maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/global_atomics.ll

1227 lines
55 KiB
LLVM
Raw Normal View History

[AMDGPU] Add an AMDGPU specific atomic optimizer. This commit adds a new IR level pass to the AMDGPU backend to perform atomic optimizations. It works by: - Running through a function and finding atomicrmw add/sub or uses of the atomic buffer intrinsics for add/sub. - If all arguments except the value to be added/subtracted are uniform, record the value to be optimized. - Run through the atomic operations we can optimize and, depending on whether the value is uniform/divergent use wavefront wide operations (DPP in the divergent case) to calculate the total amount to be atomically added/subtracted. - Then let only a single lane of each wavefront perform the atomic operation, reducing the total number of atomic operations in flight. - Lastly we recombine the result from the single lane to each lane of the wavefront, and calculate our individual lanes offset into the final result. Differential Revision: https://reviews.llvm.org/D51969 llvm-svn: 343973
2018-10-08 23:49:19 +08:00
; RUN: llc -march=amdgcn -amdgpu-atomic-optimizations=false -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,SI,SIVI %s
; RUN: llc -march=amdgcn -mcpu=tonga -amdgpu-atomic-optimizations=false -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI,SIVI %s
; RUN: llc -march=amdgcn -mcpu=gfx900 -amdgpu-atomic-optimizations=false -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9 %s
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_offset:
; SIVI: buffer_atomic_add v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_add v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_max_neg_offset:
; GFX9: global_atomic_add v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off offset:-4096{{$}}
define amdgpu_kernel void @atomic_add_i32_max_neg_offset(i32 addrspace(1)* %out, i32 %in) {
entry:
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 -1024
%val = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
ret void
}
; GCN-LABEL: {{^}}atomic_add_i32_soffset:
; SIVI: s_mov_b32 [[SREG:s[0-9]+]], 0x8ca0
; SIVI: buffer_atomic_add v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], [[SREG]]{{$}}
; GFX9: global_atomic_add v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_soffset(i32 addrspace(1)* %out, i32 %in) {
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
entry:
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 9000
%val = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_huge_offset:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
; SI-DAG: v_mov_b32_e32 v[[PTRLO:[0-9]+]], 0xdeac
; SI-DAG: v_mov_b32_e32 v[[PTRHI:[0-9]+]], 0xabcd
; SI: buffer_atomic_add v{{[0-9]+}}, v{{\[}}[[PTRLO]]:[[PTRHI]]{{\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
; VI: flat_atomic_add
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_add v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_huge_offset(i32 addrspace(1)* %out, i32 %in) {
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
entry:
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 47224239175595
%val = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_ret_offset:
; SIVI: buffer_atomic_add [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_add v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Fix/add tests for atomicrmw add The previous tests claimed to test constant offsets in the function name, but the tests weren't actually testing them. Clone the tests, and do testing of all combinations of the following: 1) with/without constant pointer offset 2) 32/64-bit addressing modes 3) Usage and non-usage of the return value from the atomicrmw Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220103
2014-10-18 07:32:50 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Fix/add tests for atomicrmw add The previous tests claimed to test constant offsets in the function name, but the tests weren't actually testing them. Clone the tests, and do testing of all combinations of the following: 1) with/without constant pointer offset 2) 32/64-bit addressing modes 3) Usage and non-usage of the return value from the atomicrmw Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220103
2014-10-18 07:32:50 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_add v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Fix/add tests for atomicrmw add The previous tests claimed to test constant offsets in the function name, but the tests weren't actually testing them. Clone the tests, and do testing of all combinations of the following: 1) with/without constant pointer offset 2) 32/64-bit addressing modes 3) Usage and non-usage of the return value from the atomicrmw Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220103
2014-10-18 07:32:50 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Fix/add tests for atomicrmw add The previous tests claimed to test constant offsets in the function name, but the tests weren't actually testing them. Clone the tests, and do testing of all combinations of the following: 1) with/without constant pointer offset 2) 32/64-bit addressing modes 3) Usage and non-usage of the return value from the atomicrmw Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220103
2014-10-18 07:32:50 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
; GFX9: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Fix/add tests for atomicrmw add The previous tests claimed to test constant offsets in the function name, but the tests weren't actually testing them. Clone the tests, and do testing of all combinations of the following: 1) with/without constant pointer offset 2) 32/64-bit addressing modes 3) Usage and non-usage of the return value from the atomicrmw Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220103
2014-10-18 07:32:50 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Fix/add tests for atomicrmw add The previous tests claimed to test constant offsets in the function name, but the tests weren't actually testing them. Clone the tests, and do testing of all combinations of the following: 1) with/without constant pointer offset 2) 32/64-bit addressing modes 3) Usage and non-usage of the return value from the atomicrmw Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220103
2014-10-18 07:32:50 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32:
; SIVI: buffer_atomic_add v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_add v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Fix/add tests for atomicrmw add The previous tests claimed to test constant offsets in the function name, but the tests weren't actually testing them. Clone the tests, and do testing of all combinations of the following: 1) with/without constant pointer offset 2) 32/64-bit addressing modes 3) Usage and non-usage of the return value from the atomicrmw Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220103
2014-10-18 07:32:50 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile add i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Fix/add tests for atomicrmw add The previous tests claimed to test constant offsets in the function name, but the tests weren't actually testing them. Clone the tests, and do testing of all combinations of the following: 1) with/without constant pointer offset 2) 32/64-bit addressing modes 3) Usage and non-usage of the return value from the atomicrmw Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220103
2014-10-18 07:32:50 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_ret:
; SIVI: buffer_atomic_add [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
; GFX9: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile add i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_add v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile add i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_add_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_add_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile add i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add support for global atomic add llvm-svn: 218457
2014-09-26 02:30:26 +08:00
ret void
}
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_and_i32_offset:
; SIVI: buffer_atomic_and v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_and_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile and i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_and_i32_ret_offset:
; SIVI: buffer_atomic_and [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_and_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile and i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_and_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_and v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_and_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile and i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_and_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_and [[RET:v[0-9]]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_and [[RET:v[0-9]]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_and_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile and i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_and_i32:
; SIVI: buffer_atomic_and v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_and v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_and_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile and i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_and_i32_ret:
; SIVI: buffer_atomic_and [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_and v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_and_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile and i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_and_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_and v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_and_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile and i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_and_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_and_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile and i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw and v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220105
2014-10-18 07:32:54 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_sub_i32_offset:
; SIVI: buffer_atomic_sub v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_sub_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_sub_i32_ret_offset:
; SIVI: buffer_atomic_sub [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_sub v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_sub_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_sub_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_sub v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_sub_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_sub_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_sub_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_sub_i32:
; SIVI: buffer_atomic_sub v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_sub v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_sub_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile sub i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_sub_i32_ret:
; SIVI: buffer_atomic_sub [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_sub_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile sub i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_sub_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_sub v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_sub v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_sub_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile sub i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_sub_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_sub_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile sub i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw sub v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220104
2014-10-18 07:32:52 +08:00
ret void
}
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_max_i32_offset:
; SIVI: buffer_atomic_smax v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_max_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile max i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_max_i32_ret_offset:
; SIVI: buffer_atomic_smax [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_max_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile max i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_max_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_smax v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_max_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile max i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_max_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_max_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile max i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_max_i32:
; SIVI: buffer_atomic_smax v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_max_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile max i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_max_i32_ret:
; SIVI: buffer_atomic_smax [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_max_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile max i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_max_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_smax v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_max_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile max i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_max_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_max_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile max i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umax_i32_offset:
; SIVI: buffer_atomic_umax v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umax_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile umax i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umax_i32_ret_offset:
; SIVI: buffer_atomic_umax [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umax_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile umax i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umax_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_umax v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umax_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile umax i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umax_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umax_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile umax i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umax_i32:
; SIVI: buffer_atomic_umax v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umax_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile umax i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umax_i32_ret:
; SIVI: buffer_atomic_umax [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umax_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile umax i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umax_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_umax v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umax_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile umax i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umax_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umax_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile umax i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw max/umax v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220106
2014-10-18 07:32:56 +08:00
ret void
}
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_min_i32_offset:
; SIVI: buffer_atomic_smin v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_min_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile min i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_min_i32_ret_offset:
; SIVI: buffer_atomic_smin [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_min_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile min i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_min_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_smin v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_min_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile min i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_min_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_min_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile min i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_min_i32:
; SIVI: buffer_atomic_smin v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_min_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile min i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_min_i32_ret:
; SIVI: buffer_atomic_smin [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_min_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile min i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_min_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_smin v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_min_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile min i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_min_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_min_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile min i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umin_i32_offset:
; SIVI: buffer_atomic_umin v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umin_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile umin i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umin_i32_ret_offset:
; SIVI: buffer_atomic_umin [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umin_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile umin i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umin_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_umin v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umin_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile umin i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umin_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umin_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile umin i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umin_i32:
; SIVI: buffer_atomic_umin v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umin_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile umin i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umin_i32_ret:
; SIVI: buffer_atomic_umin [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umin_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile umin i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umin_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_umin v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umin_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile umin i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_umin_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_umin_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile umin i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw min/umin v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220107
2014-10-18 07:32:57 +08:00
ret void
}
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_or_i32_offset:
; SIVI: buffer_atomic_or v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_or_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_or_i32_ret_offset:
; SIVI: buffer_atomic_or [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_or_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_or_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_or v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_or_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_or_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_or_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_or_i32:
; SIVI: buffer_atomic_or v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_or_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile or i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_or_i32_ret:
; SIVI: buffer_atomic_or [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_or_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile or i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_or_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_or v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_or_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile or i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_or_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_or_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile or i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw or v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220108
2014-10-18 07:32:59 +08:00
ret void
}
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xchg_i32_offset:
; SIVI: buffer_atomic_swap v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xchg_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile xchg i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xchg_i32_ret_offset:
; SIVI: buffer_atomic_swap [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xchg_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile xchg i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xchg_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_swap v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; VI: flat_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
; GFX9: global_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xchg_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile xchg i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xchg_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xchg_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile xchg i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xchg_i32:
; SIVI: buffer_atomic_swap v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xchg_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile xchg i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xchg_i32_ret:
; SIVI: buffer_atomic_swap [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xchg_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile xchg i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xchg_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_swap v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xchg_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile xchg i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xchg_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xchg_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile xchg i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw xchg v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220110
2014-10-18 07:33:03 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_cmpxchg_i32_offset:
; SIVI: buffer_atomic_cmpswap v[{{[0-9]+}}:{{[0-9]+}}], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_cmpswap v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_cmpxchg_i32_offset(i32 addrspace(1)* %out, i32 %in, i32 %old) {
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = cmpxchg volatile i32 addrspace(1)* %gep, i32 %old, i32 %in seq_cst seq_cst
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_cmpxchg_i32_ret_offset:
; SIVI: buffer_atomic_cmpswap v{{\[}}[[RET:[0-9]+]]{{:[0-9]+}}], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword v[[RET]]
; GFX9: global_atomic_cmpswap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_cmpxchg_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i32 %old) {
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = cmpxchg volatile i32 addrspace(1)* %gep, i32 %old, i32 %in seq_cst seq_cst
%extract0 = extractvalue { i32, i1 } %val, 0
store i32 %extract0, i32 addrspace(1)* %out2
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_cmpxchg_i32_addr64_offset:
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
; SI: buffer_atomic_cmpswap v[{{[0-9]+\:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
; VI: flat_atomic_cmpswap v[{{[0-9]+\:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_cmpswap v[{{[0-9]+\:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}], off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_cmpxchg_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index, i32 %old) {
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
entry:
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = cmpxchg volatile i32 addrspace(1)* %gep, i32 %old, i32 %in seq_cst seq_cst
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_cmpxchg_i32_ret_addr64_offset:
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
; SI: buffer_atomic_cmpswap v{{\[}}[[RET:[0-9]+]]:{{[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
; VI: flat_atomic_cmpswap v[[RET:[0-9]+]], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}] glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword v[[RET]]
; GFX9: global_atomic_cmpswap v[[RET:[0-9]+]], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_cmpxchg_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index, i32 %old) {
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
entry:
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = cmpxchg volatile i32 addrspace(1)* %gep, i32 %old, i32 %in seq_cst seq_cst
%extract0 = extractvalue { i32, i1 } %val, 0
store i32 %extract0, i32 addrspace(1)* %out2
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_cmpxchg_i32:
; SIVI: buffer_atomic_cmpswap v[{{[0-9]+:[0-9]+}}], off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_cmpswap v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_cmpxchg_i32(i32 addrspace(1)* %out, i32 %in, i32 %old) {
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = cmpxchg volatile i32 addrspace(1)* %out, i32 %old, i32 %in seq_cst seq_cst
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_cmpxchg_i32_ret:
; SIVI: buffer_atomic_cmpswap v{{\[}}[[RET:[0-9]+]]:{{[0-9]+}}], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword v[[RET]]
; GFX9: global_atomic_cmpswap [[RET:v[0-9]+]], v[{{[0-9]+\:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}], off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_cmpxchg_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i32 %old) {
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = cmpxchg volatile i32 addrspace(1)* %out, i32 %old, i32 %in seq_cst seq_cst
%extract0 = extractvalue { i32, i1 } %val, 0
store i32 %extract0, i32 addrspace(1)* %out2
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_cmpxchg_i32_addr64:
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
; SI: buffer_atomic_cmpswap v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
; VI: flat_atomic_cmpswap v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_cmpswap v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_cmpxchg_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index, i32 %old) {
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
entry:
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = cmpxchg volatile i32 addrspace(1)* %ptr, i32 %old, i32 %in seq_cst seq_cst
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_cmpxchg_i32_ret_addr64:
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
; SI: buffer_atomic_cmpswap v{{\[}}[[RET:[0-9]+]]:{{[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
; VI: flat_atomic_cmpswap v[[RET:[0-9]+]], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}] glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword v[[RET]]
; GFX9: global_atomic_cmpswap v[[RET:[0-9]+]], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_cmpxchg_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index, i32 %old) {
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
entry:
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = cmpxchg volatile i32 addrspace(1)* %ptr, i32 %old, i32 %in seq_cst seq_cst
%extract0 = extractvalue { i32, i1 } %val, 0
store i32 %extract0, i32 addrspace(1)* %out2
AMDGPU: Implement {BUFFER,FLAT}_ATOMIC_CMPSWAP{,_X2} Summary: Implement BUFFER_ATOMIC_CMPSWAP{,_X2} instructions on all GCN targets, and FLAT_ATOMIC_CMPSWAP{,_X2} on CI+. 32-bit instruction variants tested manually on Kabini and Bonaire. Tests and parts of code provided by Jan Veselý. Patch by: Vedran Miletić Reviewers: arsenm, tstellarAMD, nhaehnle Subscribers: jvesely, scchan, kanarayan, arsenm Differential Revision: http://reviews.llvm.org/D17280 llvm-svn: 265170
2016-04-02 02:27:37 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xor_i32_offset:
; SIVI: buffer_atomic_xor v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
; GFX9: global_atomic_xor v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xor_i32_offset(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile xor i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xor_i32_ret_offset:
; SIVI: buffer_atomic_xor [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_xor v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xor_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
%val = atomicrmw volatile xor i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xor_i32_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_xor v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xor_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile xor i32 addrspace(1)* %gep, i32 %in seq_cst
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xor_i32_ret_addr64_offset:
R600/SI: Change mubuf offsets to print as decimal This matches SC's behavior. llvm-svn: 223194
2014-12-03 11:12:13 +08:00
; SI: buffer_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xor_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = atomicrmw volatile xor i32 addrspace(1)* %gep, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xor_i32:
; SIVI: buffer_atomic_xor v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
; GFX9: global_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xor_i32(i32 addrspace(1)* %out, i32 %in) {
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile xor i32 addrspace(1)* %out, i32 %in seq_cst
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xor_i32_ret:
; SIVI: buffer_atomic_xor [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xor_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile xor i32 addrspace(1)* %out, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xor_i32_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_xor v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GFX9: global_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xor_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile xor i32 addrspace(1)* %ptr, i32 %in seq_cst
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_xor_i32_ret_addr64:
R600/SI: Change all instruction assembly names to lowercase. This matches the format produced by the AMD proprietary driver. //==================================================================// // Shell script for converting .ll test cases: (Pass the .ll files you want to convert to this script as arguments). //==================================================================// ; This was necessary on my system so that A-Z in sed would match only ; upper case. I'm not sure why. export LC_ALL='C' TEST_FILES="$*" MATCHES=`grep -v Patterns SIInstructions.td | grep -o '"[A-Z0-9_]\+["e]' | grep -o '[A-Z0-9_]\+' | sort -r` for f in $TEST_FILES; do # Check that there are SI tests: grep -q -e 'verde' -e 'bonaire' -e 'SI' -e 'tahiti' $f if [ $? -eq 0 ]; then for match in $MATCHES; do sed -i -e "s/\([ :]$match\)/\L\1/" $f done # Try to get check lines with partial instruction names sed -i 's/\(;[ ]*SI[A-Z\\-]*: \)\([A-Z_0-9]\+\)/\1\L\2/' $f fi done sed -i -e 's/bb0_1/BB0_1/g' ../../../test/CodeGen/R600/infinite-loop.ll sed -i -e 's/SI-NOT: bfe/SI-NOT: {{[^@]}}bfe/g'../../../test/CodeGen/R600/llvm.AMDGPU.bfe.*32.ll ../../../test/CodeGen/R600/sext-in-reg.ll sed -i -e 's/exp_IEEE/EXP_IEEE/g' ../../../test/CodeGen/R600/llvm.exp2.ll sed -i -e 's/numVgprs/NumVgprs/g' ../../../test/CodeGen/R600/register-count-comments.ll sed -i 's/\(; CHECK[-NOT]*: \)\([A-Z_0-9]\+\)/\1\L\2/' ../../../test/CodeGen/R600/select64.ll ../../../test/CodeGen/R600/sgpr-copy.ll //==================================================================// // Shell script for converting .td files (run this last) //==================================================================// export LC_ALL='C' sed -i -e '/Patterns/!s/\("[A-Z0-9_]\+[ "e]\)/\L\1/g' SIInstructions.td sed -i -e 's/"EXP/"exp/g' SIInstrInfo.td llvm-svn: 221350
2014-11-05 22:50:53 +08:00
; SI: buffer_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops Summary: The MUBUF addr64 bit has been removed on VI, so we must use FLAT instructions when the pointer is stored in VGPRs. Reviewers: arsenm Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11067 llvm-svn: 242673
2015-07-20 22:28:41 +08:00
; VI: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_xor_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
entry:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = atomicrmw volatile xor i32 addrspace(1)* %ptr, i32 %in seq_cst
store i32 %val, i32 addrspace(1)* %out2
R600/SI: Add global atomicrmw xor v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Matt Arsenault <matthew.arsenault@amd.com> llvm-svn: 220109
2014-10-18 07:33:01 +08:00
ret void
}
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_load_i32_offset:
AMDGPU/SI: Assembler: Unify parsing/printing of operands. Summary: The goal is for each operand type to have its own parse function and at the same time share common code for tracking state as different instruction types share operand types (e.g. glc/glc_flat, etc). Introduce parseAMDGPUOperand which can parse any optional operand. DPP and Clamp/OMod have custom handling for now. Sam also suggested to have class hierarchy for operand types instead of table. This can be done in separate change. Remove parseVOP3OptionalOps, parseDS*OptionalOps, parseFlatOptionalOps, parseMubufOptionalOps, parseDPPOptionalOps. Reduce number of definitions of AsmOperand's and MatchClasses' by using common base class. Rename AsmMatcher/InstPrinter methods accordingly. Print immediate type when printing parsed immediate operand. Use 'off' if offset/index register is unused instead of skipping it to make it more readable (also agreed with SP3). Update tests. Reviewers: tstellarAMD, SamWot, artem.tamazov Subscribers: qcolombet, arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D19584 llvm-svn: 268015
2016-04-29 17:02:30 +08:00
; SI: buffer_load_dword [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc{{$}}
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
; VI: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_load_i32_offset(i32 addrspace(1)* %in, i32 addrspace(1)* %out) {
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %in, i64 4
%val = load atomic i32, i32 addrspace(1)* %gep seq_cst, align 4
store i32 %val, i32 addrspace(1)* %out
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_load_i32:
AMDGPU/SI: Assembler: Unify parsing/printing of operands. Summary: The goal is for each operand type to have its own parse function and at the same time share common code for tracking state as different instruction types share operand types (e.g. glc/glc_flat, etc). Introduce parseAMDGPUOperand which can parse any optional operand. DPP and Clamp/OMod have custom handling for now. Sam also suggested to have class hierarchy for operand types instead of table. This can be done in separate change. Remove parseVOP3OptionalOps, parseDS*OptionalOps, parseFlatOptionalOps, parseMubufOptionalOps, parseDPPOptionalOps. Reduce number of definitions of AsmOperand's and MatchClasses' by using common base class. Rename AsmMatcher/InstPrinter methods accordingly. Print immediate type when printing parsed immediate operand. Use 'off' if offset/index register is unused instead of skipping it to make it more readable (also agreed with SP3). Update tests. Reviewers: tstellarAMD, SamWot, artem.tamazov Subscribers: qcolombet, arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D19584 llvm-svn: 268015
2016-04-29 17:02:30 +08:00
; SI: buffer_load_dword [[RET:v[0-9]+]], off, s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
; VI: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], off glc
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_load_i32(i32 addrspace(1)* %in, i32 addrspace(1)* %out) {
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = load atomic i32, i32 addrspace(1)* %in seq_cst, align 4
store i32 %val, i32 addrspace(1)* %out
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_load_i32_addr64_offset:
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
; SI: buffer_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
; VI: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}] glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_load_dword [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], off offset:16 glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_load_i32_addr64_offset(i32 addrspace(1)* %in, i32 addrspace(1)* %out, i64 %index) {
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
entry:
%ptr = getelementptr i32, i32 addrspace(1)* %in, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
%val = load atomic i32, i32 addrspace(1)* %gep seq_cst, align 4
store i32 %val, i32 addrspace(1)* %out
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_load_i32_addr64:
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
; SI: buffer_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
; VI: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}] glc{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; SIVI: buffer_store_dword [[RET]]
; GFX9: global_load_dword [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], off glc{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_load_i32_addr64(i32 addrspace(1)* %in, i32 addrspace(1)* %out, i64 %index) {
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
entry:
%ptr = getelementptr i32, i32 addrspace(1)* %in, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%val = load atomic i32, i32 addrspace(1)* %ptr seq_cst, align 4
store i32 %val, i32 addrspace(1)* %out
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_store_i32_offset:
AMDGPU: Implement memory model llvm-svn: 308781
2017-07-22 05:19:23 +08:00
; SI: buffer_store_dword {{v[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; VI: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+$}}
; GFX9: global_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_store_i32_offset(i32 %in, i32 addrspace(1)* %out) {
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
entry:
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %out, i64 4
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
store atomic i32 %in, i32 addrspace(1)* %gep seq_cst, align 4
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_store_i32:
AMDGPU: Implement memory model llvm-svn: 308781
2017-07-22 05:19:23 +08:00
; SI: buffer_store_dword {{v[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; VI: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+$}}
; GFX9: global_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_store_i32(i32 %in, i32 addrspace(1)* %out) {
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
entry:
store atomic i32 %in, i32 addrspace(1)* %out seq_cst, align 4
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_store_i32_addr64_offset:
AMDGPU: Implement memory model llvm-svn: 308781
2017-07-22 05:19:23 +08:00
; SI: buffer_store_dword {{v[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; VI: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+$}}
; GFX9: global_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}}, off offset:16{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_store_i32_addr64_offset(i32 %in, i32 addrspace(1)* %out, i64 %index) {
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
entry:
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
AMDGPU: Fix missing and broken check lines in atomic tests llvm-svn: 272343
2016-06-10 07:42:44 +08:00
%gep = getelementptr i32, i32 addrspace(1)* %ptr, i64 4
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
store atomic i32 %in, i32 addrspace(1)* %gep seq_cst, align 4
ret void
}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; GCN-LABEL: {{^}}atomic_store_i32_addr64:
AMDGPU: Implement memory model llvm-svn: 308781
2017-07-22 05:19:23 +08:00
; SI: buffer_store_dword {{v[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
AMDGPU: Start selecting global instructions llvm-svn: 309470
2017-07-29 09:03:53 +08:00
; VI: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+$}}
; GFX9: global_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}}, off{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @atomic_store_i32_addr64(i32 %in, i32 addrspace(1)* %out, i64 %index) {
AMDGPU/SI: Implement atomic load/store for i32 and i64 Standard load/store instructions with GLC bit set. Reviewers: tstellardAMD, arsenm Differential Revision: http://reviews.llvm.org/D18760 llvm-svn: 265709
2016-04-08 03:23:11 +08:00
entry:
%ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
store atomic i32 %in, i32 addrspace(1)* %ptr seq_cst, align 4
ret void
}