llvm-project/llvm/test/CodeGen/AMDGPU/min3.ll

337 lines
13 KiB
LLVM
Raw Normal View History

; RUN: llc -march=amdgcn < %s | FileCheck -check-prefixes=GCN,SI %s
; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,VI %s
; RUN: llc -march=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s
; GCN-LABEL: {{^}}v_test_imin3_slt_i32:
; GCN: v_min3_i32
define amdgpu_kernel void @v_test_imin3_slt_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr, i32 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
[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
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i32, i32 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%a = load i32, i32 addrspace(1)* %gep0
%b = load i32, i32 addrspace(1)* %gep1
%c = load i32, i32 addrspace(1)* %gep2
%icmp0 = icmp slt i32 %a, %b
%i0 = select i1 %icmp0, i32 %a, i32 %b
%icmp1 = icmp slt i32 %i0, %c
%i1 = select i1 %icmp1, i32 %i0, i32 %c
store i32 %i1, i32 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_umin3_ult_i32:
; GCN: v_min3_u32
define amdgpu_kernel void @v_test_umin3_ult_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr, i32 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
[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
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i32, i32 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%a = load i32, i32 addrspace(1)* %gep0
%b = load i32, i32 addrspace(1)* %gep1
%c = load i32, i32 addrspace(1)* %gep2
%icmp0 = icmp ult i32 %a, %b
%i0 = select i1 %icmp0, i32 %a, i32 %b
%icmp1 = icmp ult i32 %i0, %c
%i1 = select i1 %icmp1, i32 %i0, i32 %c
store i32 %i1, i32 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_umin_umin_umin:
; GCN: v_min_i32
; GCN: v_min3_i32
define amdgpu_kernel void @v_test_umin_umin_umin(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr, i32 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid2 = mul i32 %tid, 2
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i32, i32 addrspace(1)* %cptr, i32 %tid
[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
%gep3 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid2
%gep4 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid2
%gep5 = getelementptr i32, i32 addrspace(1)* %cptr, i32 %tid2
[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
%outgep0 = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%outgep1 = getelementptr i32, i32 addrspace(1)* %out, i32 %tid2
%a = load i32, i32 addrspace(1)* %gep0
%b = load i32, i32 addrspace(1)* %gep1
%c = load i32, i32 addrspace(1)* %gep2
%d = load i32, i32 addrspace(1)* %gep3
%icmp0 = icmp slt i32 %a, %b
%i0 = select i1 %icmp0, i32 %a, i32 %b
%icmp1 = icmp slt i32 %c, %d
%i1 = select i1 %icmp1, i32 %c, i32 %d
%icmp2 = icmp slt i32 %i0, %i1
%i2 = select i1 %icmp2, i32 %i0, i32 %i1
store i32 %i2, i32 addrspace(1)* %outgep1
ret void
}
; GCN-LABEL: {{^}}v_test_umin3_2_uses:
; GCN-NOT: v_min3
define amdgpu_kernel void @v_test_umin3_2_uses(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr, i32 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid2 = mul i32 %tid, 2
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i32, i32 addrspace(1)* %cptr, i32 %tid
[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
%gep3 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid2
%gep4 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid2
%gep5 = getelementptr i32, i32 addrspace(1)* %cptr, i32 %tid2
[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
%outgep0 = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%outgep1 = getelementptr i32, i32 addrspace(1)* %out, i32 %tid2
%a = load i32, i32 addrspace(1)* %gep0
%b = load i32, i32 addrspace(1)* %gep1
%c = load i32, i32 addrspace(1)* %gep2
%d = load i32, i32 addrspace(1)* %gep3
%icmp0 = icmp slt i32 %a, %b
%i0 = select i1 %icmp0, i32 %a, i32 %b
%icmp1 = icmp slt i32 %c, %d
%i1 = select i1 %icmp1, i32 %c, i32 %d
%icmp2 = icmp slt i32 %i0, %c
%i2 = select i1 %icmp2, i32 %i0, i32 %c
store i32 %i2, i32 addrspace(1)* %outgep0
store i32 %i0, i32 addrspace(1)* %outgep1
ret void
}
; GCN-LABEL: {{^}}v_test_imin3_slt_i16:
; SI: v_min3_i32
; VI: v_min_i16
; VI: v_min_i16
; GFX9: v_min3_i16
define amdgpu_kernel void @v_test_imin3_slt_i16(i16 addrspace(1)* %out, i16 addrspace(1)* %aptr, i16 addrspace(1)* %bptr, i16 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i16, i16 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i16, i16 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i16, i16 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i16, i16 addrspace(1)* %out, i32 %tid
%a = load i16, i16 addrspace(1)* %gep0
%b = load i16, i16 addrspace(1)* %gep1
%c = load i16, i16 addrspace(1)* %gep2
%icmp0 = icmp slt i16 %a, %b
%i0 = select i1 %icmp0, i16 %a, i16 %b
%icmp1 = icmp slt i16 %i0, %c
%i1 = select i1 %icmp1, i16 %i0, i16 %c
store i16 %i1, i16 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_umin3_ult_i16:
; SI: v_min3_u32
; VI: v_min_u16
; VI: v_min_u16
; GFX9: v_min3_u16
define amdgpu_kernel void @v_test_umin3_ult_i16(i16 addrspace(1)* %out, i16 addrspace(1)* %aptr, i16 addrspace(1)* %bptr, i16 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i16, i16 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i16, i16 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i16, i16 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i16, i16 addrspace(1)* %out, i32 %tid
%a = load i16, i16 addrspace(1)* %gep0
%b = load i16, i16 addrspace(1)* %gep1
%c = load i16, i16 addrspace(1)* %gep2
%icmp0 = icmp ult i16 %a, %b
%i0 = select i1 %icmp0, i16 %a, i16 %b
%icmp1 = icmp ult i16 %i0, %c
%i1 = select i1 %icmp1, i16 %i0, i16 %c
store i16 %i1, i16 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_imin3_slt_i8:
; SI: v_min3_i32
; VI: v_min_i16
; VI: v_min_i16
; GFX9: v_min3_i16
define amdgpu_kernel void @v_test_imin3_slt_i8(i8 addrspace(1)* %out, i8 addrspace(1)* %aptr, i8 addrspace(1)* %bptr, i8 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i8, i8 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i8, i8 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i8, i8 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i8, i8 addrspace(1)* %out, i32 %tid
%a = load i8, i8 addrspace(1)* %gep0
%b = load i8, i8 addrspace(1)* %gep1
%c = load i8, i8 addrspace(1)* %gep2
%icmp0 = icmp slt i8 %a, %b
%i0 = select i1 %icmp0, i8 %a, i8 %b
%icmp1 = icmp slt i8 %i0, %c
%i1 = select i1 %icmp1, i8 %i0, i8 %c
store i8 %i1, i8 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_umin3_ult_i8:
; SI: v_min3_u32
; VI: v_min_u16
; VI: v_min_u16
; GFX9: v_min3_u16
define amdgpu_kernel void @v_test_umin3_ult_i8(i8 addrspace(1)* %out, i8 addrspace(1)* %aptr, i8 addrspace(1)* %bptr, i8 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i8, i8 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i8, i8 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i8, i8 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i8, i8 addrspace(1)* %out, i32 %tid
%a = load i8, i8 addrspace(1)* %gep0
%b = load i8, i8 addrspace(1)* %gep1
%c = load i8, i8 addrspace(1)* %gep2
%icmp0 = icmp ult i8 %a, %b
%i0 = select i1 %icmp0, i8 %a, i8 %b
%icmp1 = icmp ult i8 %i0, %c
%i1 = select i1 %icmp1, i8 %i0, i8 %c
store i8 %i1, i8 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_imin3_slt_i7:
; SI: v_min3_i32
; VI: v_min_i16
; VI: v_min_i16
; GFX9: v_min3_i16
define amdgpu_kernel void @v_test_imin3_slt_i7(i7 addrspace(1)* %out, i7 addrspace(1)* %aptr, i7 addrspace(1)* %bptr, i7 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i7, i7 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i7, i7 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i7, i7 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i7, i7 addrspace(1)* %out, i32 %tid
%a = load i7, i7 addrspace(1)* %gep0
%b = load i7, i7 addrspace(1)* %gep1
%c = load i7, i7 addrspace(1)* %gep2
%icmp0 = icmp slt i7 %a, %b
%i0 = select i1 %icmp0, i7 %a, i7 %b
%icmp1 = icmp slt i7 %i0, %c
%i1 = select i1 %icmp1, i7 %i0, i7 %c
store i7 %i1, i7 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_umin3_ult_i7:
; SI: v_min3_u32
; VI: v_min_u16
; VI: v_min_u16
; GFX9: v_min3_u16
define amdgpu_kernel void @v_test_umin3_ult_i7(i7 addrspace(1)* %out, i7 addrspace(1)* %aptr, i7 addrspace(1)* %bptr, i7 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i7, i7 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i7, i7 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i7, i7 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i7, i7 addrspace(1)* %out, i32 %tid
%a = load i7, i7 addrspace(1)* %gep0
%b = load i7, i7 addrspace(1)* %gep1
%c = load i7, i7 addrspace(1)* %gep2
%icmp0 = icmp ult i7 %a, %b
%i0 = select i1 %icmp0, i7 %a, i7 %b
%icmp1 = icmp ult i7 %i0, %c
%i1 = select i1 %icmp1, i7 %i0, i7 %c
store i7 %i1, i7 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_imin3_slt_i33:
; GCN-NOT: v_min3
define amdgpu_kernel void @v_test_imin3_slt_i33(i33 addrspace(1)* %out, i33 addrspace(1)* %aptr, i33 addrspace(1)* %bptr, i33 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i33, i33 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i33, i33 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i33, i33 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i33, i33 addrspace(1)* %out, i32 %tid
%a = load i33, i33 addrspace(1)* %gep0
%b = load i33, i33 addrspace(1)* %gep1
%c = load i33, i33 addrspace(1)* %gep2
%icmp0 = icmp slt i33 %a, %b
%i0 = select i1 %icmp0, i33 %a, i33 %b
%icmp1 = icmp slt i33 %i0, %c
%i1 = select i1 %icmp1, i33 %i0, i33 %c
store i33 %i1, i33 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_umin3_ult_i33:
; GCN-NOT: v_min3
define amdgpu_kernel void @v_test_umin3_ult_i33(i33 addrspace(1)* %out, i33 addrspace(1)* %aptr, i33 addrspace(1)* %bptr, i33 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i33, i33 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i33, i33 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i33, i33 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i33, i33 addrspace(1)* %out, i32 %tid
%a = load i33, i33 addrspace(1)* %gep0
%b = load i33, i33 addrspace(1)* %gep1
%c = load i33, i33 addrspace(1)* %gep2
%icmp0 = icmp ult i33 %a, %b
%i0 = select i1 %icmp0, i33 %a, i33 %b
%icmp1 = icmp ult i33 %i0, %c
%i1 = select i1 %icmp1, i33 %i0, i33 %c
store i33 %i1, i33 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_imin3_slt_i64:
; GCN-NOT: v_min3
define amdgpu_kernel void @v_test_imin3_slt_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr, i64 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i64, i64 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i64, i64 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
%a = load i64, i64 addrspace(1)* %gep0
%b = load i64, i64 addrspace(1)* %gep1
%c = load i64, i64 addrspace(1)* %gep2
%icmp0 = icmp slt i64 %a, %b
%i0 = select i1 %icmp0, i64 %a, i64 %b
%icmp1 = icmp slt i64 %i0, %c
%i1 = select i1 %icmp1, i64 %i0, i64 %c
store i64 %i1, i64 addrspace(1)* %outgep
ret void
}
; GCN-LABEL: {{^}}v_test_umin3_ult_i64:
; GCN-NOT: v_min3
define amdgpu_kernel void @v_test_umin3_ult_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr, i64 addrspace(1)* %cptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep0 = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
%gep1 = getelementptr i64, i64 addrspace(1)* %bptr, i32 %tid
%gep2 = getelementptr i64, i64 addrspace(1)* %cptr, i32 %tid
%outgep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
%a = load i64, i64 addrspace(1)* %gep0
%b = load i64, i64 addrspace(1)* %gep1
%c = load i64, i64 addrspace(1)* %gep2
%icmp0 = icmp ult i64 %a, %b
%i0 = select i1 %icmp0, i64 %a, i64 %b
%icmp1 = icmp ult i64 %i0, %c
%i1 = select i1 %icmp1, i64 %i0, i64 %c
store i64 %i1, i64 addrspace(1)* %outgep
ret void
}
declare i32 @llvm.amdgcn.workitem.id.x() #1
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone speculatable }