max_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_max_rtn_i64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_max_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw max i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_max_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_max_rtn_i64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_max_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw max i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_umin_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_min_rtn_u64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_umin_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw umin i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_umin_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_min_rtn_u64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_umin_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw umin i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_umax_ret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_max_rtn_u64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_umax_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw umax i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_umax_ret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_max_rtn_u64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_umax_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-06-12 02:08:54 +08:00
|
|
|
%result = atomicrmw umax i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
store i64 %result, i64 addrspace(1)* %out, align 8
|
|
|
|
|
ret void
|
|
|
|
|
}
|
2014-09-08 23:07:31 +08:00
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xchg_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_wrxchg_rtn_b64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_xchg_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw xchg i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xchg_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_wrxchg_rtn_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_xchg_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw xchg i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_add_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_add_u64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_add_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw add i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_add_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI-DAG: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
|
|
|
|
; SI-DAG: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x9
|
|
|
|
|
; GFX89-DAG: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x24
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 9
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0
|
Bias physical register immediate assignments
The machine scheduler currently biases register copies to/from
physical registers to be closer to their point of use / def to
minimize their live ranges. This change extends this to also physical
register assignments from immediate values.
This causes a reduction in reduction in overall register pressure and
minor reduction in spills and indirectly fixes an out-of-registers
assertion (PR39391).
Most test changes are from minor instruction reorderings and register
name selection changes and direct consequences of that.
Reviewers: MatzeB, qcolombet, myatsina, pcc
Subscribers: nemanjai, jvesely, nhaehnle, eraman, hiraditya,
javed.absar, arphaman, jfb, jsji, llvm-commits
Differential Revision: https://reviews.llvm.org/D54218
llvm-svn: 346894
2018-11-15 05:11:53 +08:00
|
|
|
; GCN-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
|
2017-11-29 08:55:57 +08:00
|
|
|
; GCN: ds_add_u64 {{v[0-9]+}}, v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}} offset:32
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_add_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i64 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw add i64 addrspace(3)* %gep, i64 9 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_add1_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI-DAG: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 1{{$}}
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0{{$}}
|
2017-11-29 08:55:57 +08:00
|
|
|
; GCN: ds_add_u64 {{v[0-9]+}}, v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_add1_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw add i64 addrspace(3)* %ptr, i64 1 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_add1_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN: ds_add_u64 {{.*}} offset:32
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_add1_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw add i64 addrspace(3)* %gep, i64 1 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_sub_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_sub_u64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_sub_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw sub i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_sub_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_sub_u64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_sub_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw sub i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_sub1_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI-DAG: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 1{{$}}
|
|
|
|
|
; GCN-DAG: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0{{$}}
|
|
|
|
|
; GCN: ds_sub_u64 {{v[0-9]+}}, v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_sub1_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw sub i64 addrspace(3)* %ptr, i64 1 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_sub1_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN: ds_sub_u64 {{.*}} offset:32
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_sub1_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw sub i64 addrspace(3)* %gep, i64 1 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_and_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_and_b64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_and_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw and i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_and_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_and_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_and_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw and i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_or_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_or_b64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_or_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw or i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_or_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_or_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_or_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw or i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xor_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_xor_b64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_xor_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw xor i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_xor_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_xor_b64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_xor_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw xor i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; FIXME: There is no atomic nand instr
|
2016-03-24 07:23:38 +08:00
|
|
|
; XGCN-LABEL: {{^}}lds_atomic_nand_noret_i64:uction, so we somehow need to expand this.
|
2017-03-22 05:39:51 +08:00
|
|
|
; define amdgpu_kernel void @lds_atomic_nand_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
; %result = atomicrmw nand i64 addrspace(3)* %ptr, i32 4 seq_cst
|
|
|
|
|
; ret void
|
|
|
|
|
; }
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_min_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_min_i64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_min_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw min i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_min_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_min_i64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_min_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw min i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_max_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_max_i64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_max_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw max i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_max_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_max_i64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_max_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw max i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_umin_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_min_u64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_umin_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw umin i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_umin_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_min_u64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_umin_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw umin i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_umax_noret_i64:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_max_u64
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_umax_noret_i64(i64 addrspace(3)* %ptr) nounwind {
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw umax i64 addrspace(3)* %ptr, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-24 07:23:38 +08:00
|
|
|
; GCN-LABEL: {{^}}lds_atomic_umax_noret_i64_offset:
|
2017-11-29 08:55:57 +08:00
|
|
|
; SICIVI: s_mov_b32 m0
|
|
|
|
|
; GFX9-NOT: m0
|
|
|
|
|
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: ds_max_u64 {{.*}} offset:32
|
|
|
|
|
; GCN: s_endpgm
|
2017-03-22 05:39:51 +08:00
|
|
|
define amdgpu_kernel void @lds_atomic_umax_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
|
[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
|
|
|
%gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
|
2014-09-08 23:07:31 +08:00
|
|
|
%result = atomicrmw umax i64 addrspace(3)* %gep, i64 4 seq_cst
|
|
|
|
|
ret void
|
|
|
|
|
}
|